Download Computational Model for Processing Lexical Information

Computational Model for Processing Lexical Information
Zoltán Bánréti
Copyright © 2000 Zoltán Bánréti, Katalin Kiss, Gábor Rádai, Péter Rebrus, Péter Szigetvári,
Miklós Törkenczy, Beáta Gyuris, Csaba Oravecz, Gabriella Tóth, Viktor Trón
This research report was downloaded from the Research Support Scheme Electronic Library
at http://e-lib.rss.cz. The work on the report was made possible by a grant from, and was
published by, the Research Support Scheme of the Open Society Support Foundation, which
is a part of the Open Society Institute-Budapest. The digitisation of the report was supported
by the publisher.
Research Support Scheme
Bartolomějská 11
110 00 Praha 1
Czech Republic
www.rss.cz
The digitisation and conversion of the report to PDF was completed by Virtus.
Virtus
Libínská 1
150 00 Praha 5
Czech Republic
www.virtus.cz
_________________________
The information published in this work is the sole responsibility of the author and should not be construed as
representing the views of the Research Support Scheme/Open Society Support Foundation. The RSS/OSSF
takes no responsibility for the accuracy and correctness of this work. Any comments related to the contents of
this work should be directed to the author.
All rights reserved. No part of this work may be reproduced, in any form or by any means without permission in
writing from the author.
Contents
Abstract ........................................................................................................................................................... 1
Objectives ........................................................................................................................................................ 2
Findings........................................................................................................................................................... 3
List of publications........................................................................................................................................... 4
Detailed dummary of the results of the reserach ................................................................................................ 6
Neurolinguistics submodule.......................................................................................................................... 6
Open and closed class lexical items in the mental lexicon and their role in sentence processing.................. 6
Representational complexity of verb in mental lexicon and its effect on aphasics sentence production........ 8
Ellipsis in sentence processing ................................................................................................................ 10
Syntax-semantics submodule...................................................................................................................... 12
Ellipsis and the structure of the lexicon ................................................................................................... 12
The interpretation of certain logical vocabulary items in adult and child usage......................................... 13
Aspect and Argument Structure .............................................................................................................. 14
Phonology submodule ................................................................................................................................ 15
Degrees of phonotactic grammaticality: partitioning the lexicon .............................................................. 15
Phonotactics and morphological complexity............................................................................................ 17
Computational linguistics submodule.......................................................................................................... 19
Representation of linguistic knowledge ................................................................................................... 19
Extension of the GIN formalism ............................................................................................................. 20
Non-Complin References in Section 10................................................................................................... 22
Closed class lexical items in sentence processing............................................................................................ 24
Abstract ..................................................................................................................................................... 24
A time-based parser .................................................................................................................................... 27
Analysis of the repetition test...................................................................................................................... 30
References ................................................................................................................................................. 49
Egyeztetés agrammatikus afáziában................................................................................................................ 50
Absztrakt ................................................................................................................................................... 50
Irodalom .................................................................................................................................................... 61
Nyelvtan és Mentális Elemzõ Neurolingvisztikai Megközelítésben ................................................................. 63
MTA Nyelvtudományi Intézete .................................................................................................................. 63
Irodalom .................................................................................................................................................... 68
Semantic Bridging Effects in VP-Ellipsis........................................................................................................ 69
Introduction ............................................................................................................................................... 69
A preliminary overview of the data ............................................................................................................. 70
Semantic and pragmatic accounts of VP-ellipsis interpretation .................................................................... 73
A lexical semantic account of VP-ellipsis ................................................................................................... 74
Some relevant syntactic claims ............................................................................................................... 74
Ellipsis licensing with meaning postulates............................................................................................... 76
On the interaction of semantic parallelism and the organization of the grammar....................................... 78
Some further claims on the structure of the Lexicon .................................................................................... 79
Conclusion ................................................................................................................................................. 81
References ................................................................................................................................................. 81
The Interpretation of Universal Quantification in Child Language................................................................... 83
Aims and theoretical background................................................................................................................ 83
Preliminary overview of the data ................................................................................................................ 83
Philip’s (1995) experiments........................................................................................................................ 86
Experiments with Hungarian children ......................................................................................................... 88
Experiment 1.......................................................................................................................................... 88
Experiment 2.......................................................................................................................................... 90
Discussion of the Hungarian experiments and their implications.................................................................. 92
Conclusion ................................................................................................................................................. 93
References ................................................................................................................................................. 93
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian ......................................... 94
Introduction ............................................................................................................................................... 94
Comparing Hungarian ha ‘if’ and amikor ‘when’ clauses ............................................................................ 96
On the lexical meaning of akkor ‘when’ and ha ‘if’ ................................................................................. 96
Implicit quantifiers ................................................................................................................................. 97
Formalizing the semantic interpretation ...................................................................................................... 97
Basic ingredients .................................................................................................................................... 97
Formalizing the interpretation of amikor ‘when’ clauses.......................................................................... 99
Formalizing the intepretation of ha ‘if’ clauses ...................................................................................... 101
Conclusion ............................................................................................................................................... 102
References ............................................................................................................................................... 102
Coordinate Ellipsis as Phonological non-Insertion ........................................................................................ 104
Forward and backward ellipsis.................................................................................................................. 104
Ellipsis as deletion vs. reconstruction vs. anaphora.................................................................................... 106
BWE as phonological deletion — or morphological non-insertion ............................................................. 107
FWE as morphological non-insertion: The "meaning postulate" cases ....................................................... 112
Beyond meaning postulates? A further type .............................................................................................. 114
The true nature of the difference between BWE and FWE......................................................................... 116
Summary.................................................................................................................................................. 117
References ............................................................................................................................................... 117
Effect of verb complexity on agrammatic aphasic's sentence production........................................................ 119
Introduction ............................................................................................................................................. 119
Agrammatic sentence production .............................................................................................................. 120
Patterns of verb production ....................................................................................................................... 120
The present study ..................................................................................................................................... 122
Experiment I ............................................................................................................................................ 123
The structure of the verbs used in Experiment I..................................................................................... 123
Method .................................................................................................................................................... 125
Results ..................................................................................................................................................... 126
Effects of morphological complexity......................................................................................................... 128
Types of answers...................................................................................................................................... 129
Argument assignment and thematic hierarchy .......................................................................................... 131
(Agent (Experiencer(Goal/Source/Location(Theme)))).............................................................................. 131
Case assignment in isolated arguments...................................................................................................... 133
The calusal answers (Type B and C) ........................................................................................................ 133
Word order in the clausal answers............................................................................................................. 134
Summary.................................................................................................................................................. 135
Experiment II ........................................................................................................................................... 136
Test material ............................................................................................................................................ 138
Method .................................................................................................................................................... 139
Results of Experiment II. .......................................................................................................................... 140
Types of errors ......................................................................................................................................... 142
Summary.................................................................................................................................................. 142
References ............................................................................................................................................... 144
How to Cope with "Free Word Order": An Efficient Part-of-Speech Tagging Method for Hungarian ............. 146
Abstract ................................................................................................................................................... 146
Aspect and Argument Structure.................................................................................................................... 148
Introduction ............................................................................................................................................. 148
On Aspect ................................................................................................................................................ 148
Preliminaries ............................................................................................................................................ 149
The theory................................................................................................................................................ 153
Classification of verbs ............................................................................................................................. 153
Stative verbs ......................................................................................................................................... 153
Process verbs ........................................................................................................................................ 154
Accomplishment verbs.......................................................................................................................... 154
Achievement verbs ............................................................................................................................... 155
Classification of Verbs.......................................................................................................................... 156
External arguments and aspectual verb classes ...................................................................................... 157
Conclusion ............................................................................................................................................... 157
Footnotes ................................................................................................................................................. 157
References ............................................................................................................................................... 158
Deconstructing syllable structure.................................................................................................................. 159
Empty positions in the skeleton ................................................................................................................ 159
The skeleton-melody relationship.......................................................................................................... 159
Empty skeletal positions and the null hypothesis ................................................................................... 161
Syllable structure...................................................................................................................................... 163
Why have syllable structure? .................................................................................................................... 163
Problems with the standard view........................................................................................................... 164
Empty nuclei in the skeleton ..................................................................................................................... 166
Does the coda exist?................................................................................................................................. 168
Without codas .......................................................................................................................................... 170
Heavy versus light syllables .................................................................................................................. 170
Compensatory lengthening.................................................................................................................... 171
Against constituency ................................................................................................................................ 172
Conclusion ............................................................................................................................................... 174
References ............................................................................................................................................... 174
Phonotactic grammaticality and the lexicon .................................................................................................. 176
Introduction ............................................................................................................................................. 176
The SPE-a algorithm ................................................................................................................................ 178
The SPE-b algorithm ................................................................................................................................ 179
The Greenberg and Jenkins algorithm ....................................................................................................... 180
Summary.................................................................................................................................................. 184
Notes........................................................................................................................................................ 185
Bibliography ............................................................................................................................................ 186
Aspect and Argument Structure.................................................................................................................... 187
Introduction ............................................................................................................................................. 187
On Aspect ................................................................................................................................................ 187
Preliminaries ............................................................................................................................................ 188
The theory................................................................................................................................................ 191
Classification of verbs ............................................................................................................................. 192
Stative verbs ......................................................................................................................................... 192
Process verbs ........................................................................................................................................ 193
Accomplishment verbs.......................................................................................................................... 193
Achievement verbs ............................................................................................................................... 194
Classification of Verbs.......................................................................................................................... 195
External arguments and aspectual verb classes ...................................................................................... 196
Conclusion ............................................................................................................................................... 196
Footnotes ................................................................................................................................................. 196
References ............................................................................................................................................... 197
Ertékek azonossága-e az egyeztetés?............................................................................................................. 199
Bevezetés................................................................................................................................................. 199
Az unifikáció hiányosságai ....................................................................................................................... 199
Egyeztetés ............................................................................................................................................ 199
Szubkategorizáció ................................................................................................................................ 200
Általánosítás és típusrezolúció .............................................................................................................. 201
Jegy-érték párok mint tulajdonságok......................................................................................................... 203
Grammatikai viszonyok és mellérendelõ szerkezetek ................................................................................ 203
Az egyeztetés osztály-alapú elemzése ................................................................................................... 204
Határozottság koordinált NP-kben......................................................................................................... 205
Grammatikai viszonyok rekurzív definíciója ......................................................................................... 206
Összefoglalás és további lehetõségek ........................................................................................................ 207
A cikk fontosabb állításai...................................................................................................................... 207
További lehetõségek ............................................................................................................................. 207
Hivatkozások............................................................................................................................................ 208
A Magyar Igekoto Egyeztetese ..................................................................................................................... 210
Bevezetés................................................................................................................................................. 210
Alapfogalmak........................................................................................................................................... 211
Expletívum-e az igekötõ? ......................................................................................................................... 211
Milyen mondattani viszonyokról van szó? ................................................................................................ 212
A személy- és számegyeztetés hiánya ....................................................................................................... 213
Leírási kísérlet.......................................................................................................................................... 214
Hivatkozások............................................................................................................................................ 216
Representation of Linguistic Knowledge in GIN 99 ...................................................................................... 217
Introduction ............................................................................................................................................. 217
The GIN language.................................................................................................................................... 217
Motivation............................................................................................................................................ 217
Attribute/value structures...................................................................................................................... 217
AVSs and relations............................................................................................................................... 220
2.4 Representing relations in an AVS format......................................................................................... 221
Yet another fragment ............................................................................................................................ 223
Multi-AVSs and types .............................................................................................................................. 225
Multi-AVS type hierarchies .................................................................................................................. 225
The type resolution process................................................................................................................... 227
References ............................................................................................................................................... 229
Is Agreement Value Sharing? ....................................................................................................................... 230
Abstract ................................................................................................................................................... 230
Problems with unification ......................................................................................................................... 230
Agreement............................................................................................................................................ 230
Subcategorization ................................................................................................................................. 232
Generalization and type resolution ........................................................................................................ 232
Attribute/value pairs as properties............................................................................................................. 234
Grammatical relations and co-ordinate structures ...................................................................................... 235
Class-based analysis of agreement ........................................................................................................ 236
3.2 Definiteness in co-ordinate NPs....................................................................................................... 237
Recursive definition of grammatical relations........................................................................................ 237
Summary and further perspectives ............................................................................................................ 238
Main statements.................................................................................................................................... 238
Further perspectives.............................................................................................................................. 239
References ............................................................................................................................................... 241
Constructional CV phonology....................................................................................................................... 242
Abstract ................................................................................................................................................... 242
Construction phonology............................................................................................................................ 242
CV Phonology in a constraint-based setting........................................................................................... 242
Domain-final empty nuclei ................................................................................................................... 243
Intervocalic consonant clusters ................................................................................................................. 244
Domain-final consonant clusters ........................................................................................................... 246
Long vowels......................................................................................................................................... 250
Domain initial clusters.......................................................................................................................... 252
Exceptional domain final licensing........................................................................................................ 253
Constructions and the hierarchical lexicon............................................................................................. 254
Phonotactics and morphophonology.......................................................................................................... 256
Types of suffocation ............................................................................................................................. 256
General constraints on monomorphemic stems...................................................................................... 257
Epenthetic stems................................................................................................................................... 258
Lowering and exceptional licensing ...................................................................................................... 260
2.5 Synthetic suffixation ....................................................................................................................... 269
Verbal stems and synthetic suffixation .................................................................................................. 269
Pseudo-analytic suffixes........................................................................................................................ 271
References ............................................................................................................................................... 273
Kormányzás-fonológia kormányzás nélkül ................................................................................................... 275
A mássalhangzók engedélyezése............................................................................................................... 275
Valódi mássalhangzó-kapcsolatok......................................................................................................... 275
A kóda engedélyezése........................................................................................................................... 278
Az üres mag engedélyezése ...................................................................................................................... 283
Kivételes engedélyezés ......................................................................................................................... 283
Trocheikus engedélyezés ...................................................................................................................... 287
Engedélyezési tartományok................................................................................................................... 290
Köszönetnyilvánítás ................................................................................................................................. 292
Irodalom .................................................................................................................................................. 292
A helyelemek egyeztetése a CV-fonológiában .............................................................................................. 294
Bevezetés................................................................................................................................................. 294
Kormányzási tartományok ........................................................................................................................ 294
Valódi mássalhangzó-kapcsolatok......................................................................................................... 294
Szerkezeti párhuzamok ......................................................................................................................... 297
Asszimilációk........................................................................................................................................... 300
Hely-hasonulás mássalhangzók között .................................................................................................. 300
Magánhangzó-harmónia ....................................................................................................................... 301
Összegzés ................................................................................................................................................ 303
Függelék .................................................................................................................................................. 304
Általánosítások......................................................................................................................................... 308
Köszönetnyilvánítás ................................................................................................................................. 309
Irodalom .................................................................................................................................................. 309
1
Abstract
Abstract
We analyzed some syntactic, semantic and phonological phenomena that presuppose the
existence of interrelated components within the lexicon, which motivate the assumption that there
exist some sublexicons within the global lexicon of a speaker.
This result is confirmed by experimental findings in neurolinguistics. Hungarian speaking
agrammatic aphasics were tested in several ways. The results showed that the sublexicon of closed
class lexical items provides a highly automated complex device for processing surface sentence
structure.
Analysing Hungarian ellipsis data from a semantic-syntactic point of view, we established that
the lexicon is best conceived of as split into at least two main sublexicons: the store of semanticsyntactic feature bundles, and a separate store of sound forms, and proposed a format for representing
open-class lexical items whose meanings are connected via certain semantic relations. We proposed a
new classification of verbs to account for the contribution of the aspectual reading of the sentence
depending on the referential type of their arguments, and a new account of the syntactic and semantic
behaviour of aspectual prefixes.
The partitioned sets of lexical items are sublexicons on phonological grounds. These
sublexicons differ in terms of phonotactic grammaticality. The degrees of phonotactic grammaticality
are tied up with the problem of psychological reality: how many degrees of phonological
grammaticality are native speakers sensitive to.
We implemented a hierarchical construction network as an extension of the original General
Inheritance Network formalism. This framework was used as a platform for the implementation of the
grammar fragments.
Keywords: mental lexicon, sublexicon, neurolinguistics, syntax, semantics, morphology,
phonological grammaticality, construction network, inheritance, GIN.
Objectives
2
Objectives
The aim of the research was to give a description and analysis of the mental lexicon with
particular reference to sentence processing, in an interdisciplinary framework consisting of three
submodules. The main tasks were the following:
Neurolinguistics submodule:
Collection of spontaneous speech samples of agrammatic aphasics, construction of off-line test
materials, testing aphasics and control subjects. Evaluation of the test results. Construction of models
for lexical access and retrieval. Answering questions on the organisation of the mental lexicon, and as
to nature of the most economical type of lexical access in an agglutinative language like Hungarian.
Two articles.
Theoretical submodule:
Semantics:
The description of what semantic principles, if any, determine the storage of certain lexical
items, what semantically based strategies speakers use to substitute the inaccessible lexical items
under normal conditions and in aphasia. Collecting child language data on logical vocabulary.
Morphophonology:
Answering questions on how we process forms showing stem and suffix alternations, how the
phonological-morphological subsystems are organised, whether there are phenomena in language
processing which support the concept of analytic–synthetic suffixation, whether phonological,
morphological and lexical information has a role in selecting the grammatical form. Two articles.
Computational linguistics submodule:
Discussing the feasibility of the implementation of experimental results and theoretical analyses
in Generalized Inheritance Networks. Computational and formal aspects of content addressable
inheritance systems developed for the encoding of hierarchical construction networks with relations.
Two articles.
The objectives have been successfully achieved. Responding to the need for investigations in
the domain of syntax on the part of other submodules, the scope of research in the original Semantics
Submodule was extended in the course of the project to include topics on syntax as well, thus, the four
submodules working in the framework of the project were the following: neurolinguistics submodule,
syntax-semantics submodule, phonology submodule, and computational linguistics submodule. (This
change in the organisation of the project was indicated in the Interim Report.)
Findings
3
Findings
Importance:
Due to the structural properties of Hungarian (agglutinative morphology, relatively free word
order) certain basic assumptions of the traditional models of the mental lexicon may well be called
into question. We have elicited data from normal native and non-native speakers of Hungarian and
agrammatic aphasic patients as well.
Scientific significance, innovative character:
We analyzed some syntactic semantic and phonological phenomena that presuppose the
existence of interrelated components within the lexicon, which motivate the assumption on the
existence of sublexicons within the global lexicon of a speaker.
1. This finding was confirmed by experimental results in neurolinguistics. We demonstrated
that the sublexicon of closed class lexical items (store of grammatical formatives, inflectional endings,
suffixes in the mental lexicon) is critical for Hungarian speaking agrammatic aphasics. Speakers
access open class words (content words) and closed class items by two distinct access systems. The
interaction of the two access systems provides a highly automated complex device for processing
sentence structure. We characterised the representational complexity of the verbs in the mental
lexicon and its effect on aphasics’ sentence production. The internal temporal structure of verbs and
argument selection from the lexicon was also analysed.
2. We found that the two directional types of co-ordinate ellipsis, although displaying certain
different properties, can be treated by the same syntactic mechanism: the non-insertion (rather than
deletion or reconstruction) of phonological shapes to the terminal nodes in the structure. This
presupposes a model of syntax with split lexicon and late vocabulary insertion. We also isolated a
subclass of cases of forward VP-ellipsis which is licensed by the semantic relations of meaning
equivalence and entailment between propositions, the conditions of which are encoded in the
representation of the individual lexical items in the lexicon.
We propose a classification based on three aspectual classes of verbs, which gives the right
predictions for the obligatory presence of certain arguments and for the interaction of the verbs and
the different referential types of the arguments in the event structure.
3. The partitioned sets of lexical items are sublexicons on phonological grounds. These
sublexicons differ in terms of phonotactic grammaticality. The phonotactic grammaticality of a string
of segments is a measure that refers to the extent to which a given string is a potential/actual lexical
item. The question of how many degrees of phonotactic grammaticality are to be recognised
phonologically is tied up with the problem of psychological reality: how many degrees of
phonological grammaticality are native speakers sensitive to; and the possible partitioning of the
lexicon into sublexicons on phonological grounds.
4. The primary goal was to model a hierarchical construction network enriched with the
representation of relations that are the formal counterparts of correspondence relations in construction
grammar. The implementation is the extended update version of the original GIN (Generalized
Inheritance Network) framework.
4
List of publications
List of publications
Neurolinguistics submodule
Bánréti, Zoltán 2000a. ‘Closed class lexical items in sentence processing. A neurolinguistic approach’. ms p 38.
Submitted to Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian
Academy of Sciences, Budapest.
Bánréti, Zoltán 2000b. ‘Nyelvtan és mentális elemző neurolingvisztikai megközelítésben’ (Grammar and Parser
from the point of view of neurolinguistics), ms p11. Accepted for publication: 50 éves a Nyelvtudományi Intézet
(50th anniversary of the Research Institute for Linguistics). Ed: Mária Gósy, Budapest.
Bánréti, Zoltán 2000c. ‘Egyeztetés agrammatikus afáziában. A szintaktikai fa metszése’ (‘Agreement in
agrammatic Broca’s aphasia. The syntactic tree pruning.’) ms p15. Submitted to: Néprajz és nyelvtudomány
(Ethnography and Linguistics), Eds: M. Maleczki and L. Büky. Szeged.
Kiss, Katalin 2000. ‘Representational complexity of verbs and sentence production in agrammatic aphasics’, ms.
Submitted to Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian
Academy of Sciences, Budapest.
Syntax-semantics submodule
Bartos, Huba and Beáta Gyuris 2000. ‘Coordinate ellipsis as phonological non-insertion’, ms. Submitted to
Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of
Sciences, Budapest.
Gyuris, Beáta. 2000a. ‘Semantic Bridging Effects in VP-Ellipsis’. Submitted to Huba Bartos (ed.) Papers on the
mental lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest.
Gyuris, Beáta. 2000b. ‘The Interpretation of Universal Quantification in Child Language’. Submitted to Huba
Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences,
Budapest.
Gyuris, Beáta. 2000c. ‘On the semantic interpretation of WHEN and IF clauses in
Hungarian.’ To appear in Approaches to Hungarian VII. Ed: István Kenesei, JPTE: Pécs.
Tóth, Gabriella. 2000. ‘Aspect and Argument Structure’. Submitted to Huba Bartos (ed.) Papers on the mental
lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest.
Phonology submodule
Rebrus, Péter. 2000a. „Kormányzás fonológia kormányzás nélkül”. [„Government phonology without
government”]. Submitted to Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics,
Hungarian Academy of Sciences, Budapest.
Rebrus, Péter. 2000b. „A helyelemek egyeztetése a CV-fonológiában”. [„Agreement of place elements in CVphonology”]. Submitted to Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics,
Hungarian Academy of Sciences, Budapest.
Rebrus, Péter and Viktor Trón. 2000. „Constructional CV Phonology”. Submitted to Huba Bartos (ed.) Papers
on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest.
Szigetvári, Péter. 2000. „Deconstructing syllable structure”. Ms., Eötvös Loránd University. Submitted to
Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of
Sciences, Budapest.
Törkenczy, Miklós. 2000. „Phonotactic grammaticality and the lexicon”. Submitted to Huba Bartos (ed.)
Papers on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest.
Computational submodule
Rádai, Gábor. 2000. „Implementing Construction Grammars in GIN”. Submitted to Huba Bartos (ed.) Papers
on the mental lexicon, Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest.
5
List of publications
Csaba Oravecz, Péter Dienes, Zoltán Alexin and Tibor Gyimóthy. 2000. ``How to Cope with "Free Word
Order": An Efficient Part-of-Speech Tagging Method for Hungarian'', abstract accepted for poster session at the
Second International Conference on Language Resources and Evaluation, Athens, Greece, 31 May-2 June 2000.
Final paper due on Apr 2 and will appear in the proceedings.
László Kálmán and Viktor Trón. 2000. Is Agreement Value Sharing? Submitted to Natural Language and
Linguistic Theory. English version of the paper presented at the conference titled ``A Magyar Nyelv leírásának
újabb módszerei IV.'' [Modern methods in the description of Hungarian IV.], Version of January, 2000.
László Kálmán and Viktor Trón. 1999a. Értékek azonossága-e az egyeztetés? [Is Agreement Value Sharing?]
(co-authored with László Kálmán) To Appear. In: Proceedings of the Conference ``A Magyar Nyelv leírásának
újabb módszerei IV'' [Modern methods in the description of Hungarian IV.]
László Kálmán and Viktor Trón. 1999b. A magyar igevivő egyeztetése. [Agreement of the Hungarian Verb
Carrier.] (co-authored with László Kálmán) To Appear. In: Proceedings of the Conference ``A Magyar Nyelv
leírásának újabb módszerei IV.'' [Modern methods in the description of Hungarian IV.]
László Kálmán and Viktor Trón. 1999c. Linguistic Representations in GIN (Unpublished manuscript)
Detailed dummary of the results of the reserach
6
Detailed dummary of the results of the reserach
The main research findings are going to be presented individually for the four research groups,
including the Neurolinguistics submodule, the Syntax-semantics submodule, the Phonology
submodule, and the Computational linguistics submodule.
Neurolinguistics submodule
Participants: Zoltán Bánréti and Katalin Kiss.
Types of sublexicons in the mental lexicon
Open and closed class lexical items in the mental lexicon and their role in sentence
processing
Sentence repetition tests
In the course of sentence repetition tests our agrammatic aphasic patient gave answers that were
suggestive of initial structure building operations. With respect to stress patterns, each target sentence
was neutral in the test. Hungarian is an inflectional language where the verb assigns case to noun
phrases by means of case endings that mark theta roles in surface structure. Our patient was tested
with the help of the strategy of monitored repetition. The patient processed the sentence both
syntactically and semantically, then attempted to produce an utterance which matched the
phonological, syntactic, and semantic properties of the original utterance.
The performance of our patient’s parser can be characterised as follows. In comparison with the
target sentence,
1. it is possible for the parser to approximate the class of the target predicate, and its case
frame is retrievable;
2. if a different predicate is retrieved, then the suffixes are those appropriate to the case frame
of the "original" predicate;
3. if the predicate is missing, the parser stops; for instance, it cannot list only the NPs from the
target sentence;
4. it is possible to fill one slot from the argument frame of the predicate with selectional
restrictions that are the same as (or very much like) the original;
5. knowledge about missing, lexically or phonologically null arguments is manifest in further
search attempts that either mention case endings without a content word, or link them to
pronouns or neologisms, in repetition of case endings, or in compensatory speech.
Grammaticality judgement tests
We tested a total of five Hungarian Broca's aphasics. Our grammaticality judgement tests
covered some relevant features of Hungarian syntax and the lexicon. Three interesting cases are worth
attention in this respect: (i) there were some easy tasks, where the acceptability judgements of the
patients coincided with the expected answers in 100 per cent of the cases; (ii) in some cases we
witnessed guessing, since judgements turned out to be essentially random and chaotic from a
statistical point of view; (iii) in some other cases we faced systematic misjudgement of the data,
which means that acceptable sentences were judged as good in 100 percent of the cases but
unacceptable counterparts were also judged as good in 100 percent of the cases or at least close to 100
percent. The distribution of judgements supports a time-based approach to a parser.
The plausibility of an account based on asynchrony between syntactic and lexical processes is
motivated in the following way. The parser produces a structural frame for all possible sentences. This
syntactic frame contains categorised slots. When the configuration of surface case endings assigned
7
Detailed dummary of the results of the reserach
by category of Verb to its complements and the configuration of other closed class items are in their
active phase in the working memory, they define and open up syntactic slots for the content word
filler. Content words would be generated by the lexicon and would be inserted into their slots in the
syntactic frame.
Impairments of the accessibility of closed class morphemes create syntactic difficulties. Normal
activation happens at the expense of fast decay and, vice versa, normal decay happens at the expense
of slow activation. Applying this theory to our data we find the following. Specific features of
syntactic subcategories and closed class morphemes can be activated at a normal rate, but then they
decay very fast, too early from working memory; or they can be retained at a normal rate, at expense
of their slow activation in working memory. In the former case other specific lexical information has
also not been investigated previously, when it was needed. In the slow activation case other specific
lexical information in working memory is already gone when needed.
The fast decay or slow activation of grammatical features and subfeatures causes a
desynchronization in the building of syntactic structure. Syntactic slots are opened up too late or too
early for the content word filler; when specific lexical information in working memory had not been
activated yet or is already gone when needed. Therefore patients are not able to complete the analysis
of stimuli, processing operations result in a merely sketchy and unfinished structure. Patients were
aware of their unfinished analysis, they often made comments on it. This could lead to guessing
responses on complex, non-local relations.
Patients were able to use initial structure building operations involved in first pass parse for the
correct judgements of easy tasks. In the case of normals, first pass parse must be tightly synchronised
with a second major parsing module which extracts detailed and specific features of the category of
arguments and the predicate. But fast decay or slow activation of specific, unprotected information in
working memory can cause desynchronization between processing modules. The consequences are
systematic misjudgements or guessing responses, depending on the type of grammatical error and the
complexity of sentence to be judged.
Since closed class items have to be integrated with their categorised slots in the syntactic frame,
and open class (content) words have to be inserted into their categorised slots in the syntactic frame as
well, these two kinds of integration require synchronisation, the synchronised activation of structure
building elements in working memory for language. The slow activation or fast decay of closed class
items leads to a desynchronization between syntactic slots opened up by closed class items and the
active phase of content word fillers.
Theoretical results
Differences between memory time for open-class and closed class items are important for
accessing items in the mental lexicon. Closed-class items may fade away so fast from memory that the
construction of a proper NP or Sentence (for instance) is doubtful.
Temporal deficits do not affect the initial structure building operations. Our tests present
empirical evidence for the fact that syntactic and lexical processes are partially autonomous routines.
This becomes apparent in the case of a working memory deficit. The type of elements affected by the
temporal deficit do make a difference, however. When function word nodes are affected, the required
pattern does not emerge. It appears only when phrasal category nodes are impaired.
Although the patient’s restricted working memory time may not be sufficient to produce a full
sentence representation, it is nevertheless sufficient for the judgement of a verb and a string of
inflectional endings (related to that verb). This is compatible with the assumption that the patient has
to trade the processing of surface form against lexical access. (Inflection is part of the surface parser
module but we do not claim that this (sub)module would not be impaired.)
Therefore we apply the first-pass parse hypothesis. The hypothesis of initial structure building
operations has been proposed by a number of psycholinguists. In accordance with this hypothesis we
Detailed dummary of the results of the reserach
8
assume that in the sentence repetition test an initial structural analysis is computed and is
subsequently interpreted. This is followed by later processing operations involving constraints on the
indexing of structures involving content lexical items. The first-pass parser protects some of the
processed lexical and syntactic information during first-pass parse and a working memory deficit
restricts further processing operations.
The patients' performance in the repetition tasks showed that the verb is the starting point for
the surface syntactic analysis. The patients never made both inflectional errors and errors in the choice
of the main verb in the same sentence. Patients made inflectional errors when they were unable to
retrieve any verb. If the patients approximated the class of the target verb, however, then its surface
case frame was retrievable for them. The case frame assigns suffixes to associated nouns; and it does
so even if the nouns to which the endings are to be attached cannot be correctly reproduced by the
patients.
It can be assumed that in the first phase of processing the parser selects surface syntactic
information (subcategorizational frame of the verb, surface case frame, word order). Closed class
elements provide a syntactic frame into which open class items are inserted in the course of sentence
processing. In non-fluent aphasia the surface syntactic parser is too slow in processing closed class
lexical items, so lexical information in the working memory is already gone when needed. The
subjects are unable to integrate the output of the syntactic parser with the segments of the lexical
process.
Publications resulting from the research: Bánréti (2000a,b,c) in the List of publications
Representational complexity of verb in mental lexicon and its effect on aphasics
sentence production
"Picture description/action naming" tests
We have investigated how Hungarian Broca’s aphasic patients can lexically select and retrieve
verbs which differ in their representational complexity and how are they able to construe Verb
Phrases and simple sentences using the target predicate.
As elicitation task an off-line method, a „picture description/action naming” test was used; the
data were interpreted in the theoretical framework of Government and Binding Theory.
The structure of the verbs used in the tests
Based on their argument structure complexities, the tested verbs of the present study formed
three main groups. Thus, one-place intransitive predicates which take only one Agent or Experient
argument, two-place verbs, and three-place predicates were involved in the tests.
The verbal performance of two agrammatic Broca’s aphasic patients was analysed. Both
patients are native speakers of Hungarian. Our elicitation method was an ‘action naming’ /’picture
description’ test. The pictures represented the target verbs/actions. We regarded an answer to be
‘complete’ if the patients were able to build the whole Verb Phrase or sentence. It means that the verb
and its complements were lexically accessible, the argument NPs were supplied with the appropriate
overt case marker, noun-verb agreement was intact and nonterminal node deletion did not occur.
Analysis of experimental results
Comparing the distribution of the retrievable target verbs within each verb group we found the
following ‘verb difficulty order’:
simple 1-place > morphologically complex 1-place = transitive (2-place) > 3-place (with
locative and dative complement) > 2-place with locative complement
9
Detailed dummary of the results of the reserach
Access to the ‘simple one-place’ verbs was outstandingly successful. The lexical selection of
the ‘two-place verbs with locative complement’ proved to be the most difficult for the patients (they
could not retrieve any verb in this group). These predicates were directional motion verbs. The lexical
representations of these verbs integrate mental knowledge related to the cognitive representation of
space or spatial relations. These verbs include such contents as direction of the motion, place-coordinates, starting point and end point. This information is encoded in the semantic representation and
thematic roles of the predicate. Processing of this information seemed to be more difficult for our
patients, they produced marked selection disorder when attempting to produce these verbs. The ratio
of the three-place verbs was lower in the sample than the proportion of the one-place and two-place
verbs.
The empirical results show that the ‘representational complexity’ of the predicate has a direct
effect on the lexical accessibility of the verb for agrammatic aphasics. The complexity of the
argument-structure of the verb (number of obligatory arguments) plays an important role in verb
retrieval but it is not the only factor. The morphological and semantic representational complexity of
the ‘one-place derived verbs’ and the semantic representational complexity of the ‘two-place locative’
verbs also had an effect on the lexical-semantic selection of the predicates.
According to the ‘thematic hierarchy’ hypothesis, the argument structure of the verb is not only
a set of arguments. It has its own internal structure which represents prominent relations that are
determined by the thematic information of the predicate. Grimshaw suggested a protoargumentstructure which is a structured representation of arguments based on thematic hierarchy: (Agent
(Experiencer (Goal/Source/Location (Theme)))) The subjects were able to produce arguments of
every type (Agent, Theme, Goal, Benefactive) but a difference was found in the distribution of the
type of arguments activated first.
Activation of the arguments lower in the thematic hierarchy was more frequent than that of the
more prominent arguments of a given predicate (e.g. Theme > Benefactive > Agent ; Theme > Agent;
Goal > Agent). Two exceptions were found to this tendency, namely the Agent >Goal order in the
‘three-place locative’ group and the Agent > Theme order in the ‘transitive [+animate]’ group.
Comparing the proportions of arguments, an outstanding contrast was found between the activation of
Agent and Theme arguments in the ‘transitive [-animate]’ and ‘3-place dative’ verb groups. In the
case of the ‘3-place locative’ group the Agent><Theme contrast was not so sharp, rather, the
Goal/Source><Theme and the Agent ><Goal/Source contrasts were considerable. The contrast was
also less sharp between the Agent and Goal arguments in the ‘two-place locative’ type of verbs. The
data show that the less prominent Theme argument was activated faster than the other arguments if
the predicate assigned the thematic role of Theme mapped to an object NP specified as [-animate].
Activation of the Theme argument fell behind the Agent only if the verb was reversible (if the Theme
thematic role was mapped into an object specified as [+animate, +human]). The Theme [-animate]
argument seems to be a preferred one for agrammatic aphasics.
Theoretical results
The representational complexity of the verbs had a direct effect on the accessibility of the
predicates. ‘Morphologically simple one-place predicates’ were produced in the highest number.
Much lower proportions of the ‘morphologically complex one-place' predicates and in the transitive
verbs were found, and only some verbs were activated in the 3-place verb groups. Production of the
directional motion verbs proved to be the most difficult for the patients. This data showed that the
argument structure complexity of the verb is important but not the only factor in the lexical selection
of predicates. The semantic representational and morphological complexity of the predicate is also
relevant in the lexical-semantic selection of the verbs.
Dysfunction of the syntactic structure building mechanisms had a connection with the lexical
accessibility of the formatives and the nominal elements of the phrase structure. The reduced capacity
to preserve the previously activated argument Ns or NPs had a role in the unsuccessful structure
building operations.
Detailed dummary of the results of the reserach
10
Agrammatic performance can be interpreted by those asynchronic mechanisms that cannot
function simultaneously between the level of semantic operations (activation of argument-structure
and thematic information) and syntactic processing (procedures that construct the syntactic phrase
structure and map the arguments/thematic roles into the syntactic frame).
Publication resulting from the research: Kiss (2000), in the List of Publications
Ellipsis in sentence processing
Agrammatic aphasics overuse the linguistic option of ellipsis in free conversation. Ellipsis is
understood as the omission of the projection of V after a focused DP or a quantified DP in a surface
syntactic string . The use of elliptic sentences in spontaneous speech can be considered as a kind of
adaptation or reaction. Broca’s aphasic patients are aware of their reduced linguistic capacity, that is
why they employ elliptical constructions. The role of this strategy is to prevent computation overload
in the linguistic system. Employment of this strategy (among others) is optional rather than
obligatory.
Hungarian grammar allows forward and backward types of Verb Phrase ellipsis. Forward VP
Ellipsis (FVPE) is interpretively dependent on its antecedent. The syntactic tree is complete.
Syntactic and semantic features of lexical items are present in the ellipsis site, it is the phonological
form that is not inserted. there is no need to assume the deletion of lexical items. Backward VP
Ellipsis (BVPE) sites result from a deletion of lexical forms. There are identification asymmetries
between forward and backward ellipses. Forward ellipsis sites contain lexical content throughout the
derivation, but fail to undergo phonological form-insertion. Backward ellipsis sites result from
deletion after form-insertion.
Sentence repetition tests
We tested the neurolinguistic reality of the identification asymmetries with respect to the
direction of VPE mentioned above. Our subjects were agrammatic Broca’s aphasics. The test material
involved co-ordinated sentences with VP ellipsis sites. Each test contained 15 sentences containing
forward VP ellipsis and 15 sentences containing backward VP ellipsis. Two subjects were given the
test three different times. Sentence patterns were filled with different (though equally frequent) words
in each test but we did not change the sentence structures themselves. To repeat sentences patients
were pursuing the strategy of monitored repetition involving two basic operations: (1) processing the
heard utterances both syntactically and semantically, then storing them; (2) attempting to produce an
utterance which matches the phonological, syntactic, and semantic properties of the original utterance.
Analysis of data
Identification asymmetries between FVPE and BVPE are relevant for real sentence processing
operations as well. Repetition of BVPE imposed syntax/phonology interface requirements that
exceeded the impaired capacity of the language processor with agrammatic aphasics.
Producing co-ordinated sentences with FVPE in a repetition test requires the patients to store a
content-based representation of the heard co-ordinated sentence, then convert it into a surface
syntactic and phonological form. Supposing the processor builds structures from left to right, there
was no built-in delay in the processes because of direction of ellipsis. Patients often mentioned elided
VPs in overt phonological form at its correct position in the second conjunct. It was easy for them to
reconstruct FVPE in overt phonological form.
To produce a co-ordinated sentence with a BVPE in repetition test, it is necessary to recover the
deleted lexical forms in the first conjunct. If structures are built from left to right, there is a built-in
delay in the operations because of the direction and identification level of the backward VPE.
Recovering is delayed, because the deletion is located in the first conjunct, but the phonologically
realised licensing string is found in the second conjunct. Patients were able to repeat only the second
conjunct in its correct grammatical form. The first elliptic conjuncts were often fragmented and
11
Detailed dummary of the results of the reserach
ungrammatical. The elided VP in the first conjunct was rarely mentioned in its overt phonological
form. Patients didn’t reconstruct the BVPE overtly in the first conjunct. This is because the demands
of the task increased to the extent that exceeded the capacity of the impaired processor.
Theoretical results
Properties of sentence repetition in aphasia result from the operations of normal processing
principles under exceptional circumstances, in the face of impaired computational resources.
The data characterised above are relevant from the point of view of a time based parser. The
differences are related to timing. Generally speaking, when a non-empty Verb appears, most of the
pieces to properly assemble the clause are available. In case of backward ellipsis the question is how
to build a sketchy structure for the FIRST clause without lexical material of a non-empty V-bar.
In the case of forward ellipsis, the parser processes the lexical material of the antecedent V-bar
in the first place, then tries to analyse an empty category later. The parser is able to determine an
empty (elided) category with the help of surface structural parallelism.
Forward ellipsis is easier for impaired speech, because the same category is used twice in two
parallel structures but an overt lexical form is mentioned only once.
Backward ellipsis is harder for a time based parser. Backward ellipsis means that an empty
category is detected at first. At the very moment when a parser detects an empty V-bar there is no
information about its subtype and no information about the lexical material of the V-bar. The
necessary decision is postponed. The parser must put that empty category into the memory buffer and
wait for a posterior lexical item, namely lexical material of the posterior V-bar in the second clause.
After processing the posterior V-bar the parser tries to determine the identity of the phonological
form of the posterior V-bar and the elided V-bar and copy back the semantic/syntactic features.
Backward ellipsis must cause a delay in structure building operations with the first co-ordinated
clause.
A hypothesis on the structure of the mental parser
Suppose the following structure-building operations. The mental parser must produce a
structural frame for all possible sentences. This syntactic frame contains categorised slots. When the
configuration of surface case endings assigned by the category of Verb to its complements and the
configuration of other closed class items are in their active phase in working memory, they define and
open up syntactic slots for the content word filler. Content-words would be generated by the lexicon
and would be inserted into their slots in the syntactic frame. Because closed class items have to be
integrated with their categorised slots in the syntactic frame, and open class (content) words have to
be inserted into their categorised slots in the syntactic frame as well, these two kinds of integration
require synchronisation, a synchronised activation of structure building elements in working memory
for language. The slow activation or fast decay of closed class items leads to a desynchronization
between syntactic slots opened up by closed class items and the active phase of content word fillers.
We define the mental parser as an automaton which becomes specialised in the processing of
categories and features involved in the grammatical representation of sentences. Under this view the
parser is a device which transfers information between a grammatical representation and a message
level representation. The parser computes the grammatical representations of sentences and
transforms them into a message level representation (at which the „what is to be said” is represented).
The category and feature system is hierarchical in the grammatical representation. It has various levels
of sub- and sub-sub categories, from the bare category to the individual lexical item and from the
closed class category to the fully specific features of that closed class item. Then, it is the question of
capacity and synchronisation how far down the hierarchy in grammatical representation the parser
goes on its search for information. The distribution of patients’ performance in tests reflects the
limitations on the interface between the impaired parser and the grammatical representation
containing a hierarchy of categories and their features.
Detailed dummary of the results of the reserach
12
For publications resulting from the research see the paper Bánréti (2000a,b) in the List of
publications
Syntax-semantics submodule
Participants: Beáta Gyuris, Gabriella Tóth
Ellipsis and the structure of the lexicon
The aim of the syntax-semantic research group was to contribute to the general research aim of
the project, the investigation of the internal structuring of the mental lexicon, from a syntactic and
semantic perspective. More particularly, this meant isolating some of the sub-lexicons which are
distinguished from the others in terms of syntactic and semantic behaviour.
Structures containing ellipsis have always been in the forefront of the interests of syntacticians
and psycholinguists, since the production and the understanding of structures which contain a
„missing element” can shed a lot of light on the principles of mental computation, and the knowledge
of grammar. Since the neurolinguistics subteam of our project group decided to investigate the mental
lexicon of aphasic speakers by testing their comprehension and production of sentences containing
ellipsis, in order to foster internal communication and co-operation within the project, instead of
dealing on the three individual research topics as specified in the proposal, we concentrated, from the
beginning, on the semantic analysis of Hungarian sentences containing ellipsis as a means of fulfilling
our original research objectives. (The change of perspective was already indicated in the Interim
Report.)
In the domain of syntax, we examined a wide range of ellipsis types, in order to explore the
mechanisms underlying ellipsis phenomena. In particular, we wished to test a grammar model based
on the theory of Distributed Morphology (Halle & Marantz 1993), in which ellipsis is implemented as
the non-insertion of phonological shapes to certain (strings of) lexical elements. This approach,
stemming from work by Wilder (1997) and Bartos (to appear), abandons the traditional analyses of
ellipsis, which were set in terms of either phonological deletion, or the content
reconstruction/interpretation of syntactically null elements, and places it in the context of a model, in
which what used to be seen as a unitary lexicon is split up into two (or even three) separate lexical
modules: one containing the lexical items as semantic-syntactic feature bundles, and another one
which contains sound forms, which are to be associated with the formal featural lexical items after the
syntactic derivation. (A third possible lexical module would contain idiomatic meaning units, to be
associated with the lexical items postsyntactically.)
In the first phase, we carefully studied the relevant literature within the generative grammatical
tradition, to have a firm grasp on existing analyses of ellipsis (study trip by B. Gyuris to Edinburgh).
On the basis of a summary of this, we examined data from Hungarian, to establish what languagespecific properties are involved, so as to be able to abstract away from such particularities. These data
also served as the primary area of our own analyses. The results and findings of this stage were
presented at a group meeting, for the whole project group.
We analysed VP-ellipsis data from Hungarian and English (taking this type of ellipsis to be
representative), where we found that functional items do not behave uniformly under ellipsis: formal
agreement items show much wider variability between ellipsis targets and licensers than tense/mood
markers, which essentially pattern with open-class items, i.e. content words.
In the second phase, we focused our attention on the comparison of backward and forward
ellipsis (i.e. whether the elided material is in the first or the last conjunct in a co-ordinate structure).
These two subtypes have often been treated by different analyses, due to the fact that they display
characteristically different properties: while in backward ellipsis the elided part must be fully formidentical to the parallel part of the final conjunct, for forward ellipsis the constraints are less strict —
e.g. full form-identity is not (always) required. Instead of following the mainstream, which attributes
this difference to different executions of elision (deletion under identity vs. reconstruction under
13
Detailed dummary of the results of the reserach
semantic/pragmatic equivalence or entailment, or anaphor interpretation) (a review of these is found
in Gyuris 2000a), we attempted to assimilate the two types to the greatest possible extent. We have
shown (Bartos–Gyuris 2000) that some seemingly complicated non-identity cases of forward ellipsis
can be handled without recourse to less well-defined notions like reconstruction/interpretation, with
the simple licensing mechanisms for phonological non-insertion used for strict form-identity ellipsis,
by relying on meaning postulates, i.e. equivalence/entailment relations between lexical items (such as
inversion of argument linking, or detransitivization), which can be encoded in the lexicon in a simple
and exact fashion.
The scientific significance of our findings consists in two main points: (a) the success of these
analyses gives strong support to the split lexicon hypothesis (see above), by virtue of the fact that our
efficient theory of ellipsis is set in the context of such a model of grammar; and (b) our approach to
the problem of ellipsis opens up a perspective to incorporate further particular problematic ellipsis
types into the non-insertion analysis, thereby facilitating the understanding of the phenomena in order
to arrive at a more unified treatment for ellipsis types. Furthermore, the application of meaning
postulates (cf. Gyuris 2000a, on the semantic aspects of ellipsis) presupposes that there exist
sublexical networks linking lexical items via such equivalence/entailment relations — another piece
of contribution to the understanding of the structure of the so-called “mental lexicon”.
The papers resulting from the research are Bartos and Gyuris (2000), and Gyuris (2000a) in
the List of Publications
The interpretation of certain logical vocabulary items in adult and child usage
It was discovered by Philip (1995) that the interpretation of certain determiner quantifiers
differs in the adult and child language of English and Japanese speakers, and suggested that the reason
for this is that as opposed to adults, children seem to quantify over minimal events, just like it is done
in the case of adverbial quantifiers by all speakers.
We carried out two experiments with children aged between 4 and 7 year in the framework of
the project to find out whether Philip’s (1995) predictions would also hold for Hungarian speakers.
Although we found recurring patterns among the data produced by Philip’s subjects and ours, we did
not find evidence for the same developmental stages in employing event quantification as Philip
claimed to characterise his subjects. His arguments for the more nature of event quantification also
did not seem to be convincing enough, due to the obvious conceptual differences between the tangible
objects adults are supposed to quantify over and the more abstract events children are supposed to
operate with.
Consequently, we proposed an alternative account of the facts, according to which in child
language all determiner quantifiers are represented in a similar way, namely, as insensitive to the
order of their arguments, while, at a later stage, the conceptual distinction between strong and weak
quantifiers develops. We believe that the account proposed by us gives a much more credible
explanation for the data than assuming the conceptually complex event quantification to operate in
child usage.
The paper resulting from the research is Gyuris (2000b).
We investigated the meaning of a different pair of logical vocabulary items, namely the
connectives amikor ‘when’ and ha ‘if’ as well, in order to find out how the similarities and differences
in the interpretation of sentences containing them can be accounted for in terms of their lexical
representations.
It is important to note that the use of vocabulary items corresponding to amikor ‘when’ and ha
‘if’ shows some crosslinguistic differences, for example, the so-called „atemporal” use of whenclauses is impossible in Hungarian. Our investigations therefore aimed at discovering the universal
and the individual aspects of the meaning of these items.
Detailed dummary of the results of the reserach
14
After having studied the interpretation of sentences with subordinate clauses introduced by the
above connectives with or without adverbs of quantification in their main clauses we established that
amikor ‘when’ and ha ‘if’ clauses are always interpreted as expressing quantification over sets of
eventualities and n-tuples of possible worlds and eventualities or individuals, respectively. This
difference accounts well for the factivity of amikor ‘when’ clauses and the lack of it in ha ‘if’ clauses.
The lack of atemporal readings for amikor ‘when’ clauses was explained by the fact that the
particular temporal relations which determine the temporal ordering between the runtimes of
eventualities satisfying the descriptions in the main and subordinate clauses are not defined for a
certain subclass of states, which property does not extend to cross-lingusitic equivalents of these
words.
The paper resulting from the research is Gyuris (2000c).
Aspect and Argument Structure
The goal of the project was twofold. On the one hand, we wished to examine the different types
of events with respect to the internal temporal structure of the predicate, the referential types of their
arguments and the thematic roles that relate the arguments to the event. On the other hand, we aimed
to study the argument structure on the basis of the internal temporal structure of the verb
It has been proposed by several linguists that telicity and atelicity are not the property of the
sentence but the property of the verbs. Vendler and Dowty classify verbs into four major aspectual
classes (statives, processes, accomplishments and achievements) while Verkuyl and Krifka
distinguish stative events from dynamic events. Events are composed on the basis of the temporal
property of he verb and the referential type of the arguments.
Verkuyl (1989, 1993) observed that the Dowty-Vendler classification is a highly problematic
because a well defined group of verbs, namely process verbs have to be listed in the lexicon twice as
process verbs and as accomplishment verbs depending on the referential type of their arguments (cf.
John ate an apple vs. John ate apples). Verkuyl-Krifka approach can account for the above problems
by simply claiming that there is one verb listed in the lexicon and telic events are composed of a
dynamic verb and quantified nominal arguments while atelic events are composed by either stative
verbs and their arguments or dynamic verbs and their bare existential arguments. There is, however,
also a serious problem with the compositional approach developed by Verkuyl, since in some cases
bare nominals can occur in telic events if they are combined with achievement verbs (e.g. John found
money in the garden).
To avoid the above mentioned problems, we propose a new classification of aspectual verb
classes on the basis of the internal temporal structure of verbs. We claim that verbal predicates are
classified into three basic aspectual classes in the lexicon. Stative verbs do not consist of atomic
events, process verb do, but the number of atomic events in their case is unlimited, while achievement
verbs consist of exactly two atomic events. Accomplishments are composed of process verbs and
quantified nominal arguments in the spirit of Verkuyl (1993).
We propose that only dynamic verbs can be delimited by introducing a direct internal argument
of a certain type, that by undergoing some change of state delimits the event. Statives cannot have this
type of argument at all. Statives can never be delimited, process verbs can, but, as the number of
atomic events process verbs can contain is unlimited, they need a well-defined direct internal
argument, a quantified DP that can delimit the event temporally. Achievement verbs differ from
process verbs with respect to their internal temporal structure as they consist of exactly two atomic
events. They can temporally delimit the event, but to encode the change of state achievement verbs
need a direct internal argument. As the verbs can temporally delimit the event, the internal argument
need not be quantified, it can be a bare existential.
We have also established process verbs may, but achievement verbs must have an internal
argument. If process verbs occur in a telic event, the presence of the direct internal argument is
Detailed dummary of the results of the reserach
15
obligatory. Neither achievement verbs nor process verbs form one homogeneous class. Only those
direct internal arguments can delimit an event whose subparts have a unique relation with the atomic
events (to eat, to drink, to destroy, to create, to paint), those direct internal arguments, whose subparts
do not have this type of relation (to hammer, to run) cannot delimit the event. These verbs can have
another indirect internal argument (a small clause or a directional prepositional phrase).
External arguments can occur only with dynamic verbs. We assume Kratzer (1995)'s analysis
on the status of the external argument. They claim that the external argument is not the argument of
the verb. It is argument of a predicate they call CAUSE that can be combined with the verb in the
syntactic representation. We will assume that this combination is possible if the verb is dynamic.
The major significance of our findings lies in the proposal for a new classification of verbs on
the basis of their internal temporal structure. This new classification eliminates the redundancy of the
Vendler-Dowty classification, and gives the right results with the achievement verbs (problems that
arise in the Verkuyl-Krifka theory) that heavily relies on the referential type of the internal argument
in the calculation of (a)telic events. The theory further predicts the presence of external, direct internal
and indirect internal arguments with minimal information about the verbal predicate.
The paper resulting from the research is Tóth (2000) in the List of Publications.
Phonology submodule
Participants: Péter Rebrus, Péter Szigetvári, and Miklós Törkenczy
The research conducted clusters under three topics. Firstly, we have examined theories about
degrees of phonotactic grammaticality, including an excursus on a related issue, paradigmatic
defectivity. Secondly, we have improved a theoretical framework of phonological representation that
is adequately restricted, but, nevertheless, is more or less capable of capturing phonotactic patterns in
an abstract model. Thirdly, by proposing a formalization of the theoretical model the framework can
be embedded in a concept of the hierarchical lexicon.
Degrees of phonotactic grammaticality: partitioning the lexicon
The question of how many degrees of phonotactic grammaticality are to be recognized
phonologically is tied up with the problem of psychological reality (How many degrees of
phonological grammaticality are native speakers sensitive to?) and the possible partitioning of the
lexicon into sublexicons on phonological grounds (Are there partitioned sets of lexical items that
differ in terms of phonotactic grammaticality?). These are the main research questions of this topic.
While phonotactic graduality is conceivable in a homogeneous lexicon, a theory of sublexicons
constituting the full lexicon of a language changes the picture.
The properties of the existing two-level and multi-level models of phonotactic grammaticality
were explored. These models were all found unsatisfactory in various ways. An attempt was made to
delimit the theoretical dimensions of a phonotactic grammar, i.e. to theoretically identify the factors
and their interaction involved in phonotactic grammaticality. The factors considered were: the direct
vs. indirect nature of the algorithm, the role of the number of violations in the evaluated string, the
role of (type and/or token) frequency, and the "structuredness" of the lexicon. The results were
reported in Törkenczy (2000).
Examining Japanese, Ito & Mester (1995) conclude that the lexicon is organized by partly
overlapping constraint domains, lexical strata are defined by sets of items that are subject to different
phonotactic constraints and phonological rules. The existence of such sublexicons within the global
lexicon of a speaker may be manifested in differences in the speed of activation, therefore the analysis
of the speech of aphasiacs provides empirical evidence for detecting the existence and extent of the
different sections of the lexicon.
It follows from assuming such an organization of the lexicon that phonotactic constraints
cannot result in simple yes/no choices in grammaticality, i.e., the assumption that a given string is
Detailed dummary of the results of the reserach
16
either fully well-formed or fully ill-formed has to be abandoned. The phonotactic grammaticality of a
string of segments is a measure that refers to the extent to which a given string is a potential/actual
lexical item. In fact, unless a sublexicon is always a subset or a superset of another sublexicon (which
is not what Ito & Mester assume), phonotactic graduality cannot be implemented. Within such a
model the phonotactic grammaticality of a given item can only be given relative to some sublexicon,
i.e., a form is not grammatical or ungrammatical and not more or less grammatical, but grammatical
or not according to the constraints that prevail in a given sublexicon.
Phonotactic constraints may even restrict the operation of apparently independent
morphological processes. In the first half of the research period it became increasingly clear that
verbs are a phonotactically identifiable subgroup of the Hungarian lexicon. There are several
indicators of this:
a)
verb stems can only end in a set of consonants clusters which is a subset of the clusters
permitted at the end of stems in general;
b)
(synthetic) suffixation may only create verb-final clusters that are a subset of the clusters
that can result from (synthetic) suffixation at the end of words in general (compare
suffixation of nouns by accusative -t with the suffixation of verb by past tense -t);
c)
verb-stems may end in clusters that are not permitted word-finally. These stems are not
only bound, but lack all forms that contain analytic suffixes. Thus, these verbs are
defective in that they have a defective paradigm. Furthermore, the defectivity of this
group of stems is phonologically motivated, i.e. not arbitrary.
Defectivity is thus an important characteristic of a subset of items in the Hungarian lexicon.
The behavior of these stems was analysed within the framework of a nonderivational phonological
model (Government Phonology) and it was proposed that the paradigmatic gaps are due to the fact
that defective verbs do not have a lexical stem, while their synthetically suffixed forms are stored in
the lexicon. These results were presented in Rebrus & Törkenczy (1998, 1999).
In the second half of the research period the phenomenon of defectivity was explored.
Phonologically motivated defectivity is not a well-researched topic and at first appears to be a rare
phenomenon. Nevertheless, a thorough search in the international phonological literature revealed
that there are several examples of phonologically motivated defectivity in a variety of languages (e.g.
Tagalog: um- infixation/prefixation; Catalan: hypochoristics by truncation; Turkish: minimal size
constraint on suffixed forms; Swedish: paradigms of /dd/-final adjectives and verbs; English: -ize
suffixation, verb-forming -en suffixation; Russian: 2nd conjugation verbs, etc.), and more
importantly, that there are different types of phonologically motivated defectivity. The Hungarian
case is especially interesting because the motivating factor is a general phonotactic constraint which
true of the possible phonological forms in the language across the board. Compare, for example the
productive English verb-forming suffix -en, which can only attach to obstruent-final stems (dampen,
redden, loosen, etc) while sonorant-final adjectival stems will not have corresponding -en verbs
(*coolen, *greyen, *thinnen, etc.). This is a case of phonologically motivated defectivity which is
different from the Hungarian case because the excluded forms in English are not phonotactically illformed (compare woollen, swollen etc). Our research suggests that defectivity in general, and the
case of Hungarian defective verbs in particular, is problematic for Optimality Theory, even when the
theory is "fixed up" with devices specifically designed to handle absolute ill-formedness. The results
of this research were presented in Törkenczy 1999.
VC phonology
Dienes & Szigetvári (1999), Szigetvári (1999) and Dienes (in prep.) are devising a theory for
explaining the plausible sites and direction of consonantal lenition and its absence and phonotactic
constraints.
17
Detailed dummary of the results of the reserach
Lowenstamm (1996) claims that the phonological skeleton consists of strictly alternating C and
V positions, i.e., any consonant or vowel cluster encountered in the interpretation of a phonological
string is separated by a vocalic or consonantal position, respectively, in its representation which fails
to be interpreted phonetically. The theory presupposes a fairly abstract definition of skeletal position,
one which allows it to remain unmanifested in pronunciation, there is, nevertheless, reason to accept
this stance, some arguments are offered in Szigetvári 2000.
Ségéral & Scheer's (1999) offer a theory of consonant lenition involving CV skeletons. Their
aim is to apply the partly redefined notions of government and licensing to explain why word-final
and intervocalic consonants are more prone to lenition than those in word-initial and postconsonantal
position. The theory suffers from some weak points though, the set of governing and licensing
relationships proposed to hold between units of the skeleton appears to be incomplete, not providing
for a number of well-attested cases (like long vowels or the type of falling-sonority cluster that occurs
word-finally in languages like English or Hungarian, as opposed to other clusters that do not).
These are among the problems that are addressed by VC Phonology. The novelties of the
theory are: (i) an explicit formulation of what consonantalness and vocalicness means in CV
frameworks, (ii) an explicit definition of the two prosodic forces, licensing and government and (iii)
the claim that the skeleton of each domain in phonological representations begins with a V position
and ends with a C position. It is claimed that the inherent property of C positions is that they are
mute, thus remain unpronounced unless external influence forces them to be phonetically interpreted.
As opposed to this, V positions are inherently loud, that is, it takes an effort to silence V positions on
the skeleton which fail to be pronounced. The notion of government is redefined: instead of inhibiting
the expression of melodic content (an effect identical to the absence of licensing} as in Ségéral &
Scheer's theory, it is claimed to destroy the inherent properties of its target. Thus a governed
consonant becomes louder, a governed vowel becomes muter. Lowenstamm's CV framework has
inherited the concept of word-final empty nucleus, an empty vocalic position posited at the end of
consonant-final words, from Government Phonology. This position, however, remains silent only
because of a stipulative clause, which, furthermore, is language specific. VC Phonology denies the
existence of this position, instead, it posits an empty consonantal position at the end of vowel-final
words. The superiority of this solution is that the noninterpretation of a consonantal position needs no
special care.
The relevance of the theory for the current project is the fact that it provides a relatively
coherent theory of phonological representation, which aims at explaining phonotactic constraints, a
feature of the lexicon that appears to have an important role in defining the sublexicons. The
parameters that influence the grammaticality or lack thereof of phonological strings are often seen to
hold in certain subcomponents of the lexicon and reset in others.
Phonotactics and morphological complexity
On the basis of phonotactics languages can be divided into different types. In our view this
typology of languages is based on the inventory of specific phonological constructions used by the
given language. For instance the prosodic typology roughly presents the following types of languages:
(i) CV-languages (ones which only have words containing CV sequences), (ii) the so-called Princelanguages (in which a limited set of homorganic consonant clusters are allowed), (iii) languages in
which almost every cluster is allowed intervocalically, (iv) languages with long monophthongs or
diphthongs, (v) languages which allow domain-final consonants, (vi) languages which allow a limited
set of consonant clusters domain-initially (or postconsonantally), (vii) languages which have vowels
in the beginning of the word (or in hiatus). A language can be a member of one or more of the types
mentioned above. For a language, however, these properties cannot be selected randomly: certain
properties must involve others (these relations are called universal implications). For instance (iii)
naturally involves (ii), and (iv) involves (ii), also (i.e. if there are long vowels, there must be closed
syllable). Between (ii) and (v), however, there is no such implication; the same as between (vii) and
(ii)/(iii). Such implications limit the number of the possible language types, and a
18
Detailed dummary of the results of the reserach
morphophonological theory has to include a proper formalism which expresses these implicational
relations.
Markedness relations can be easily expressed by phonological constructions. The availability of
a certain construction makes a language more marked than one lacking it. This way we can establish
markedness relations between languages in several dimensions. These markedness relations, however,
usually appear within a language, too. The results of certain morphological processes can be different
from monomorphemic forms (stems) in their phonotactics. In the simplest case there is a twofold
distinction: some morphologically complex (e.g. suffixed) forms have the same phonotactics as
monomorphemic forms. Other morphological complex forms, however, have looser phonotactics than
monomorphemic forms, hence these forms could not be stems. This latter type of suffixation is
exemplified by analytic suffixation. This does not mean that the phonotactics of analytic forms is
unlimited: there are constraints which cannot be violated in analytic forms, either. These constraints
are either static (i.e. restricting the range of possible morphemes) or dynamic (i.e. the so-called
postlexical processes which are automatically triggered).
In languages with more complex morphological structure, the simple twofold distinction
between analytically and non-analytically suffixed forms is to be refined. In Hungarian, for instance,
different non-analytic forms suggest a different phonotactics. This can be explained if one assumes
different types of specific classes of suffixes may select a different range of phonological
constructions as available for the entire suffixed form. Moreover, there are stems which are irregular
in terms of their non-productive alternations in their non-analytic suffixed forms. Similarly to the
suffixes, specific "irregular" classes of morphemes can "refer" to a more or less arbitrary range of
constructions which determine their suffixed forms. With these assumptions we obtain a highly
structured lexicon, where the phonological representation of different suffixed forms are parsed by
constructions, the inventory of which may be limited by the individual suffix- as well as stemmorpheme classes.
Our approach to the formalisation of phonological representations relies on the results of CV
phonology (Szigetvári 2000). Rebrus (in press) gives a detailed analysis of a great deal of Hungarian
morphophonological phenomena in Hungarian within a CV phonology framework. Rebrus & Trón
(2000) is an attempt to reformulate some results of CV phonology in the context of construction
grammar. Phonological constructions referred to above are thought to reflect various licensing
configurations in CV phonology. Motivated by our findings in Hungarian morphophonology, it seems
worthwhile to integrate a framework of sufficiently abstract phonological representation with the
conception of a hierarchical lexicon.
This conception of an intricate lexical hierarchy fits well the overall architecture of grammar
advocated by lexicalist grammars, such as Construction Grammar (cf. Rádai 2000). Also, it lends
itself to a plausible implementation in the GIN (Generalized Inheritance Network) framework of
linguistic representations (see the section on the computational submodule).
On the other hand, the grammatical architecture reflects the structure of the mental lexicon: the
existence of various subsystems of available phonological constructions parallels the gradual nature of
phonotactic grammaticality judgments (Törkenczy 2000). The different phonotactics of
morphologically complex forms can give essential clues needed in morphological parsing processes.
Apart from theoretical considerations the importance of such a representation of phonological forms
lies exactly in this fact.
Detailed dummary of the results of the reserach
19
Computational linguistics submodule
Participants: Csaba Oravecz, Gábor Rádai, Viktor Trón
Representation of linguistic knowledge
The architecture of the lexicon
A thorough investigation into mainstream ideas on lexical matters as well as a critical
examination of various treatments of all kinds of grammatical phenomena suggesting lexical
restrictions have led us to serious conclusions concerning a plausible organisation of the lexicon and
its relation to the language faculty as a whole.
We take a radically lexicalist stance. We claim that if the lexicon is needed to store the
unpredictable sound-meaning correspondences, anyway, then it is all we need. Instead of the rules of
grammar, or derivations, what is crucial here is solely the organisation of information. This
parallels with the standard requirement for scientific theories, that is the assumption of a minimal
ontology.
Construction Grammars in GIN
It has been around for a while that syntactic constructions previously thought of as being
generated by phrase-structure rules or the like are nothing more (or less) than correspondences
between form and meaning ((Goldberg 1995), (Jackendoff 1997), (Fillmore and Kay 1993)). With the
obvious difference that here items are more complex (more structured) in the general case than that of,
say, monomorphemic stems. Given that the representational language of the interface modules should
be able to express phonological and semantic information, (constituency relations: the analysis of the
item into parts) as well as the devices for underspecification, the representation of more complex
items is nearly trivial.
Now the concept that has traditionally been associated with the term `lexical item' suddenly
dissolves and what we are left with instead is a set of sound-meaning correspondence-rules. The
generality of a correspondence statement is no more a categorical attribute of a construction, as in the
case of a syntactic rule, rather, it is best thought of as attaining a measure of degree. On one end of the
scale there are the most general constructions that involve as their „constituents” very generally
characterised (lexical) classes, possibly including open classes. On the other end of the scale we have
fully lexicalised patterns that — even if they fit in the lexicon as a special case of some general
construction, and therefore have similarly structured relatives — have idiosyncratic, (unparalleled,
exceptional, unpredictable) features.
What we are provided with now is one quite intricately woven hierarchical network of
grammatical constructions, one which establishes the inextricable link between the interface levels of
language, on the one hand, while still retaining their relative autonomy of the representational levels,
on the other. This autonomy manifests itself in the fact that the interface levels are representational
modules complying with the principle of representational modularity.
Sublexicons
We have found that some phonological phenomena presuppose the existence of various
interrelated compartments within the lexicon, which motivate the assumption of sublexicons in the
mental lexicon. This finding is confirmed by experimental results in neurolinguistics.
This phenomenon defies an adequate description, let alone an explanation, given a not very
carefully worked-out notion of the lexicon as a non-redundant list of idiosyncratic features of atomic
items.
Instead, this problem lends itself to a solution within a hierarchically structured lexicon, where
morphosyntactic, semantic and phonological information is equally present and related by
20
Detailed dummary of the results of the reserach
correspondence constructions. These correspondence constructions, being underspecified to an
arbitrary extent, represent lexical classes. Their partial subsumption order in turn can express the
relevant generalisations across these lexical classes or can allow special subcases within each. With
this emerging view of lexical architecture, we hope to have provided the formal setting for the
intuitive idea of sublexicons.
Our primary goal then was to model such a hierarchical construction network but to allow for
the representation of relations that are the formal counterparts of correspondence constructions.
Therefore we chose to implement it as an extension of the original GIN formalism.
Extension of the GIN formalism
AVSs and relations
Inheritance networks have become the major tool of knowledge representation in AI over the
past decades. This naturally lead to the assumption - realised in the framework of Head-Driven Phrase
Structure Grammar (HPSG) - that linguistic knowledge should also be stored in this form. This
formalism had an important new feature over former frameworks such as DATR: whereas in the latter
one could query the network in the old fashioned AI form - namely by naming the object and the
property one wished to have information on -, the former framework made an important concession to
the different way in which linguistic knowledge is put to use. HPSG is based on the assumption that
our linguistic knowledge has the form of constraints - namely which form goes together with which
meaning. Thus all linguistic tasks - such as generation and parsing - consist of finding and
instantiating the constraints a given structure is subject to. This is to say that we query the network via
certain keys given in the - syntactic or semantic - description of the structure in question, i.e., carry
out a content oriented search of a database.
In this framework the above concept of a relation can not be properly encoded. The standard
„tricks” usually employed to get around this problem (such as the introduction of set values or
indeterminate feature names) all lead to extensions of the formalism in which the basic operation on
AVSs, unification is not well-defined and/or the algorithm deciding the equivalence of two AVSs has
a rather high complexity (Kálmán and Trón 1999a).
The formalism worked out under the RSS grant had the goal to overcome some of the inherent
limitations of the above system. Most importantly the drawback that HPSG networks are a
continuation of the AI tradition inasmuch as they only allow information about (classes of) objects to
be encoded. We have come to the conclusion that in linguistic applications it is often necessary to
represent relational information and extend inheritance to relational nodes (see Kálmán and Trón
1999a, 1999b). This led to the birth of the formalism of Generalised Inheritance Networks (GIN). The
first year of the project was taken up by designing and implementing the above system.
As we pointed out above, we considered it to be the major limitation of systems using Typed
Feature Structures (TFS) - such as HPSG - that they do not allow for the direct representation of
relational information that plays an important role in representing linguistic knowledge. Thus we had
to come up with a system that fits this purpose. Since TFS's have proved an efficient representational
tool for inheritance relations in the form of type declarations, we chose to use this formalism as a
point of departure and to enrich it for our purposes. The basic idea was to interpret all nodes of the
network as a description of n-ary relations — thus the old class-descriptions simply become border
cases in terms of being unary relations, i.e., properties. This gives us the opportunity to describe
relations and inheritance relations on them directly. The language of TFS's had to be enriched to allow
for n-ary features and n-ary type declarations. The expressive power of the RAVS (Relational
Attribute Value Structure) language is between the AVS language and positive first order logic,
enables the expression of relational information. The algorithm deciding the identity of two AVSs
having <n nodes attains a complexity of O(n2). The operation of unification is trivial and well defined.
21
Detailed dummary of the results of the reserach
Type resolution
Partially specified RAVSs are structured according to subsumption in a type hierarchy.
Standardly presuming the closed world assumption whereby general types stand for the disjunction of
their most specific instances. Naturally the values of attributes as well as relations can be
(under)specified and can refer to a type.
The network can then be thought of as a generative grammar, in which the well-formed
structures are those in which all referenced types are resolved to at least one most specific instance.
The advantages of such a highly homogeneous and declarative representation of linguistic
knowledge is exactly that however rigid it may be with respect to formal representations, it is highly
flexible with respect to a software implementation.
We believe that the details of the actual type-resolution algorithm could be adjusted in a way
so as to enable experimental research on modelling at least some aspects of human language
processing and parsing. We can index the network and use pointers to drive the search whenever we
know that some part of it will certainly play no role in answering a query or whenever we know at
which point it is most economical to start looking for an answer. Yet this indexing has more to do
with database technologies than with the specific nature of linguistic knowledge.
An investigation can be launched to reveal the relevant aspects of a type resolution algorithm
that could play a role in modelling psycholinguistic effects in the course of processing. This includes
the theoretical investigations on how incremental parsing can be simulated in the algorithm (see
Future prospects).
Lazy evaluation
We found that in such a homogeneous architecture one cannot very easily find the way to block
the calculation of irrelevant but still associated information, for a given input. A remedy to this has
been evolved which blocks the resolution of types that do not contain information potentially
contradicting the hypothesis processed. This notion of potential contradiction can be given a formal
definition and implemented in the processing algorithm, it can influence type resolution procedure,
rendering it lazy in the general case.
The implementation of the whole framework involved a considerable amount of programming
work to implement the processes for the handling of such databases. Once the system has been
implemented we could turn to problems relating to it. On the more practical side we had results
relating to the representation of certain fragments of grammar (Rebrus and Trón 2000, Rádai 2000),
the automatic acquisition of such networks (Oravecz 2000). On the theoretical side we have
investigated what consequences the above approach has on our view of the modularity of mental
grammars - since we consider the above system to be a psychologically plausible implementation of
the linguistic knowledge of speakers of Natural Languages.
Implementing Grammars
Once we had a working version of GIN available we started to investigate concrete
implementational issues. Papers discuss in what ways results of existing frameworks can be
reformulated and integrated in this conception of grammar. This work resulted in descriptions of
various phenomena of English and Hungarian. This comprises the following topics:
1. As we pointed out earlier our work has been mainly motivated by the conception of grammar
embodied in Construction Grammars. Specific analyses of linguistic phenomena presented in this
framework, however intuitive and appealing, failed to attain the degree of exactness required for a
direct encoding. Rádai (2000) provides a correct reformulation of the framework in GIN which
establishes the way to the computational implementation of CGs.
22
Detailed dummary of the results of the reserach
2. Kálmán and Trón (1999a) show that agreement phenomena can not be adequately captured
in existing constraint-based formalisms. They claim that a remedy for this can be provided if one
adopts the relational hierarchy (GIN) as a grammatical framework. They argue that an analysis
advocating the direct encoding of grammatical congruence relations such as agreement can capture a
great deal of phenomena more adequately than previous ones.
3. Kálmán and Trón (1999b) show that the strange behaviour of some verbal prefixes in
Hungarian verbs which defy an elegant solution in traditional frameworks, argues for relational
hierarchy. It is argued that a lexicalist theory implemented in relational inheritance networks is much
superior to previous ones in being able to cope with the range of related phenomena.
4. Rebrus and Trón (2000) advocates a novel approach that integrates current trends in
declarative phonology with the lexicalist grammatical framework of Construction Grammars. It is
showed how the elegant treatments of CV phonology can be encoded in inheritance construction
networks. The restricted and universal phonological constructions and the abstract representation of
the former together with a conception of hierarchical lexicon provided by the latter offers a promising
perspective in the explanation of a wide range of morphologically (and lexically) motivated
phonological phenomena.
Automatic Acquisition and Lexical Databases
An important challenge for a plausible GIN implementation that encodes large fragments of
natural language phenomena is to apply some machine learning technique on large amounts of
annotated corpora to automatically extract information and build working GIN fragments out of them.
As a by-product of this procedure a statistical extension to GIN could be worked out for the modelling
of lexical processing with GIN type-resolution. However, to make the automatic induction of such
information and the construction of large-scale computational lexicons feasible one needs significant
amount of language corpora annotated at least at the morphosyntactic level. This need has led to the
fact that considerable work has been devoted within the context of the COMPLIN project to the
development of such an annotated database. This work has revealed several technical as well as
theoretical problems specific to Hungarian, which problems do not normally arise if the same task is
taken for western languages, for most of which the construction of such a database has long been
considered a solved issue. Details of the problems encountered and solutions proposed are discussed
in Váradi and Oravecz (1999), and Oravecz et. al. (2000).
Non-Complin References in Section 10.
Bartos, Huba. To appear. „VP-ellipsis and Verbal Inflection in Hungarian”. Acta Linguistica Hungarica.
Dienes, Péter. in prep. VC Phonology. MA thesis, Eötvös Loránd University.
Dienes, Péter and Péter Szigetvári. 1999. Repartitioning the skeleton: VC Phonology. Ms., Eötvös Loránd
University.
Dowty, David. 1979. Word meaning and Montague Grammar. The Semantics of Verbs and Times in
Generative Semantics and in Montague's PTQ. Reidel: Dordrecht.
Ch.J. Fillmore and P. Kay. 1993.Construction Grammar Coursebook.
Reading materials for Linguistics X20, University of California, Berkeley CA.
A.E. Goldberg. 1995. Constructions: A Construction Grammar Approach to Argument Structure. The
University of Chicago Press, Chicago.
Halle, M. and A. Marantz 1993. Distributed Morphology and the Pieces of Inflection. In: Hale, K. and S. J.
Keyser (eds.) The View from Building 20: Essays in Honor of Sylvain Bromberger. The MIT Press, Cambridge,
MA: 111-176
Ray Jackendoff. 1997. The architecture of the language faculty. Linguistic Inquiry monographs. MIT Press,
1997.
23
Detailed dummary of the results of the reserach
Itô, Junko and Armin R. Mester. 1995. „Japanese phonology”. In John A. Goldsmith (ed.). The handbook of
phonological theory. Cambridge, Mass. & Oxford: Blackwell. 817-838.
Kratzer, Angelika, 1994. The event argument and the semantics of Voice. ms. University of Massachusetts at
Amhurst.
Krifka, Manfred. 1989. Nominal Reference, Temporal Constitution and Quantification in Event Semantics. In
R Bartsch, J. van Bethem, and P. van Emde Boas (eds.) Semantics and Contextual Expression. Foris Dordrecht,
75-115.
Lowenstamm, Jean. 1996. „CV as the only syllable type”. Ms., Université Paris 7. Appeared in Jacques Durand
and Bernard Laks (eds.). Current Trends in Phonology: Models and Methods. European Studies Research
Institute, University of Salford Publications. 419-442. Academy of Sciences.
Csaba Oravecz and Tamás Váradi. 1999. „Morphosyntactic ambiguity and tagset design for Hungarian”. in
Proceedings of the Workshop on Linguistically Interpreted Corpora, EACL'99, pp. 8-13. Association for
Computational Linguistics
Philip, William (1995), Event Quantification in the Acquisition of Universal Quantification. Doctoral dissertation,
University of Massachusetts at Amherst.
Rebrus, Péter. in press. „Morfofonológiai jelenségek". In Kiefer Ferenc (ed.) Strukturális magyar nyelvtan 3.
Morfológia. Budapest: Akadémiai Kiadó.
Ségéral, Philippe and Tobias Scheer. 1999. The Coda mirror. Ms., Université de Paris 7 and Université de Nice.
Szigetvári, Péter. 1999. VC Phonology: a theory of consonant lenition and phonotactics. PhD dissertation,
Eötvös Loránd University
Vendler, Zeno. 1967. Verbs and Times. Linguistics in Philosophy, 97-121. Ithaca: Cornell University Press.
Verkuyl, Henk. 1989. Aspectual Classes and Aspectual Composition. Linguistics and Philosophy 12, 39-95.
Verkuyl, Henk (1993) A Theory of Aspectuality. Cambridge: Cambridge University Press.
Wilder, Chris (1997) Some Properties of Ellipsis in Coordination, in: A. Alexiadou-T. Hall (eds): Studies on
Universal Grammar and Typological variation, J. Benjamins, 59-107.
Closed class lexical items in sentence processing
24
Closed class lexical items in sentence processing
A neurolinguistic approach
Zoltán Bánréti
Abstract
Non-fluent agrammatic aphasics were tested. The impaired mental parser protects some of the
processed syntactic information during first-pass parse and a working memory deficit restricts further
processing operations.
Closed class elements provide a syntactic frame. Open class items are inserted into the syntactic
frame in the course of sentence processing. In non-fluent aphasia the syntactic parser is too slow in
processing closed class lexical items, so lexical information in the working memory is already gone
when needed. According to the results of sentence repetition tests and grammaticality judgement tests
the subjects are unable to integrate the output of the syntactic parser with the fully specified segments
of the lexical process. The distribution of patients’ performance in tests reflects the limitations on the
interface between the impaired parser and the complexity of the grammatical representation
containing a hierarchy of categories and features.
1.1. According to Chomsky (1997), the faculty of language is embedded within the broader
architecture of the mind/brain. It interacts with other systems, which impose conditions that language
must satisfy if it is to be usable at all. We might think of these as “legibility conditions“. The systems
within which the language faculty is embedded must be able to “read“ the expressions of the language
and use them as “instructions“ for thought and action. The sensorimotor systems have to be able to
read the instructions having to do with sound. The articulatory and perceptual apparatus have specific
design that enables them to interpret certain properties, not others. These systems thus impose
legibility conditions on the generative processes of the faculty of language, which must provide
expressions with the proper “phonetic representation.“
The same is true of conceptual and other systems that make use of the resources of the faculty
of language. They have their intrinsic properties, which require that the expressions generated by the
language have certain kinds of semantic representations, not others. (Chomsky, 1997. 12-14).
1.2. Brain/mind functions to externalize language will be called human language processor.
This paper presents some empirical evidence on relations between grammar and the human language
processor. The evidence is based on results of sentence repetition tests performed by Broca’s
aphasics.
2. Natural languages tend to contain two quite different sorts of morphemes, those that are
primarily of the world (open class items: nouns, adjectives, adverbs with their own lexical-semantic
content) and those that are primarily of the grammar (closed class items). The closed class is generally
taken to include case endings, prepositions, determiners, pronouns, conjunctions, auxiliaries,
inflectional affixes and a variety of other expressions. (cf. Biassou- Tyler- Nespoulous- DordainHarris 1997). Linguistic symptoms of Broca's aphasia are sometimes defined as the impairment of
access to closed class morphemes. (Linebarger 1995). The fragmentation or agrammaticality of
spontaneous speech, poor sentence repeating skills and good sentence comprehension skills may be
correlated with this fact. Closed class morphemes are the elements of a structure-analysing and
structure-building complex in on-line speech comprehension and production (Bock 1989). Closed
class morphemes can be used as indicators for the speaker since these formatives mark the beginning
and the end of noun phrases and other phrases, the units of constituent structure, boundaries of main
and subordinate clauses, word order, etc. They impose structure on strings of words. These
morphemes are members of computational vocabulary.
Accessing closed-class morphemes influences access to open-class words (words that refer to
entities in the world) as well. Formatives can radically reduce search time in open class vocabulary, if
25
Closed class lexical items in sentence processing
formal information is available as to whether one has to search for a noun, an adverb or an adjective,
for example. Speakers access open class words and closed class morphemes by two distinct access
systems. The two access systems have to interact, especially during on-line sentence comprehension.
(Martin-Saffran 1997). From the point of view of the mental lexicon, there is a level at which theta
assigning predicates, like verbs, are members of the computational vocabulary (Frazier-McNamara,
1995). Verbs and their subcategorizational frames, which include surface case endings constitute
complex lexical entries. Surface case endings are parts of subcategorizational frames of verbs and
mark theta role assigned by the verb on the complements.
3. The interaction of open and closed class lexical items in sentence processing is important for
Hungarian speaking agrammatic Broca’s aphasics. In case of Hungarian the inflectional endings,
especially surface case ending frames subcategorized for by verbs (predicates) provide an automatic
complex device for processing surface sentence structure.
There are a number of competing theories of agrammatic aphasia, which appear to cluster
around two fundamental hypotheses. One popular assumption is that although in essence the grammar
in aphasia is intact, one module of it or other is impaired, some specific rules of the grammar are
disappeared. The alternative group can be characterised as the capacity limitation approach. The
fundamental claim here is that patient's grammar is wholly preserved, and the deficiency lies in some
aspect of the performance system.
We refer to the first type of theories as grammar based theory, and to the second type of
theories as capacity limitation theory.
3.1. According to the grammar-based theories, although in essence the grammar in aphasia is
intact, one aspect of it or other is deficient, which is the source of breakdown of patients' quality of
performance. There is no absolute loss of competence, what we face is a disappearance or impairment
of some specific rules or principles of the grammar.
In the framework of the Minimalist Program (Chomksy 1997), the verb (and the nominal
constituents) is already inflected for agreement and tense in the stage of the sentence derivation,
where the lexical insertion into the phrase markers takes place. (The agreement and tense features of
the verb have to be checked in the different functional nodes. It is the checking of the feature that
triggers the verb movement into the functional nodes.) The features are realised as inflectional
suffixes/affixes in the Phonetic Form. Thus any substitution of the number/person marker of the
subject and the tense inflectional marker of the verb in the Phonetic Form can be interpreted as the
changing of the number/person features of the subject, and the tense feature of the verb in the Surface
Structure.
3.1.2. An interesting example for grammar based theory is the "Pruning Tree“ Hypothesis.
Friedman and Grodzinsky (1995) found tense and agreement errors in the sentence production of the
agrammatic aphasic patients in Hebrew. They employ the Split inflection theory elaborated by
Pollock (1989) and the Checking Theory of the Minimalist Program to explain these errors.
Friedmann and Grodzinsky suppose that the functional categories C, T and AgrP are underspecified in
agrammatism and that an underspecified node cannot be projected. Moreover, functional nodes might
be selectively impaired in agrammatism.
According to their " Pruning Tree“ Hypothesis, the impairment of a lower functional node
implicates the impairment of the higher functional nodes in the syntactic tree of a sentence.
(Impairment of the functional node Agreement implicates the impairment of Tense phrase and CompP
nodes in Hebrew). The feature checking is not carried out in an impaired functional node in the
higher functional nodes.
3.1.3. Hungarian is “problematic“ language for this hypothesis. For instance, the Pruning-tree
Hypothesis (PTH) makes different predictions in connection with two types of the ungrammatical
answers produced by a Hungarian agrammatic patient in sentence repetition tests (Mészáros, 1999).
The PTH makes a good prediction of the responses containing only agreement (1). It assumes that
Closed class lexical items in sentence processing
26
only AgrSP functional node is impaired but TenseP is intact. Fig 1. shows the postulated syntactic
structure of the answer (1).
(1) E: A fiú
a könyveket
nézte,
a lányok
pedig
a hirdetéseket
the boy-NOMsg. the books-pl-ACC lookPASTsg.3. the girls-pl.-NOM however advertisements-pl.ACC
It is the books that the boy was looking at, and it is the advertisement that the girls were.
P:* A fiúk
a könyveket
nézte,
a lányok
pedig
a hirdetéseket.
the boys-PL.-NOM the books-pl-ACC look-PAST -SG.3, the girls-pl-NOM however the
advertisements-pl.-ACC
*It is the books that the boys looks at, and it is the advertisements that the girls
*[a fiúk k [a könyveket j
[ a hirdetéseket [........ ]
Top
FocP
F
[nézte i [ t j
AgrS
ti [
ti
[ tj
ti
[ ti tj tk ]]]]] [a lányok ][ pedig]
TenseP AgrOP VP
TopP
Foc
F’
Postulated syntactic tree of the answer by the patient
Fig.1 shows: the checking of the number (and person) feature of the subject failed to come
because of impaired AgrSP. At the same time the tense feature of the verb is checked.
A functional node above an impaired functional node is not accessible. In (2) agreement
functional node is intact resulting in correctly inflected verb for person and number. The agreement
inflection is preserved but the tense inflection is not. The answers contain only tense errors.
(2) E: Tegnap Ákos
fogják.
szidta meg
Marit,
ma
pedig
a gyerekek
yesterday Ákos-NOM sg. scolded-PASTsg3. Mary-ACC., today however the childrenNOMpl3.
will-pl3.
'It is John who scolded Mary yesterday, and it is the children who will do so today'.
P:*Tegnap Ákosék
fogják
szidni Marit, holnap
pedig
a gyerekek fogják.
*Yesterday Ákos-pl-NOM will-pl.3. scold Mary tomorrow, however, the children will do so
In accordance with PTH we can assume that the incorrect tense affixation results from the
impaired tense functional node. (The tense marker is not omitted, only substituted for another one.)
The response (2) is inconsistent with the "Pruning tree hypothesis". The impaired Tense node is lower
than the intact Agreement functional node in the syntactic tree. The verb cannot skip over an impaired
functional node; that is, the impairment of the TenseP node implicates the impairment of the
agreement node. As it can be seen, the Pruning-tree hypothesis makes a wrong prediction to these
answers. We presume that this theory alone cannot provide a satisfactory explanation for the answers
27
Closed class lexical items in sentence processing
containing tense agreement error but not agreement error. Moreover it cannot account for the fact that
the occurrence of the agreement or tense error is more frequent when the subject or the adverb of time
is topicalised but not focused.
According to Mészáros (1999) the errors mentioned above can be interpreted as a consequence
of the two different operations, first one is used in the reconstruction of the predicate part of the
sentence containing focused constituents of the sentence and the second one is used in attachment of
topic constituent to the predicate part.
The patient is able to reactivate the VP in the antecedent-containing clause. It means that he is
able to interpret the semantic and syntactic features in the place of the gap. Mészáros (1999) claims
that the reconstruction of the gap has precedence in the course of the construction of the answer. The
low performance in the recalling of the sentences supposes that the reconstruction of the VP ellipsis
overloads the sentence processor resulting in agrammatical production.
A time-based parser
3.2. We will argue for a different theory that can be characterised as the capacity limitation
theory, or the time based approach. The fundamental claim here is that patient's grammar is wholly
preserved, and the deficiency lies in some aspect of the mental system.
3.2.1. We take notion of parser as a kind of automatic memory function that becomes
specialized in processing of categories and features, involved in grammatical representation of
sentences. The parser is a device that transfers information between grammatical representation and
message level representation. At the message level the "what-is-to-be-said" is represented. The parser
as a whole is adapted to its limited capacity. According to Kolk (1995) paper, complex message
structures in working memory are "fine-tuned" to creative linguistic competence. This fine-tuning is
related to the size of temporal window produced by parser for syntactic computation and syntactolexical integration during a given time period). The parser based on activation and synchronization
functions. Grammatical representation of a sentence is derived from the model of grammar as the
output of grammar. This representation contains a hierarchy of relevant grammatical categories and
features of sentence. Parser computes the hierarchy of categories and features involved in
grammatical representations and transforms them into message level representation. This requires an
interaction between parser and grammatical representation of sentence.
3.2.2. Impairments of the syntactic parser appear to include the slowing down of critical
functions. According to Haarman and Kolk (1994), Broca's aphasia affects sentence processing by
either slowing down the rate at which new elements are constructed or increasing the rate at which
they decay. But not both at the same time. Kolk (1995) argues for computational simultaneity or
synchrony. His computational model, SYNCHRON, simulates the temporal course of building up a
sentence structure representation. Simultaneity or synchrony is associated with bottom-up features.
Two critical parameters are involved. In the "slow activation" case, it takes longer for the parser to
begin processing of an item. The critical activation level is reached too late, thus the item does not
become available for further processing tasks. On the other hand, "Fast decay makes elements
unavailable when they fall below their critical level too soon to be combined with other elements..."
(284).
Cornell (1995) introduced a new computational model, GENCHRON, based on Haarman and
Kolk's model. GENCHRON produces semantic representations in accordance with the double
dependence hypothesis (Mauner et al. 1993). The grammar used by GENCHRON is a constraint
based phrase structure grammar in which rules combine both syntactic and semantic constraints.
Cornell's computational model is bottom-up, parallel, and it has the property of simultaneity. The
Extended Simultaneity Condition is the following: "Construct a superordinate constituent node, and
solve its associated constraints, only if there is a point in time at which all of its subordinate
constituent nodes are simultaneously available in memory" (306).
Closed class lexical items in sentence processing
28
In addition to a component of grammar, GENCHRON system has parameter files to control the
rate at which nodes become available in memory and with which they decay away.
According to Cornell (1995) retrieval time models represent the following deficit: lengthening
the time period that it takes to process a new element "increases the likelihood that earlier arriving
constituents will have faded from working memory by the time the later arriving constituents are
finally constructed" (316).
In processing simulation, however, Cornell used a memory time model. This refers to the
period during which an element is available in working memory. "Shortening this time period
increases the likelihood that earlier arriving constituents will have faded from working memory before
later arriving constituents are made available." (Cornell 1995:316)
In processing simulation memory-time parameters were varied according to the openclass/closed class distinction. Cornell made the following parameter settings:
Open-class items persist for: 6 clock cycles;
Closed-class items persist for: 3 clock cycles;
Retrieval time for all items: 1 clock cycle.
(Cornell 1995:317.)
3.2.3. Differences between memory time for open-class and closed class items are important.
According to the parameter settings above, closed-class items fade away so fast from memory that the
construction of a proper NP (for instance) is doubtful.
Cornell supposes that a processing account of asyntactic comprehension should make
predictions for correct/incorrect grammaticality judgements as well. He suggests as a next step that
"The version of GENCHRON used in these simulations is subject to the extended simultaneity
condition: it waits until all subtrees have been parsed and then attempts to solve all of the constraint at
once.
Generalized Simultaneity Condition:
The output of a particular task only becomes available when and if the output of all of its
subtasks is available at some point in time. At that point in time the superordinate task begins to make
its output available" (323).
3.2.4. To access a closed class item is to retrieve the structure building operations that are
associated with that closed class item. The case ending frame assigned by the category of the verb and
other inflectional endings open up a syntactic slot for integration with a content word filler.
Impairments on access system of closed class items cause a delay in opening up syntactic slots. In the
sense of Kolk (1995), this means that the point in time at which closed class morphemes deliver a
syntactic slot for an open class lexical filler is in synchrony with the late phase of lexical selection, at
the end of activation of a content word, when "the amount of activation is relatively low, competition
from alternative lexical candidates is relatively high" (Kolk 1995, 290.).
4. Sentence repetition test
In the course of a sentence repetition test the patient gave answers that were suggestive of
structure building operations of the mental parser. We will demonstrate this below.
The patient was 37 years of age, right handed, a car mechanic, suffering from a stroke, which
resulted in extended fronto-parietal hypodensity in left hemisphere. He performed some repetition
tests.
(E stands for the examiner who utters the sentence to be repeated. P stands for the patient's
replies. The test was in Hungarian; the glosses below contain the relevant details only):
Closed class lexical items in sentence processing
29
(3) E: Péter
beszélgetett
Marival.
Peter-nom talk-3sg/past Mary-with
'Peter talked to Mary'.
P: Péterrel beszél
..iinná ........
-val
Peter-with talk-3sg/present..nonsense-word -with
(4) E: Marival
találkozott
János.
Mary-with meet-3sg/past John-nom
'John met Mary'
P: Marival.... beszélgetett
volna
vele.
Mary-with talk -3sg/past would have her-with
Ő beszélgetett
vele....... Marival.
He talk-3sg/past
her-with... Mary-with.
(5) E: Mari
megcsinálta
az ágyat
és lefeküdt.
Mary-nom make-3sg/past/def the bed-acc and (she) go-3sg/past to bed.
‘Mary made the bed and (she) went to the bed.‘
P: Mara...... Mara......Mara......mmmmmm
Mara-nom ..Mara-nom Mara-nom .mmmmmmm
(6) E: Sándor küldött
egy képeslapot Marinak.
Alex send-3.sg/past a postcard-acc Mary-dat.
'Alex sent Mary a postcard.'
P: Sándor jött
Alex
és akkor írta‚
és
azt...
azt
come-3sg/past and then write-3sg/past/def and that-acc that-acc
Closed class lexical items in sentence processing
30
akkor ment
hozta....
a .. mi az
a....mit??
then go-3sg/past bring-3sg/past/def the what is that what-acc?
E: Képeslap!
Postcard-nom!
P: Épetlapot,
épeslapot
édeslapot.
Nonsense word-acc nonsense word-acc sweetcard-acc
E: Mit csinált
vele?
What did he do with it?
P: Képeslapot adott
a kis gyereknek
adott oda
Postcard-acc give-3sg/past the little child-dat give-3sg/past to
‘He gave a postcard to the little child...gave to and'
... és akkor ment haza
... and then go-3sg/past home
...'and then he went home'.
Analysis of the repetition test
4.1. In principle, the task of repeating someone else's words could be accomplished in several
ways:
(i) Purely phonological repetition: no syntactic or semantic processing is performed; the subject
simply repeats what he/she hears.
(ii) Surface syntactic repetition: the input sentence is processed up to the level of surface
syntactic form, which is then repeated without any semantic processing. This requires the subject to
process the surface syntactic structure, derive a phonological representation, and then produce the
phonological form thus derived.
(iii) Unmonitored semantic repetition: the input sentence is processed to extract the semantic
gist; the subject then repeats that gist without endeavouring to use the same syntactic structures or
phonological forms.
(iv) Monitored repetition: the patient processes the sentence both syntactically and
semantically, then attempts to produce an utterance, which matches the phonological, syntactic, and
semantic properties of the original utterance. Our patient was pursuing the strategy of monitored
repetition (iv).
4.2. Temporal asynchrony between accessing case endings and content NPs is shown in
example (3). The examiner produced an utterance in which the first NP was marked for nominative
31
Closed class lexical items in sentence processing
with a zero suffix (Péter) and the second NP was marked with instrumental case ending (Mari+val,
Mary-with) in sentence-final position. The patient produced an utterance in which the first NP was
marked with instrumental case ending (Péter-rel, Peter-with) and the final NP was not mentioned at
all. That is to say, the patient attached the case ending of the final NP to the first NP. He retrieved a
case ending that was heard later and attached it to a NP that was heard earlier.
In example (4) the target sentence contained a sentence-initial NP marked with instrumental
case ending (Mari-val, Mary-with) and a final NP marked for nominative with a zero suffix (János,
John-nom). In his first attempt, the patient repeated the sentence-initial NP marked with instrumental
case (Mari-val). He was unable to retrieve the sentence-final NP marked for nominative case with a
zero suffix (János, John-nom) rather he produced a grammatical pronoun marked with instrumental
case, i.e., he attached the instrumental case ending that has been retrieved to the pronoun (vel-e, withher). In his second attempt, the patient was able to retrieve the first case ending without the content
NP: he produced a grammatical pronoun marked for nominative case with a zero suffix (Ő, He-nom)
then produced a grammatical pronoun marked with instrumental case ending (vel-e, with-her) and
finally, after a pause, he repeated the content NP marked with instrumental case ending (Mari-val,
Mary-with). To sum up: by the end of the second attempt, the patient produced the complete surface
case ending frame of the target verb (NP-nominative, NP-instrumental), he tried to attach case
endings to NPs, during this process he used grammatical pronouns
(marked for nominative and instrumental case as well). He was able to attach a case ending that
was heard earlier to a NP that was heard earlier. He was able to retrieve a case ending that was heard
later and was unable to attach it to a NP that was heard later.
In our repetition test the parsing mechanism could not proceed unless a verb was produced.
This is shown in example (5). The target utterance contained two conjoined verbs with their different
case frames. The patient was not able to retrieve either of the verbs and was even unable to "list" only
the nouns with correct case endings. He also failed to use any inflections (see example (5)). But he
never made both inflectional errors and errors in the choice of the main verb in the same sentence.
This is compatible with the assumption that the patient has to trade processing of surface form against
lexical access. (Inflection is part of the surface parser module but I do not claim that this (sub)module
would not be impaired).
In example (6), the patient was attempting to repeat the Hungarian equivalent of Alex sent Mary
a postcard. He made several false starts: notably, they were semantically related to the intended
message. First, he tried the Hungarian equivalent of the verb came (semantically a motion verb, like
sent, but intransitive). Next he tried the Hungarian equivalent of the verb write-3sg/past/def (with
3.pers.-suffix referring to direct object as well), correctly transitive but more closely related
semantically to postcard than to sent). However he was not able to retrieve postcard itself. He
mentioned the accusative case ending (-t) of postcard without the content word (postcard), and linked
the accusative case ending to pronouns. (az-t: that-acc, mi-t: what-acc.) Next he tried the Hungarian
equivalent of went (which is again, incorrectly, intransitive). Next he came up with the Hungarian
equivalent of bring3sg/past/def (with 3.pers-suffix referring to direct object) which is both syntactically and
semantically close to sent. But by that time he was completely unable to retrieve what the object was
supposed to be. Next he heard the original noun marked for nominative with a zero suffix (the
Hungarian equivalent of postcard-nom) and he returned a nonsense word marked with an accusative
case ending! Next he heard a Hungarian pronoun marked with instrumental case ending (What did he
do with-it ?) and again he returned an accusative case ending but by that time he was able to repeat the
original content word (postcard-acc) linking accusative case ending to this content word.
Hungarian has a very rich inflectional system for nouns. It is remarkable that the patient did not
make purely inflectional errors in the repetition task. If he approximated the class of the target verb,
then its surface case frame was retrievable. Utterances in (6) show that the surface case ending of a
noun was mentioned earlier than the noun itself (with that case ending). See in (6) for instance the
Closed class lexical items in sentence processing
32
temporal relation between the Hungarian accusative case ending and the Hungarian equivalent of
postcard, and the temporal relation between the Hungarian dative case ending and the Hungarian
equivalent of Mary/little child, nouns in the dative in the patient's responses.
It is remarkable that the patient did not make purely inflectional errors during repetition task. If
he approximated the class of target verb, then its surface case frame was retrievable. The patient's
responses exhibit the features of temporal desynchronization. Restarting, however, improves patient’s
performance and repairs desynchronization in repetitions.
4.3. Restarting
Our patient restarted his responses many times. Restarting can improve aphasics' performance
and repair desynchronization. The normal speaker, after error identification, can simply restart the
production of the sentence, or part of it. Restarting automatically leads to repairing. In the aphasic
speaker the situation is different. The question is how can restart lead to improvement. Suppose the
critical limitation would consist in a reduction of the size of a syntactic buffer. If a particular
structure would be too complex for this small buffer, restarting would have no benefit: a
structure of given complexity that does not fit into the buffer will never fit. When the limitation is a
temporal one, the situation is different. The basic difficulty that results from a timing deficit is the
following. Particular representational elements are activated too slowly, and reach their active
memory phase in working memory too late, when other elements, with which they have to be in
synchrony, are already gone out of working memory. No possibilities for synchronization. But
restarting does offer an advantage. After decay, the critical elements may still have a relatively high
level of activation, because of the fact that they have just been activated. Reactivation can occur from
this higher level, rather than from the rest level, and the critical element will reach the active memory
phase sooner. In this way, restarting leads to faster processing and to the possibility of
synchronization of structure building elements in working memory.
5. Grammaticality judgements
Agrammatic Broca's aphasics can correctly judge the grammaticality of certain sentences
(Linebarger 1995), while they are unable to produce them correctly. This has been interpreted in
various ways. We present some samples of Hungarian data that support a time-based approach to
sentence processing.
Hungarian is a more or less free phrase-order language. (Bánréti, 1994, É.Kiss-Kiefer,1994,
Pléh 1998). Syntactic functions and/or thematic roles are expressed by attaching case suffixes to NPs.
The possible subcategorization by verbs involves at least 17 cases expressed by surface case ending
forms. Suffixes of a finite verb express number and person of subject and definiteness of direct
object. Another set of suffixes indicates tense and mood.
5.1. The partial process
The grammaticality judgement tasks do not involve the kind of extended simultaneity. These
tasks are easier than comprehension tasks in aphasia. Grammaticality judgements require shorter
availability of the syntactic representation in memory than comprehension tasks and are therefore less
easily disrupted. Solving judgement tasks does not require that the parser waits "until all subtrees
have been parsed and attempts to solve all of the constraint at once". It is not necessary that a
syntactic tree for a full sentence should be available. Judgement of grammaticality is possible as soon
as minimally sufficient structural information has been made available. Patients' performance in
judgements depends on the type of grammatical error hidden in the task, i.e., on the availability of the
minimally sufficient structural information, which is necessary for correct judgement.
5.2. The initial structure building operations
In what follows we apply the first-pass parse hypothesis. The hypothesis of initial structure
building operations has been proposed by a number of psycholinguists (e.g., Linebarger 1990). In
accordance with this hypothesis I assume that in the case of grammaticality judgements an initial
Closed class lexical items in sentence processing
33
structural analysis is computed and is subsequently interpreted. This is followed by later processing
operations involving constraints on the indexing of structures. In the sense of Saffran (1985), the firstpass parser protects some of the processed syntactic information during first-pass parse and a working
memory deficit restricts further processing operations.
The solution of a grammaticality judgement task is based on a minimally sufficient structural
representation. (For aphasic subjects, grammaticality judgement tasks are easier than comprehension
tasks). What counts as a minimally sufficient structure, within a given language, will change from
task to task. 'Easy-to-judge task' means that minimally sufficient structure is available and 'hard-tojudge task' means that minimally sufficient structure is not available.
As for Hungarian speaking aphasics, we assume that the first-pass parser is based on the verb,
its subcategorizational selections for syntactic category of complements and for case endings (that
marks theta role on surface structure). This constitutes important syntactic information for the
possible syntactic structure, the possible linear order of categorized syntactic slots and the hierarchy
of nodes of the structure. According to the Projection Principle, syntactic representation must be
projected from the lexicon in that they observe subcategorizational properties of lexical items.
5.3. Subjects
Five subjects had had a cerebral vascular accident (CVA) in the left hemisphere. Patients were
diagnosed as agrammatic Broca's aphasics on the basis of performance profiles on the Western
Aphasia Battery (WAB) (Kertesz 1982) and in further clinical evaluations by speech-language
pathologists and neurologists. All subjects were right handed. Patients were required to judge taperecorded sentences. The instruction was "please tell me whether this sentence is correct or incorrect"
„As you feel, and no explanation is required“. Grammatical and ungrammatical items all figured in
minimal pairs in the test. Each minimal pair stood for a particular structural category. Members of a
minimal pair were separated by intervening items.
5.4. The main experimental results
Some conditions were easy and some hard. The hard conditions break into two main groups:
systematic misjudgements and guessing. Table 1 shows the distribution of judgements according to
particular sentence structures.
Closed class lexical items in sentence processing
34
Table 1. - Patients’ responses
Grammatical sentences
Task
Judgement:
Correct
Ungrammatical sentences
Judgement:
Wrong
Correct
Wrong
EASY TASKS:
Case endings
in the sentence
30
0
30
0
V-anaphora
30
0
30
0
________________________________________________________________________________
HARD TASKS:
SYSTEMATIC MISJUDGEMENTS
Unfocussable sentence
adverbial in focus
30
0
3
27
All 3 arguments precede the verb
30
0
2
28
____________________________________________________________________
HARD TASKS: GUESSING
pro-Subject
16
14
17
13
Sentential intertwining
13
17
9
21
Anaphora + case
11
19
14
16
Aspect
18
12
14
16
____________________________________________________________________
Closed class lexical items in sentence processing
35
5.4.1. Examples for easy tasks
In easy conditions patients’ judgements were correct for grammatical and ungrammatical
sentences, alike.
Case endings in the sentence:
The acceptable version contains a grammatical configuration of surface case endings assigned
by Verb to its arguments, and agreement of verbal suffixes expressing person and number of Subject
and Object. The ungrammatical versions involved errors with surface case endings attached to NPs
and errors with person-number suffixes of a finite Verb. In these items only one argument NP
preceded the verb, the others followed it in the surface string.
Examples:
(7) a. A papá-nak
kölcsönadott
a fiú
egy könyv-et .
the father-dat lend- past/3sg the boy-nom a book-acc
'The boy lent a book to the father.'
b. * A papá-*ra
kölcsönadott a fiú
egy könyv-et .
the father-*on lend-past/3sg the boy-nom a book-acc
(8) a. Róbert
nézi
a könyvet.
Robert-nom look-3.sg/present/def the book-acc
'Robert looks at the book'
b. * Te
nézi
a könyvet.
You-nom look-3.sg/present/def the book-acc
(9) a. A gyerek-et
elküld-te
a bolt-ba
a mama.
the child-acc send-past/3sg the shop-to the mother-nom
'The mother sent the child to the shop.'
b. * A gyerek
elküld-te
a bolt-ba
a mama.
the child-nom send-past/3sg the shop-to the mother-nom
V-anaphora: copying only bare V
Another example for easy conditions is task V-anaphora, which requires the judgement of the
category of Verb itself, whether it is an attribute predicate or an action verb.
Example:
Closed class lexical items in sentence processing
36
(10) a. János magas volt és Mari is.
John tall
was and Mary too
'John was tall and Mary too.'
b. * János magas volt és
John tall
ezt
csinálta Mari is.
was and this-acc did
Mary too
*'John was tall and so did Mary.'
5.4.2. Examples for systematic misjudgements
The hard conditions break into two main groups. The first one is the group of systematic
misjudgements. In this case, acceptable sentences were judged as good with 100%, but unacceptable
counterparts were judged as good with close to 100%. These tasks contain errors, which cannot be
detected by means of surface case frame of the verb.
Unfocussable sentence adverbial in focus:
In these tasks the surface case ending frame is the same in the grammatical and ungrammatical
sentences, alike. In Hungarian there is a distinct syntactic position for the focused constituent,
accompanied by heavy stress, before the Verb. Perhaps-type sentence adverbial can occur in several
syntactic positions, except the position of focus. Perhaps-type adverbial is an unfocussable category
in Hungarian syntax. If this adverbial is put into syntactic position of focus, the sentence will be
ungrammatical. Patients, however, accepted this unfocussable category in the position of focus.
Example:
Perhaps-type of unfocussable adverbial is put into the syntactic position of Hungarian focus in
front of the verb. Capitals and " stand for the focus position and heavy stress.
(11) a. János talán
elkésett.
'John perhaps came late.'
b. * János "TALÁN késett el.
It is PERHAPS that John came late.
All three arguments precede the verb:
Other example for systematic misjudgements is task All three arguments precede the verb. This
condition involved errors with surface case endings. It is well known: the Verb assigns the surface
case frame attached to NPs. Because, in these conditions, verb was the last syntactic constituent in
surface string, the correctness of case endings assigned to NP's may be assessed without any
knowledge of verb or, once verb becomes known the entire string can be recalled and case endings
verified. Under these conditions patients failed to detect errors with surface case endings.
Examples:
Judgements of case endings and agreement of person and number suffixes between NPs and
Verb are required. All three NPs precede the Verb.
(12) a. A gyerek-et
a bolt-ba
a mama
elküld-te.
Closed class lexical items in sentence processing
37
the child-acc
the shop-to the mother-nom
send-past/3sg
'The mother sent the child to the shop.'
b. * A gyerekről
a bolt-ba
a mama
elküld-te.
the child-about the shop-to the mother-nom
(13) a. A papá-nak
a fiú
egy könyv-et
the father-dat the boy-nom a book-acc
send-past/3sg
kölcsönadott.
lend-past/3sg
'The boy lent a book to the father.'
b. * A papá-*ra
a fiú
egy könyv-et
the father-*on the boy-nom a book-acc
kölcsönadott.
lend-past/3sg
5.4.3. Examples for Guessing
The second main group of hard conditions is Guessing: Judgements were random or chaotic.
With a subtype of these tasks patients were required to judge syntactic dependencies, which involved
two clauses, two different verbs and dependencies between arguments of these two verbs (tasks of
VP- anaphora, Gapping, Sentential intertwining, pro-Subject).
pro-Subject
In the example below the Subject is the Hungarian equivalent of my mother. According to the
grammatical rules of Hungarian, the overt lexical item of Subject from the first clause can not be
repeated in overt form in subject position of the second (subordinate) clause. A phonologically empty
pronoun must be found in the position of the repeated Subject (indicated by pro). If this position is
filled with the repeated Subject in overt lexical form, the whole sentence becomes ungrammatical.
Example:
pro is found in the position of repeated Subject. Judgements are required for pro and overt
lexical material in the syntactic position of the repeated Subject:
(14) a. Anyukám
azt gondolta, hogy
'My mother thought
i
megkapta az állást.
that [pro] had got the job.'
i
b. * Anyukám azt gondolta, hogy Anyukám
* 'My mother
i
megkapta az állást.
thought that my mother had got the job.'
i
Closed class lexical items in sentence processing
38
Aspect
A different example for type Guessing is task Aspect. In the grammatical version of example
below, the verb in itself denotes progressive aspect. This is compatible with the ‘duration’ meaning of
a time adverbial in the sentence (Hungarian equivalent of for two days). The ungrammatical version
contains a verbal prefix that has [perfective] feature, indicating perfective aspect. This verbal prefix is
Hungarian EL. And there is a time adverbial with duration meaning -- (for two days) -- in the
sentence, again. [Perfective] feature and [duration] feature are incompatible with each other,
therefore the sentence will be ungrammatical. Patients failed to detect this feature clash.
Example:
Judgements are required for compatibility of aspect of verb and time adverbial.
(15) a. Két napon át
for two days
készítette
az ebédet.
(she) was making the dinner-acc.
'She was making dinner for two days.'
b. * Két napon át
az ebédet.
elkészítette
for two days (she) has made (='completed making') the dinner-acc
Sentential intertwining:
Judgements are required for lexical material in the syntactic position of a trace of moved NP.
The constituent THE BOOK was moved from the subordinate clause into the main clause. Its trace is
marked by (trace). Capitals and " stand for heavy stress-bearing Focus position)
(16) a. Maria KÖNYVET mondta, hogy
Mary the book-acc
said
megveszi Jánosnak.
that (she) buys (it) John-dat
'As for Mary, it was the book that she said she would buy (trace) for John.'
i
i
b. * Mari a KÖNYVET mondta hogy a kabátot
Mary the book-acc said
megveszi Jánosnak.
that the coat-acc (she)buys John-dat.
* 'As for Mary, it was the book that she said she would buy *the coat for John.'
Anaphora + Case:
Judgements are required for case assignments to anaphora and its antecedent. Word order is
free, the case assignment, however, is bound. Antecedent must be marked with a zero suffix for
nominative and anaphora must be marked with accusative case ending. Therefore NP-nom and
himself-acc are grammatical but the NP-acc and himself-nom are not grammatical.
(17) a. A vezető
látta
önmagá-t
the driver-nom see-past/3sg/def self-3sg/acc
'The driver saw himself in the mirror.'
a tükörben.
the mirror-in
Closed class lexical items in sentence processing
39
b. * a vezető-t
önmaga
látta
a tükörben.
the driver-acc self-3sg/nom see-past/3sg/def the mirror-in
* 'Himself saw the driver in the mirror'.
5.5. A post-hoc analysis of data
How can parser and grammatical representation of a sentence interact with each other? I
suppose that patients' performance in judgements depends 1/ on the capacity of the impaired parser,
and 2/ on the type of grammatical error hidden in the grammatical representation of sentence. I accept
the hypothesis of a distinct first stage of sentence parsing, which is called first-pass parse or initial
structure building operations.
5.5.1. Judgements in easy conditions can be based on initial structure building operations and
the first-pass parse can protect information that was analysed
The first pass parse is supposed to be a more or less phrase structure parse, sensitive only to the
basic features of syntactic categories of input. Initial structure building operations process only the
scheme or “gestalt“ of sentence. Grammatical errors involved in our easy conditions were related to
basic features of syntactic categories. Therefore judgements in easy conditions can be based on
dependencies, which are computed during first-pass parse, when initial structure building operations
take place. According to easy tasks in our test, the crucial relations are the following: the local
dependencies between the category of verb and its subcategorization frame, its surface case ending
frame and tense and mood inflections.
Processes can be effected in stepwise checks on surface inflectional endings: "what it is seeking
to match what in their basic features“. Parser is orientated by sets of structural expectations. These
dependencies are carried along as alterations of the internal state of the parser, therefore this
information can be protected from temporal or memory deficit provided that first-pass parse works.
The specific semantic/pragmatic features are not available to the first pass parse.
5.5.2. Systematic misjudgements can be related to erasure of specific features during processing
and to slowing down of parser
Some syntactic information, which is not encoded in the internal state of parser, after first pass
parse, is unprotected. Unprotected information decays more rapidly in aphasics than in normals. This
can cause the erasure of the specific features of syntactic categories (like category of unfocussable
adverbial) and the underspecification of features of closed class categories (like case endings). The
erasure of specific features of syntactic categories and closed class categories result in an incomplete
processing of sentence structure. The grammaticality judgements can be based on partial processing.
In systematic misjudgements patients made a partial analysis and were unable to detect the
feature clash lurking there, when the critical features were not expressed by surface forms, rather they
were “hidden“ into the properties of a syntactic category. (Features of unfocussable sentence
adverbial or features of reciprocal anaphora). This can lead to a strong Yes-bias type of poor
performance.
With task All three arguments precede the verb, grammatical errors were related to the
configuration of case endings. Patients' performance, however, deteriorated to Yes-bias. The relevant
data for judgements are relations between specific category of verb, which is the last constituent, and
a NP, which is the first constituent. The correctness of case ending attached to the first NP can be
judged after the verb has processed. In other words, the correct judgement of the first NP requires the
Verb and its subcategorization frame as a starting point. When the verb becomes known the entire NP
string preceded the verb must be recalled and the case ending attached to first NP can be verified. In
this condition patient’s performance reflected the slowing down of the parser: because Verb was in
Closed class lexical items in sentence processing
40
the last position of the surface string, the slow parser was too late to receive its starting point: the
category of Verb and its surface case ending frame. The reason why patients neglected the
configuration of case endings is the lack of the safe starting point in due time. Patients’ incorrect
judgements reflect the fact that NPs, one by one, were correct in themselves, if they were considered
independently of Verb. Patients may follow a strategy like this. It is remarkable: patients' performance
was 100 % correct in easy tasks Case endings in the sentence at which Verb was not the last but the
second constituent in surface string.
5.5.3. Guessing responses reflect a sketchy and unfinished analysis of stimuli because of
desynchronization of parsing modules
With Guessing conditions, subjects not only failed to detect ungrammatical features of
stimulus, but also mysanalyzed fully grammatical sentences. How is it possible?
To judge these tasks, syntactic and lexical processes should have been integrated. In compound
sentences, patients should have judged whether syntactic positions (like Subject, Hungarian Focus or
a position of a trace of moved constituent) and their content word fillers were associated with each
other grammatically or were not (tasks pro-Subject, Sentential intertwining). Or, patients should have
judged whether features of a closed class item were compatible with features of a content word when
they were in distinct syntactic positions (tasks Aspect and Anaphora + case).
Impairments on accessibility of closed class morphemes create syntactic difficulties. The results
in our grammaticality judgement tests are compatible with findings in Haarmann and Kolk (1994) and
Kolk (1995): agrammatic aphasia may show either slow activation or fast decay but not both at the
same time. The normal activation goes at the expense of fast decay and, vice versa, normal decay goes
at the expense of slow activation. Applying this theory to our data we find the following. Specific
features of syntactic subcategories and closed class morphemes can be activated at a normal rate, but
then they decay very fast, too early from working memory; or they can be retained at a normal rate, at
expense of slow activation into working memory. In the fast decay case other specific lexical
information had not been activated yet, when needed. In the slow activation case other specific lexical
information in working memory is already gone when needed.
The fast decay or slow activation of grammatical features and subfeatures causes a
desynchronization in the building of syntactic structure. Syntactic slots are opened up too late or too
early for content word filler; specific lexical information in working memory had not been activated
yet or is already gone when needed. Therefore patients are not able to complete the analysis of
stimuli, processing operations result in a merely sketchy and unfinished structure. Patients were aware
of their unfinished analysis; they often made comments on it. This could lead to guessing responses
on complex, non-local relations.
5.5.4. Summary
Patients were able to use of initial structure building operations involved in first pass parse for
correct judgements of easy tasks. In normals first pass parse must be tightly synchronized with a
second major parsing module that extracts detailed and specific features of category of arguments and
predicate. But fast decay or slow activation of specific, unprotected information in working memory
can cause desynchronization between processing modules. The consequences are: systematic
misjudgements or guessing responses, depending on the type of grammatical error and the complexity
of sentence to be judged.
6. ellipsis and sentence processing
6.1. In our view ellipsis is essentially non-insertion of phonologically based Vocabulary items
into the nodes of syntactic structures. From the point of view of syntax-lexicon interface, what used to
be referred to as the lexicon is in fact at least two distinct lists of items: one serves as the store of
initial syntactic inputs, in the form of syntactic/semantic feature bundles, while the other (called
Vocabulary) contains the phonological shapes, to be associated with the featural nodes postsyntactically. Only the semantic-syntactic feature bundles participate in the syntactic computation,
Closed class lexical items in sentence processing
41
and one output of the computation is associated with the elements of the phonologically based
Vocabulary. The way ellipsis resolution must be conceived of is likely to support this view of
distributed lexicon. A new account for Hungarian ellipsis data in the frame of the distributed lexicon
theory is found in Bartos 1998.
6.2. Ellipsis in agrammatic aphasia
6.2.1. According to Kolk- Hofstede 1994, agrammatics overuse linguistic option of ellipsis in
free conversation. Telegraphic, fragmented speech shares properties with normal ellipsis, when the
missing elements can be derived from the context. Kolk - Hofstede 1994 presented a list of congruent
properties of normal ellipsis and agrammatic, fragmented spontaneous speech.
The use of elliptic sentences in spontaneous speech is a kind of adaptation. Broca's aphasic
patients are aware of their reduced capacity in linguistic system, therefore they employ elliptic
constructions. The role of this strategy is to prevent computation overload in the linguistic system.
Employment of this strategy (among others) is optional rather than obligatory.
6.2.2. Hungarian word order is free; thus any permutation of subject, verb and object is
grammatical. Hungarian is a discourse configurational language: the ordering of constituents has
discourse functions. Syntactic positions like topic position and focus position contain key constituents
of sentence. These syntactic positions serve to express discourse functions: topic is GIVEN, focus is
NEW, CONTRASTED, compared to previous discourse. Topic and focus are not restricted with
respect to case. A DP in topic position can be marked with accusative, dative, instrumental case
endings. The same is true for a DP in focus position. A sentence can contain more than one DP in
topic position and only one DP in focus position.
Ellipsis is understood as the non-insertion of the phonological shape of a V’ (or a VP) after
focused DP or quantifier DP in a surface syntactic string. Ellipsis does not coincide with VP
anaphora. In the case of ellipsis there is NO overt anaphoric expression in the position of elided string
(like. …did too). We refer to the elided V -projection (V-bar) as an empty category of V-bar.
Hungarian grammar allows forward and backward types of Verb Phrase ellipsis.
Some examples:
In the examples: [………..] stands for the position of elided string, which is a V-bar (Cf. [ V DP ].
Diagrams contain relevant details only:
DP/t = DP in topic position
DP/q = quantifier DP (NP)
AdvP = Adverbial Phrase
DP/focus= focused DP (NP)
A = Adjective (Phrase)
V = verb
DP-acc, DP-dat, DP-poss:= DP marked for accusative case, or dative case or possessive ,
respectively. DP-nom= marked with a zero suffix for nominative. pro=pronoun
Hungarian forward ellipsis in co-ordinated sentence:
Forward ellipsis after quantifier position:
DP/t DP/q V DP
and DP/t DP/q [ V DP ]:
(18)
Mária
" minden fiúnak odaadta az ajándékot,
Mary/nom every
boy-to gave
majd János minden lánynak
the present-ACC, then John/nom every girl-to
[ V DP ].
Closed class lexical items in sentence processing
42
DP/t
DP/q
V
DP
DP/t
[ V DP ].
conj DP/q
'MARY gave the present to EACH BOY, then JOHN to EVERY GIRL’.
Forward ellipsis after DP/focus position:
DP/t DP/focus V DP and DP/t DP/focus [ V DP ]:
(19)
Péter
"Annának mutatta be
Peter
Ann-to
DP/t
DP/focus
Róbertet , Jen pedig “Marinak [ V DP ].
introduced(in) Robert-acc , Eugene, however Mary-to
V
DP
DP/t
conj
DP/focus [ V DP ].
'PETER introduced Robert to ANN but EUGENE to MARY.'
Forward Sluicing saves a Wh-word:
DP/t DP/focus V DP but DP/t not V that when [ S1 ]
(20)
[S2…….………]
Erzsi "London-ban tanult angolul, de
én nem tudom, hogy "mikor [ S1 ].
Liz
I not know
London-in learned English but
DP/t DP/focus
V
DP
conj DP-t neg V
that when
comp Wh/focus [ S1 ].
'Liz has learned English in London but I do not know when.'
Hungarian backward ellipsis in co-ordinated sentence:
Backward ellipsis after quantifier position:
DP/t DP/q [V DP ] and DP/t DP/q V DP
(21)
Mária
"minden lánynak , [ V DP ], János
Mary/nom every girl-to
DP/t
DP/q
pedig
minden fiúnak odaadta az ajándékot.
John/nom whereas every boy-to gave the present-acc
[V
DP] DP/t
conj
DP/q
V
DP .
'MARY [gave the present] to EACH GIRL , and JOHN gave the present to EVERY BOY'.
Backward ellipsis after DP/focus position:
DP/t DP/focus [V DP] and DP/t DP/focus V DP
(22)
Péter
"Annának, [ V DP ], Jen
Peter/nom Ann-to,
pedig “Marinak mutatta be
Róbertet.
Eugene/nom however Mary-to introduced(in) Robert-acc
Closed class lexical items in sentence processing
43
DP/focus [ V DP ],
DP/t
DP/t
conj
DP/focus V
DP
'PETER [introduced Robert ]to ANN but EUGENE introduced Robert to MARY.'
Backward Sluicing saves a Wh-word:
DP/t not V that when [ S2 ], but DP DP/focus V
DP .
[S2…………..……………..…]
(23)
Én nem tudom, hogy "mikor [ S2 ]
I
not know
DP/t not V
that when
de .Erzsi "London-ban tanult angolul,
. but Liz
London-in
comp wh-focus [ S2 ] conj DP/t DP-focus
learned English
V
DP
' I do not know when, but Liz has learned English in London.'
6.2.3. Forward VP Ellipsis (FVPE) is dependent on its antecedent. The syntactic tree is
complete. Syntactic and semantic features of lexical items are present in ellipsis site; it is the
phonological form that is not inserted. No need to assume deletion of lexical items. Backward VP
Ellipsis (BVPE) sites result from non-insertion of phonological shapes of lexical forms. In the frame
of minimalist program Wilder (1997) adopts “split lexicon“ approach, following Halle & Marantz
(1993). According to this approach there are lexical items containing syntactic and semantic features
<Syn, Sem>, but lacking phonological content <Pho> of their over counterparts. The usual structure
derivation is fed by lexical items comprising <Syn, Sem>, but lacking <Pho>. Corresponding
phonological features <Pho> are inserted by a post-S-structure operation of Vocabulary Insertion
applying at Spell-Out. Vocabulary Insertion feeds PF but not LF. Wilder (1997) proposes the
following. Forward ellipsis sites contain lexical content throughout the derivation, but fail to undergo
phonological form-insertion. Backward ellipsis sites result from deletion after form-insertion.
Identification asymmetries between forward and backward ellipses fall out as a consequence of the
different levels at which identity is checked: forward ellipsis is licensed at Logical Form and
backward ellipsis is licensed at Phonetic Form.
6.3. Sentence repetition tests
We tested neurolinguistic reality of identification asymmetries with direction of VPE
mentioned above. Our subjects were aphasics. They were diagnosed as agrammatic Broca's aphasics
having good comprehension skill and a kind of syntactic and phonological (motor) impairment in
speech. Their speech output is characterized by difficulties in using inflectional affixes and
grammatical formatives, leading to fragmented telegraphic speech, (“syntactic“ impairments) and
difficulties in phonological/phonetic output: substitutions, omissions or distortions of sound (“motor“
impairments).
6.3.1. Test material involved co-ordinated sentences with VP ellipsis sites. Each test contained
15 sentences containing forward VP ellipsis and 15 sentences containing backward VP ellipsis. Two
subjects were given the test three different times. Sentence patterns were filled with different (though
equally frequent) words in each test but we did not change the sentence structures themselves. To
repeat sentences patients were pursuing the strategy of monitored repetition involving two basic
operations: (1) processing the heard utterances both syntactically and semantically, then storing them;
(2) attempting to produce an utterance, which matches the phonological, syntactic, and
semantic properties of the original utterance.
6.3.2. The main experimental results:
Closed class lexical items in sentence processing
44
Table 1. - Forward ellipsis
Total:
90
Correct responses:
52
Elided V in overt phonological form in the correct responses: 24
Incorrect responses:
38
The main types of the errors in the incorrect responses
(One response can contain more than one error)
Plural inflection attached to NP:
9
Agreement inflection (number, person) attached to Verb:
12
Tense inflection:
14
Case ending:
15
Second elliptic clause is fragmented:
4
Table 2. - Backward ellipsis
Total:
90
Correct responses:
21
Elided V in overt phonological form in the correct responses: 3
Incorrect responses:
69
The main types of the errors in the incorrect responses:
(One response can contain more than one error)
Plural inflection attached to NP:
14
Agreement inflection (number, person) attached to Verb:
28
Tense inflection:
21
Case ending:
20
First elliptic clause is fragmented:
29
Ungrammatical copy of person-number features
of elliptic conjunct into the licensing clause:
12
6.3.3. Examples from the material of the tests:
Correct response:
(24)
E: János a "repülőgéppel érkezett meg Londonba, Mari pedig a "kocsival [
John the plane-by
V PP
arrived perf London-to Mary, however the car-by
' John arrived in London by plane Mary however (arrived in London) by car.'
P: János a repülőgéppel érkezett meg Londonba, és Mari a kocsival [
John the plane-by
arrived perf London-to
V PP
and Mary the car-by
]
]
Closed class lexical items in sentence processing
45
'John arrived in London by plane and Mary (arrived in London) by car.'
The response is grammatical but not identical with the target:
(25)
E: Mária
minden fiúnak odaadta az ajándékot,
Mary/nom every
boy-to gave
majd János minden lánynak
[ V DP ].
the present-ACC, then John/nom every girl-to
'MARY gave the present to EACH BOY, then JOHN to EVERY GIRL’.
P: Mária
a fiúknak odadta az ajándékot,
Mary/nom the boys-to gave
János
meg a lányoknak.
the present-ACC, John/nom then the girls-to
The response is grammatical. Two constituents of elliptic clause are neglected and elided verb is
repeated in overt phonological form:
(26)
E: Erzsi "London-ban tanult angolul, de én nem tudom, hogy "mikor [ S1 ].
Liz
London-in learned English but
I not know
that when
'Liz has learned English in London but I do not know when.'
P : Erzsi "London-ban tanult angolul, hát... valamikor tanult ott [ S1 ].
Liz
London-in learned English well
once learned there
'Liz has learned English in London well once upon a time. (she) learned there.'
Distribution of case endings is different from the target. The elided verb is repeated in overt
phonological form:
(27)
E: Péter
Peter
"Annának mutatta be
Ann-to
Róbertet , Jen pedig “Marinak [ V DP ].
introduced(in) Robert-acc , Eugene, however Mary-to
'PETER introduced Robert to ANN but EUGENE to MARY.'
P: Péter
Peter
"Annának mutatta be
Ann-to
Róbertet , Marinak pedig Jen t [ V DP ].
introduced(in) Robert-acc , Mary-to however Eugene-acc
'Peter introduced Robert to ANN and Eugene to MARY.'
First elliptic clause is fragmented and mixed with the second one. Ungrammatical copy of
person-number features of elliptic conjunct into the licensing clause
(28)
E: Mária
"minden lánynak , [ V DP ], János
pedig
minden fiúnak odaadta az ajándékot.
Closed class lexical items in sentence processing
46
Mary/nom every
girl-to
John/nom whereas every boy-to gave the present-acc
'MARY [gave the present] to EACH GIRL , and JOHN gave the present to EVERY BOY'.
P: * A lány
a Jánosnak.. meg
The girl-sing
John-to
a fiúk is
hát.. odaadta
neki
az ajándékot.
and the boy-pl too well gave3sg/past to him/her the present-acc
Ungrammatical person-number inflection on the Verb. Ungrammatical copy of person-number
features of elliptic conjunct into the licensing clause
(29)
E: A gyerek a ‘’villamosra,[
] a férfiak pedig a ‘’buszra szálltak fel.
the child-NOM sg. at the train-LOC , men-pl.NOM got-pl.PAST up
It is the train that the child (got on), it is the bus that the men got on.
P: A gyerek a ‘’villamosra, [
] a férfi pedig a ‘’buszra szállt fel.
the child-NOM sg. at the train-LOC, man-sg NOM at the busz-LOC got on-3sg..up
In the response: ungrammatical Aux and Tense marker attached to the Verb. First and second
clause are mixed with each other: Ungrammatical copy of person-number features of elliptic clause
into the licensing clause
(30)
E: Holnap Mari
fog,
tegnap meg
tomorrow Mary-NOMsg3 will-sg3 [
írt
egy verset.
], yesterday then she-NOMsg3 wrote-PASTsg3 a poem-ACC
It is Mary who will (write a poem), it is he who wrote a poem yesterday.
*P: Tegnap Mari
fog ....
tegnap Mari,
tegnap
pedig
fog
írni
egy számlát.
yesterday Mary-NOM will-sg3 yesterday Mary-NOM, yesterday however he-NOM will-sg.3 writeInf
an invoice-ACC.
First elliptic clause is completely ungrammatical and fragmented. Ungrammatical distribution of case
endings
(31)
E: Péter
"Annának, [ V DP ], Jen
Peter/nom Ann-to,
pedig “Marinak mutatta be
Róbertet.
Eugene/nom however Mary-to introduced(in) Robert-acc
'PETER [introduced Robert ]to ANN but EUGENE introduced Robert to MARY.'
P: Róbertnek, Róbertet
Annát
bemutatják,
Closed class lexical items in sentence processing
47
Robert-dat Robert-acc Ann-acc introduced (in) –3sg/pres
First elliptic clause is neglected
(32)
E: Én nem tudom, hogy "mikor [ S2 ]
I
not know
that when
de .Erzsi "London-ban tanult angolul,
. but Liz
London-in
learned English
' I do not know when, but Liz has learned English in London.'
P: Tanult angolul Londonban .. valaki...
(he/she) learned English in London, someone...
6.4. A post-hoc analysis of data
6.4.1. Identification asymmetries between FVPE and BVPE (the different levels at which
identity is checked) are relevant for the real sentence processing operations.
Repetition of BVPE imposed Syntax/Phonology interface requirements that exceeded the
impaired capacity of language processor with agrammatic aphasics.
At least two types of operations are required in the processing of ellipsis. The processor needs
to be able to look ahead and look back over some number of successive items or structures, to store a
linear array of lexical items. Processes based on rapidly assigned surface syntactic representations.
There must also be means of storing items or structures that are not yet complete, to which additional
surface material can subsequently be attached. This is one of the functions that is performed by order
preserving phonological buffer, a type of short-term memory (Martin and Saffran (1997).
6.4.2. Producing co-ordinated sentences with FVPE in a repetition test requires the patients to
store content-based representation of co-ordinated sentence heard, then convert it into surface
syntactic and phonological form. Supposing the processor builds structures incrementally, from left to
right, there was no built-in delay in processes because of direction of ellipsis. Patients often
mentioned elided VP in overt phonological form at its correct position in second conjunct. It was easy
for them to reconstruct FVPE in overt phonological form. Order preserving phonological buffer
linked phonologically overt antecedent to dependent lexical items with lacking phonological form in
the second conjunct.
To produce co-ordinated sentence with BVPE in repetition test, it is necessary to recover the
omitted lexical items in first conjunct. Supposing the processor builds structures incrementally, from
left to right, there is a built-in delay in operations because of direction and identification level of
backward VPE. (Recovering is delayed, because the elided material is located in the first clause, and
the phonologically overt licensing string is found in the second clause). Order preserving
phonological buffer should serve a place holding function for VP ellipsis site, ensuring that elided
material is recovered and order information is maintained despite backward direction. Patients were
able to repeat mostly the second clause in correct grammatical form. The first elliptic clause often was
fragmented and ungrammatical. Elided VP in first clause rarely was mentioned in its overt
phonological form. This is because the demands of task increased to the extent that exceeded the
capacity of impaired processor.
48
Closed class lexical items in sentence processing
6.4.3. The data characterized above are relevant from the point of view of a time-based parser.
The differences are related to timing. Generally speaking, when a non-empty Verb appears, most of
the pieces to properly assemble the clause are available.
In case of backward ellipsis the question is the following: how to build a sketchy structure for
the FIRST clause without the lexical material of a V-bar.
In case of forward ellipsis, the parser processes the lexical material of antecedent V-bar in the
first place, then tries to analyse an empty category later. An empty (elided) category can be detected
by means of surface structural parallelism. According to parallelism, a „later“ empty category is the
same type as its antecedent category and both of them must occur in same type of syntactic positions
in the first and the second co-ordinated clause, respectively.
The role of parallelism is important. At the very moment when a parser detects an empty
category of V-bar, the parser tries to find an antecedent V-bar whose lexical material has been
processed earlier and makes use of the processed lexical material of that V-bar once again in the
building of the second clause. Forward ellipsis is easier for impaired speech, because the same
category is used twice in two parallel structures but the overt lexical form is mentioned only once.
Backward ellipsis can be harder for the time-based parser. Backward ellipsis means: an empty
category is detected at first. At the very moment when a parser detects an empty V-bar there is no
information about the subtype of it and no information about the lexical material of V-bar. The
decision is postponed. The parser must put that empty category into memory buffer and wait for a
posterior lexical item: the licensing lexical material of a posterior V-bar in the second clause. After
processing the posterior V-bar the parser tries to determine the identity of the phonological form of
the posterior V-bar and the elided V-bar and copy back semantic-syntactic features associated to the
phonological form of the posterior licenser. At this stage, the direction of operations contradicts to the
incremental structure building, which is a fundamental processing principle.
6.6. A hypothesis on the structure of mental parser
Suppose the following structure-building operations. The mental parser must produce structural
frame for all possible sentences. This syntactic frame contains categorized slots. When the
configuration of surface case endings assigned by category of Verb to its complements and the
configuration of other closed class items are in their active phase in working memory, they define and
open up syntactic slots for content word filler. Content-words would be generated by lexicon and
would be inserted into their slots in the syntactic frame.
Because, closed class items have to be integrated with their categorized slots in the syntactic
frame, and open class (content) words have to be inserted into their categorized slots in the syntactic
frame as well, these two kinds of integration require synchronization, synchronized activation of
structure building elements in working memory for language. The slow activation or fast decay of
closed class items leads to a desynchronization between syntactic slots opened up by closed class
items and active phase of content word fillers.
We define mental parser as an automatic device that becomes specialized in processing of
categories and features, involved in grammatical representation of sentences. Under this view parser is
a device that transfers information between grammatical representation and message level
representation. Parser computes grammatical representations of sentences and transforms them into
message level representation (at which the „what is to be said“ is represented). The category and
feature system is hierarchical in grammatical representation. It has various levels of sub- and sub-sub
categories, from the bare category to the individual lexical item and from the closed class category to
the fully specific features of that closed class item. Then, it is the question of capacity and
synchronization how far down the hierarchy in grammatical representation, the parser goes on its
search for information. The distribution of patients’ performance in tests reflects the limitations on the
interface between impaired parser and grammatical representation containing a hierarchy of
categories and their features.
49
Closed class lexical items in sentence processing
References
Bartos, H. 1998. VP-Ellipsis and Verbal Inflection in Hungarian, ms.
Bánréti, Z. 1994. Coordination and Ellipsis, in The Syntactic Structure of Hungarian. Syntax and Semantics (
É.Kiss K. -Kiefer F.(Eds.)) Volume 27, 355 - 414. New York, Academic Press.
Biassou, N, Tyler, L.K., Nespoulous, L.J., Dordain, M., Harris, K.S. 1997. Dual Processing of Open and
Closed-Class Words, Brain and Language Vol 57, 360-373.
Bock, K. 1989. Closed class immanence in sentence production. Cognition, 31. , 163-186.
É.Kiss, K.- Kiefer, F. (eds.) 1994 The Syntactic Structure of Hungarian. Syntax and Semantics Volume 27,
New York, Academic Press.
Caramazza, A. 1990. (ed.): Cognitive Neuropsychology and Neurolinguistics, Lawrence Earlbum Associates,
Publishers. New Jersey.
Chomsky, N 1995. The Minimalist program, Cambridge, Mass.: MIT Press.
Chomsky N. 1997. Language and Mind: Current Thoughts on Ancient Problems, Universidade de Brasilia,
Pesuisa Linguistica, 3,4, 21 p.
Cornell, T. L. 1995. On the Relation between Representational and Processing Models of Asyntactic
Comprehension. Brain and Language Volume 50, 304-324.
Frazier, L. - McNamara, P. 1995. Favor Referential Representations. Brain and Language 49, 224-240.
Friedmann, N. -Grodzinsky, Y. 1997. Tense and Agreement in Agrammatic Production: Pruning the Syntactic
Tree, Brain and Language, 56, 397-425.
Haarmann, H J. and Kolk, H. J. 1994. On-line Sensitivity to Subject-Verb Agreement Violations in Broca's
Aphasics: The Role of Syntactic Complexity and Time. Brain and Language Volume 46, 493 - 516.
Halle, M - Marantz A. 1993. Distributed Morphology and the Pieces of Inflection. In: The View from Building
20, ed.: Keyser S.J. –Hale K.. MIT Press 111-176.
Kolk, H. H. J., & Hofstede, B. M. T. 1994. The choice for ellipsis: A case study of stylistic shifts in an
agrammatic speaker. Brain and Language, 47, 507--509.
Kolk, H. J. 1995. A Time-Based Approach to Agrammatic Production. Brain and Language Volume 50, 282303.
Linebarger, M.C. 1990. Neuropsychology of Sentence Parsing, In: Caramazza (1990) (ed.) , 55-122.
Linebarger, M. C. 1995. Agrammatism as Evidence about Grammar. Brain and Language, Volume 50, 52-91.
Martin N. and Saffran E.M. 1997. Language and Auditory-verbal Short-term Memory Impairments, Cognitive
Neuropsychology, 14 (5) 641-682.
Mészáros, É. (1999 ) Immediate Recalling of Sentences Containing a VP Ellipsis in a Broca’s Aphasic Patient.
ms
Pléh, Cs. 1998 A magyar morfológia pszicholingvisztikai aspektusai, (Psycholinguistic aspects of the
Hungarian morphology) in press: Strukturális magyar nyelvtan 3. Morfológia (Structural Hungarian Grammar.
3. Morphology (ed.: Kiefer F.) Akadémiai kiadó, 2000.
Pollock, J.Y. 1989 Verb movement, UG and the structure of IP. Linguistic Inquiry, 20, 365-424.
Wilder C. 1997. Some Properties of Ellipsis in Coordination, in: A. Alexiadou-T.Hall (eds): Studies on
Universal Grammar and Typological Variation, J. Benjamins, 59-107.
Egyeztetés agrammatikus afáziában
50
Egyeztetés agrammatikus afáziában
A szintaktikai fa metszése
Zoltán Bánréti
Absztrakt
A dolgozat a nyelvtanmodellek neurológiai realitásának problémájához szolgál empirikus
adalékokkal szolgálni. Az adatokat agrammatikus afáziások neurolingvisztikai teszteléseiben nyertük.
Az agrammatikus afáziában tapasztalt egyeztetési, morfoszintaktikai hibák értelmezésére, a
korlátozottságok különféle fokozatainak magyarázatára a Minimalista Program keretében a
szintaktika fa - metszés hipotézisének egy kiterjesztett változatát javasoljuk.
1. A neurolingvisztikai vizsgálatok a nyelvtant olyan környezetben írják le, melyben a nyelvtan
beágyazott az agy architektúrájába, interfész pontokon kapcsolódva az agy és az elme mentális
programjaihoz. A nyelvtanelméletek számára pedig ma már lehetséges kutatási cél olyan nyelvtan
modellek felépítése, amelyek konstrukciója a mentális nyelvtant, vagyis a beszélők agyában
reprezentált nyelvtan felépítését, tulajdonságait fejezi ki.
2. Az újabb neurolingvisztikai elméletek többsége (Kolk 1995, Linebarger 1995, Swinney Zurif 1995, Friedmann- Grodzinsky 1997) úgy tekinti a mentális grammatikát, mint egymással
összeköttetésben álló agyi területek által tárolt nyelvtani reprezentációk rendszerét, ahol az egyes agyi
területek strukturálisan is determináltak valamely specifikus nyelvtani reprezentációtípus tárolására
illetve közvetítésére.
3. Az empirikus adatok egy fontos osztálya abból adódik, hogy az agy különböző területein
jelentkező lokalizált sérülések -- a gondolkodási és más kognitív képességek épen maradása mellett -specifikus nyelvi károsodásokhoz vezethetnek, a nyelvi képesség valamely részrendszere,
részfunkciója sérüléséhez, korlátozódásához, a többi épen maradása mellett. Jobbkezes embereknél a
homloklebeny baloldalán, a homloklebeny harmadik agytekervényének hátsó területeit és a
precentrális agytekervény ezzel szomszédos alsó területét ért lokális károsodás a beszédprodukciós
képességek korlátozódását eredményezi. Az említett régiót Broca területnek nevezik, a nyelvi zavart
pedig Broca afáziának. Mai ismereteink szerint a baloldali elülső kérgi területek és a velük
szomszédos kérgi területek felelősek azokért a műveletekért, amelyek a beérkező nyelvi input
egységeinek a szintaktikai szerkezetbe történő gyors és automatikus szerkesztését végzik el.
3.1. A Broca afáziás betegek nyelvi tüneteinek számunkra legfontosabb sajátossága a
szintaktikai és a morfológiai deficit korrelációja. A Broca afáziásokat a viszonylag ép beszédértés
mellett elsősorban a beszédprodukció zavara jellemzi. Ez megmutatkozik a mondatfragmentumokat
produkáló, töredezett, lassú beszédben, gyakran elhagyott funkciószavakban, és az egyeztető
toldalékok, ragok, használatának hibáiban. A beszéd intonációja monoton, artikulációs hibák
jellemzik. Az alábbi, lejegyezett spontán beszédben aláhúzással jelöltük az explicit egyeztetési
hibákat (A szótalálási, szókeresési nehézségek hatásaitól most el kell tekintsünk).
3.2. Broca afáziás vizsgálati személy, (37 éves férfi, a lézió helye: bal frontális) lejegyzett
spontán beszédéből mutatunk be alább egy részletet:
Vizsgáló: Mi történt magával?
… egyszercsak ujjamban néz, így mi fordult meg.. nem tudo, hogy mi van? És akko így néz..
néz jó..jó.. nem bír mozdítani, nem bírtam egysz ..akkor megmozdulok a kézzel.. azt vár, vár a nem
mozdulni a kezem meg a testem. És nem tud, hogy mi van! Egy olyan fél perc.. vegy nem.. egy
órára,..hogy hát mondom, várni nem kellni, kellni.. egy orvos. És meg...beszéltük, hogy .....hát ugye ô
is, meg a másik is...mer
..mintha máskor szóval mindig a fár.. ház lejjebb, ott van a kórház... Úgy van, ismerek pár
orvos ..., hogy mindig, mindig, oda járt...., megisment, elismertem de mondom várj. De nem mehet
Egyeztetés agrammatikus afáziában
51
így, . vagy nem! Nem jól mondom, hanem ....szóval átmentünk az izébe,....tehát így a.. olyan
kényelemtől már megmondani..na. Egy bizonyos rész, egy olyan.... fél-háromnegyed tíz, tíz óra fele
amikor már éreztem..na. És aztán el is vertek, el is vertek (a szándékolt szó: el is vittek)
(Mészáros Éva lejegyzése)
3.3. A fent bemutatott adatokban látható az afáziás nyelvi korlátozottság egy fontos mozzanata,
a nyílt lexikai osztályú és a zárt lexikai osztályú egységek aktivációjának a szétválásai. A világ
objektumait jelölő tartalmas szavak bővíthető osztálya: a nyílt szóosztály. Az morfoszintaktikai
egyeztetési szabályok által közvetlen érintett objektumok a zárt szóosztályok tagjai közül kerülnek
ki, például az inflexiók, esetragok, névelő, általában a nyelvtani viszonyokat, és nem a világ
objektumait jelölő formatívumok. A zárt szóosztályú morfémák a mentális struktúraelemző és
struktúraépítő rendszer elemeit alkotják. A mondatfeldolgozás során például jelzik a főnévi szerkezet
kezdetét/végét, a fő- és alárendelt mondatok megkülönböztetését és határaikat, stb. A gyors
előhívásuk lehetővé teszi a lokális szintaktikai szerkezetről alkotott azonnali döntéseket.
A nyílt és a zárt lexikai osztályok aktivációjának valamilyen szétválása sokféle nyelvi
deficitnek összetevője lehet. Feltételezhető, hogy a nyílt és zárt szóosztálybeli egységek két
különböző “szublexikonban“ tárolódnak melyek külön pályákon érhetők el. Azonban egymáshoz
rendezve, szinkronizáltan kell őket aktiválni a mondatprodukció és feldolgozás során. Az
agrammatikus afáziások spontán beszédének töredezettsége és agrammatikussága összefüggésbe
hozható a zárt lexikai osztályok aktiválásának korlátozottságával.
3.4. Az afáziás korlátozottság nyelvtani viszonylatok mentén is jellemezhető. Ha például csakis
a munkamemória korlátozottsága okozna afáziát, akkor nehéz lenne megmagyarázni, hogy miért
éppen a lexikai egységek egyik osztályát érinti a memóriazavar, és a másikat nem, hiszen vannak nyílt
osztályú lexikai egységek, melyek rövidebbek, fonotaktikailag „könnyebbek“, mint egyes
formatívumok.
4. Az afáziás korlátozottság kifejezhetősége a nyelvtanmodellekben
4.1. Feltételezéseink a következők. A mentális nyelvtani reprezentációkat olyan
instrukcióhalmaznak tekintjük, melyek instruálják a beszédprodukciót, illetve a beszédfeldolgozást
szervező mentális programokat. Egészséges esetben a különféle agyi területeken tárolt nyelvtani
reprezentációk, aktiválhatók és (időben) szinkronizálhatók. Az aktivált nyelvtani reprezentációkat a
beszédprodukciót vagy a beszédfeldolgozást szervező valós idejű mentális programok mint
instrukciókat "olvassák el" azokon az interfész szinteken, melyek a nyelvtan és a mentális programok
között vannak. Ugyanakkor a lokalizált agysérülések a nyelvi képesség valamely részrendszere,
részfunkciója korlátozódásához vezethetnek, a többi épen maradása mellett, tehát a mentális nyelvtan
szétválhat funkcionális komponenseire.
Ennek vagy az az alapvető oka, hogy nincsen ép, „elolvasható“ nyelvtani reprezentáció, vagy
pedig az, hogy a beszédprodukciós, illetve beszédfeldogozó mentális programok korlátozódnak, egyes
nyelvtani reprezentációkat el tudnak olvasni, míg más nyelvtani reprezentációkat nem képesek
elolvasni.
Feltételezem, legalábbis a nyelvtanelméletek egy osztálya számára, hogy ezeknek a mentális
realitással bíró nyelvtani reprezentációknak a tulajdonságai, rendező elvei képezik tárgyukat.
Nem ismerek olyan közvetlen és cáfolhatatlan bizonyítékot, amely a nyelvészetben kifejlesztett
nyelvtanmodellek valamelyikének – abban a formában, ahogy azok leírtak – a mentális vagy
neurológiai realitását igazolná. Valamely nyelvtanmodell egészére nézvést nem tudhatjuk pontosan,
mennyire fedi le a neurológiai vagy mentális realitásokat.
Ettől azonban meg lehet különböztetni azt a kérdést, hogy van-e mentális vagy neurológiai
realitása annak, amit kifejez valamely nyelvtanmodell a nyelvtani reprezentációkról, azok
tulajdonságairól, melyeket például a levezetésükben felhasznál?
52
Egyeztetés agrammatikus afáziában
4.2. Ami az agrammatikus afáziás korlátozottság jellemzését illeti, ennek kifejezésére olyan
nyelvtanmodellek lehetnek alkalmasak, melyekben a szintaktikai szerkezet levezetése és a
morfológiai deriváció szigorúan feltételezi, sőt tükrözi egymást.
Ebben az esetben a
beszédprodukcióban megnyilvánuló morfoszintaktikai korlátozottságok összefüggésbe hozhatók a
szintaktikai szerkezet építésnek korlátozottságával, töredezettségével. Chomsky a Minimalista
Program keretében például ilyennek tekinti a nyelvtani reprezentációknak olyan tulajdonságait,
melyek lehetővé teszik, hogy – az interfész szinteken-- a nyelvtani reprezentációk, mint instrukciók,
tökéletesen elolvashatóak legyenek a külső szenzomotoros mentális programok számára
(beszédprodukció, beszédfeldolgozás), illetve a külső konceptuális mentális programok
(jelentés/interpretáció) számára (Chomsky 1999).
4.3. Az agrammatikus afáziának, egyebek mellett, van három olyan empirikus sajátossága,
melyek egymással összefüggésben jelennek meg. (Kolk, 1995, Linebarger, 1995, Pléh 1998.) Ezek a
következők:
4.3.1. A morfológiai zavarok. Az agrammatikus betegek hajlamosak a zárt szóosztályok
tagjainak, az inflexiók, ragok, névelők stb. elhagyására, néha helyettesítésére. Ugyanakkor a “nyitottosztályú“ szótár (tartalmas szavak) viszonylag megőrzöttek. Megnevezési feladatokban azonban a
Broca afáziásoknak több nehézségük van az igékkel, mint a főnevekkel.
4.3.2. A morfológiai deficit szintaktikai korlátozottsággal jár. Az afáziás betegek a nyelvtani
formák, szerkezetek nagyon szűk körét alkalmazzák, nagyon rövid, egyszerű kijelentő mondatokat
használnak a szokványos szórenddel. Ritkán produkálnak mondatbeágyazásokat, nagyon kevés
bővítményt használnak.
4.3.3. A zavarok harmadik osztálya a beszéd lelassulása és töredezettsége. A lelassulás
monoton intonációval jár együtt, a szintaktikai pozíciókhoz társult nyomatékokat (fókusz, kvantor
nyomaték) sem produkálják. A töredezettség pedig a „mondatdarabok“, fragmentumok
produkálásában nyilvánulhat meg. (non-fluens beszéd.)
4.4. Összefoglalóan azt mondhatjuk, hogy az agrammatikus afázia alapvető nyelvi tünete a
szintaktikai és a morfológiai korlátozottság együttjárása és összefüggése non-fluens, fragmentizált
beszédprodukcióval. Ez a megfigyelés olyan neurolingvisztikai elméleteket inspirált, melyek a
Minimalista Program keretein belül gondolkodnak. (Chomsky, 1995). Ezek az elgondolások az MP
által követett komputációs lépések neurológiai realitására keresnek adatokat és érveket. A Minimalista
Program egyik elve ugyanis éppen az, hogy a szintaktikai szerkezet levezetése és a morfológiai
deriváció feltételezik és tükrözik egymást. A szerkezetépítő műveletek a lexikonból szelektált
egységekhez nem adnak hozzá új információt vagy jegyet, hanem a lexikai egységek adott
grammatikai-kategoriális jegyeinek az ellenőrzését, egyeztetését végzik. Ehhez szükséges a kritikus
jegyek lokális szerkezeti relációkban való ellenőrzése, egyeztetése, a lexikai egységek szintaktikailag
helyes sorrendbe állítása.
5.1. Hagiwara (1995) a nyelvtani reprezentáció és a mentális feldolgozó kapacitás között a
következő összefüggést tételezi fel. Azok a szerkezetek, amelyeket a szintaktikai fastruktúrában
alacsonyabb szintű csomópontok dominálnak, olyan felépítő műveletet kívánnak meg, amely
rövidebb idő alatt elvégezhető. A lexikális egységeket összekombináló, szerkezetépítő műveletnek
(Merge, Chomsky, 1995)) tehát kevesebb lépést kell elvégeznie, mert alacsonyabb szinten levő
csomópontot kell megépítenie. Ez az agrammatikus betegek számára gazdaságosabb és könnyebb.
A mondatszerkezetben lexikai és funkcionális fejeket tartalmaz. A funkcionális fejek, mint
például a Tense, az AgrS, AgrO absztrakt morfológiai jegyeket dominálnak. A szerkezetépítő
műveleteket, a mozgatásokat az vezérli, hogy a jegyek, köztük a morfológiai jegyek megkívánják az
ellenőrzést. Minden jegy-ellenőrzés valamely szerkezeti projekció Spec-Head viszonylatában történik.
A mondatok a funkcionális projekciók sokszoros rétegzéseivel építődnek fel. A funkcionális fejek
hozzáadása vagy kombinációval (Merge) vagy a kombinációnak a csatolásával történik, amely
kombinál két fejet, mint például a Tense és az AgrS fej, és egy Tense-AgrS komplexumot formál. A
53
Egyeztetés agrammatikus afáziában
Spec pozíció helyettesítéssel megépített, például a szerkezetépítő művelet összekombinálja a (Specbeli) tárgyat és az AgrO' –t, ami projektál egy új kategóriát: az AgrOP-t.
Ezekhez és a hasonló műveletekhez, Hagiwara érvelése szerint produkálási vagy feldolgozási
költséget rendelhetünk. A költség terminussal az „economy“ elvre utal Hagiwara, abban az
értelemben, hogy egy adott nyelvben a szerkezetépítő műveletek lehetséges tárházából azt a változatot
választjuk, amely az adott nyelv típusa, inflexiós rendszere stb. alapján a minimális ráfordítást igényli,
megtoldva azzal, hogy agrammatikus afáziában ehhez a mentális memóriatárolók (procedurális,
deklaratív és munkamemória funkciók) korlátozottságát is figyelembe kell venni. Ez azt jelenti, hogy
a sérült szerkezetépítő műveletek memória tároló igénye radikálisan megnőhet a normál esethez
képest. Az agrammatikus afáziás személyek olyan komplexitású mondatreprezentációk megépítésére
törekednek, melyek még összehangolhatók a memóriatároló kapacitással, ezek lesznek a számukra
gazdaságos szerkezetek. Ezeket a feltételezéseket számos nyelven végzett empirikus kutatással
motiválták.
5.2. Hagiwara (1995) japán afáziásokkal végzett teszteléseinek eredeménye az volt, hogy az
agrammatikus afáziások hibás mondatszerkezetei mögött olyan sérült szintaxis áll, melyben a
szintaktikai fastruktúrán belül a funkcionális kategóriák károsodtak. A funkcionális kategóriák a
szintaktikai szerkezet azon csomópontjai, amelyek a grammatikai morfológiát dominálják. Hagiwara
tesztjeiben az agrammatikus afáziások TenseP-én vagy AgrP-én belül ( e csomópontok alatt) tudtak
konvergens reprezentációt létrehozni, és nem a CP -én belül. Innen a megértési nehézségek
(aszintaktikus értés) és a produkciós korlátozottságok. Az AGR csomópont az igei egyeztetés
morfológiáját dominálja, a C csomópont pedig a mondatot bevezető kötőszókat. Minden ilyen
csomópont a szintaktikai fastruktúra egy bizonyos szintjén helyezkedik el. A betegek nyelvi
korlátozottsága ezért úgy jellemezhető, mint a szintaktikai fastruktúra egy bizonyos szintjén fellépő
zavar. Azok a grammatikai morfémák, amelyek a kritikus szint alatti csomópontoktól függnek, a
betegnél megtartottak, míg azokat a morfémák, melyek a kritikus szintnél magasabb csomóponttól
függnek, már korlátozottak. Hagiwara kimutatta, hogy a japán agrammatikus betegek kevés hibát
vétenek a tagadó kifejezésekkel, az idői egyeztetéssel, és a nem alanyesetű főnevekkel. Ezek a NegPtől, illetve a TensP-től, illetve az AgrOP csomópontoktól függnek, melyek a szintaktika fastruktúra
viszonylag alacsonyabb szintjein vannak a japánban. Ugyanakkor sok hibát vétettek a
mondatbevezető kötőszavakkal, a kérdő kifejezésekkel, a kérdő partikulával és az alanyi DP-hez
kapcsolt nominatívuszt jelölő, topikot jelölő, illetve genitívuszt jelölő partikulákkal kapcsolatban.
Ezek a CP, az AgrSP, illetve a TopicP csomópontoktól függnek, melyek magasabban vannak a
szerkezeti hierarchiában, mint a Tense csomópont.
5.3. Friedmann és Grodzinsky (1997) olyan héberül beszélő afáziás betegről számol be, aki a
mondatismétlési és mondat-befejezési feladatok során az alany-ige egyeztetés morfológiájában szinte
egyáltalán nem követ el hibákat, viszont az ige idői egyeztetése, inflexiója során sok hibát követ el
(például nem kompatibilis a mondatban használt időhatározó jelentése és az ige idői inflexiója).
Friedmann és Grodzinsky magyarázata az, hogy a héber nyelv mondattana esetében a szintaktikai
fában az AGR csomópontok alacsonyabb szinten helyezkednek el, (ezek a betegük számára
elérhetőeknek bizonyultak) viszont az ige idői morfológiáját domináló Tense csomópont a szintaktikai
fastruktúrában magasabb szinten van (és az adott beteg számára nem elérhető). Ennek a betegnek a
deficitje tehát a Tense szinten van. Friedmann-Grodzinsky feltételezi: ha valamely betegnek egy adott
csomópont tekintetében károsodása van, akkor a szintaktikai fastruktúrában az ennél magasabb
szintek csomópontjai elérhetetlenek lesznek. A fastruktúra megépítésekor ugyanis az egyeztetésekhez
szükséges az ige felfelé mozgatása bizonyos csomópontokon keresztül, de az ige nem mehet át a
“károsodott“ csomópontokon.
Ennek a heurisztikusan nagyon értékes elképzelésnek hiányossága az, hogy nem ad számot
arról, hogy a baloldali frontális lebeny sérülése következtében fellépő afáziák nyelvi tünetei nagyon
sokfélék, változatosak lehetnek, a súlyosabb korlátozottságtól az enyhébb korlátozottságokig terjedő
skálát alkotva.
Egyeztetés agrammatikus afáziában
54
6. Mondatismétlési tesztek magyar anyanyelvű afáziásokkal
6.1. A tesztek során élőben vagy hangszalagról lejátszott mondatok változatlan formában
történő megismétlésére kérjük a vizsgálati személyeket. Az adatokból jól látható, hogy a teszt
teljesítése megkívánja a célmondat tényleges feldolgozását, a jelentésreprezentáció tárolását, majd
annak olyan szintaktikai szerkezetben és fonológiai formában történő produkálását, mely hasonlít a
célmondatéhoz vagy egyezik vele. A vizsgálati személyek mindegyike válaszaiban olyan stratégiát
követett, melyben megkísérelte egymásra leképezni a hallott mondathoz általa rendelt szemantikai
interpretációt, a hallott mondat eredeti szintaktikai szerkezetét, és az eredetivel lehetőleg azonos
fonológiai formát.
A következőkben három, különböző vizsgálati személlyel végzett tesztekből mutatunk be
példákat.
6.1.2. I. afáziás vizsgálati személy: S., 37 éves férfi, traumás eredetű, baloldali frontális és
temporális területű bevérzés.
A vizsgálati személy válaszai korlátozottságot mutatnak mind az esetragok, mind az idői
inflexió mind pedig a személy-szám egyeztető inflexiók produkálásában. A beteg 73 válaszából csak
2 esetben produkált olyan mondatot, melyben hibás az esetrag de ugyanakkor hibátlan az idői
valamint a személy/szám egyeztetés. Példák a fennmaradó 71-ból:
(V: vizsgáló, bold –dal szedett: az afáziás válasza)
V: Megjavítottam az autómat, de újra elromlott.
Autó…. autó ..szokott csinálni, meg szokott csinálni.. hogy menni..
V: Vendégeket vártam, de nem jöttek.
Nem bírok várni őket, hanem a nőt mond, hogy maradok. Hát mondok, akkor maradok,
maradok, mert annyira mondta szépen én. (tillik nekem)
Spontán beszéd (részlet)
V: Mikor engedik haza?
Nem szokott tudni.. Azér otthon azér csak jobb lesz azér, mer nem nem egymaga leszek, És
akkor legalább ottan míg beszél még jobban leszek mint én igaz?… (tillik: jobban leszek mint
ahogy most vagyok én
6.1.3. II. afáziás vizsgálati személy: I.N. 57 éves nő, bal frontális-parietális hypodensitas
A beteg válaszaiban grammatikus esetragokat használ, grammatikus az idői inflexió, viszont
hibás a személy/szám egyeztető inflexió az igén. A beteg kompenzációs stratégiát is alkalmaz: az
alanyt vagy elhagyja, vagy a szintaktikai fában „lejjebb“ helyezi, az ige utánra:
--az alany elhagyása
V: Az autó "elsuhant a ház előtt.
Elsuhant a ház előtt.
V: Az igazgató "elküldte vidékre a sofőrt.
Sofőrt, sofőrt akkor "vidékre küldte.
Egyeztetés agrammatikus afáziában
55
-- az alanyt az ige alá
V:A lámpák késő estig világítottak.
Késő estig világított… az.. a lámpák.
V: A biciklista semmiről sem tehet.
Sem semmiről sem evett a biciklista.
6.1.4. A prozódiai korlátozottság és szintaktikai korlátozottság összekapcsolódva, együttesen
jelenik meg a II. afáziás vizsgálati személy válaszaiban. Az agrammatikus afáziásoknál gyakran mind
a beszédprodukció, mind pedig a beszédértés egyidejűleg valamilyen módon károsodik. Egyrészt
korlátozott és jellegzetesen monoton intonációt produkálnak, másrészt a prozódiailag jelölt
szintaktikai kategóriák feldolgozása hibás lehet. Nagel és Shapiro (1994) egészséges személyek és
afáziások rögzített beszédét analizálva azt találták, hogy az afáziások nem produkálnak nyelvtani
szempontból normál dallamot a mondatokon belüli hosszabb tartományokban, és nem produkálják a
szintaktikai pozíciókhoz társítandó kontrasztív nyomatékokat, és azt megelőző dallamot és időtartam
nyújtást.
A szerkezeti pozíciókhoz társított prozódia közreműködik annak meghatározásában, miképpen
kapcsolódnak össze lexikai és szerkezeti kategóriák a mondatstruktúrában. A prozódiai információ
valószínűleg nagyon korán, a kezdeti elemzés során felhasználásra kerül annak érdekében, hogy
segítsen a mondatfeldolgozás során felmerülő csatolási kérdések eldöntésében, hogy a szerkezet
feldolgozása során fellépő bizonytalanságokat megoldja. Nagel és Shapiro szerint az afáziások nem
képesek a megfelelő időben felhasználni a prozódiai információt a hallott mondatszerkezet on-line
feldolgozásában.
A mi adataink is alátámasztják Nagel és Shapiro eredményeit. A magyar anyanyelvű afáziások
a beszédprodukcióban kikerülik az erős nyomatékot hordozó fókuszt, a tagadott fókuszt és a
kvantoros kifejezéseket, és helyettük a szintaktikai fában "lejjebb" helyezett, nem nyomatékos
változókat produkálnak. Mivel csak késve képesek vagy egyáltalán nem képesek időben felhasználni
a prozódiai információt a szerkezeti viszonylatok meghatározásában, ezért például a fókuszra olykor
hibás, töredékes szerkezet feldolgozásokat végeznek. A mondat megértésében pedig fennakadásokat,
nehézségeket okozhat a fókusz interpretációja és a kvantoros kifejezés értelmezése: Ezt tapasztaltuk a
II.-vel jelölt vizsgálati személy esetében is. Néhány példa:
elhagyás
Tagadó operátor törlése:
V: A vezetőt NEM idegesítette a zaj.
A vezetőt idegesítette a zaj.
„ lejjebb“ helyezés:
-- igetagadás képzése:
V: A moziban “nem Péterrel beszélgettem.
Péterrel ö…. Péterrel “nem beszélgettem.
Egyeztetés agrammatikus afáziában
56
-- a kérdőszó helyett hasonló változó lejjebb a szintaktikai fában:
V: Kit láttál az utcán?
..Hogy az utcán ment valaki, az utcán.. az utcán mentél.
V: Mikor érkezett meg a pécsi gyors?
Igen, a.. hogy a pécsi gyors, az.. valamikor megjött.
A fókusznyomatékot nem produkálja:
V: MARI hajtotta biciklit gyorsan és PÉTER [
].
Tehát .. mindenki hajtott gyorsan , Mari is meg Péter is.
6.1.5. III. afáziás vizsgálati személy: Sz. V. 42 éves férfi, bal oldali frontális hypodensitas.
Broca afáziás.
Sz. V. mondatismétlési tesztben nyújtott teljesítménye erős függést mutatott a célmondat
tulajdonságaitól. Az alábbiakban – az eredetileg random módon tesztelt mondatokat --- a célmondat
tulajdonságai szerint rendeztük, és bemutatjuk a vizsgált személy teljesítményét a célmondatok egyes
csoportjaira. Három csoportot találtunk.
I. oszlop: a neutrális intonációjú célmondatok. A beteg válaszaiban nincsen raghiba, de vannak
hibás egyeztető inflexiók az igén: alany-ige, tárgy-ige egyeztetésbeli hibák.
II. oszlop: olyan fókuszos célmondatok, melyekben az alany a topik pozícióban, a
tárgy/határozó pedig a fókusz pozícióban van. A beteg válaszaiban hibás ragokat és hibás egyeztető
inflexiókat produkál, de mindig sikeresen aktiválja magát az igét.
III. oszlop: olyan fókuszos célmondatok, melyekben a tárgy/határozó a topik pozícióban, az
alany pedig a fókusz pozícióban van. A beteg válaszaiban hasonló mennyiségű a hibás ragot és hibás
egyeztető inflexiót produkál, mint a II. oszlop mondatainál, viszont most a 12 mondatból 3 mondatban
nem képes aktiválni magát az igét.
Ezek szerint a beteg számára az I. oszlop célmondatai a viszonylag könnyűek, a II. oszlop
célmondatai nehezebbek, mint az I. oszlopé, és a III. oszlop célmondatai pedig nehezebbek, mint a II.
oszlopé:
I.
II.
III.
V: Te felszedted a szemetet.
V: Te a SZEMETET szedted fel.
V: A szemetet TE szedted fel.
Te fele le felszed.ted a szemetet.
Én sz.. szeme.tet szedem
* Szemetet szemetek Te
V: Ő kitisztítja a cipőt.
V: Ő a CIPŐT tisztítja ki.
V: A cipőt Ő tisztítja ki.
*Ő ktit ki tiszti.. tom a cipőt.
*Ő cipőt tisztítja ki.
*Cipőt ő tisztítja.
V: Mi szántjuk a földet.
V: Mi a FÖLDET szántjuk fel.
V: A földet MI szántjuk fel.
Mi szánt.. juk a föl..det.
Mi a földet szánt…juk ki.
*Fö… nem tudom
Egyeztetés agrammatikus afáziában
57
V: Ti gurítjátok a követ.
V: Ti a KÖVET gurítjátok el.
V: A követ TI gurítjátok el
*Én gurigá..gurit..ják a krövek
*Mi a követ gurit..já..tok ki
*Követ ki ti követ ki
V: Ők megterítik az asztalt.
V: Ők az ASZTALT terítik meg.
V: Az asztalt ŐK terítik meg.
*Ő megterít a.. az asztalt.
*Ő a.. asztal terít..t..t..
*Asz..talna asztal Ők terít meg.
V: Én megkapom a szemüveget.
V: Én a SZEMÜVEGET kapom meg.
V:A szemüveget ÉN kapom meg.
Én meg ka..pom a szemüve..get
*Én a sze.mü.ve.get rakod ki.
*Szemüveg én kapom meg.
V: Te kimosod a zoknit.
V: Te a ZOKNIT mosod ki.
V: A zoknit TE mosod ki.
*Én én mos..sok a zoknimat.
Én a zoknit mosom ki.
*Zokni Te mosol ki.
V: Ő kifizeti a számlát.
V: Ő a SZÁMLÁT fizeti ki.
V: A számlát Ő fizeti ki.
Ő ki.fi.ze.ti az számlát.
Ő a szám.lát fi.ze.ti ki
Számj számláját ő fizeti ki.
V: Mi ki dobjuk a virágot.
V: Mi a VIRÁGOT dobjuk k i.
V: A virágot MI dobjuk ki.
Én ki.do.bom. a ssz
Mi a…..
Virág…
Mi ki.dob.juk a vi.rá.got
*Mi a virág dobjuk ki.
*Virágot mi dobjuk ki.
V: Ti várjátok az ünnepeket.
V: Ti az ÜNNEPEKET várjátok.
V: Az ünnepeket TI várjátok.
Ti vár.játok a ünnepekt
Ti a ünnepeket…..
*Ünnepeket ti várjuk.
Mi a ünnepeket
*Ünnepek várjá..taok
V: Én felszálltam a buszra.
V: Én a BUSZRA szálltam fel.
V: A buszra ÉN szálltam fel.
Én fel.szál..lok a busz.ra.
Busz..ra száll fel.
*Buszba fel…szed
V: Ők kinéznek az ablakon.
V: Ők az ABLAKON néznek ki.
V: Az ablakon ŐK néznek ki.
Ő ki.néz az ablakon.
*Ő az ablakot nézek ki.
*Ablakot ő nézik ki.
ÖSSZES HIBA: 6/12
ÖSSZES HIBA: 10/12
ÖSSZES HIBA: 17/12
2 Agr + 2 Def
3 Agr + 3 Def
3 Agr + 3 Def
+ 3 igehiány
1 tárgyraghiány
3 tárgyraghiány
3 tárgyraghiány
3 hibás esetrag
1 időváltás
1 időváltás
2 időváltás
A beteg tesztválaszaiban a topik/alany és a fókusz/alany az igével való személy/szám
egyeztetés szempontjából nem mutat különbséget. Ugyanakkor az ige aktiválása szempontjából
sokkal könnyebb konfiguráció a topik/alany + fókusz/tárgy, mint a topik/tárgy + fókusz/alany. Az
esetragok kiosztása tekintetében nincs nagy különbség a fókusz/bővítmény és a topik/bővítmény
között. A fókuszos mondatoktól eltérően, a neutrális célmondatokra adott válaszaiban a beteg nem
követett el raghibát, és mindig produkálta az igét.
58
Egyeztetés agrammatikus afáziában
7. 1. A tesztjeinkben nyert adatok értelmezésére a következőket javasoljuk.
a) A bal frontális lebeny és a szomszédos agyterületek sérülése eredményeként kialakuló
agrammatikus afáziák szerkezetépítő és morfológiai korlátozottságokat egyaránt mutatnak.
b) Elfogadjuk a tükörelvet (Baker 1985). Az elv azt mondja ki, hogy a nyelvekben
(különösen a gazdag inflexiós rendszerrrel rendelkezőkben) a morfológiai deriváció ( a szó
szoros értelemében) közvetlenül tükrözi a szintaktikai derivációt. A morfémák és a morfok
abban sorrendben csatlakoznak egymáshoz, ahogyan a szintaktikai szerkezetépítő lépések,
(kombinációk, csatolások) végbemennek.
c) A tükörelvet alapvetőnek tekintjük, és feltételezzük, hogy a sérült nyelvtani
reprezentációkban -- amennyiben a sérülés mértéke nem katasztrofikus – a tükörelv
megmarad, és a korlátozottság manifesztálódásában szerepet játszik. Ezért az
agrammatikus afáziások
morfológiai korlátozottsága valójában szintaktikai
korlátozottságot tükröz.
7.2. A fenti keretben a vizsgálati személyektől a mondatismétlési tesztekben nyert
eredményeink értelmezhetők a szintaktikai fa metszésének hipotézise keretében.
A szintaktikai fa metszésének hipotézise a következőket jelenti. Az agrammatikus afáziában:
a) A szintaktikai reprezentációban vagy a Case vagy a Tense vagy az Agr csomópontja a
fastruktúrában alulspecifikált lehet.
b) Egy alulspecifikált csomópont nem terjeszthető ki (nem projektálható) tovább.
c) Mennél lejjebb van a sérült, alulspecifikált csomópont, annál több a felette levő és már el
NEM érhető funkcionális csomópontok száma, így annál súlyosabb a korlátozottság.
Vagyis az “enyhe“ agrammatizmus csak a magas csomópontokat érinti.
Emlékeztetünk arra, hogy a szerkezetépítő műveleteket a lexikai egységek jegyei orientálják,
ezeknek az összeillését, egyezését ellenőrzik, és csakis interpretálható jegyeket hagynak meg az ép
(konvergens) a reprezentációban. A már levezetett szerkezetbe történik a fonológiai alak beillesztése.
Valamely funkcionális fej alulspecifikáltsága morfoszintaktikai hibákban mutatkozhat meg. Ha
például alulspecifikált a Tense fej, ahova az ige mozog, akkor ez azzal járhat, hogy a komputációs
rendszer nem képes a „szándékolt“ idő (a Tense fejben specifikált jegy) és az ige idői inflexiójának az
eltérését észlelni. Így inflexió hibák fordulhatnak elő: a komputáció nem észleli a rossz inflexiót vagy
a hiányzó inflexiót.
7.3. Három, eltérő súlyosságú agrammatikus afázia következményeit mutatták a
teszteredményeink. (A „súlyosság“ kifejezéssel nem általában a beteg klinikai állapotára utalunk,
hanem az agrammatikussság mértékére a beteg nyelvi produkciójában). Kiindulva a magyar
mondatszerkezet leírására, Szabolcsi (1997) által a Minimalista Program keretében javasolt
struktúrából, a vizsgálati személyek szintaktikai-morfológiai korlátozottságait következőképpen
fejezhetjük ki a szintaktikai fa metszéseivel: 1. ÁBRA1
1
Az 1. ábra csakis a jelen tárgyalás szempontjából releváns csomópontokat tartalmazza. Az AgrSP és az AgrOP
csomópontoknak a tükörelv által empirikusan csak részben motivált sorrendjének problémájára, valamint az idői
és a mód funkcionális projekciói sorrendjének a problémájára nem térünk ki. Jelenleg nincsen olyan afáziás
adatunk, melyek ezekkel a csomópontokkal kapcsolatosan a szintaktikai fa metszésére utalnának, miszerint a
AgrSP és az AgrOP csomópontok közül egyik csomópont és a kapcsolódó morfoszintaktikai egyeztetés ép
lenne, de a másik csomópont és a hozzá kapcsolódó morfoszintaktikai egyeztetés egyidejűleg súlyosan
korlátozott lenne. Nincsen erre utaló adatunk a Mod és Finit csomópontokra és a kapcsolódó moroszintaktikai
egyeztetésre sem. Ezeket a kérdéseket tehát a további kutatás számára nyitva kell hagyjuk.
Egyeztetés agrammatikus afáziában
59
A SZINTAKTIKAI FA METSZÉSE
CP
ToP
DPi
Top`
QP
Q’
NegP
FP
DPj
F`
III.
NegP
AgrSP
AgrS’
_i,_j
II.
TenseP
[Múlt]
Tense`
CaseP
Case`
I.
DPk
VP
V
_i
_j
_k
I. afáziás:
A CaseP alatt, és a VP felett van a fa metszése: az esetragok, az idői inflexió és az egyeztetőinflexiók együttes korlátozódása.
Egyeztetés agrammatikus afáziában
60
II. afáziás:
Az TenseP-nél van a fa metszése, (felette korlátozott): ekkor grammatikusak az esetragok,
grammatikus az idői inflexió, de hibás a személy/szám egyeztető inflexió az igén. Kompenzáció:
alany elhagyása vagy lejjebb helyezése a szintaktikai fában az ige utánra.
III. afáziás:
Az AgrP-nél van a fa metszése: ekkor neutrális mondatban grammatikusak az esetragok,
grammatikus az idői inflexió és a személy/szám egyeztető-inflexió. Viszont a fókuszos mondatokban
esetrag, idői inflexió és személy/szám egyeztetés hibák vannak, nagyjából egyforma mennyiségben a
fókuszra és topikra.
A topik/tárgy és fókusz/alany konfiguráció mellett igeaktiválási zavar keletkezik.
7.4. Az AgrP-nél metszést tartalmazó hibás reprezentációból adódhatnak a csakis fókuszos
mondatokban keletkező nyelvtani hibák. Ezek magyarázatára javasoljuk az újraelemzésre vonatkozó
hipotézisünket.
A hipotézist a következő jelenség magyarázatára javasoljuk. Ha feltételezzük, hogy a kritikus
morfoszintaktikai egyeztetések, ellenőrzések az AgrSP-ig bezárólag már megtörténtek, akkor
károsodhatnak-e újra az egyeztető végződések, vagy a ragok például a topik vagy a fókusz
pozícióban? A lehetőség egyáltalán nem abszurd. Először is erre utal a tesztelt személy
válaszmondatai grammatikusságának az erős függése attól, hogy neutrális vagy fókuszos
célmondatokra adta-e őket.
Figyelembe kell venni továbbá azt, hogy a fókusz és topik feldolgozása csakis a prozódiai
információk hatékony felhasználásával történhet (A magyarban nincsen topikot vagy fókuszt jelölő
végződés). Említettük, hogy Nagel és Shapiro adatai szerint az agrammatikus afáziások nem képesek
a megfelelő időben felhasználni a prozódiai információt a szerkezeti viszonylatok meghatározásában.
Láttuk, hogy a magyar anyanyelvű afáziásoknak számára nehézségeket okozhat a nyomatékos fókusz
feldolgozása. Ha a prozódiai információ nem használható fel időben a szerkezeti viszonylatok
meghatározására, akkor olyan helyezet áll elő, melyben a rendelkezésre álló reprezentáció
strukturálisan homályos, nem egyértelmű.
Hahne és Friederici (1999) kiváltott agypotenciál vizsgálatai2 azt bizonyították, hogy ilyen
esetekben a szerkezet-feldolgozást javító, újraelemző műveletek aktiválódnak. Eközben pedig olyan
(újabb) grammatikai hibák keletkezhetnek, melyeket a korábbi reprezentáció nem tartalmazott. A
kiváltott agypotenciál vizsgálatok erre neurológiai magyarázatot adnak. Az újraelemző műveletek
ugyanis nem azokon az agykérgi területeken mennek végbe, ahol az elsődleges feldolgozás.
A mondatfeldolgozással kapcsolatosan négyféle agyi elektromos potenciál adatot találtak,
melyek három eltérő idő-ablakkal jellemezhetők. Először a nyelvi inger kezdete után 100-200ms
körül, a bal oldali agyfélteke elülső részére kiterjedő néhány millivolt negatív töltésű elektromos
potenciált észleltek, amely a lokális frázis-szerkezeti információk feldolgozásával volt kapcsolatos.
Másodszor kétféle, egyaránt 400ms körüli, ám eltérő eloszlású negatív bioelektromos potenciált
mértek: (a) a nyelvi inger kezdete után 400ms körül, egy bal oldali elülső kérgi területű negatív
potenciált, amely az igei alkategorizációs információk és az egyeztető morfológia feldolgozásával
hozható kapcsolatba, és (b) a lexikai-szemantikai feldolgozás által kiváltott negatív potenciált, mértek,
mely az agykéreg baloldali hátulsó területein volt észlelhető széles kiterjedésben.
2
A kiváltott agypotenciál vizsgálatok. során mérik az agy bioelektromos aktivitását. A betegek fejbőrére --ismert meghatározott beállítási poziciók szerint --- elektródákat tesznek, és mérik a különböző nyelvi ingerek
által kiváltott változásokat az agy bioelektromos aktivitásában, amely a nyelvi inger megjelenésével bizonyos
idői kapcsolatban áll. A világ különböző laboratóriumaiban lényegében egybehangzó eredményekhez jutottak.
Eszerint a mondatfeldolgozási műveleteknek három fázisa határolható el; ezek közül kettő főként szintaktikai
természetű.
Egyeztetés agrammatikus afáziában
61
Ha a szerkezeti viszonylatok nem egyértelműek, például a prozódiai és a szerkezeti
információk egymásra leképezése nem volt sikeres, akkor újraelemzés válik szükségessé. Friederici
meggyőző kísérleti és empirikus adatokat szolgáltat arra, hogy az ilyen újraelemzés az előzőktől
eltérő agykérgi területekhez kötött és eltérő jellegű: a nyelvi inger kezdete után 600-800 ms között,
néhány millivoltos pozitív töltésű bioelektromos aktivitást mutatnak a középső és a hátulsó agykérgi
területek, viszonylag széles kiterjedésben mindkét agyféltekén.
Az elsődleges feldolgozás és az újraelemző műveletek közti különbségek megteremtik a
lehetőséget az újabb nyelvtani hibák keletkezésére: egy korábbi, korrekt morfoszintaktikai és
szerkezeti reprezentációt, az őt strukturálisan, kiterjeszteni, egyértelműsíteni kívánó újrafeldolgozás
elronthat.
Ez a magyarázat összefér mind I.N.-nek
teszteredményeivel.
mind pedig Sz. V. –nek a bemutatott
8. Összefoglalás
Az agrammatikus afáziában tapasztalt egyeztetési, morfoszintaktikai hibák értelmezésére, a
korlátozottságok enyhébb és súlyosabb fokozatainak leírására a szintaktika fa - metszés hipotézisének
egy kiterjesztett változatát javasoltuk. Eszerint:
1. A szintaktikai csomópontok sorrendjének megvan a maga komputációs költsége. Az
alacsonyabb csomópontok elérése kevésbé költséges, mint a magasabban levő csomópontoké.
2. A tükörelvet úgy egészíthetjük ki, hogy agrammatikus afáziában a
morfológiai korlátozottság szerkezetépítõ korlátozottságot tükröz, a szerkezeti pozíciókhoz
rendelt prozódia feldolgozásának korlátozottsága pedig morfológiai hibákhoz vezet.
3. A szerkezeti (frazális) kategóriáknak mennél kisebb számú kombinációja megy végbe, annál
gazdaságosabb a szerkezet és annál könnyebben elérhető az agrammatikus afáziások számára.
4. Az agrammatikus afáziások képesek produkálni és megérteni a sérült nyelvtan számára
gazdaságos szerkezetet. Így: mennél alacsonyabban van egy funkcionális fej és a projekciója a
mondatszerkezeti hierarchiában, annál könnyebben elérhető az agrammatikus afáziás számára.
5. A szintaktikai fastruktúra azon tartományában, amely a diskurzus konfigurációs jellegű, és a
nyelvtani alapú nyomaték és intonáció viszonyoktól függő, az agrammatikus afázia prozódiai
korlátozottsága olyan újraelemző műveleteket válthat ki, melyek az alacsonyabb tartományban
felépített, korrekt morfoszintaktikai reprezentációkat is elronthatnak.
Irodalom
Baker, M. 1985. The Mirror Principle and Morphosyntactic Explanation. Linguistic Inquiry 16, 373-415.
Chomsky, N 1995. The Minimalist program, Cambridge, Mass.: MIT Press.
Chomsky, N. 1999. Linguistics and Brain science, Working Papers in LingisticsVol 8, University of Maryland
104-117.
Friederici, A. D,.1995. The Time Course of Syntactic Activation during Language Processing: A Model Based
on Neuropsychological and Neurophysiological Data, Brain and Language, Vol 50, 259-281.
Hahne A. - Friederici A.D. 1999. Electrophysiological Evidence for Two Steps in Syntactic Analysis: Early
Automatic and Late Controll Process. Journal of Cognitive Neuroscience 11:2, 194-205.
Friedmann, N. -Grodzinsky, Y. 1997. Tense and Agreement in Agrammatic Production: Pruning the Syntactic
Tree, Brain and Language, 56, 397-425.
Hagiwara, H. 1995. The breakdown of functional categories and the economy of derivation. Brain and
Language, 50. 92-116.
Kolk, H. 1995. A Time-Based Approach to Agrammatic Production. Brain and Language 50. 282-303.
62
Egyeztetés agrammatikus afáziában
Linebarger, M. C. 1995. Agrammatism as Evidence about Grammar. Brain and Language, 50. 52-91.
Nagel, N., & Shapiro, L. P. 1994. Prosody and the processing of filler-gap sentences. Journal of
Psycholinguistic Research. [Special Issue]
Pléh, Cs. 1998 A magyar morfológia pszicholingvisztikai aspektusai, sajtó alatt: Strukturális magyar nyelvtan
3. Morfológia (szerk.: Kiefer F.) Akadémiai kiadó, 2000.
Swinney D., - Zurif E., 1995. Syntactic Processing in Aphasia, Brain and Language, Vol 50, 225-239.
Szabolcsi, A. Strategies of Scope Taking, in: A. szablcsi, Ed: Way of Scope Taking, 109-155. Dordrecht,
Kluwer.
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
63
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
Bánréti Zoltán
MTA Nyelvtudományi Intézete
1.1. A neurolingvisztika alapvető kérdésfeltevései közé tartoznak a következők. Milyen
szerkezetű a nyelvtan reprezentációja az agyban? Milyen kapcsolat van a nyelvtan moduljai (lexikon,
szintaxis, fonológia) és az agy szerkezeti, térbeli felépítése között? Milyen sajátosságokkal bírnak
azok a
neuropszichológiai mechanizmusok, mentális programok, amelyek a nyelvtani
reprezentációkat aktiválják a beszédprodukció és beszédértés folyamatai számára?
Másik oldalról, a nyelvtanelméletek számára pedig ma már lehetséges kutatási cél olyan
nyelvtan modellek felépítése, amelyek konstrukciója a mentális nyelvtant, vagyis a beszélők agyában
reprezentált nyelvtan felépítését, tulajdonságait fejezi ki. A kurrens neurolingvisztikai elméletek
többsége (Kolk 1995, Friedmann-Grodzinsky 1997, Linebarger 1995, Zurif 1994) úgy tekinti a
mentális grammatikát, mint számos, egymással összeköttetésben álló agyi terület által tárolt nyelvtani
reprezentációk rendszerét, ahol az egyes agyi területek strukturálisan is determináltak valamely
specifikus nyelvtani reprezentációtípus tárolására illetve közvetítésére.
1.2. Az empirikus adatok egy fontos osztálya abból adódik, hogy az agy különböző területein
jelentkező lokalizált sérülések -- a gondolkodási és más kognitív képességek épen maradása mellett -specifikus nyelvi károsodásokhoz vezethetnek, a nyelvi képesség valamely részrendszere,
részfunkciója korlátozódásához, a többi épen maradása mellett. Napjainkra már nagyon sok adat
gyűlt össze arról, hogy jobbkezes embereknél a homloklebeny baloldalán, a homloklebeny harmadik
agytekervényének hátsó területeit és a precentrális agytekervény ezzel szomszédos alsó területét ért
lokális károsodás a beszéd produkciós képességek korlátozódását eredményezi. Az említett régiót
Broca területnek nevezik, a nyelvi zavart pedig Broca afáziának: 1.ábra
A Broca-afáziás betegeket elsősorban a beszédprodukció zavara, mondatfragmentumokat
produkáló, töredezett, lassú beszéd, szótalálási nehézségek, gyakran elhagyott funkciószavak,
hiányzó toldalékok és ragok, monoton intonáció, artikulációs hibák jellemzik. Ugyanakkor viszonylag
ép a beszéd megértés. Az alábbiakban részletet mutatunk be egy afáziás vizsgálati személy
lejegyezett spontán beszédéből. A vizsgálati személy: 37 éves férfi, a lézió helye: bal frontális. A
tipikus nyelvtani hibákat aláhúzással jelöltük meg.
64
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
Vizsgáló: Mi történt magával, hogy került a kórházba?
… egyszercsak ujjamban néz, így mi fordult meg.. nem tudo, hogy mi van? És akko így néz..
néz jó..jó.. nem bír mozdítani, nem bírtam egysz ..akkor megmozdulok a kézzel.. azt vár, vár a nem
mozdulni a kezem meg a testem. És nem tud, hogy mi van! Egy olyan fél perc.. vegy nem.. egy
órára,..hogy hát mondom, várni nem kellni, kellni.. egy orvos. És meg...beszéltük, hogy .....hát ugye ô
is, meg a másik is...mer ..mintha máskor szóval mindig a fár.. ház lejjebb, ott van a kórház... Úgy
van, ismerek pár orvos ..., hogy mindig, mindig, oda járt...., megisment, elismertem de mondom várj.
De nem mehet így, . vagy nem! Nem jól mondom, hanem ....szóval átmentünk az izébe,....tehát így a..
olyan kényelemtől már megmondani..na. Egy bizonyos rész, egy olyan.... fél-háromnegyed tíz, tíz óra
fele amikor már éreztem..na. És aztán el is vertek, el is vertek (a szándékolt szó: el is vittek)
(Mészáros Éva lejegyzése)
1.3. A Broca afázia lényegét a mondatszerkezet tervezésében és beszédbeli produkálásában
mutatkozó zavarok adják. Viszont a Wernicke-terület sérülése, vagyis a baloldali halántéklebeny első
agytekervényének hátulsó területét és a szomszédos kérgi területeket ért lokális sérülés más típusú
nyelvi károsodást eredményez. A Wernicke afáziás betegek folyamatos viszonylag gyors beszédet
produkálnak, amely gyakran értelmetlen, vagy halandzsaszerű szavakból is áll, zavarokat mutatnak a
szavak hangalakjának és morfológiai szerkezetének a produkálásában, téves szóhelyettesítéseket,
fonemikus, morfológiai és szemantikai jellegű, hibás felcseréléseket produkálnak. Mindehhez társul a
hangzó beszéd megértésének erős zavara. A Wernicke afázia legfontosabb jegye a lexikai egységek
reprezentációjának, mindenekelőtt a hangformájának korlátozott elérhetősége vagy részleges
törlődése adja.
A Wernicke afáziások beszédét jellemzi az ún. zsargon-afázia. Például egy Wernicke-afáziás
beteg fluens, de nehezen értelmezhető megnyilatkozásai. Lejegyzett spontán beszéd. A vizsgálati
személy: 62 éves férfi, Wernicke afáziás.
A Zsigulikat, a legislegelső országokat mikor megvették a magyaroknak a magyaroktól, az úgy
volt csinálva először, hogy első godás volt. De ez a kis krebekó Trabant az tizen volt vagy tizenegy
rágerős, ahová a frajók a rág után jutottak. Eleltem mindent, vároztam országoltam, moszkat, kutást.
Mentem hazafelé, azt leesett a lábam. Innen tarboltam le a lábam. De most nem vagyok jól, mert nem
tudom észben tartani az eszemből az eszemnek, hogy észben tartsam egészen a szemembe, ami köztünk
van.
1.4. Látható a bemutatott adatokban az afáziás nyelvi korlátozottság egy fontos mozzanata, a
nyílt lexikai osztályú és a zárt lexikai osztályú egységek aktivációjának a szétválása.
A világ objektumait jelölő tartalmas szavak bővíthető osztálya: a nyílt szóosztály. Míg a zárt
szóosztály, a nyelvtani viszonyokat jelölő egységeket, formatívumokat tartalmazza, például az
esetragok, toldalékok, prepozíciók, determinánsok, névmások, kötőszók, stb. A zárt szóosztályú
morfémák egy struktúraelemző és struktúraépítő rendszer elemeit alkotják. A mondatfeldolgozás
során például jelzik a főnévi szerkezet kezdetét/végét, a fő- és alárendelt mondatok
megkülönböztetését és határaikat, stb. A gyors előhívásuk lehetővé teszi a lokális szintaktikai
szerkezetről alkotott azonnali döntéseket.
A nyílt és a zárt lexikai osztályok aktivációjának valamilyen szétválása sokféle nyelvi
deficitnek összetevője lehet. Például egy szerzett dyszlexiában (olvasászavarban) szenvedő beteg
olyan hibákat vétett a tartalmas szavak olvasásakor, melyek egyrészt kifejezik a vizsgálati személy
öntudatlan belső tudását a nyílt és zárt szóosztályok különbségéről, másrészt megmutatják azt is, hogy
nem tudja zavartalanul használni a lexikonnak azon részét, amely a nyelvtani morfémákat
tartalmazza. A vizsgálati személyt arra kérték, hogy olvassa el hangosan a kártyákon egyenként
bemutatott szavakat. A beteg a kiolvasott tartalmas szavakról gyakran elhagyta a nyelvtani viszonyt
jelölő morfémákat, amikor pedig a nyelvtani toldalékok kiolvasására kérték, a toldalék helyett
gyakran tartalmas szót olvasott. Néhány példa:
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
65
Mély diszlexiás paralexiák, zárt szósztály elérési zavarokkal, elhagyásokkal:
Nyílt szóosztály olvasása:
Zárt szóosztály olvasása:
CÉLSZÓ KIOLVASOTT SZÓ
CÉLSZÓ
KIOLVASOTT SZÓ
kocsiért >>>> kocsi
-on >>>> otthon
tanárként >>>> orvos
-ba >>>> arra
elrepülök >>>> repülő
-ben >>>> bent
fiaid >>>> fiú
-nál >>>> már
nyaranta >>>> nyár
-tól >>>> ól
ötször >>>> öt
-hoz >>>> ház
tanult >>>>tanuló
-hez >>>> kihez
átszalad >>>> szakadék
-kor >>>> kör
balul>>>> bal váll
-ul>>>>tulajdon
(Kiss Katalin lejegyzése.)
1.5. Feltételezhető, hogy a nyílt és zárt szóosztálybeli egységek két különböző szublexikonban
tárolódnak. Ezek külön pályákon érhetők el, azonban egymáshoz rendezve, időben szinkronizáltan
kell őket aktiválni a mondatprodukció és feldolgozás során. Sok afáziás esetet jellemez a zárt
szóosztályú elemek elérhetőségének a lelassulása. A spontán beszéd töredezettsége és
agrammatikussága összefüggésbe hozható ilyen korlátozottsággal. A baloldali elülső kérgi területek és
a velük szomszédos kérgi területek felelősek azokért a műveletekért, amelyek a beérkező nyelvi input
egységeinek a szintaktikai szerkezetbe történő gyors és automatikus szerkesztését végzik el. Az
afáziában azonban a zárt és a nyílt szóosztály aktivációja deszinkronizálódik. Ennek illusztrálására
néhány részletet mutatunk be mondatismétlési tesztekből:
A vizsgálati személy: 59 éves férfi, a sérülés: baloldali artéria cer. mediális területi
hypodensitas. Az alábbiakban V-vel jelöljük a vizsgáló által adott, megismétlendő célmondatokat, és
bold-dal jelöljük a vizsgálati személy válaszait.
Azonos esetrag több tartalmas szóhoz csatolása:
V: A bácsi a NÉNIVEL[
], a fiú meg a LÁNNYAL sétált.
A bácsi a nénivel, a fiúval meg a bácsival, mentél . A gyerekkel ment.
A vizsgálati személy: 37 éves férfi, traumás eredetű, baloldali temporális területű bevérzés.
Az esetragok elhagyása, vagy hibás igetoldalékolás, hibás egyeztetések
V: Megjavítottam az autómat, de újra elromlott.
Autó, autó szokott csinálni, maga. Meg szokott csinálni, hogy akkor...
V: Vendégeket vártam, de nem jöttek.
Nem bírok várni őket, hanem a nőt mondta, hogy maradok. Hát mondok, akkor
maradok, mert annyira mondta szépen én.
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
66
Spontán beszéd -- részlet
V: Mikor engedik haza?
Nem szokott tudni.. Azér otthon azér csak jobb lesz azér, mer nem nem egymaga leszek,
És akkor legalább ottan míg beszélek beszél még jobban leszek mint én igaz?…
2.1. A mondatfeldolgozás szakaszainak megértése szempontjából különösen érdekesek a
kiváltott agypotenciál vizsgálatok. Ennek során mérik az agy bioelektromos aktivitását. A betegek
fejbőrére --- ismert meghatározott beállítási pozíciók szerint --- elektródákat tesznek, és mérik a
különböző nyelvi ingerek által kiváltott változásokat az agy bioelektromos aktivitásában, amely a
nyelvi inger megjelenésével bizonyos idői kapcsolatban áll. Ez lehet negatív vagy pozitív töltésű,
néhány millivolt értékű elektromos aktivitás. A világ különböző laboratóriumaiban lényegében
egybehangzó eredményekhez jutottak. Eszerint a mondatfeldolgozási műveleteknek három fázisa
határolható el; ezek közül kettő főként szintaktikai természetű.
Friederici (1995) és Friederici és Hahn (1999) kiváltott agypotenciál kísérleteiben grammatikus
és agrammatikus mondatokat vizuálisan, szavanként mutattak be. Az agrammatikus mondatok olyan
nyelvtani hibákat tartalmaztak, melyek vagy a szókategóriákkal, vagy a szintaktikai szerkezettel és a
szórenddel, vagy az összetevõk lexikai jelentésével, vagy pedig az egész mondat szemantikai
interpretációjával függtek össze. Mindig tesztelték a megfelelõ grammatikus változatokat is.
A szavak 100ms-ig voltak láthatók egy képernyőn, minden szónál 100ms-os szünetet tartottak,
ami előkészítette a kontextust a mondatvégi célszónak, amelyet 200ms-ig mutattak be. A vizsgált
személyeknek ebben a feladatban egy késleltetett lexikai döntést kellett hozniuk a mondat utolsó
szavára vonatkozóan. A mondatfeldolgozással kapcsolatosan négyféle agyi elektromos potenciál
adatot találtak, melyek három eltérő idő-ablakkal jellemezhetők. Először a nyelvi inger kezdete után
200ms körül, a bal oldali agyfélteke elülső részére kiterjedő negatív potenciált észleltek, amely a
lokális frázis-szerkezeti információk feldolgozásával volt kapcsolatos. A strukturálisan hibás
mondatszerkezetek feldolgozásakor ez a terület maximálisan aktiválódott. Másodszor kétféle,
egyaránt 400ms körüli, ám eltérő eloszlású negatív potenciált mértek: (a) a lexikai-szemantikai
anomáliák- hibák által kiváltott negatív potenciált, mely az agykéreg hátulsó területein volt látható
széles kiterjedésben, nemcsak a baloldalon, hanem a jobboldalon is; (b) ugyancsak a nyelvi inger
kezdete után 400ms körül, egy bal oldali elülső területű negatív potenciált mértek, amely az igei
argumentum szerkezeti információk, igei alkategorizációs információk feldolgozásával hozható
kapcsolatba. Harmadszor, a nyelvi inger kezdete után 600-800 ms között pozitív potenciált mértek a
középső és a hátulsó kérgi területeken, mindkét agyféltekén nagy kiterjedésben, amely valószínűleg a
szerkezeti újraelemzés folyamatával függ össze. Ez akkor válik szükségessé, ha a felépített szerkezet
és a lexikai-szemantikai információk összekapcsolása nem volt teljesen sikeres, például a mondat
interpretációja nem egyértelmű. Az ún. "becsali" (garden path), mondatok feldolgozásakor történhet.
(v.ö. Nevettek az árbocon a matrózok a fedélzeten).
2.2. Az eredmények azt jelzik, hogy a bal oldali elülső agykérgi területek elsősorban a
szintaktikai szerkezet valós idejű kijelöléséért felelősek. A lexikai egységeket feldolgozó műveletek
korlátozódása gyakran egybeesik a bal oldali hátulsó agyterületek sérülésével, míg az alkategorizációs
információk feldolgozásában a baloldali elülső kérgi területek vesznek részt.
3.1. Az agrammatikus afáziásoknál gyakran mind a beszédprodukció, mind pedig a beszédértés
egyidejűleg valamilyen módon károsodik, és ezeket a károsodásokat nyelvtani terminusokban lehet
leírni. Számos kutató szerint a produkció és a megértés egyidejű zavara már magának a mentális
nyelvtannak a közvetlen sérülését jelzi. Linebarger (1995) szerint ilyenkor a nyelvtan szintaktikai
komponense veszik el. Friedmann-Grodzinsky (1997) szerint viszont a szintaktikai feldolgozás
meghatározott almoduljának elvesztéséről van szó, melynek következtében például a szintaktikai
összetevők mozgatása utáni nyomok törlődnek a szintaktikai reprezentációból.
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
67
A beszédprodukció és feldolgozás egyidejű deficitjére azonban más magyarázat is lehetséges.
Az afáziások beszédintonációja gyakran erősen korlátozott, jellegzetesen monoton intonációt
produkálnak. Nagel és Shapiro (1994) egészséges személyek és afáziások rögzített beszédét
analizálva azt találta, hogy az afáziások nem produkálnak nyelvtani szempontból normál dallamot a
mondatokon belüli „hosszabb tartományok”-ban, és nem produkálnak sem bizonyos időtartam
jegyeket, sem szintaktikai pozíciókhoz társítandó kontrasztív nyomatékokat.
A szerkezeti pozíciókhoz társított prozódia közreműködik annak meghatározásában, miképpen
kapcsolódnak össze lexikai és szerkezeti kategóriák a mondatstruktúrában. A prozódiai információ
valószínűleg nagyon korán, a kezdeti elemzés során felhasználásra kerül annak érdekében, hogy
segítsen a mondatfeldolgozás során felmerülő csatolási kérdések eldöntésében, hogy a szerkezet
feldolgozása során fellépő bizonytalanságokat megoldja. Nagel és Shapiro szerint az afáziások nem
képesek a megfelelő időben felhasználni a prozódiai információt a mondatszerkezet feldolgozásában.
3.2. A magyar nyelvű afáziások a beszédprodukcióban kikerülik az erős nyomatékot hordozó
fókuszt, a tagadott fókuszt és a kvantoros kifejezéseket. Helyettük a szintaktikai fában "lejjebb"
helyezett, nem nyomatékos változókat produkálnak. Mivel csak késve képesek vagy egyáltalán nem
képesek időben felhasználni a prozódiai információt a szerkezeti viszonylatok meghatározásában,
ezért olykor hibás, töredékes szerkezet feldolgozásokat végeznek. A mondat megértésében pedig
fennakadásokat, nehézségeket okozhat a fókusz interpretációja és a kvantoros kifejezés értelmezése:
Adatok mondatismétlési tesztekből. A vizsgálati személy:
parietális hypodensitas.
57 éves nő, sérülése: fronto-
a fel nem dolgozott információ elhagyása:
Tagadó operátor törlése
V: A vezetőt NEM idegesítette a zaj.
A vezetőt idegesítette a zaj.
A feldolgozott információ korlátozott produkciója: lejjebb helyezés a szintaktikai fában:
-- igetagadás képzése:
V: A moziban “nem Péterrel beszélgettem.
Péterrel ö…. Péterrel “nem beszélgettem.
-- a kérdőszó helyett hasonló változó lejjebb a szintaktikai fában:
V: Kit láttál az utcán?
..Hogy az utcán ment valaki, az utcán.. az utcán mentél.
A fókusznyomatékot nem produkálja. Tárgy elhagyása:
V: MARI hajtotta biciklit gyorsan és PÉTER [
].
Tehát .. mindenki hajtott gyorsan , Mari is meg Péter is.
4. Egészséges esetben a különféle agyi területeken tárolt nyelvtani reprezentációk, időzített
sorrendben, aktiválhatók. Az éppen aktivált nyelvtani reprezentációkat a beszédprodukciót és a
beszédfeldolgozást szervező valós idejű mentális programok, mint instrukciókat "olvassák el". A
lokalizált agysérülések a nyelvi képesség valamely részrendszere, részfunkciója korlátozódásához
68
Nyelvtan és Mentális Elemző Neurolingvisztikai Megközelítésben
vezethetnek, a többi épen maradása mellett. Vagyis a nyelv és használata válhat szét funkcionális
komponenseire.
A dolgozatunkban tárgyalt adatokkal és összefüggésekkel azt kívántuk bemutatni, hogy a
neurolingvisztikai vizsgálatok a nyelvtant olyan környezetben írják le, melyben a nyelvtan beágyazott
az agy architektúrájába, interfész pontokon kapcsolódva az agy és az elme mentális programjaihoz.
Ezáltal megnyílik az út a nyelvtan modellek neurológiai realitásának a kutatásához is.
Irodalom
Friederici, A. D,.1995. The Time Course of Syntactic Activation during Language Processing: A Model Based
on Neuropsychological and Neurophysiological Data, Brain and Language, Vol 50, 259-281.
Hahne A. - Friederici A.D. 1999. Electrophysiological Evidence for Two Steps in Syntactic Analysis: Early
Automatic and Late Controll Process. Journal of Cognitive Neuroscience 11:2, 194-205.
Friedmann, N. -Grodzinsky, Y. 1997. Tense and Agreement in Agrammatic Production: Pruning the Syntactic
Tree, Brain and Language, 56, 397-425.
Hagiwara, H. 1995. The breakdown of functional categories and the economy of derivation. Brain and
Language, 50. 92-116.
Kolk, H. 1995. A Time-Based Approach to Agrammatic Production. Brain and Language 50. 282-303.
Linebarger, M. C. 1995. Agrammatism as Evidence about Grammar. Brain and Language, 50. 52-91.
Nagel, N., & Shapiro, L. P. 1994. Prosody and the processing of filler-gap sentences. Journal of
Psycholinguistic Research. [Special Issue]
Swinney D., - Zurif E., 1995. Syntactic Processing in Aphasia, Brain and Language, Vol 50, 225-239.
Semantic Bridging Effects in VP-Ellipsis
69
Semantic Bridging Effects in VP-Ellipsis
Beáta Gyuris
Introduction
The aim of this paper is to investigate the semantic processes underlying the interpretation of a
certain subgroup of Hungarian sentences containing VP-ellipsis. The data and the analysis will give
support to the claim made by Bánréti (1999, 2000) and Bartos (to appear), according to which the
mental lexicon consists of sublexicons with their own, sometimes independent, phonological,
syntactic and semantic principles which are not always extendable to the other sublexicons and to
structural complexes containing items from more than one sublexicons. The study will also contribute
to the characterization of differences between the linguistic processes operating in backwards versus
forward VP-ellipsis, as argued by Wilder (1997), Bánréti (1999, 2000) and Bartos (to appear).
The Hungarian data to be presented and analysed exemplifies a less frequently studied subtype
of VP ellipsis, a case where the antecedent and the ellipsis site contain material which cannot be
equivalent to each other syntactically, regardless of which level of syntax (LF, surface structure, etc.)
is considered, like, for example, in (1) below, where the intended content of the ellipsis site is crossed
out3:
(1)
Vili
szeretne
Marival
Bill
like-cond-3sg Mari-with
[VP táncolni/ Vilivel
dance-inf
táncolni,
de
Mari
nem szeretne
dance-inf
but
Mari
not
like-cond-3sg
táncolni].
Bill-with
dance-inf
‘Bill would like to dance with Mary but Mary wouldn’t like to [dance/dance with Bill].’
The necessary non-identity between the ellipsis site and the antecedent in the case of (1) can be
proved by the ill-formedness of (2) below, where the same syntactic constituents appear in the
antecedent and at the ellipsis site:
(2)
*
szeretne
Vili
Bill
like-cond-3sg
[VP
Mari-with
szeretne
Marival
like-cond-3sg Mary-with
Marival
táncolni,
de
Mari
nem
dance-inf
but
Mari
not
táncolni].
dance-inf
‘Bill would like to dance with Mary but Mary wouldn’t like to [dance with Mary]’.
As opposed to accounts of the above and related phenomena (found in, for example, Asher
(1993), Hardt (1993), etc.), which consider the interpretation of VP-ellipsis with syntactic nonidentity as belonging to the domain of pragmatics, we are going to argue in this paper that a large
portion of the cases manifesting lack of syntactic identity between the antecedent and the elided
3
In what follows, acceptability judgments for sentences with ellipsis will also take into account the assumed
content of the ellipsis site, where indicated.
Semantic Bridging Effects in VP-Ellipsis
70
material can in fact be accounted for on the basis of purely lexical semantic principles, which can also
give us an insight into the mental representation of individual lexical items.
A preliminary overview of the data
In this paper, we will concentrate only on instances of VP ellipsis occurring in coordinate
structures. By forward ellipsis we refer to those cases where the ellipsis site appears in a non-initial
conjunct, and it is preceded by its antecedent. Conversely, the term backwards ellipsis will be used to
refer to those cases where the ellipsis site appears in a non-final conjunct, and it is followed by its
antecedent.
Sentence (3) below illustrates a case of forward ellipsis which is non-interpretable if the elided
material is considered to be identical to its antecedent. It, however, can be made acceptable, as (4)
shows, if the ellipsis site is taken to have an interpretation non-identical to, but semantically related to
that of the antecedent in such a way that the meanings of the clauses containing the antecedent and the
ellipsis site can be considered synonymous.
(3)*
Viki
és
Gabi
szeretnének
összeházasodni, de
Gabi
nem
Viki
and
Gabi
like-cond-3pl
marry-inf
Gabi
not
mer
[VP összeházasodni], mert
dare-3sg marry-inf
because
but
az
apja
utálja
the
father-3sgposs hate-3sg
Vikit.
Viki-acc
* ‘Viki and Gabi would like to get married, but Gabi does not dare to [marry], since her father
hates Viki.’
(4)
Viki és
Gabi
szeretnének
összeházasodni,
de
Gabi
nem
Viki and
Gabi
like-cond-3pl
marry-inf
but
Gabi
not
mer
[VP összeházasodni
Vikivel],
mert
dare3-sg
marry-inf
Viki-with
because the
az
apja
utálja
Vikit.
father-3sgposs
hate-3sg Viki-acc
‘Viki and Gabi would like to get married, but Gabi does not dare to [marry Viki], since her
father hates Viki.’
Similarly, in the case of (1) above, repeated here as (5), those cases of ellipsis reconstruction
lead to felicitous interpretations where the meaning of the clause contianing the ellipsis site is either
entailed by (as in Mari nem szeretne táncolni ‘Mary would not like to dance’) or synonymous with (as
in Mari nem szeretne Vilivel táncolni ‘Mary would not like to dance with Bill’) the clause containing
the antecedent:
(5)
Vili
szeretne
Marival
Bill
like-cond-3sg Mari-with
táncolni,
de
dance-inf but
Mari
nem
szeretne
Mari
not
like-cond-3sg
Semantic Bridging Effects in VP-Ellipsis
71
[VP táncolni/
dance-inf
Vilivel
táncolni].
Bill-with
dance-inf
‘Bill would like to dance with Mary but Mary wouldn’t like to [dance/dance with Bill].’
As opposed to the instances of forward ellipsis illustrated above, which tolerate lack of
syntactic identity in the case of semantic equivalence or entailment between the clauses containing the
antecedent and the consequent, backwards ellipsis seems to be based on full phonological identity
between the antecedent and the elided material, since the backwards counterparts of (4) and (5) are
infelicitous, as (6) and (7) illustrate:
(6)
* Mari
nem szeretne
Mary
not
[VP Vilivel
like-cond-3sg Bill-with
szeretne
Marival
táncolni.
like-cond-3sg
Mary-with
dance-inf
táncolni/táncolni],
de
Vili
dance-inf/dance-inf
but
Bill
‘Mary would not like to [dance with Bill/dance], but Bill would like to dance with Mary.’
(7)
* Gabi nem
Gabi not
mer
[VP összeházasodni
Vikivel],
bár
dare-3sg
marry-inf
Viki-with
though Viki
szeretnének
összeházasodni.
like-cond-3pl
marry-inf
Viki
és
Gabi
and
Gabi
‘Gabi doesn’t dare [to marry Viki], although Viki and Gabi would like to marry.’
Incidentally, the elliptical strings themselves cannot be made felicitous at all, even under a
different construal of the elided VP. (8) and (9) below illustrate that an interpretation according to
which the ellipsis site contains material identical to its antecedent also fails, just like the
corresponding interpretations of (2) and (3) do. (8) is ill-formed both due to the irreflexivity of the
predicate expressed by the verb táncol ‘dance’ and Principle C of the Binding Theory, while (9) fails
since the only argument of the verb összeházasodik ‘marry’ denotes an individual an not a plural
entity, as required by the semantics of the verb (to be discussed more throroughly in Section 4.3) :
(8)
* Mari
Mary
nem
szeretne
[VP Marival
táncolni],
de
Vili
szeretne
not
like-cond-3sg Mary-with
dance-inf
but
Bill
like-cond-3sg
Marival
táncolni.
Mary-with
dance-inf
‘Mary would not like to [dance with Mary], but Bill would like to dance with Mary.’
Semantic Bridging Effects in VP-Ellipsis
72
(9)
* Gabi
Gabi
nem
mer
[VP összeházasodni], bár
not
dare-3sg
marry-inf
szeretnének
összeházasodni.
like-cond-3pl
marry-inf
Viki és
though Viki and
Gabi
Gabi
‘Gabi doesn’t dare [to marry], although Viki and Gabi would like to marry.’
The above examples indicate, on the one hand, that VP ellipsis can be possible even if syntactic
identity is ruled out between the content of the ellipsis site and its antecedent, while, on the other
hand, they argue for distinguishing between the processes and the licensing mechanisms operating in
the case of backwards versus forward ellipsis. The data seem to support the claim made by Wilder
(1997) and Bartos (to appear) that backwards ellipsis is licensed under phonological identity, while
the licensing mechanism of forward ellipsis allows for certain degrees of mismatch between
antecedent and elided material. (A detailed discussion about the allowed degrees of dissimilarity
between antecedent and ellipsis site is found in Bartos (to appear) and Bartos & Gyuris (2000)). This
freedom, however, is not without its limits, since, while in (10)-(11) we could in theory successfully
infer the contents of the ellipsis site on the basis of the „synonymy-strategy” or the „entailmentstrategy” mentioned above, they are markedly worse than (2)-(3), if not downright rejectable.
(10)
?? Iván
táncolna
Mártával,
de
Ivan
dance-cond Márta-with but
Márta nem [VP táncolna/táncolna
Márta not
Ivánnal].
dance-cond/dance-cond Ivan-with
?? ‘Ivan would dance with Márta, but Márta not.’
(11) ?? Péter
Peter
szeretne
táncolni
Marival,
de
like-cond
dance-inf Mary-with but
Mari
nem [VP szeretne táncolni
Mary
not like-cond
dance-inf
Péterrel].
Peter-with
?? ‘Peter would have liked to dance with Mary, but Mary not.’
The last two examples illustate that the interpretation of the site of forward VP-ellipsis,
although less constrained than backwards ellipsis, cannot be based only on the availability of a
feasible semantic relation connecting the antecedent and the ellipsis site, but has to obey certain
constraints of the grammar as well, which are going to be discussed below in Section 4.
The structure of the rest of the paper is as follows. In Section 3 we will summarize and point
out the difficulties with existing treatments of the phenomena of semantic bridging in ellipsis
illustrated above. In Section 4 will show, following Bartos (to appear) and Bartos & Gyuris (2000)
that, in order to make the theory more constrained and to incorporate the restrictions imposed by the
grammar on the interpretation of VP-ellipsis, the semantic relations making possible the interpretation
of sentences with ellipsis have to be encoded in the lexicon, and not relegated to the LF or semantic
component of the grammar. This view will naturally have some repercussions on the structure of the
lexicon, which are discussed in Section 5. The paper ends with the conclusions in Section 6.
Semantic Bridging Effects in VP-Ellipsis
73
Semantic and pragmatic accounts of VP-ellipsis interpretation
Traditionally, investigations of the relation between the antecedent and the elided material in
VP-ellipsis (as well as other types of ellipsis) concentrated on finding the level of syntax where the
above mentioned two constituents could be found identical to each other (e.g., Sag 1976, Williams
1977, Lappin 1996, Tomioka 1997, etc.). Webber (1979) was the first to call attention to examples of
VP-ellipsis like those reproduced here as (12) and (13) (Webber’s exx. (67) and (68)), where syntactic
identity between the antecedent and target of ellipsis is clearly impossible:
(12)
Irv and Martha wanted to dance together, but Martha’s mother said that she couldn’t ∅.
∅ = dance with Irv
(13) Irv and Martha wanted to dance with each other, but Martha’s mother said that she
couldn’t ∅.
∅ = dance with Irv
Webber (1979) argues for allowing inferences to determine the appropriate content of the
ellipsis site. For example, the reasoning procedure giving the interpretation of the elided material in
(12) and (13) would be the following. If a reciprocal predicate P applies to two arguments a and b,
then the predicate has to apply to any two arguments one of which is a or b, while the other is
represented by a variable. This rule is formalized by Webber as follows:
(14) If <a> and <b> do <P> together (or with each other) then <a> does <P> with <b> and <b>
does <P> with <a>.
λ-abstraction over the variable serving as an argument of the predicate results in two possible
logical representations of the elided VP, which are the following: λ(r) [Dance r, Irv] and λ(s)[Dance s,
Martha]. Since in (12) and (13) only the former can be meaningfully applied to the denotation of the
noun Martha, the contents of the elided VP, as indicated above, can be determined unambigously.
At this point, I can see two problems with Webber’s approach. First, it is not clear at which
level of grammar the above mentioned inferences take place, and whether they help to give the
interpretation of a VP slot generated empty (which would contradict the minimalist assumption
against generating empty places), or help justify the deletion of material which appears overtly in the
second VP at some level of grammar. In the latter case, the place of the inferences would presumably
either by the LF or the semantic component, from where, assuming the Chomskyan minimalist
organization of grammar, there is no input to PF, where the actual pronunciation of the sentence is
licensed. The second problem concerns the formulation of the inference in (14), since, as it stands, it
is not clear whether the inference holds between the lexical items themselves (and, in this case, what
the independent representation of the word together would be), or their logical representations
(begging the question how the compositionally-based translation between lexical items and their
logical representations takes place).
Webber’s semantic account of (12) and (13) was given a more pragmatically-oriented
development by Hardt (1993), who accounts for the preferred interpretation of Webber’s (1979)
example, repeated here as (15) in the following way:
(15) Irv and Martha wanted to dance with each other, but Martha couldn’t, because her husband
was there.
Semantic Bridging Effects in VP-Ellipsis
74
Hardt (1993) claims, in accordance with the general assumption that reciprocal expressions
apply to a predicate distributively (argued for, for example, by Bennett (1974), Heim, Lasnik & May
(1991)), that the predicate missing form the second clause of (15) is λx dance (x, y), where the value
of the free variable y is fixed by the context. The same predicate is applied distributively to Irv and
Martha in the clause containing the antecedent and to Martha in the ellipsis target, where Irv is a
salient referent for the parameter y. The acceptability of (15) above contrasts with (16), where,
according to Hardt, the referent corresponding to Susan’s partner is not salient, making the sentence
infelicitous.
(16) *Irv and Martha wanted to dance with each other. Susan couldn’t, because her husband was
there.
I am not certain, however, that salience alone, that is the presence of a possible partner for
Susan either in the discourse or in the real world situation would be enough to make the above short
text well-formed, as the following extension of (16) might indicate:
(17) *Irv and Martha wanted to dance with each other. In spite of Peter being present, Susan
couldn’t, because her husband was there.
Example (18), incidentally, presents no problem for the theory formulated above, since it
involves „intransitive” dance, which, according to Hardt, which is free from the requirement that the
„partner” be salient:
(18)
Irv and Martha wanted to dance. Susan couldn’t, because her husband was there.
To sum up, Hardt (1993) argues for the existence of a pragmatic constraint operating in the case
of reciprocals, according to which the free variable in the logical representation of the predicate must
have a salient referent, which constraint, being less stringent than the syntactic constraints, only
becomes observable under ellipsis.
The reason why I am not satisfied enough with the above mentioned semantico-pragmatic
theories of ellipsis-interpretation might be clear by now. In addition to the possible difficulties which
the reconciliation of the above claims with the current minimalist framework of syntax we are
assuming here, Webber’s and Hardt’s accounts face some theory-internal organizational problems and
fail to account for the degraded acceptability of some of our data, like examples (11) and (12) above.
In the next section we will thus consider the possibility of accounting for our data in a syntactically
more constrained framework. (For detailed arguments for a more syntactically-based ellipsis
resolution procedure see Bartos & Gyuris (2000).)
A lexical semantic account of VP-ellipsis
Some relevant syntactic claims
It is argued in Bartos & Gyuris (2000) that in spite of the observable differences between the
principles regulating backwards and forward ellipsis, theoretical elegance and predictive power would
require that the two processes are directed by the same principles. The essential differences between
Semantic Bridging Effects in VP-Ellipsis
75
the two types of ellipsis are, according to Wilder (1997) and Bartos (to appear), that while the former
can only be licensed under full (phonological) identity with the antecedent, the latter can be
acceptable even if the φ-features of the antecedent and elided verb are different from each other.
Similarly, a tensed verb in the antecedent is able to license the ellipsis of an infinitival form in
forward ellipsis, but not vica versa. Bartos & Gyuris (2000), however, argue that in spite of the above
differences, backwards and forward ellipsis can equally well be accounted for within the framework
of Halle and Marantz’s Distributed Morphology, where ellipsis is taken to be non-insertion of
phonological forms corresponding to the feature bundles characterising the lexical items, under
identity with some other form or feature in the structure.
In this section we will show that the licensing of VP ellipsis in coordinate structures where the
antecedent and the elided VP are not equivalent syntactically, no matter what level of syntactic
identity is to be assumed, can also be accounted for in terms of the Distributed Morphology model in
the following way. It was observed earlier that in cases of syntactic non-identity between the
antecedent and elided VP, the propositions expressed by these VPs together with their required
arguments either have to be synonymous or the first has to entail the second. The meaning relations
expressed by VPs and their arguments can be reflected in the lexicon if verbs are represented together
with specifications on the number and case of their required arguments, and with an indication what
other verbs, provided the necessary arguments are also present, can express the same proposition, or
semantically related propositions.
A practical realization of the above proposal would look as follows. Since, assuming the theory
of late morphological insertion by Halle and Marantz (1993), adopted for Hungarian by Bartos (to
appear), verbs giving rise to identical propositions could be represented in terms of the same abstract
features or feature bundles. Non-insertion of one verb form at MS could thus be licensed by the
presence of a verb non-identical to its antecedent phonologically, but represented in terms of the same
features or feature bundles, or features from which the contents of the features at the ellipsis site could
be recovered. This could explain the licensing of the ellipsis in the second clause of example (3)
above.
A further question concerns the actual representation of the lexical items which would allow the
above mentioned inferences to go through. The best way to do that, we believe, would be to represent
the meaning relations between lexical items (or, more precisely, between propositions formed by
lexical items) in terms of meaning postulates, that is, devices accommodating analytical truths based
on the meaning of words in a formal way (Johnson-Laird 1981). Johnson-Laird gives (19) as an
example for representing the meaning relation between the English verbs sell and buy in terms of
meaning postulates:
(19)
for all x, y, and z, x sells y to z ≡ z buys y from x
A detailed discussion on exactly what format the above mentioned meaning postulates should
appear in is found in Section 4.2 below. Here we only wish to point out that the representation of
lexical items together with their arguments and meaning relations is also useful from the point of view
of explaining backwards ellipsis, since, as it often goes unnoticed by syntactic accounts, the licensing
of the latter phenomenon does not only require phonological identity, but also sense identity. That is,
an antecedent verb occurring in a different argument structure, even if it is phonologically identical to
a verb appearing in a previous conjunct in a different argument structure, cannot license the ellipsis of
the latter, as (20)-(21) below indicate:
Semantic Bridging Effects in VP-Ellipsis
76
(20) *Vili
Bill
szeretne
Marival
would-like Mary-with
[VP táncolni], de
Mari
nem fog
táncolni.
dance-inf
Mary
not
dance-inf
but
will
‘Bill would like [to dance] with Mary, but Mary won’t dance.’
(21) *Vili
Bill
szeretne
Marival
[VP táncolni], de
would-like Mary-with dance-inf
but
Mari
nem fog
vele
Mary
not
he-with dance-inf
will
táncolni.
‘Bill would like with Mary [to dance], but Mari will not dance with him.’4
Since the verb táncolni ‘dance’ appears in (20) in the final conjunct in its one argument version
as opposed to its two-argument version in the initial conjunct, the ellipsis of the former cannot be
licensed by the presence of the latter, in spite of phonological identity of the two verb forms. This
suggests that not only phonological but also sense identity is required for backwards ellipsis, which
somehow has to be specified in the lexicon. More surprisingly, the ellipsis of the verb in the initial
conjunct is not licensed in (21) either, probably because the case requirements of the arguments of the
two verb forms are not identical. This suggests that information about the required cases of NP
arguments and the meaning relations between verbs appearing in different case-frames also have to be
specified in the lexicon. In the next subsection we will suggest a device which will make possible the
lexical representation of all the information which has turned out to be required for the explanation of
ellipsis licensing.
Ellipsis licensing with meaning postulates
It was already mentioned in Section 1 that in order to make our theory more constrained and
account for the apparent contrasts in acceptability between various types of sentences with VP
ellipsis, it seems advisable to try and explain the majority of ellipsis licensing on the basis of
principles of syntax instead of leaving all cases for the semantic component or pragmatics.
We have argued above that non-insertion of a VP at Morphological Structure might be possible
even if its antecedent appears in a different argument structure, provided all the features of the elided
VP are recoverable from the construction, not necessarily on the basis of the antecedent. The
situation, we believe, is very similar to the licensing of the ellipsis of the second V in Bartos &
Gyuris’s (2000) example (1), repeated here as (22) below:
(22)
John
drinks wine and his kids[drink] coke.
3sg
non-3sg
In (22) above, the person and number features of the elided verb are not identical to those of its
antecedent, but they can be recoverable from the structure due to agreement with the corresponding
features of the subject NP. It therefore seems plausible to claim that in cases of ellipsis under syntactic
non-identity, the verb itself would be licensed under identity with the antecedent verb stem, and the
requirements for the number and cases of arguments, being part of the features of the verb, would be
licensed through the specification of meaning relations in the narrow lexicon, while the non-insertion
of arguments would be licensed under identity with semantically corresponding arguments in the
antecedent VP (with no requirement on identity of case).
4
These sentences, however, are impeccable with the right-node raising intonation pattern, that is, with a rise on
fog and vele.
Semantic Bridging Effects in VP-Ellipsis
77
Let us now consider step-by-step the licensing of the ellipsis in sentence (4) above, repeated
here as (23):
(23) Viki
és
Gabi
szeretnének
összeházasodni, de
Viki
and
Gabi
want-cond-3pl marry-inf
mer
[VP összeházasodni
Vikivel],
mert
dares
marry-inf
Viki-with
because the father-3sgposs
but
az
Gabi nem
Gabi not
apja
utálja
Vikit.
hates
Viki-acc
‘Viki and Gabi would like to get married, but Gabi does not dare to [marry Viki] , since her
father hates Viki.’
Let us suppose that the lexical representations of the two senses of the verb összeházasodik
‘marry’ look as in (24a), where, the number of arguments, together with their required cases in
Hungarian are also indicated:
(24) a.
összeházasodik [(x és y)-nom]
‘marry’ [(x and y)-nom]
⇔
összeházasodik [x-nom, y-comitative]
‘marry’ [x–nom, y-comitative]
In (24) above, the parentheses [ and ] enclose the required arguments of the verb, while the
brackets ( and ) enclose larger constituents which function as one argument. The sign ⇔ denotes
meaning equivalence. As a result, the first sense of the verb should be represented as requiring one
argument, which, however, refers to a group of two individuals, as the following examples show:
(25)
A
fiatalok összeházasodtak.
The
youth
marry-past-pl
‘The young people got married.’
(26)
A
pár
összeházasodott.
the
couple marry-past-sg
‘The couple got married.’
The meaning equivalence between the two senses in (24) only holds if the only argument of the
verb used in its first sense is expressed with the help of two conjoined NPs. When lexical insertion
takes place in (23), the non-insertion of the verb and the second argument in the second conjunct is
licensed by the presence of the different, related sense of the verb in the first conjunct, and that of its
arguments.
Semantic Bridging Effects in VP-Ellipsis
78
The licensing of the two possible cases of ellipsis in example (27) takes place in the following
way:
(27)
Vili
és
Mari
szeretnének
Bill
and
Mary
like-cond-3pl dance-inf
tud
[VP táncolni/
can
Vilivel
dance-inf Bill-with
táncolni,
de
Mari
nem
but
Mary
not
táncolni].
dance-inf
‘Bill and Mary would like to dance but Mary can’t [dance/dance with Bill].’
In the lexicon, the following meaning relations would be associated with the various senses of
táncol ‘dance’:
(28) a.
táncol [(x és y) - nom]
⇔
‘dance’ [(x and y) - nom]
b.
táncol [(x és y) - nom]
‘dance’ [(x and y) - nom]
táncol [x-nom, y-comitative]
‘dance’ [x–nom, y-comitative]
⇒
táncol [x-nom]
‘dance’ [x–nom]
(28) shows, on the one hand, that the proposition containing the sense of táncol ‘dance’ used
with an argument referring to a plural entity is equivalent in meaning to the one where two arguments
appear, one in the nominative and one in the comitative case. A proposition where the same verb
appears with only one, singular argument, is entailed by any of the above propositions. Therefore, the
ellipsis in the second clause of (27) is again licensed by the antecedent verb and its arguments in the
first clause and the meaning relations specified in the lexicon, shown in (28).
On the interaction of semantic parallelism and the organization of the grammar
In the preceding sections we saw a couple of examples for the licensing of VP ellipsis based on
semantic principles. In this section we will discuss the interaction of these semantic principles and
syntactic operations, in particular, the effect of the presence of verbal suffixes encoding person-,
number- and tense-features on the semantic licensing of ellipsis.
The following examples illustrate that while deletion of the infinitival form can be licensed by a
tensed form due to semantic reasons in both forward and backwards ellipsis, as in (29)-(30), semantic
parallelism alone cannot license the ellipsis of tensed forms, like that in (31):
(29)
Vili
táncolna
Marival,
de Mari nem akar [VP táncolni/ táncolni
Bill
dance-cond Mary-with but Mary not
want dance-inf/
dance-inf
‘Bill would dance with Mary but Mary doesn’t want to [dance/dance with Bill].’
Vilivel].
Bill-with
Semantic Bridging Effects in VP-Ellipsis
79
(30) Vili
most
nem
fog
[VPtáncolni Marival],
Bill
now
not
will
dance-inf
táncolt
Marival.
dance-past-3sg
Mary-with
mivel tegnap
Mary-with since
már
yesterday already
‘Bill is not going to [dance with Mary] now, since he already danced with Mary yesterday.’
(31)
?? Iván táncolna
Iván
Mártával,
de Márta nem [VP táncolna/táncolna
dance-cond Márta-with but Márta not
Ivánnal].
dance-cond/dance-cond Iván-with
?? ‘Ivan would dance with Márta, but Márta not.’
An explanation for the degraded acceptability of (31), as opposed to, for example, the wellformedness of (29), could be given along the following lines. According to Bartos (to appear), the
licensing of the ellipsis of infinitival forms by finite forms is made possible either by the fact that the
V+ni infinitival form realizes a syntactically simplex unit underspecified for tense, the non-insertion
of which at MS can be licensed by the V-stem alone in the parallel clause, or due to selection of the
infinitival form by a matrix V or Aux. Since the meaning postulates coding the meaning equivalence
or entailment relations between different senses of the same verb or related verbs are formulated for
base forms, the ellipsis of an infinitival form (which is a base form appearing in the lexicon, and for
which the meaning relations are also specified) is licensed by either an infinitival form or a tensed
form. The reason why the licensing of the non-insertion of a tensed verb seems to be less acceptable is
that non-insertion at MS proceeds in a step-by-step fashion. The licensing of the ellipsis of the verb
itself in (29) and (31), cannot take place before the other constituents of the whole VP are assessed.
Since the conditional suffix only belongs to the „domain” of the verb and not of the whole VP,
circularity is violated if its non-insertion is considered after the non-insertion of the whole VP. (For a
syntactic explanation of the contrast between (29) and (31) see Bartos & Gyuris (2000).)
In the next section some further examples are provided for semantic relationships between
lexical items to be represented in the Lexicon.
Some further claims on the structure of the Lexicon
In this section we will consider some further examples for the approach to the licensing of VP
ellipsis based on meaning postulates, as it was described above, and try to isolate those classes of
lexical items for which the ellipsis licensing principles could prove to be applicable in an automatic
way.
One set of lexical items which behave more or less as táncol ‘dance’ in the above mentioned
examples denote such bodily, verbal and other, miscellaneous activities which are reciprocal, that is,
require the presence of at least two, active participants, such as csókolózik ‘kiss’, vitatkozik ‘argue’,
beszélget ‘chat’, levelez ‘correspond’, and the like. The meaning relationship between the two
possible senses of the above verbs can be characterized as follows:
(33) csókolózik [(x és y) - nom]
‘kiss’ [(x and y)-nom]
⇔
csókolózik [x-nom, y-comitative]
‘kiss’ [x–nom, y-comitative]
Semantic Bridging Effects in VP-Ellipsis
80
The other, quite productive meaning relation which has a large effect on the licensing of VP
ellipsis is the entailment between the propositions containing morphologically related verbs with a
causative and intransitive meaning, like megtanít - megtanul ‘teach-learn’, megsüt - megsül ‘bake (tr)
- bake (intr)’, megfürdet - megfürdik ‘bathe (tr) - bathe (intr)’, illustrated in (34)-(36):
(34) Pista
meg
akarja
tanítani
Pista
pv
want-3sg
teach-inf Mari-acc dance-inf but Mari not want-3g
[VP
megtanulni
táncolni].
learn-inf
dance-inf
Marit
táncolni,
de Mari nem akar
‘Pista wants to teach Mari to dance but Mari doesn’t want to [learn to dance].
(35) Pista
meg
akarja
Pista
pv
want-3sg bake-inf the cake-acc
mert
nem
jó
a
sütő.
good
the
oven
because not
sütni
a
tortát,
de
az
nem fog
[VP megsülni],
but
that
not
bake-inf
will
‘Pista wants to bake the case but it won’t [bake] since the oven is broken.’
(36)
Pista
meg
akarja
fürdetni
Pista
pv
want-3sg bathe-inf
a
kutyát,
de az
nem akar [VP megfürödni].
the dog-acc but that not
want bathe-inf
‘Pista wants to bathe the dog but it doesn’t want to [bathe].
The licensing of the above cases of ellipsis are based on the following type of meaning
postulate encoded in the lexicon:
(37)
süt [x nom, y-acc]
‘bake’ [x-nom, y-acc]
⇒
sül [y-nom]
‘bake’ [y–nom]
It is interesting to notice that the ellipsis licensing mechanism illustrated in (34)-(36) above is
only operational in case of verb forms which are derived from the same root. (For the discussion of a
similar relation between ellipted noun phrases and their antecedents see Bartos & Gyuris (2000).) A
parallel relation between non-related forms is impossible:
Semantic Bridging Effects in VP-Ellipsis
81
(38) * Mari meg akarja
Mari pv
kötni
a kutyát,
de az
nem akar
want-3sg tie-inf a dog-acc but that not
[VP megkötve lenni].
want-3sg pv-tied
be-inf
* ‘Mari wants to tie up the dog but it does not want to [be tied up].
(39) * Mari
Mari
megpróbálta megölni
Jánost,
de
pv-tried-3sg pv-kill-inf János-acc but
az
nem akart
that
not
[VP meghalni].
wanted-3sg die-inf
* ‘Mari tried to kill János but he didn’t want to [die].’
In this section, some further types of lexical relations between verb were illustrated which can
make the licensing of VP-ellipsis possible in cases of non-surface identity between the antecedent and
the ellipsis site. The results of the study are summarized in the following section.
Conclusion
This paper has argued for a semantically based licensing mechanism of VP ellipsis in the
framework of Halle and Marantz’s Distributed Morphology. It was claimed that in case of lack of
syntactic identity between the elided VP and its antecedent, forward VP ellipsis is possible if there
exist some systematic meaning relations between the propositions expressed by the clauses containing
the two verbs and their required arguments. We have argued for the possibility of representing the
allowed meaning relationships between word senses in the narrow Lexicon, in terms of meaning
postulates. The significance of the approach suggested here lies in the fact that it contributes to
making the organization of the grammar more constrained, by incorporating phenomena usually
treated in pragmatics by previous approaches into grammar proper.
References
Bánréti, Zoltán (1994) Coordination and Ellipsis, in: É. Kiss-Kiefer (eds): The Syntactic Structure of Hungarian,
New York, Academic Press, 355-414.
Bánréti, Zoltán (1999) Directionality of VP Ellipsis in Sentence Processing: A Neurolinguistic Approach. Ms,
Hungarian Academy of Sciences.
Bánréti, Zoltán 2000. ‘Closed class lexical items in sentence processing. A neurolinguistic approach’. ms p 38.
Submitted to Huba Bartos (ed.) Papers on the mental lexicon, Research Institute for Linguistics, Hungarian
Academy of Sciences, Budapest.
Bartos, Huba (to appear) VP-ellipsis and Verbal Inflection in Hungarian. Acta Linguistica Hungarica.
Bartos, Huba and Beáta Gyuris (2000) Coordinate Ellipsis as Phonological Non-Insertion. Ms. Submitted to:
Bartos Huba (ed.) Papers on the Mental Lexicon, Research Institute for Linguistics of the Hungarian Academy
of Sciences, Budapest.
Bennett, Michael (1974) Some Extensions of a Montague Fragment of English. Doctoral dissertation, UCLA,
Los Angeles, California.
Gardent, Claire (1991) Gapping and VP Ellipsis in a Unification-Based Grammar. Doctoral dissertation,
University of Edinburgh
Halle, M. and A. Marantz (1993) Distributed Morphology and the Pieces of Inflection. In: Hale, K. and S. J.
Keyser (eds.) The View from Building 20: Essays in Honor of Sylvain Bromberger. The MIT Press, Cambridge,
MA: 111-176
Hardt, Daniel (1993) VPEllipsis and Semantic Identity. University of Pennsylvania Report.
Heim, Irene, Howard Lasnik & Robert May (1991) Reciprocality and Plurality. Linguistic Inquiry 22: 63-101.
82
Semantic Bridging Effects in VP-Ellipsis
Johnson-Laird, P. N. (1981) Mental Models of Meaning. In: Aravind K. Joshi, Bonnie Webber and Ivan A. Sag
(eds.) Elements of Discourse Understanding. Cambridge University Press.
Lappin, Shalom (1993) Ellipsis resolution at S-structure. NELS 23: 255-269.
Lappin, Shalom (1996) The Interpretation of Ellipsis. In: Shalom Lappin (ed.) The Handbook of Contemporary
Semantic Theory. Blackwell, Oxford.
Sag, Ivan (1976) Deletion and Logical Form. Doctoral dissertation, MIT.
Tomioka, Satoshi (1997) Focusing Effects and NP-Interpretation in VP-Ellipsis. Doctoral dissertation, UMass,
Amherst.
Webber, Bonnie Lynn (1979) A Formal Approach to Discourse Anaphora. Garland: New York and London.
Wilder, Chris (1997) Some Properties of Ellipsis in Coordination, in: A. Alexiadou-T. Hall (eds): Studies on
Universal Grammar and Typological variation, J. Benjamins, 59-107.
Williams, Edwin S. (1977) Discourse and Logical Form. Linguistic Inquiry 8: 101-140.
The Interpretation of Universal Quantification in Child Language
83
The Interpretation of Universal Quantification in Child Language
Beáta Gyuris
Aims and theoretical background
The aim of the present study is to bring evidence for the differences between the semantic
interpretation of certain logical vocabulary items, in particular, natural language quantifiers, in child
and adult usage, and to investigate whether they stem from the differences in the mental
representation of these lexical items, different cognitive abilities, or the different levels of language
competence characteristic of these two age groups.
The present study will exclusively deal with the interpretational differences characteristic of the
adult and child use of the universal quantifier minden ‘every’, for the following reasons. As is wellknown from the literature on generalized quantifiers (Barwise and Cooper 1981, Bach et al. 1995,
etc.), natural language quantifiers can be divided into two subgroups, depending on whether they can
appear in existential sentences or not. The members of the first group, illustrated in (1) below, are
referred to in the literature usually as ‘weak’ quantifiers, while the members of the second one,
illustrated in (2), as ‘strong’, or ‘essentially quantificational’ quantifiers.
(1). There are many/few/two/at least five/no boys in the garden.
(2). There are most/every/two thirds of the boys in the garden.
A further difference between the above mentioned two subtypes is that while the ‘weak’
quantifiers are symmetric, that is, the set-theoretical relation they represent is not sensitive to the
ordering of the two arguments, the ‘strong’quantifiers are not symmetric. This means that while the
logical meaning of the sentence Two boys are hungry is equivalent to that of the sentence Two hungry
beings/entities are boys, the members of the pair Every boy is hungry and Every hungry being/entity is
a boy are not necessarily equivalent. This also means that in order to be able to interpret sentences
with ‘strong’ quantifiers we not only need to identify the type of the relations they denote, but also
need to determine the order of the arguments of the relation, that is, the interpretation of the latter
might involve one more cognitive operation.
Preliminary overview of the data
In order to illustrate the main problems to be explained with respect to the child usage of the
word minden ‘every’, let us consider some of the answers I received from children aged between 3
and 7 for the question in (3), which was accompanied by the picture shown in Figure 1 of the
Appendix.
(3)
Minden
kisfiú
autót
vezet?
every
boy
car-acc
drive-3sg
‘Does every boy drive a car?’5
5
The scope ambiguity inherent in the English translation could not have contributed to the oddness of the
children’s responses, sicne the corresponding sentences in Hungarian have only one scopal reading.
The Interpretation of Universal Quantification in Child Language
84
(4)
a. Nem.
no
Ebben
az
autóban
nincs
kisfiú.6
this-inessive
the
car-inessive
no
boy
‘No. There is no boy in this car.’
b.
Nem. Mert
no
a
kislány nem ül
because the girl
not
sit-3sg
abban
a
piros kocsiban.
that-inessive
the red
car-inessive
‘No. Because the girl is not sitting in that red car.’
c.
Nem.
Mert
itt
egy
no
because here one
lány, meg
itt
az
autója.
girl
here
the
car-3sgposs
and
‘No. Because here is a girl and here is her car.’
d.
Nem.
Mert
ebben
nincsen
no
because this-inessive no
senki
sem.
nobody not
‘No. Because there is nobody in this one.’
As the examples in (4) indicate, some (in fact a large portion, to be discussed below) of the
answers given by the children to question (3) differed from the affirmative answer adult speakers of
the language would be expected to give. It seems as if the children considered minden ‘every’ as a
symmetrical quantifier, since for an affirmative answer to question (3) they seem to require that both
of the statements Every boy is driving a car and Every car is driven by a boy be true on the basis of
the situation depicted by the picture. A futher interesting aspect of the answers is that they repeatedly
mention the presence of the girl in the picture, who, however, is not referred to in any way in the
sentence. According to Philip (1995), who performed similar experiments with child speakers of
English and Japanese, this kind of phenomenon is the result of the fact that the child does not take the
determiner universal quantifier to quantify over the set of objects denoted by the noun following the
determiner but over events of the type described in the sentence. This issue will be further discussed
below.
Even more interesting findings are connected to Figure 2, and the accompanying question in
(5):
(5)
Minden
cica
alszik?
every
cat
sleep-3sg
‘Is every cat asleep?’
6
This answer was accompanied by a pointing gesture at the empty car.
The Interpretation of Universal Quantification in Child Language
85
(6)
a.
Nem.
Itt
nincs cica.
no
here
no
cat
‘No. There is no cat here.’7
b.
Nem.
Az
egyik elment.
no
the
one
pv-left-3sg
‘No. One of them has left.’
c.
Nem.
Ez
nem alszik,
itt
nincs
macska.
no
this
not
here
no
cat
sleep-3sg
‘No. This one here is not asleep, there is no cat here.’
d.
Nem.
Három alszik csak,
az
egyik
fel
van
kelve.
no
three
the
one
pv
be
raise-partic.
sleep-3sg only
‘No. There is only three that are asleep, one of them is up.’
Philip’s above mentioned claims seem to be supported by data like those in (6) since here the
children again make reference to objects in the universe of the picture which do not appear in the
question at all.
In order to show that the above cited answers are not the result of the complete lack of
knowledge of the meaning of the quantifier minden ‘every’ on the part of the children, let us consider
the questions in (7)−(8) accompanying Figures 3 and 4, respectively, for which we received answers
identical to those expected from adult speakers.
(7)
Minden
maci
újságot
olvas?
Every
bear
newspaper-acc
read
‘Is every bear reading a newspaper?’
(8)
Minden
elefánt
zászlót
lenget?
Every
elephant
flag-acc
wave-3sg
‘Is every elephant waving a flag?’
In the next section we will give an overview of related research reported by Philip (1995), and
his explanations of the fact that the interpretation of quantifiers might differ in child and adult usage.
7
This sentence was accompanied by a pointing gesture at the empty pillow.
The Interpretation of Universal Quantification in Child Language
86
This is going to be followed by the description of the experiments we carried out with Hungarian
children in the preschool age, and our analysis of the results.
Philip’s (1995) experiments
Philip (1995), describing experiments similar to the ones we have carried out and briefly
illustrated in Section 2 above, claims that the reason why children do not assent to questions like Is
every boy riding a pony? at the site of a picture where there is a pony without a rider but otherwise
every boy is on top of a pony is that children aged 3-5 do not consider every to be quantifying over
objects but over events.
Philip (1995) defines events as entities with an inner structure, consisting of set of objects, one
of which can be a distinguished one, which are connected to each other via certain relations.
Subevents are those events whose participants are among the participants of the original (larger)
event. No participant in a subevent can play the role of the distinguished participant unless it is also
the distinguished participant of the larger event. Philip argues that in a sentence containing a universal
determiner quantifier the quantifier will also mark the distinguished participant of the event described
by the sentence.8 He considers an event of a boy driving a car as involving two participants, among
which the boy is the distinguished participant. The event of a car standing alone (waiting for the boy
to appear) can be a subevent of this one, but the one where a girl is driving the car cannot.
Among the explanations given by the children for their negative answer to the question Is every
boy driving a car? at the sight of Figure 1, shown in (4), we can find some which emphasize the
presence of the empty car, while others refer to the picture of the girl. According to Philip (1995),
these two types of answers manifest two strategies, which he refers to as the ‘symmetry’ and
‘perfectionist’ strategies, respectively. These latter two strategies for assessing the truth of a sentence
with a universal determiner quantifier signal the presence of two different developmental stages. The
chronologically first, ‘perfectionist’ strategy manifests itself in checking about each object on the
picture whether it is a participant of an event described by the sentence. According to the second,
„more advanced”, ‘symmetry’ strategy, the child only checks the participation of those objects in the
event under discussion, which were referred to in the corresponding sentence. Thus, the answers cited
in (6) should be all evidences for the ‘perfectionist’ interpretation strategy, since the even described in
the sentence has only one participant, the cat, but the children repeatedly refer to the presence of the
pillow in one way or another.
In (10) below the logical (tripartite) structure is reproduced, which, according to Philip’s (1995)
proposal would be assigned to sentence (9) under the ‘symmetry’ interpretation, where PART(x, e)
would refer to the fact that x is a participant of event e.
(9)
Minden
kisfiú
autót
vezet.
Every
boy
car-acc drive-3sg
‘Every boy drives a car.’
8
It is easily shown that there is an independent necessity for quantification over events or eventualities, since
most accounts of the meaning of adverbial quantifiers, like always never, twice, etc. (e.g., de Swart 1991,
Johnston 1994, etc.) assume that they denote quantification over events.
The Interpretation of Universal Quantification in Child Language
87
(10)
Sentence
Quantifier
∀e1
Restrictor
Nuclear scope
∃ e2[e1≤e2 & drive(boy,car,e2) & PART(boy, e1)] a boy is driving
or
a car (e1)
∃ e2[e1≤e2 & drive(boy,car,e2) & PART(car,e1)]
Formula (10) above thus means that all minimal events which are subevents of an event of a
boy driving a car and have a car or a boy as participants are such that they are at the same time events
of a boy driving a car. Philip claims that generally, according to the ‘symmetry’ interpretation, the
domain of quantification is constituted by those minimal events which have a participant either acting
as a distinguished participant of an event containing this minimal event, or an entity referred to by a
referential expression in the sentence.
The logical structure of the sentence according to the ‘perfectionist’ interpretation would be as
follows:
(11)
Sentence
Quantifier
∀e1
Restrictor
Nuclear scope
∃ e2[e1≤e2 & drive(boy,car,e2) & PART(boy,e1)] a boy is driving
or
a car (e1)
∃ e2[e1≤e2 & drive(boy,car,e2) & PART(car,e1)]
or
∃ e2[e1≤e2 & drive(boy,car,e2) & PART(visible obj,e1)]
(11) above means that all minimal events which are subevents of an event of a boy driving a
car, and which have a boy, a car (or both) or another object in the picture as participants are events of
a boy driving a car.
The account summarized above naturally presents the question why children should prefer
quantification over events as opposed to individuals, and if this is indeed the reason for their non-adult
responses, what makes them gradually switch to the „adult” interpretation. Studies on natural
language quantifiers in various languages (e.g., Bach et al. 1995) have proved that although there are
languages which do not contain determiner quantifiers (of the form every, some, at least five, etc.), all
languages have adverbial quantifiers corresponding, for example, to English always, usually, never.
Since the latter types of quantifiers have been argued to quantify over eventualities, according to
Philip, it seems reasonable to say that quantification based on eventualities is the more general option,
the only one available in Universal Grammar, which is only supplemented in some languages with the
The Interpretation of Universal Quantification in Child Language
88
more marked option of quantifying over objects. The reason why children, after a certain age, still
start to prefer quantification over objects, Philip claims, is of pragmatic nature: they realize that via
the interpretation of quantification based on events they do not always gain correct information about
the world.
Experiments with Hungarian children
In the experiments to be described below we were aiming to discover whether the Hungarian
speaking children would react in a similar way to the sentences containing ‘strong’ determiner
quantifiers as their English speaking peers studied by Philip, and if they do, whether the features of
their performance are really due to the fact that they employ quantification over events, or the result
of some other cognitive or linguistic operation.
Experiment 1
a) Participants
This experiment took place with the participation of 20 children in the preschool age. Among
them, 18 gave acceptable answers, whose average age was 5 years and 7 months. The youngest
among them was 4 years and 7 months old, while the oldest was 6 years and 7 months old.
b) Description
During the experiment the children were asked to answer 16 questions with yes or no,
depending on whether the question correctly described the contents of the picture or not. The coloured
pictures, the black-and white equivalents of some of which are found in the Appendix to this paper,
showed human and animal figures in traditional or slightly unusual situtations. If a child gave a
negative answer to the question, he or she always had to give a reason for his choice.
c) Test sentences
Table 1 below shows the test questions, the ‘adult’ answers to these questions, the number of
answers given by the children which were not identical to the ‘adult’ answer, and the number of the
‘perfectionist’ answers among them.
The structure of the test sentences followed the two patterns in (12):
(12)
a.
Subject Object
Verb
minden ‘every’+ noun (denoting a human or animal)
b.
noun
Subject Verb
minden ‘every’ + noun (denoting a human or animal)
intransitive
transitive
The Interpretation of Universal Quantification in Child Language
89
Table 1
Question
1
Minden kisfiú autót vezet?
Adult
Non-identical
Perfectionists
response
child responses
Yes
11
5
6
5
16
-
6
3
4
2
7
6
2
-
5
5
13
13
5
5
1
-
2
-
2
-
1
-
1
-
2
-
‘Is every boy driving a car?’
Minden süni almát eszik?
‘Is every hedgehog eating an apple?’
Minden kismalac kocsit húz?
‘Is every pig pulling a cart?’
2
Minden kisfiú autót vezet?
Yes
‘Is every boy driving a car?
Minden kutya csontot rág?
‘Is every dog eating a bone?’
Minden maci újságot olvas?
‘Is every bear reading a paper?’
Minden süni almát eszik?
‘Is every hedgehog eating an apple?’
3
Minden kisfiú integet?
Yes
‘Is every boy waving?’
Minden cica alszik?
‘Is every cat asleep?’
Minden kutya fut?
‘Is every dog running?’
4
Minden kutya fehér?
No
‘Is every dog white?’
Minden kutya fut?
‘Is every dog running?’
Minden elefánt zászlót lenget?
‘Is every elephant waving a flag?’
Minden kislány cicát simogat?
‘Is every girl petting a cat?’
Minden kacsa úszik?
‘Is every duch swimming?’
5
Minden maci újságot olvas?
‘Is every bear reading a paper?’
Yes
90
The Interpretation of Universal Quantification in Child Language
The pictures used together with the first set of questions depicted several smaller situations
involving two participants (e.g., Figure 1), a figure corresponding to the object term standing alone,
and a figure which was not mentioned in the sentence (the picture of a girl in Figure 1). If the child
referred to the presence of the object not mentioned in the sentence as a reason for his negative
answer, he or she was put down as a ‘perfectionist’. If he referred on to the object mentioned in the
sentence in his explanation, he was assumed to be following the ‘symmetry’ strategy.
The second group of pictures also showed several smaller situations involving two participants
(e.g., Figure 5), which were complemented by pictures of objects not mentioned in the sentence,
which were either shown alone, or together with figures corresponding to the object terms of the
sentence. If a child gave a negative answer to the test question in this case, he definitely had to follow
the ‘perfectionist’ strategy. This latter expectation was also supported by their reasoning.
The third group of sentences referred to events with one participants, but the pictures always
included other objects not mentioned in the sentence (e.g., Figure 2). A negative answer followed by a
reason referring to the unmentioned object was also sign of a ‘perfectionist’ strategy.
The fourth group of pictures was supposed to test whether the children have acquired the most
important component of the meaning of the quantifier minden ‘every’, which requires that each object
in a set should possess a certain feature (e.g., Figure 4). These sentences had to be answered in the
negative no matter whether somebody was following the ‘adult’, the ‘perfectionist’ or the ‘symmetric’
strategy.
The last sentence was supposed to serve the same purpose, with the only difference that this
required an affirmative answer independent of the applied strategy (e.g., Figure 3).
Experiment 2
a) Participants
Due to the large portion of the ‘perfectionist’ answers in the first experiment, we chose 14,
slightly older children for the second experiment, who, however, were still in the kindergarten. Two
of them also gave unacceptable answers, who were left out from the evaluation. The average age of
the children here was 6 years and 7 months, the youngest being 6 years old, and the oldest 6 years and
9 months old.
b) Description
This time we were talking about 11 pictures with the children, chosen from among the 16
pictures used in the first experiment. This time they were presented with statements, and they had to
determine whether they were true or false on the basis of the pictures.
c) Test sentences
Table 2 contains the test sentences, the expected ‘adult’ responses, and the number of answers
given by the children which were not identical to the expected adult one. The test sentences of the first
experiment were altered on 4 occassions, in such a way that the universal quantifier minden ‘every’
was attached to the object bare noun, and placed into sentence-initial position. This NP was followed
by the subject bare noun and the verb, as (13) below illustrates:
The Interpretation of Universal Quantification in Child Language
91
(13)
a.
Minden
újságot
maci
every
newspaper-acc
olvas.
bear
read-3sg
‘Every newspaper is read by a bear.’
b.
Minden
every cat-acc
cicát
kislány
girl
pet-3sg
simogat.
‘Every cat is petted by a girl.’
Table 2
Question
1
Minden kisfiú autót vezet?
Adult
response
Non-identical
Perfectionists
child responses
Yes
4
-
Yes
2
-
No
1
-
No
-
-
1
-
4
4
3
3
-
-
No
-
-
No
-
-
Yes
6
-
‘Is every boy driving a car?’
Minden süni almát eszik?
‘Is every hedgehog eating an apple?’
Minden kocsit kismalac húz?
‘Is every cart pulled by a pig?’
2
Minden újságot maci olvas?
‘Is every paper read by a bear?’
Minden almát süni eszik?
‘Is every apple eaten by a hedgehog?’
3
Minden kisfiú integet?
Yes
‘Is every boy waving?’
Minden cica alszik?
‘Is every cat asleep?’
Minden kutya fut?
‘Is every dog running?’
4
Minden kutya fehér?
‘Is every dog white?’
Minden elefánt zászlót lenget?
‘Is every elephant waving a flag?’
Minden cicát kislány simogat?
‘Is every cat petted by a girl?’
92
The Interpretation of Universal Quantification in Child Language
Discussion of the Hungarian experiments and their implications
In the course of the two experiments described above, among the 30 children aged between 4
years and 7 months and 6 years and 9 months there were 4 (13%) who gave answers always identical
to the expected adult answers. There were 21 (70%) children who ever gave answers of the kind
which would be expected under the ‘perfectionist’ strategy. There were 5 children (17%) who never
seemed to apply the ‘perfectionist’ strategy, only the ‘symmetry’ one. This means that in our
experimental group 83% of the children gave answers from which Philip would deduce that they
interpret sentences containing determiner quantifiers as if they denoted quantification over events.
The ratio of children giving ‘perfectionist’ and ‘symmetry’ answers in his experiments was 85%,
which is very close to our results, the only difference being that he tested much younger children,
aged between 3 years and 3 months to 5 years and 10 months. Furthermore, in his experiments only
53% of children ever gave ‘perfectionist’ answers, as opposed to our 70%.
If the application of the ‘symmetry’ and ‘perfectionist’ answers characterized a developmental
phase as claimed by Philip (1995), we would expect the much older Hungarian children to give less
non-adult answers than their American peers, and would not find the oldest children perform
according to the postulated ‘perfectionist’ strategy at all. It can therefore be concluded that our study
did not prove satsifactorily that there is in fact a developmental tendency from ‘perfectionist’ event
quantification through ‘symmetry’ event quantification to adult object quantification, and there are
some other possible alternative explanations to be given for the phenomena, which will be detailed
below. It also has to be emphasized that Philip’s reasons for the universality of event quantification
and for its more ‘basic’ nature have not been given enough psycholinguistic support yet. Although we
do not know of any studies on the topic, it would be hard to believe that events, being rather abstract
objects, would be easier to conceptualize than tangible objects.
How can we then explain the findings of the experiments described above? Instead of claiming
that universal quantifiers like every denote quantification over events in child language, let us
entertain an alternative proposal. As mentioned in Section 1, ‘weak’ natural language quantifiers are
distinguished from ‘strong’ ones in that the former are insensitive to the order of their arguments,
since they denote a symmetric relation. Although, in a logical sense, the symmetry of ‘weak’
quantifiers is an additional, special property which does not characterize every natural language
quantifier, we could also take it as a preferred option and describe ‘strong’ quantifiers as
psychologically more complex ones, since they are also sensitive to the order of their arguments. If
symmetric quantifiers are taken to be unmarked case then we have an explanation why all quantifiers
are interpreted as symmetric ones by young children. Consequently, the different interpretations of
determiner quantifiers in child and adult usage are not to be taken as the sign of lack of necessary
cognitive or linguistic abilities but due to the different mental representation of natural language
quantifiers by children and adults.
There is, however, one remaining mystery under the interpretation of every as a symmetric
quantifier, namely, why children list among their reasons for rejecting a sentence the presence of an
object on the picture which is not mentioned in the sentence. Since, however, the number of children
who mention reasons of this kind varies greatly with respect to the individual pictures, I would
attribute it either to the fact that the unmentioned object was easily associated with the activity (as int
he case of Figure 2), and its presence indicated that a participant involved in the particular activity is
missing from the picture. In other cases (like in Figure 1), the unmentioned object could in principle
enter into the relation referred to by the verb, in which case the symmetry of the relation would almost
be achieved. In other cases the surprising answers might be due to the test situation itself. It was
pointed out to me by Olga Bársony (p.c.), that children in the kindergarten receive a lot of exercises
where their task is to point out from among almost identical figures the one which is slightly different
from the others in pictures similar to the ones used int eh experiment. If this is the case, rejection of
the statement to the presence of an unmentioned object in the picture can also be the sign that the
child discovered the ‘odd one out’ among the individual figures in the picture.
93
The Interpretation of Universal Quantification in Child Language
Conclusion
This paper has argued against the view that the universal quantifier every denotes quantification
over events in child language. On the basis of experiments carried out with Hungarian children we
have claimed that Philip’s (1995) arguments for event quantification being the developmentally first
stage in the interpretation of determiner quantifiers do not seem to hold cross-linguistically. We
argued instead, that the results are equally compatible with a framework which postulates that all
natural language quantifiers are interpreted as symmetric up to a certain age, and differences between
the child and adult usage of such quantifiers do not stem from cognitive or language deficits but from
a different mental representation of the lexical items in question.
References
Bach, Emmon et al. (eds.) 1995. Quantification in Natural Languages. Dordrecht: Reidel.
Barwise, Jon, and Robin Cooper. 1981. Generalized quantifiers and natural language. Linguistics and
Philosophy 4:159–219.
Johnston, Michael. 1994. The syntax and semantics of adverbial adjuncts. Doctoral dissertation, University of
California, Santa Cruz.
Kiefer Ferenc (ed.) 1993. Strukturális magyar nyelvtan. Budapest: Akadémiai Kiadó.
Parsons, Terence 1994. Events in the Semantics of English. Cambridge, Mass.: The MIT Press
Philip, William 1995. Event Quantification in the Acquisition of Universal Quantification. Doctoral dissertation,
University of Massachusetts at Amherst.
De Swart, Henriëtte. 1991. Adverbs of quantification: A generalized quantifier approach. Doctoral dissertation,
Rijksuniversiteit Groningen.
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
94
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in
Hungarian
Beáta Gyuris
Introduction
The aim of this paper is to investigate and formally represent the differences between the
semantic interpretation of Hungarian complex sentences containing a temporal subordinate clause
introduced by the connective amikor ‘when’ and that of conditionals, which contain a subordinate
clause introduced by the connective ha ‘if’. The discussion relies on contemporary semantic theories
on the nature of adverbial quantification and on the interpretation of English if and when clauses. The
observations are accounted for in an event-based semantic framework.
Sentences (1)–(8) illustrate the relevant classes of examples we will concentrate on in what
follows:
(1)
János
mindig
megmossa a
kocsit,
amikor
szabadnapja
János
always
washes
car-ACC
when
day_off-3SGPOSS is
the
van.
‘János always washes the car when he has a day off.’
(2)
János
mindig
megmossa a
kocsit,
ha
szabadnapja
van.
János
always
washes
car-acc
if
day_off-3sgposs
is
the
‘János always washes the car if he has a day off.’
(3)
Amikor Kati
hazamegy, Péter
elkészíti
a
vacsorát.
When
goes_home Péter
prepares
the
dinner-acc
Kati
‘When Kati goes home, Péter prepares dinner.’
(4)
Ha
Kati
hazamegy, Péter
elkészíti
a
vacsorát.
If
Kati
goes_home Péter
prepares
the
dinner-acc
‘If Kati goes home, Péter prepares dinner.’
(5)
János
megmosta
a
kocsit,
amikor
tegnap
szabadnapja
János
washed
the
car-acc
when
yesterday day_off-3sgposs
volt.
was
‘János washed the car yesterday, when he had a day off.’
(6)
János
megmosta a
kocsit,
ha
tegnap
szabadnapja
volt.
János
washed
car-acc
if
yesterday
day_off-3sgposs
was
the
‘János washed the car yesterday if he had a day off.’
(7)
* János
tegnap
megmosta volna a
* János
yesterdaywashed
kocsit,
amikor szabadnapja
would the car-acc when
day_off-3sgposs lett volna.
became would
*‘János would have washed the car yesterday when he had had a day off.’
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
95
(8)
János tegnap
megmosta
János yesterdaywashed
lett
volna a
kocsit,
ha
szabadnapja
would the
car-acc
if
day_off-3sgposs
volna.
became
would
‘János would have washed the car yesterday if he had had a day off.’
The reason why it seems fascinating to compare the above two constructions is that in certain
types of sentences their distributions and interpretations are judged more or less identical, as in (1)
versus (2) and (3) versus (4) above, which contain overt and covert adverbs of quantification,
respectively, while in certain other types of contexts they can give rise to completely different
interpretations. For example, the connective amikor ‘when’ gives a factive reading to (5) above, while
sentence (6) containing ha ‘if’ is interpreted as expressing a hypothetical statement. In a third type of
context, illustrated in (7) and (8) above, the use of the amikor ‘when’ clause is unacceptable. (9)
below illustrates that a Hungarian sentence with an amikor ‘when’ clause can be semantically illformed (at least in the standard dialect) when its English counterpart (Farkas and Sugioka’s (1983)
example) is not, but its counterpart with ha ‘if’, as in (10), is acceptable:
(9)
# Amikor
when
egy
gyerek
tud
angolul, akkor
a
child
knows English then
jól
tud.
weil
knows
jól
tud.
well
knows
‘When a child knows English, he knows it well.’
(10)
Ha
egy
gyerek
tud
angolul, akkor
if
a
child
knows English then
‘If a child knows English, he knows it well.’
In view of the objectives described above, the following specific questions will be addressed in
the present paper:
a) How can the evident similarity of the interpretations of complex sentences with temporal
versus conditional subordinate clauses and an overt or covert adverb of quantification,
manifest in (1)–(2) and (3)–(4) above, be formally captured?
b) What is the domain of the implicit or explicit quantification in the examples (1)–(4)?
c) What mechanisms can enable a quantifierless structure to acquire a quantificational meaning
in examples like (3)–(8)?
d) How can the mismatch between English when and Hungarian amikor ‘when’ clauses, which
gives rise to the interpretational difference indicated in (9), be explained?
Before we try to answer any of the above questions I wish to emphasize that in order to cut the
discussion to a manageable size we will disregard many of the semantic features of temporal clauses
and conditionals here and only try to account for the similarities and differences which are inherent in
the interpretation of examples (1)–(10).
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
96
Comparing Hungarian ha ‘if’ and amikor ‘when’ clauses
As the examples in (1)–(10) above have shown, the distribution of Hungarian amikor ‘when’
and ha ‘if’ clauses shows a remarkable similarity to English when and if clauses with one notable
exception, namely that amikor ‘when’ clauses do not lend themselves to an atemporal reading at all.
This fact suggests that Hungarian amikor ‘when’ clauses, as opposed to their English
counterparts, always carry temporal information, in other words, they state a certain relationship
between the times of the eventualities (in the sense of Bach 1986) described by the when clause and
the main clause.
On the lexical meaning of akkor ‘when’ and ha ‘if’
The claim made in Heim 1982 and Kratzer 1986 that when and if clauses are devices for
restricting the domain of quantifiers, seems to hold for their Hungarian counterparts as well. If they
appear together with a main clause which contains an adverb of quantification, like in (1) and (2)
above, they define the set of entities quantified over, for example, these sentences mean that in all
cases/events/times when Pista has a day off, he washes the car. The same subordinate clauses can also
be used to restrict the domain of an invisible universal quantifier (generic operator), like in (3) and (4)
above, which mean that in all cases/events/times when Kati goes home, Péter prepares dinner.
The major difference between the interpretations of sentences with amikor ‘when’ and those
with ha ‘if’ clauses is that the connective amikor ‘when’ indicates that certain temporal relations, such
as precedence or temporal overlap hold between the eventualities satisfying the eventuality
descriptions in the main and subordinate clauses (for similar views on the meaning of English when
see Heinämäki 1974 and Johnston 1994), which is not required to hold between the eventualities in
the two clauses connected by ha ‘if’. The latter claim can be illustrated with examples (11)–(12)
below:
(11) Mari
mindig
vidám,
amikor
esik
az
eső.
Mari
always
happy
when
falls
the
rain
‘Mari is always happy when it rains.’
(12) Mari
mindig
vidám,
ha
esik
az
eső.
Mari
always
happy
if
falls
the
rain
‘Mari is always happy if it rains.’
Interestingly enough, (11) above can only be considered true if Mari is happy during each
interval when it rains, while (12) can be true in a wider range of occassions, that is, if there is an
inverval during which Mari is happy, which is sufficiently close in time to (i.e., shortly precedes,
follows or overlaps) or in any other, predictable way related to, each interval during which it rains.
In most cases, speakers expect the eventualities satisfying the descriptions in the main and the
subordinate clauses of conditionals be causally related, as suggested by Declerck (1988). I would,
however, follow Heinämäki (1974) in this matter who claims that the causal relation is an implicature
and does not constitute part of the lexical meaning of the connective.
Having discussed some aspects of the lexical meaning of amikor ‘when’ and ha ‘if’, in the next
subsection we take a look at how generic interpretation is acquired in the presence of amikor ‘when’
and ha ‘if’ clauses.
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
97
Implicit quantifiers
Hungarian amikor ‘when’ and ha ‘if’ clauses restricting the domain of implicit universal
quantifiers or generic operators were illustrated by (3) and (4) above. (3) and (4) are most readily
taken to mean that all eventualities of Kati going home are followed by or associated with an
eventuality of Péter preparing dinner, respectively, where the number of eventualities of of the former
kind in the actual world is typically more than one. As opposed to this, the subordinate clauses of (13)
and (14), which contain past tense verbs and time adverbials specifying an interval, seem to refer to
one eventuality each in the actual world:
(13)
Amikor Kati hazament
when
tegnap
este,
Péter
Kati went_home yesterday evening Péter
elkészítette a
vacsorát.
prepared
dinner-ACC
the
‘When Kati went home yesterday evening, Péter prepared dinner.’
(14)
Ha
Kati
hazament
tegnap
este,
Péter
elkészítette
a
vacsorát.
If
Kati
went_home
yesterday
evening
Péter
prepared
the dinner-ACC
‘If Kati went home yesterday evening, Péter prepared dinner.’
In other words, (13) presupposes that there is one eventuality which satisfies the eventuality
description in the subordinate clause and it makes an assertion to the effect that the eventuality which
satisfies the former eventuality description is in a specific temporal relationship with an eventuality of
the type described in the main clause. (14), however, is perfectly acceptable even if no eventuality
satisfies the eventuality description in the subordinate clause in the actual world, provided that if any
do, they have to associated with an eventuality of the type described by the main clause. There is a
subset of if-clauses called counterfactual if clauses, of which (8) above is a prime example, which are
only felicitous if there is no actual-world eventuality satisfying the description they contain. The
problem of how to extend the account based on the implicit generic operator to these cases will be the
subject of Section 3.3.
With these observations we conclude the informal discussion of the semantic properties of
Hungarian amikor ‘when’ and ha ‘if’ clauses. The findings are going to be formalized in Section 3
below.
Formalizing the semantic interpretation
Basic ingredients
The formal representation of the meanings of the Hungarian sentences illustrated above will
contain the runtime function, defined in (15) below, which assigns to each eventuality its runtime, that
is, the time interval which passes between its starting point and endpoint (based on Johnston 1994):
(15)
f:
E → I (runtime function)
E:
the set of eventualities
I:
the set of intervals
The set of eventualities E will be partitioned into EE, the set of events, ES, the set of states, and
EP, the set of processes. We will also distinguish important subsets of the latter two, namely ESMax, the
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
98
set of maximal states, EPMax, the set of maximal processes (i.e., those which are not part of a larger
state or process of the same kind), and ES*, the set of those states which are not denoted by individuallevel predicates (i.e., those which do not by definition last through the lifetime of an individual).
ES*Max, naturally, is the intersection of ESMax and ES*.
It was mentioned above that complex sentences with a temporal subordinate clause introduced
by amikor ‘when’ denote a temporal relation between the two types of eventualities satisfying the
eventuality descriptions in the two clauses, which is defined in (16) below (based on Gyuris 1998).
Note that the relation is not defined for states corresponding to individual-level predicates.
(16)
WHEN
⊆E×E
esub , emain: eventualities satisfying the eventuality descriptions in the subordinate and main
clauses, respectively
(esub, emain) ∈ WHEN iff one of the following conditions are fulfilled:
a)
esub ∈ (ES* ∪EP ∪ EE) ∧ emain ∈ (ES* ∪ EP) ∧ f(esub) ⊆ f(emain)
b)
esub ∈ (ES*Max ∪ EPMax) ∧ emain ∈ EE ∧ (f(emain) ⊆ f(esub) ∨ f(esub) < f(emain))
c)
esub ∈ EE ∧ emain ∈ EE ∧ f(esub) < f(emain)
Informally, (16) means that the temporal relations between the eventualities satisfying the
eventuality descriptions in the amikor ‘when’ and the main clauses of a complex sentence vary
according to the type of the eventualities as follows. Whenever the subordinate clause describes a
‘stage-level state’ (a state which does not hold for the whole lifetime of an individual), a process or an
event, and the main clause decribes a ‘stage-level state’ or a process, the interval assigned to the first
is included in the interval assigned to the second. This situation obtains in the case of (17a, b), where
the times of the eventualities of István’s sadness and those of János’s walking have to include those of
raining and the sun starting to shine, respectively, in order for the sentence to be felicitous:
(17) a)
István
szomorú,
amikor esik
az
eső.
István
sad
when
the
rain
sétált,
amikor
falls
‘István is sad when it rains.’
b)
János
a
parkban
János
the
park-INESSIVE walked when
kisütött
a
PREVERB-shone
nap.
thesun
‘János was walking in the park when the sun started to shine.’
When the subordinate clause eventuality is a maximal ‘stage-level state’ or a maximal process,
and the main clause describes an event, then the containment relation between the intervals is either
reversed, or the interval assigned to the subordinate clause eventuality has to precede the interval
assigned to the main clause eventuality. Thus, in order for (18) be felicitous, the sun has to start
shining either within the interval during János is walking or after it is over:
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
99
(18) Kisütött
a
PREVERB-shone
nap,
the sun
amikor János a
when
parkban
sétált.
János the park-INESSIVE
walked
‘The sun started to shine when János was walking in the park.’
When the predicates describe events in both clauses, as in (19), the interval assigned to the
subordinate clause eventuality has to precede the interval assigned to the main clause eventuality, that
is, the eventuality of Péter’s paper reading have to occur after his arrival:
(19) Péter
Péter
elolvasta
az
újságot,
amikor
megérkezett.
PREVERB-read
the
paper-ACC
when
PREVERB-arrived
‘Péter read the paper when he arrived.’
Having laid all the foundations, the next subsections will deal with the quesiton how the formal
apparatus can produce the interpretations of sentences with amikor ‘when’ and ha ‘if’ clauses.
Formalizing the interpretation of amikor ‘when’ clauses
The formal interpretations of sentences with amikor ‘when’ clauses will be given in terms of
tripartite structures (following Heim 1982, de Swart 1991, Johnston 1994, Bach et al. 1995), which
consist of the interpretation of the quantifier and the two sets between which the quantifier defines a
relation. The particular relations which individual quantifiers of Hungarian will be assumed to
represent correspond to those assigned to their equivalents in other natural languages in the
framework of Generalized Quantifier Theory (Barwise and Cooper 1981), where quantifiers represent
relations between sets of entities of the same type.
It was argued above that in complex sentences containing subordinate clauses introduced by the
connectives amikor ‘when’ or ha ‘if’, the first member of the relation defined by the overt or covert
quantifier is provided by the subordinate clause, and, since sentences were assumed to contain
eventuality descriptions, its type will be a set of eventualities. The second argument of the relation has
to be determined in a way which enables us to define the required set-theoretical relations between
them, for example, the subset relation, corresponding to the meaning of the universal quantifier.9
The only way to stay faithful to the spirit of Generalized Quantifier Theory in this matter would
be to take the second argument of the relation to consist of eventualities of the type described by the
subordinate clause which are somehow paired up with eventualities of the type described in the main
clause. A convenient way of pairing up eventualities would be through the WHEN relation, which was
argued to be inherent in the meaning of the connective amikor ‘when’. This procedure is illustrated in
the logical representation (1’) of (1) above (where the restrictor and the nuclear scope sets are
distinguished with the help of the different bracketing):
(1’)
alwayse’ {λe (have’(János, day_off, e))}
[λe (have’(János, day_off, e) ∧ ∃e’ (wash’(János, the car, e’) ∧
∧ WHEN (e, e’)))]
9
Sets of different types of eventualities cannot have common members. Sentence (1), for example, cannot be
taken to mean that all eventualities of János having a day off are eventualities of János washing the car.
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
100
The tripartite structure in (1’) is generated in the spirit of Davidson (1967), according to whom
the sentence expresses a relation between its various NP arguments and an event argument. (We will
slightly digress here from the Davidsonian tradition, and follow De Swart 1991 and Johnston 1994 in
allowing the logical representation of all and not only action sentences to contain an extra argument
position, to be called an eventuality argument.) In (1’), the adverbial quantifier mindig ‘always’
quantifies over one variable, the eventuality variable e, and the formula means that the set of all
eventualities of János having a day off is a subset of the set of those eventualities of János having a
day off which are associated with eventualities of him washing the car through the WHEN relation. In
other words, all eventualities of János having a day off are associated with an eventuality of him
washing the car, which is as close to the intuitive meaning of the sentence as possible.10
The logical representation of the generic sentence (3) above with no overt adverbial quantifier
will be assumed, following Chierchia 1995, to contain the implicit generic operator GEN, quantifying
over the eventuality variable:
(3’) GENe {λe (goes’ (Kati, home, e))}
[λe (goes’ (Kati, home, e)) ∧ ∃e’ (prepares’ (Péter, dinner, e’) ∧
∧ WHEN (e, e’)))]
(3’) therefore means that the set of eventualities of Kati going home is a subset of the set of
those eventualities of Kati going home which are associated through the WHEN relation with an
eventuality of Péter preparing dinner, in other words, each eventuality of the former type is associated
with an eventuality of the latter type, which is more or less equivalent to the intuitive meaning of the
above sentence.
The question why atemporal readings of amikor ‘when’ clauses like (9) above are lacking in
Hungarian can now be answered in two possible ways. We could claim, on the one hand, that the
implicit generic operator can only bind eventuality variables, and, since both clauses of (9) contain
individual-level predicates, which (following Chierchia 1995) have bound eventuality arguments,
there would be no variable for the main generic operator to bind. Since, however, there are generic
sentences with individual-level predicates, like (20), where the generic operator necessarily binds an
individual variable, some extra stipulations would be needed to explain why this option in not
available for sentences with when clauses.
(20) Egy
A
gyerek
játékos.
child
playful.
‘A child is playful.’
The other possible solution, which we have already opted for in the course of defining the
relation in (17) above, is that the when clause is required to single out proper parts of the
lifetime of individuals about which the clause makes an assertion, that is, the corresponding relation is
not defined for states which denote an individual-level property.
WHEN
10
The format of the nuclear scope set in (1’) above is slightly reminiscent of the conservativity feature associated
with natural language determiners: Q AB <=> Q A (A B)
101
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
Formalizing the intepretation of ha ‘if’ clauses
Unfortunately, there are many interesting features of conditionals in natural languages which
cannot even be mentioned in the course of a discussion like the present one. In the following analyses
we will only concentrate on features of the semantics of ha ‘if’ clauses which are markedly different
from those of the corresponding amikor ‘when’ clauses, established above.
Here we will follow Kratzer (1986) in saying that sentences containing ha ‘if’ clauses are
epistemically modalized, that is, they express quantification over those possible worlds w’ which are
sufficiently similar to the actual world w in view of the evidence available. In this vein, sentence (2)
can be taken to mean that in all possible worlds which are sufficiently similar to the actual one, all
eventualities of János having a day off are associated with an eventuality of him washing the car. The
carrier of the quantificational information in the logical representation of the meaning of (2), given in
(2’) below, will be the generic operator GEN (Chierchia 1995). Since it is able to bind world,
eventuality and individual variables unselectively, and there is nothing else in the formula to bind the
eventuality variable, the generic operator will be assumed to bind both world and eventuality variables
here, and thus, denote quantification over pairs of worlds and eventualities.
The ‘association’ of the two eventualities described in the two clauses of the sentence will be
expressed with the help of the function M, Rothstein’s (1995) matching function, which maps
eventualities onto eventualities, or, more precisely, eventualities of the type satisfying the description
in the main clause onto those which satisfy the description in the subordinate clause. This matching
function will take care of the requirement that to make sentence (2) felicitous, each eventuality of
John having a day off has to be connected to a different eventuality of him washing the car.11
(2’) GENe, w’ {λe λw’ (have’ (János, day_off, e, w’))}
[λe λw’(have’(János, day_off, e, w’) ∧
∧ ∃e’ (wash’ (János, the car, e’, w’))) ∧ M(e’) = e))]
The subordinate clause of (6) above, repeated here as (21), contains an eventuality description
which can be satisfied by at most one eventuality in the actual world:
(21) János megmosta a
János washed
the
kocsit,
ha tegnap
car-ACC if
szabadnapja
volt.
yesterday day_off-3SGPOSS was
‘János washed the car yesterday if he had a day off.’
(21’) GENe, w’’{λw’ λe (have'(János, day_off, e, w’) ∧ f(e) ⊆ yesterday)}
[λw’ λe (have’ (János, day_off, e, w’) ∧ f(e) ⊆ yesterday ∧
∧ ∃e’ (wash’ (János, the car, e’, w’) ∧ M(e’) = e))]
11
If the domain and the range of the matching function was swapped in (2’), it would allow identical
eventualities of washing the car to be assigned to two eventualities of having a day off. In its present form, it
allows more car-washing to belong to a single eventuality of having a day off, which does not seem to contradict
intuitions.
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
102
(21’) means that the set of world-eventuality pairs where the world is sufficiently similar to the
actual one, and the eventuality of type János having a day off (whose runtime is part of yesterday), is
a subset of those world-eventuality pairs of the same type such that the eventuality is associated with
the help of the matching relation with an eventuality of János washing the car. In other words, in all
worlds of the required type, all eventualities of János having a day off are associated with an
eventuality of him washing the car.
The difference between the semantic interpretation of indicative conditionals like (21) above
and counterfactual ones like (8) above can then be captured by saying that in the latter case the set of
worlds w’ quantified over does not contain the actual world (von Fintel 1998). In other respects, the
logical representation of (8) would be identical to (21’) above.
It is perhaps worth mentioning that the generic operator in the logical representation of
conditionals can quantify over a single world variable, as it would do in the case of (22), which
contains an individual-level predicate, or a pair of a world and a individual variable, as in (23):
(22) Ha
If
János tud
angolul,
akkor
jól
tud.
János knows
English
then
well
knows
‘If János knows English, he knows it well.’
(23) Ha
If
egy
gyerek
tud
angolul, akkor jó
a
one
child
knows
English then
the pronunciation-3SGPOSS
good
kiejtése.
‘If a child knows English, he has a good pronunciation.’
This would close the discussion on the interpretation of ha ‘if’ clauses in quantified contexts. In
the Conclusion the main results of the study are summarized.
Conclusion
In this paper a formal way of representing the meaning of Hungarian complex sentences with
subordinate clauses introduced by the connectives amikor ‘when’ and ha ‘if’ was suggested in the
framework of Generalized Quantifier Theory.
Hungarian amikor ‘when’ and ha ‘if’ clauses were argued to provide the domain for
unrestricted quantification expressed by an adverb of quantification in the main clause of the sentence,
or, in the lack of it, a covert generic operator. It was claimed that sentences containing amikor ‘when’
clauses and ha ‘if’ clauses express quantification at least over an eventuality variable and a possible
world variable, respectively. Amikor ‘when’ clauses were claimed to be factive and, as opposed to
their English counterparts, strictly temporal.
References
Bach, Emmon. 1986. The algebra of events. Linguistics and Philosophy 9:5–16.
Bach, Emmon, Eloise Jelinek, Angelika Kratzer and Barbara Partee, eds. 1995. Quantification in natural
languages. Dordrecht: Kluwer.
Barwise, Jon, and Robin Cooper. 1981. Generalized quantifiers and natural language. Linguistics and
Philosophy 4:159–219.
Carlson, Greg N., and Pelletier, Francis J., eds. 1995. The generic book. Chicago and London: University of
Chicago Press.
103
On the Semantic Interpretation of amikor ‘when’ and ha ‘if’ Clauses in Hungarian
Chierchia, Gennaro. 1995. Individual-level predicates as inherent generics. In Carlson, and Pelletier, eds., 125–
175.
Davidson, Donald. 1967. The logical form of action sentences. In The Logic of Decision and Action, ed. N.
Rescher, 81–95. Pittsburgh: University of Pittsburgh Press.
Declerck, Renaat. 1988. Restrictive when-clauses. Linguistics and Philosophy 11:131–168.
Farkas, Donka, and Yoko Sugioka. 1983. Restrictive if/when clauses. Linguistics and Philosophy 6:225–258.
von Fintel, Kai. 1998. Indicative conditionals. Talk given at the Bridges and Interfaces Conference, March 12–
15, 1998, Prague.
Gyuris, Beáta (1998) Temporális kvantorok a magyarban. [Temporal quantifiers in Hungarian]. In A mai
magyar nyelv leírásának újabb módszerei III. [New Methods in the Description of Hungarian III], eds. L. Büky
and M. Maleczki, 41–58. Szeged: JATE.
Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. Doctoral dissertation, University of
Massachusetts, Amherst.
Heinämäki, Orvokki. 1974. Semantics of English temporal Connectives. Doctoral Dissertation, University of
Texas, Austin.
Johnston, Michael. 1994. The syntax and semantics of adverbial adjuncts. Doctoral dissertation, University of
California, Santa Cruz.
Kratzer, Angelika. 1986. Conditionals. In Papers from the Parasession on Pragmatics and Grammatical Theory,
eds. A. M. Farley, P. Farley, and K. E. McCollough, 1–15. Chicago: Chicago Linguistic Society.
Kratzer, Angelika. 1988. Stage-level and individual-level predicates. In Carlson and Pelletier, eds., 125–175.
Rothstein, Susan. 1995. Adverbial quantification over events. Natural Language Semantics 3:1–31.
De Swart, Henriëtte. 1991. Adverbs of quantification: A generalized quantifier approach. Doctoral dissertation,
Rijksuniversiteit Groningen.
Coordinate Ellipsis as Phonological non-Insertion
104
Coordinate Ellipsis as Phonological non-Insertion
Huba Bartos & Beáta Gyuris
The aim of this paper is, on the one hand, to carry further the analyses of Bartos (to appear), in
the domain of VP-type ellipsis phenomena, and on the other, to explore the licensing (interpretational)
strategies involved in ellipsis. In the course of the latter, we will argue that certain cases of ellipis
where semantic bridging effects are at play, and which have therefore been primarily analysed in
terms of semantic interpretation of syntactic null-forms, can (and in fact: must) be treated under the
general framework of syntactic ellipsis construed as the non-insertion of phonological material for the
formal feature matrices of the syntactic structure — as conceived in the model of Distributed
Morphology (DM) laid out by Halle & Marantz (1993, 1994), and Marantz (1997).
One of the major consequences of our findings is that the split lexicon model of DM gains
support, thereby corroborating the view of "the" lexicon as a cover term for a group (or groups) of
sublexicons (which, in turn, also consist of sub-sublexicons, or more precisely: lexical networks, see
e.g. Rebrus & Trón 2000, Bánréti 2000). In the DM model a "narrow lexicon" of semantic-syntactic
features bundles (i.e. morphemes) serves as the input for syntactic computation, the two outputs of
which are associated with two other "lexicons": (i) en route to phonology, the syntactically assembled
structures must be fleshed out with phonological forms (morphs) corresponding to the abstract,
featural morphemes — this involves the repository of such forms, termed the "Vocabulary"; (ii) in the
Logical Form as the output representation towards semantic interpretation the structural units must
acquire their individual, potentially idiomatic meanings — the "special meanings", to be identified
here, are stored in the so-called Encyclopedia.
Forward and backward ellipsis
Of the different types of ellipsis we will exclusively deal with those occurring in coordinate
structures. By forward ellipsis (FWE) we mean those cases where the ellipsis site is in a non-initial
conjunct, i.e. when the licensing antecedent precedes it. Accordingly, when the ellipsis site is in a
non-final conjunct, and the ellipsis site precedes the licensing antecedent overt in a later conjuct, we
have a case of backward ellipsis (BWE).
As discussed by Wilder (1997) for certain types of ellipsis (but crucially not for VP-type
ellipsis), FWE and BWE display strikingly different properties, the most salient and important of
which is the fact that in BWE the elided material must be literally (i.e. phonologically) identical to its
parallel antecedent in the later conjunct, while in the case of FWE different sorts and degrees of
"sloppiness" (i.e. non-identity) are possible. For example, in Gapping, the elided verb need not be
either phonologically or morphosyntactically identical to its antecedent — at least as fas as its
grammatical features are concerned:
(1)
John drinks wine and his kids [drink] coke.
3sg
non-3sg
However, as (2) shows, this sloppiness only holds for φ-features, but not for tense:
(2) * John arrives tomorrow, and his wife [arrived] yesterday.
In some other cases, the elided material in FWE is distinct from its antecedent in the former
conjunct, but in some well-defined way. E.g. in (3), the two instances (antecedent and elision) are
argument-structural variants of each other:
105
(3a)
Coordinate Ellipsis as Phonological non-Insertion
Iván szeretne Marival táncolni, de Mari nem szeretne [Ivánnal táncolni]/*[Marival
táncolni].
Ivan would-like Mary-with dance-inf, but Mary not would-like [Ivan-with dance-inf] /
[Mary-with dance-inf]
‘Ivan would like to dance with Mary, but Mary wouldn’t like to [dance with
Ivan] /*[dance with Mary].’
(3b)
Iván szeretne Marival táncolni, de Mari nem szeretne [táncolni].
Ivan would-like Mary-with dance-inf but Mary not would-like [dance-inf]
‘Ivan would like to dance with Mary, but Mary wouldn’t like to [dance].’
In (3a) the identity requirement between the antecedent and the elided material is observed in
such a way that the two occurrences of the same verb (táncolni ‘dance’) are argument-swapped
variants of each other. Likewise, in (3b) the elided verb is used as a one-place predicate, unlike its
two-place antecedent. Such differences are apparently within the limits of "content identity" between
antecedent and elided part.
But even greater differences are tolerated sometimes. For instance, in (4) the two related items
are different derivatives of the same root:12
(4)
Péter egykor kiváló operaénekes volt, de ma már nem megy neki [az operaéneklés],
mert tönkrementek a hangszálai.
Peter once excellent opera-singer was, but today already not goes to-him [the operasinging], because went-damaged the vocal_cords-poss-pl
‘Peter once was an excellent opera singer, but now he can’t [sing opera], because his
vocal cords have gone wrong.’
Nevertheless, even here it is not the case that "anything goes". The category of the items may
not vary:
(5) * Péter kiváló operaénekes volt, de ma már nem tud [(operát) énekelni], mert ...
Peter excellent opera-singer was, but today already not can [sing (opera)], because
‘Peter was an excellent opera singer, but now he can’t [sing (opera)], because ...’
The question obviously arises: how to account for these cases? What could be a meaningful
licensing condition for ellipsis which allows for such disparities between the antecedent and the elided
12
The exact construction with a nominal phrase missing cannot be reproduced in the English translation, so it
has been replaced with a verbal phrase elided.
106
Coordinate Ellipsis as Phonological non-Insertion
part, but which still rules out "blatant" differences? And: Which module of the grammar should be
made responsible for this?
Ellipsis as deletion vs. reconstruction vs. anaphora
In the literature, three major conceptions exist as to the true nature of what is perceived as
ellipsis, i.e. the "omission", the "absence" of certain elements from a syntactic unit. Perhaps the
earliest way to deal with such cases was in terms of deletion (of whole syntactic constituents, or
merely of their phonological features): in the generative literature some of the most important
proponents of this view have been Sag (1976), Lasnik (1995a, 1995b), and Wilder (1995, 1997).
Under these analyses, the elided items are base generated as proper, full fledged elements in the
syntactic structure, but are later subject to some sort of "removal" from there, so that in the end, they
will not be present (in some sense) in the output representation to be fed into phonology (i.e.
pronunciation). There are minor excutional differences as to whether whole branches are cut off of the
syntactic tree, or just the pronunciation of the items will be changed to null, or, as in Wilder (1997),
Bartos (to appear), the phonological material fails to get inserted into the structure entirely.
Another widely held view sees ellipsis as reconstruction, i.e. the copying (or generating) of
some lexical material into the place of empty elements — items which have been generated as null
forms, and left empty throughout the syntactic derivation, up to the point of the reconstructional
operation, at s-structure (Lappin 1996), or at LF (Williams 1977, May 1985, Fiengo & May 1994), or
maybe even outside syntax, in the domain of semantics (Tomioka 1997).
Eventually, it has been proposed that ellipsis (or at least ceartain cases of ellipsis) belongs to
the larger class of anaphora (Williams 1997), and thus should be treated in terms of the general
pattern and procedures of anaphora resolution. This approach is akin to the reconstructional one, but
also differs from it insofar as the classic "identity" requirements to hold between ellipsis and
antecedent are not evoked here in any form: ellipses are null forms which are interpreted as any other
anaphoric element.
Obviously, the latter two theories are more likely to be capable of handling the "non-indetical"
type of ellipsis, illustrated in (3, 4) above, than the deletional approach. On the other hand, an
identity-based deletional model is much more constrained, at first blush, not having to make reference
to somewhat hazy/fuzzy notions like "anaphoric equivalence" or "interpretive identity".
We will here adopt Wilder's (1997) generalization that BWE is really confined to literal
identity, while FWE ellipsis obeys looser requirements. At the same time, we propose that many
instances of FWE can (and in fact, should) be handled by the very same mechanisms and constraints
as BWE, though clearly there exist types where this approach is totally hopeless. In doing so, we
adopt the following heuristic idea: Since the "deletion under identity" mechanisms (to be made
precise below) are needed anyway (for BWE), they should be utilized to the maximum. i.e. to the
largest possible domain of FWE, too, whereas the less clear-cut, less easily definable (hence more
"dangerous") notions of "reconstruction/interpretation under semantic equivalence/compatibility"
should be confined to cases escaping the deletional treatment. In other words, let syntax and/or
morphology and/or phonology do as much of ellipsis as they can — in a Chomskyan framework these
computational modules apply prior to LF-operations or semantic interpretation. And the remaining
cases can be left to some reconstructional/interpretive machinery.
Naturally, for this model to work, we have to assume that all instances of ellipsis are deletional
at heart, otherwise we would have to allow empty elements to be base-generated (i.e. inserted directly
from the "narrow" lexicon), thereby fixing them for a reconstructional/interpretive treatment
beforehand — something we wish to exclude. We want to ensure that pre-LF/semantics mechanisms
take care of every case of ellipsis they can, leaving only the remainder to "wander on" to other, later
modules of the computational system.
Also, there is some evidence that coordinate FWE cannot be seen as wholesale semantic
reconstruction. As attested in (2) above, and demonstrated here in (6) for Hungarian, tense switches
107
Coordinate Ellipsis as Phonological non-Insertion
are strictly disallowed, even though some elements in the conjuncts (time adverbials) would make it
perfectly possible for the semantics to make the necessary alterations between the antecedent and the
elided V-form — in fact, no ambiguity could arise in these caes:
(6a) * Péter ma énekel, Laci pedig tegnap [énekelt].
Peter today sings Laci and yesterday [sang]
‘Peter is singing today, and Laci did yesterday.’
(6b)
Péter tegnap énekelt, Laci pedig ma [énekelt]/*[énekel].
Peter yesterday sang Laci and today [sang] / [ sings]
‘Peter sang yesterday, and Laci sang/*is singing today.’
As (6b) shows, when the elliptical string is well-formed at all, its only available interpretation is
where the tense specifications of the two conjuncts are matching.
Nor is it clear, why the lexical difference between the antecedent and the elided form in (4) can
be bridged (operaénekes 'opera singer' → operaéneklés 'opera singing'), while the same fails in (5):
operaénekes 'opera singer' –/→ operát énekel 'sing opera / opera-sing'.13 Before we may dare to leave
such matters to semantics, we would like to know how semantics works, e.g. in such cases — or we
should at least be certain that the non-interpretive ellipsis mechanisms are truly insufficient here.
BWE as phonological deletion — or morphological non-insertion
Wilder (1997) treats BWE as the deletion of phonological forms of elements in non-final
conjuncts under strict (phonological) form-identity. He distinguishes this from the situation in FWE,
where the absence of pronunciation for elements in non-initial conjuncts is the result of their
phonological shapes not being inserted at Morphological Structure (MS, Halle & Marantz 1993), i.e.
they are left without their pronunciations. Also, FWE must be licensed at LF — being subject to much
more complicated restrictions than mere form-identity, so Wilder applies a marker feature for these
items the "meaning" of which is that items merked with this must not be paired with their
phonological features at MS, and must be content-licensed ("reconstructed") at LF. While Wilder's
arguments and main ideas are quite convincing, we will not adopt his mechanisms, mainly for the
following two reasons: (i) the application of the marker feature is clearly an ad hoc means to achieve
the purported distinction; (ii) the distinction between phonological form non-insertion and
phonological deletion is far from obvious, i.e. one cannot make any theory-external or empirical
difference between the two. Whether an item fails to get its phonological features, or it gets them, but
loses them in the immediately subsequent step does not seem to be a meaningful distinction.
If we discount this latter, dubious, device, then what remains of Wilder's model is the necessity
of LF licensing for FWE, and the absence of any such condition for BWE.
Bartos (to appear) argues that simple cases of FWE (more precisely: forward VP-ellipsis) in
Hungaian arise by not inserting the phonological shapes of items which are licensed for this by the
general recoverability condition on ellipsis: If the content of the elided material is recoverable by
formal means, ellipsis is applicable. In the particular analyses it was shown that verb roots and non-
13
The bare N(P) object + V sequences have often been treated in Hungarian as lexical units, despite the overt
case marking on N(P), cf. Kiefer 1998 — if this is true, it is even more unclear, why/how this case differs from
the one in (4).
108
Coordinate Ellipsis as Phonological non-Insertion
null (i.e. non-infinitival) tense markers are recoverable only from parallel identical elements in
another conjunct, while agreement markers can be recovered more locally, via the agreement relation
with the anchor of agreement (subject, object), and infinitival, null tense markers are likewise licensed
for ellipsis without reference to a parallel clause, either because infinitives are in fact bare forms,
consisting of just the verb root, or because inifinitives are confined to contexts which explicitly select
for them, i.e. the selectional properties of the non-elided items help recover the infinitival nature of an
elided verb unambiguously.
If we compare FWE and BWE cases of the same type of VP-ellipsis (where the elided VPs are
minimal, containing just the verb itself, except in (10c)), we find the same general difference as in
Wilder (1997): BWE is good only under strict form-identity between the ellipsis site and the parallel
string:14
(7a)
Péter tegnap vásárolt, én pedig holnap fogok [vásárolni].
FWE
Peter yesterday shopped I and tomorrow will [shop]
‘Peter went shopping yesterday, and I will [go shopping] tomorrow.’
(7b) ?? Péter holnap fog [vásárolni], én pedig tegnap vásároltam.
BWE
Peter tomorrow will [shop] I and yesterday shopped
‘Peter will [go shoppping] tomorrow, and I went shopping yesterday.’
(8a)
Péter tegnap vásárolt, én pedig tegnapelőtt [vásároltam]. FWE
Peter yesterday shopped I and yesterday-before [shopped]
‘Peter went shopping yesterday, and I did [go shopping] the day before yesterday.’
(8b) ?? Péter tegnap [vásárolt], én pedig tegnapelőtt vásároltam. BWE
Peter yesterday [shopped] I and yesterday-before shopped
‘Peter [went shopping] yesterday, and I went shopping the day before yesterday.’
(9a)
Én tegnap vásároltam, Péter pedig tegnapelőtt [vásárolt].
FWE
I yesterday shopped Peter and yesterday-before [shopped]
‘I went shopping yesterday, and Peter did [go shopping] the day before yesterday.’
(9b) * Én tegnap [vásároltam], Péter pedig tegnapelőtt vásárolt.
BWE
I yesterday [shopped] Peter and yesterday-before shopped
‘I [went shopping] yesterday, and Peter went shopping the day before yesterday.’
14
In the glosses, def and indef mark the definite vs. indefinite object agreement on the verb.
109
Coordinate Ellipsis as Phonological non-Insertion
(10a) Péter két filmet látott, Mari pedig az összeset [látta].
FWE
Peter two movie saw-indef Mary and the all [saw-def]
‘Peter saw two movies, and Mary [saw] all.’
(10b)??Péter két filmet [látott], Mari pedig az összeset látta.
BWE
Peter two movie [saw-indef] Mary and the all saw-def
‘Peter [saw] two movies, and Mary saw all.’
(10c) * Péter két [filmet látott], Mari pedig az összes filmet látta.
BWE
Peter two [movie saw-indef] Mary and the all movie saw-def
‘Peter [saw] two [movies], and Mary saw all of the movies.’
If the same sorts of non-parallel licensing mechanisms were available as in FWE cases (viz.
agreement, selection), the (b) examples should be no worse than the (a) examples — which is not the
case. As the following examples show, there is nothing wrong with BWE itself in strict identity cases:
(11a) Péter tegnap [vásárolt], Mari pedig tegnapelőtt vásárolt.
Peter yesterday [shopped] Mari and yesterday-before shopped
‘Peter [shopped] yesterday, and Mary shopped the day before yesterday.’
(11b) Péter két filmet [látott], Mari pedig négyet látott.
Peter two movie [saw-indef] Mary and four saw-indef
‘Peter [saw] two movies, and Mary saw four.’
(11c) Péter két [filmet látott], Mari pedig négy filmet látott.
Peter two [movie saw-indef] Mary and four saw-indef
‘Peter [saw] two [movies], and Mary saw four movies.’
Thus we must conclude that BWE (as Wilder observed) is subject to the requirement of full
identity between the phonological shapes of licensor and licensee. However (contra Wilder) this need
not lead to the view that phonological deletion (rather than morphological non-insertion) is at play
here — we must simply ensure somehow that non-insertion of the phonological shapes cannot be
licensed by local relations (agreement or selection) in BWE, unlike in FWE. Of course, this can only
be stipulated now (just as Wilder has to stipulate deletion vs. non-insertion — something empirically
unconfirmable), but one may wish to find some deeper explanation for this.
Let us say then that all instances of BWE, and at least some instances of FWE arise by noninsertion at MS, albeit under different conditions. This difference may be traced back to the different
110
Coordinate Ellipsis as Phonological non-Insertion
nature of the two kinds of ellipsis: while FWE is anaphoric, BWE is not. There are both general, and
particular arguments for such a distinction.
Anaphoric dependencies fall under a general pattern, as adopted from Williams (1997:588):
(12)
General Pattern of Anaphoric Dependencies (GPAD)
Anaphoric dependencies are forward and/or downward.
By "forward" we mean that the antecedent/licensor must linearly precede the anaphor — in our
particular cases: the ellipsis site. "Downward", in turn, means that the antecedent must c-command the
anaphor, or at least the antecedent must be in a clause superordinate to the clause containing the the
anaphor. It follows then, that if the "antecedent" neither precedes, nor is higher (in the relevant sense
just given) than the alleged anaphor (which is exactly the case with BWE, provided we take the
identical string in the parallel conjunct to be the intended antecedent)15, then there can be no anaphoric
dependency established, the anaphora fails. In other words: it is not anaphoric. FWE, on the other
hand, obviously satisfies the GPAD-criterion by virtue of the fact that the antecedent precedes the
ellipsis site, by definition.
Secondly, there are BWE cases where no anaphoric relation is intended to hold, whereas similar
cases are ill-formed as FWE:
(13a) Tegnap Péter vett [egy könyvet]x, Mari pedig elovasott [egy könyvet]y.
x≠y
yesterday Peter bought [a book] Mary and read(past) [a book].
‘Yesterday Peter bought, and Mary read a book.’
(13b) *Tegnap Péter vett [egy könyvet]x, Mari pedig elolvasott [egy könyvet]y.
(14a) Péter egy tíz-[betűs szót nézett meg a szótárban], Mari pedig egy húszbetűs szót nézett
a szótárban.
meg
Peter a ten- [lettered word looked up the dictionary-in] Mary and a twenty-lettered word looked
up the dictionary-in
‘Peter [looked up] a ten-, and Mary looked up a twenty-letter word in the dictionary.’
(14b) *Péter egy tízbetűs szót nézett meg a szótárban, Mari pedig egy húsz-[betűs szót nézett
a szótárban].
meg
So we can establish that a key difference between BWE and FWE is that the latter, but not the
former, is anaphoric, hence the licensing conditions on the two are conceivably (or even probably)
different.
15
The failure of BWE falling under the GPAD is even stronger under a Kaynean asymmetric view of
coordination (Kayne 1994), where coordination follows the general X’-schema, the conjunctive element being
the X0 head, the first conjunct its specifier, and the second conjunct its complement. Thus the first conjunct
asymmetrically c-commands the second one, not leaving even the faintest chance for the second conjunct to
properly “antecede” anything in the first one.
Coordinate Ellipsis as Phonological non-Insertion
111
Eventually, there are arguments from the domain of the so-called "strict" vs. "sloppy"
pronominal readings that BWE cannot be analysed as LF-copying or LF-reconstruction. Compare the
following cases, w.r.t. pronoun interpretational possibilities:
(15a) Oszkárx szereti a prox/*y kutyáját, de Maxy nem [szereti a prox/y kutyáját].
Oscar likes the pro dog-poss but Max not [likes the pro dog-poss]
‘Oscar likes his dog, but Max doesn’t,’
(15b) Oszkárx nem [szereti a prox/*y kutyáját], de Maxy szereti a proy/*x kutyáját.
Oscar not [likes the pro dog-poss] but Max likes the pro dog-poss
‘Oscar doesn’t [like his dog], but Max likes his dog.’
If the BWE case (15b) was handled by LF-copying the parallel structure from the second clause
into the ellipsis site, then it would be surprising why the index 'y' cannot be copied over, since this is
exactly what happens in the FWE case (under a recosntructional approach): the strict-identity reading
of (15a) can arise only if pro is copied over with is index linked to that of 'Oszkár' — the subject of
the first, non-elliptical, clause. So obviously an LF-copy procedure for BWE would do the same in
(15b), giving rise to the strict reading there: 'Oscar does not like Max's dog'. But such a reading is
unavailable for (15b), which means that this BWE is not backward anaphora or backward copying at
LF. The straightforward solution is then to regard BWE as an effect on the PF-branch of the grammar
— morphological non-insertion.
Note, also, that morphological non-insertion is a sufficient means to handle the FWE
pronominal readings, too, as shown by Bartos (to appear): In this setting, (15a) is in fact two different
sentences, which fall together only on the surface, due to the ellipsis. In one sentence, pro in the
second conjunct bears the index of its local antecedent (Max), which is a universally available option
of local pronominal anaphora (provided the antecedent is outside the local domain of the pronoun,
which is probably the DP of the possessive construction, as ensured by any version of the binding
theory, e.g. Chomsky 1981). In the other sentence underlying (15a), the pro in the second clause bears
the index of the subject of the first clause (Oscar), which is also an unproblematic possibility.16 The
crux of the matter is ellipsis licensing, but anaphoric recoverability holds for both cases: whether pro
is linked to Max or Oscar, recoverability is ensured because both dependencies conform to the GPAD,
given in (12).17
16
Just like it would be possible for this pro to bear an index different from both Oscar's and Max's, i.e. referring
to a third, unmentioned party — so in fact (15a) covers not two but three sentences.
17
Note, incidentally, that the whole issue of pro-drop, i.e. the "silencing" of personal pronouns may fall under
the general notion of non-insertion, too: pro is none else than a featurally (fully) specified personal pronoun left
unpronounced, so at the levels of syntax preceding MS it is indistinguishable from "ordinary", pronounced
personal pronouns, i.e. the ones that are due to be paired with their sound forms. However, the anaphoric
relations governing "pro-drop" are different from those of proper ellipsis. For instance, pro-drop can be
backward and not downward without form identity:
(i) A prox kutyája csodálja Oszkártx.
Coordinate Ellipsis as Phonological non-Insertion
112
FWE as morphological non-insertion: The "meaning postulate" cases
As promised in section 2, we will show here how far the non-insertion analysis of ellipsis can
be pushed, to incorporate ellipses where there is no strict form-identity between the elided string and
the licensing one in a parallel conjunct.
Cases like (1) and (2), repeated here, are handled readily, as has been demonstrated by Bartos
(to appear): the φ-feature difference in (1) is licit because the V-root and the inflectional items (Tense,
Agr) are to be licensed for ellipsis separately, thus the root and tense-marking are licensed by the
corresponding identical items in the parallel conjunct, while Agr can be rightfully elided because it is
identified by the local agreement relation with the (plural) subject of the clause containing the ellipsis
site:
(1)
John drinks wine and his kids [drink] coke.
3sg
non-3sg
Things are different in (2): here the elided tense-marking is not properly identifiable, since it is
neither identical to is counterpart in the first clause, nor is it in any formal agreement relation with any
potential identifier in its local clause domain.
(2) * John arrives tomorrow, and his wife [arrived] yesterday.
Observe that the time adverbial ('yesterday') does not count as a proper licensor, even though it
has past reference, because there is no formal agreement between it and the tense-marker.18
Cases like (3, 4) — also repeated here — are more of a problem for a "simple", nonreconstructional theory of ellipsis:
(3a)
Iván szeretne Marival táncolni, de Mari nem szeretne [Ivánnal táncolni]/*[Marival
táncolni].
Ivan would-like Mary-with dance-inf, but Mary not would-like [Ivan-with dance-inf] /
[Mary-with dance-inf]
‘Ivan would like to dance with Mary, but Mary wouldn’t like to [dance with
Ivan] /*[dance with Mary].’
(3b)
Iván szeretne Marival táncolni, de Mari nem szeretne [táncolni].
Ivan would-like Mary-with dance-inf but Mary not would-like [dance-inf]
18
Although not clearly demonstrable in English, the lack of formal agreement between time adverbials and
tense-marking is clearly manifest in e.g. Hungarian, where a future time adverbial is compatible with past tense,
and vice versa:
(i) Péter a múlt héten azt ígérte, hogy tegnap már itthon lesz.
(ii) ?Péter a jövő héten azt fogja hazudni, hogy már holnap készen volt a munkával.
Coordinate Ellipsis as Phonological non-Insertion
113
‘Ivan would like to dance with Mary, but Mary wouldn’t like to [dance].’
(4)
Péter egykor kiváló operaénekes volt, de ma már nem megy neki [az operaéneklés],
mert tönkrementek a hangszálai.
Peter once excellent opera-singer was, but today already not goes to-him [the operasinging], because went-damaged the vocal_cords-poss-pl
‘Peter once was an excellent opera singer, but now he can’t [sing opera], because his
vocal cords have gone wrong.’
In these examples the elided part is not the exact copy of the licensing antcecedent — in fact,
an exact copy would lead to ungrammaticality in (3a, b), violating Principle C of the binding theory,
besides yielding an infelicitous interpretation: the elliptical sentences are definitely not understood
that way. Instead, these sentences are interpreted with a version of the antecedent manipulated in a
well-defined way, as pointed out by Gyuris (2000): in (3a), the elided V has an inverse argument
linking compared to its antecedent, while (3b) illustrates the truncation of the argument structure —
the elided V is a detransitivized variant of the antecedent V.
(16a) táncol(x, y-nal)
dance(x, with y)
Iván Mari
(16b) táncol(x, y-nal)
dance(x, with y) →
táncol(x, y-nal)
→
dance(x, with y)
Mari
Iván
táncol(x)
dance(x)
(17) is a further example illustrating the phenomenon of argument structure and argument
linking manipulations, though the variance possibilities are different:
(17)
Viki és Gabi szeretnének összeházasodni, de Gabi nem mer [összeházasodni Vikivel]/
*[összeházasodni], mert az apja utálja Vikit.
Vic and Gaby would-like-3pl together-marry but Gaby not dares [together-marry
Vic-with]/[together marry] because the father-poss hates Vic
‘Vic and Gaby would like to marry each other but Gaby doesn’t dare to [marry Vic] /
[get married] because her father hates Vic.’
Here the two argument roles of összeházasodik 'marry' are distrubuted between members of a
coordination in the antecedent clause, while they are separated in the elliptical one.
Gyuris (2000) proposes to capture the relations between the argument structure/linking variants
via meaning postulates (meaning equivalence, entailment), to be encoded in the lexicon. More
precisely, certain items in the "narrow" lexicon (e.g. táncol, összeházasodik) will be marked for being
affected by these postulated meaning relations. Put formally, for the verb táncol ‘dance’ there will
exist a Vinv, whose property is that its arguments will be swapped w.r.t. the basic V, and there will also
114
Coordinate Ellipsis as Phonological non-Insertion
be a Vdetr, whose property is that it lacks one (in fact, either one) of the arguments of the basic V.
Likewise, for összeházasodik ‘marry’ there will be a formal equivalence between a V with
asymmetric arguments: marry(x, with y), and a Vsymm with symmetrized arguments: marry(x and y).
With this technical device in our hands, we can simply rely on the non-insertion strategy for
this type of non-identical FWE. Recall that in this model all lexical items, including the to-be-elided
ones (in all cases of BWE, and in many cases of FWE), are present in the syntactic structure, and none
of them bears its phonological features before MS. So (3a, b) appear as (18a) and (18b), respectively.
(CAPITALs denote syntactic-sematic feature bundles without phonological forms.)
(18a) IVÁN SZERETNE MARIVAL TÁNCOLNI, DE MARI NEM SZERETNE IVÁNNAL
TÁNCOLNI.
Ivan would-like Mary-with dance-inf but Mary not would-like Ivan-with dance-inf
‘Ivan would like to dance with Mary, but Mary wouldn’t like to dance with Ivan.’
(18b) IVÁN SZERETNE MARIVAL TÁNCOLNI, DE MARI NEM SZERETNE
TÁNCOLNI.
Ivan would-like Mary-with dance-inf but Mary not would-like dance-inf
‘Ivan would like to dance with Mary, but Mary wouldn’t like to dance.’
Note that the impossible "reconstructions" (Mari nem szeretne Marival táncolni ‘Mary
wouldn’t like to dance with Mary’) cannot exist at all, not being well-formed syntactic constructions.
Now, when it must be determined at LF which items may undergo ellipsis, i.e. non-insertion of sound
forms, licensing conditions must be checked. The parallel antecedent clause licenses the ellipsis of the
infinitival clause on the following grounds: the verbs in the two conjuncts are identical except for
argument structure/linking, but they are equivalent in this respect, too, via the meaning postulates
lexically associated with them. Since the infinitival clause of the second conjunct in (18a, b) contains
the equivalent of the one in the first conjunct (i.e. the argument swap in (18a) and the argument loss in
(18b) are taken care of as recoverable), parallelity as a sufficient condition renders non-insertion licit
for the infinitival clauses. In a formalized fashion:
(19a) V licenses the ellipsis of a structurally parallel Vinv together with its arguments.
(19b) V licenses the ellipsis of a structurally parallel Vdetr.
This way we arrive at the well-formed (3a, b). (17) works in exactly the same way, modulo the
nature of the equivalence, encoded in the meaning postulate. Thus we have been able to derive one
type of ellipis licensing between non-identical items without recourse to the less well-defined notions
of LF-reconstruction and anaphor interpretation, by applying the independently needed non-insertion
strategy at MS.
Beyond meaning postulates? A further type
Let us return now to yet another type of non-identity FWE — that illustrated in (4) (repeated
here):
Coordinate Ellipsis as Phonological non-Insertion
115
(4)
Péter egykor kiváló operaénekes volt, de ma már nem megy neki [az operaéneklés],
mert tönkrementek a hangszálai.
Peter once excellent opera-singer was, but today already not goes to-him [the operasinging], because went-damaged the vocal_cords-poss-pl
‘Peter once was an excellent opera singer, but now he can’t [sing opera], because his
vocal cords have gone wrong.’
Here some relation is supposed to be established between the antecedent noun operaénekes
'opera singer' and the elided one operaéneklés 'opera singing', for the ellipsis to be recoverable, hence
licit. First note that it is not some very loose interpretive relationship, witness the fact that it cannot
cross categorial boundaries, as was shown by (5), and it has to be a relatively strict correspondence of
derivational patterns, as seen in (20):
(5) * Péter kiváló operaénekes volt, de ma már nem tud [(operát) énekelni], mert ...
Peter excellent opera-singer was, but today already not can [sing (opera)], because
‘Peter was an excellent opera singer, but now he can’t [sing (opera)], because ...’
(20a) Péter egykor kiváló kutató/autószerelő/orvos volt, de ma már nem megy neki [a(z)
kutatás/autószerelés/?orvoslás], mert ...
Peter once excellent researcher/car-mechanic/doctor was but today already not goes to-him [the
research/car-repairing/medicine] because
‘Peter once was an excellent researcher/car-mechanic/doctor, but now he can’t [do research/repair
cars/practice medicine] because…’
(20b) *Péter egykor kiváló tanár/lakatos/tudós volt, de ma már nem megy neki [a
tanítás/munka/???], mert ...
Peter once excellent teacher/locksmith/scientist was but today already not goes to-him [the
teaching/work/???] because
‘Peter once was an excellent teacher/locksmith/scientist but now he can’t [teach/work/???],
because…’
(20b) displays cases where no straightforward morphological connection exists between the two
relevant nouns, and the ellipsis fails. So we conclude that here some sort of lexical relational rule is
needed, much in the mood of meaning postulates, which states the equivalence (for purposes of
ellipsis recovery) of two derivatives of the same root, such as (21):
(21)
[[ Rootx] -Ó/-Os]N ↔ [[ Rootx] -Ás]N
This formulation is of course an oversimplification of the matter, but we do not wish to pursue
here this issue in depth — suffice it to say that it seems possible to formulate the exact nature of the
necessary relationship, to facilitate the handling of ellipsis by ensuring recoverability on grounds of
equivalence/entailment.
So when it comes to deciding, at MS, whether it is possible to leave the phonological forms of
the -Ás nouns uninserted, the morphologically related antecedents will act as proper licensors. To
preclude the proliferation of lexical marking, this regularity must obviously be encoded in the entry of
the -Ás suffix, the "common denominator" of the exponents of this pattern.
Coordinate Ellipsis as Phonological non-Insertion
116
As regards cases of FWE (and, for that matter, intersentential ellipsis, which, by its very nature,
cannot be handled by sentence grammar) where the antecedent and the elided part are even more
loosely connected, so that sometimes even semantic interpretation is at a loss when the elided content
has to be “recovered” and quite often pragmatics is called for, there seems to be no alternative to the
null-element analysis, i.e. these ellipses must be analysed as entirely (not just phonetically) empty
elements, appearing as such throughout the derivations, to be paired with semantically interpretable
content at/after LF, or possibly only in the semantic component, while syntax will always only “see”
them as void of any content (except, conceivably, categorial specification, or else their insertion into
the syntactic phrase markers would be unconstrained or impossible). Presumably, however,
anaphoricity is a property of these ellipsis types, too, so whantever conditions, rules, or mechanisms
apply to anaphors in general, are supposedly valid for these as well. Such cases include antecedent
contained ellipsis (see e.g. Fiengo & May 1994, Brody 1995, Hornstein 1995) — which possibly
involves the operation of ‘vehicle change’, a conversion between quantifiers and variables, or Rexpressions and pronouns, and this is obviously something totally outside the potential application
domain of our non-insertion analysis.
The true nature of the difference between BWE and FWE
To summarize the main points of our discussion of ellipsis types so far: On the basis of
directionality, we can distinguish between BWE and FWE, and on the basis of the nature of
emptiness, we can distinguish phonetic deficiency (PD; arising by non-insertion of sound forms) from
full formal deficiency (FFD; i.e. the presence of a null-element in the structure). BWE is always PD,
while FWE is divided into two subtypes: PD and FFD. The major properties of these types (cf. also
Wilder 1997) should be derivable:
BWE
- not necessarily a constituent
- must be contiguous and right-peripheral
- condition: total form identity with licensor (in a subsequent parallel conjunct)
PD FWE
- necessarily a (string of) constituent(s)
- must be contiguous
- condition: equivalence with (or derivability from) licensor (in a preceding conjunct)
FFD FWE
- purely anaphoric
- not necessarily identical to, equivalent with, or derivable from a parallel licensor
- necessarily a constituent (represented by a null-element in the syntactic structure)
The properties of FFD FWE follow directly from its nature as a null category node: as a
semantically empty category, it has to be in an anaphoric relation to be interpretable, but anaphora
does not necessarily involve form-equivalence.
Since BWE does not qualify as an anaphoric relation, not conforming to the GPAD (see (12)
above), the decision of insertion vs. non-insertion of sound forms at MS cannot depend anaphorically
117
Coordinate Ellipsis as Phonological non-Insertion
on the licensor, so the grammar has no other choice than checking for form identity, thereby making
content recovery possible relying on parallel identity in the phase of processing on behalf of the
addressee of communication. This form identity checking also precludes local, caluse-internal
recoverability licensing, perceived in FWE cases, so “sloppy identity” is not allowed in BWE. Right
peripherality must be stipulated, cf. the peripherality condition of Wilder (1997:92). The contiguity of
elided words follows now: if the alleged ellipsis site was discontinuous, then any of its segments other
than the rightmost one could not be right peripheral, in violation of the peripherality condition.
What remains needed now is an account of the properties of PD FWE. Recall that this is
phonological non-insertion, just like BWE, with an item-by-item licensing, falling under the general
notion of recoverability. Peripherality is not imposed by any condition. However, due to the anaphoric
potential of FWE, also present in these cases, not just with FFD FWE, recoverability licensing may be
both local and non-local — the former comprising selectional and agreement relations, as was seen in
(7a)–(10a). In other words: recoverability is based on formal features, rather than sound forms: if the
formal features of an item are identical to those of either a local relational licensor, or of a
corresponding item in a parallel structure, that item can be subject to phonological non-insertion, i.e.
ellipsis. The only thing our analysis remains lacking is an explanation for the obligatory contiguity of
PD FWE — which must therefore be stipulated.
Finally, with respect to the marginal acceptability of (7b), (8b) and (10b), we note that in BWE
the local licensing widely available for FWE can be evoked, too, although the result will be definitely
degraded grammatically. This suggests that the blocking on the application of these licensing
mechanisms (agreement, selection) in the case of BWE (as opposed to FWE) must be some very lowlevel condition, not any deep-rooted principle, as it can obviously be suspended under pressure from
the linguistic context. In other words, the strength of the non-elided material can facilitate the forceful
overlooking of the form-identity criterion, allowing for local licensing to step in its place.
Summary
In this paper we argued that all cases of ellipsis, whether backward or forward, are best
analysed as non-insertion of sound shapes at the level of morphology (in a late-insertion framework,
such as Halle & Marantz’s (1993) Distributed Morphology), the key difference between BWE and
FWE being that the latter is anaphoric, hence it allows for “sloppiness” by local identification of the
feature content of the elided material, while the former is based on the strict identity of sound forms
between the elided sequence and its licensor in a parallel conjunct, which is the only option, as BWE
is non-anaphoric. Furthermore, we have shown that many cases of FWE, seemingly escaping a
treatment in terms of identity, can be incorporated in our analyses by the involvement of lexical
equivalence/entailment relations encoded in the form of meaning postulates. This helps narrow down
the scope of interpretive and pragmatic ellipsis resolution to an inevitable minimum.
References
Bánréti, Zoltán. 2000. ‘Closed class lexical items in sentence processing. A neurolinguistic
approach.’ Submitted to Bartos, H. (ed.), Papers on the mental lexicon. Research Institute for Linguistics,
Budapest.
Bartos Huba. To appear. ‘VP-Ellipsis and Verbal Affixation in Hungarian.’ Acta Linguistica Hungarica.
Brody, Michael. 1995. Lexico-Logical Form. The MIT Press, Cambridge, MA.
Chomsky, Noam. 1981. Lectures on Government and Binding. Foris, Dordrecht.
Fiengo, Robert & Robert May. 1994. Indices and Identity. The MIT Press, Cambridge, MA.
Gyuris Beáta. 2000. ‘Semantic bridging effects in VP-ellipsis.’ Ms., to appear in: Bartos H. (ed.), Papers on the
Mental Lexicon, Research Institute for Linguistics, Budapest.
118
Coordinate Ellipsis as Phonological non-Insertion
Halle, Morris & Alec Marantz. 1993. ‘Distributed Morphology and the Pieces of Inflection.’ In: S.J. Keyser &
K. Hale (eds.), The View from Building 20: Essays in Honor of Sylvain Bromberger. The MIT Press,
Cambridge, MA., 111–176.
Halle, Morris & Alec Marantz. 1994. ‘Some Key Features of Distributed Morphology.’ MIT Working Papers in
Linguistics 21: 275–288.
Hornstein, Norbert. 1995. Logical Form. Blackwell, Oxford.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. The MIT Press, Cambridge, MA.
Kiefer Ferenc. 1998. ‘Alaktan. [Morphology]’ In: K. É.Kiss, F. Kiefer, P. Siptár: Új magyar nyelvtan [A new
Hungarian grammar], Osiris, Budapest.
Lappin, Shalom. 1996. ‘The Interpretation of Ellipsis.’ In: Sh. Lappin (ed.), The Handbook of Contemporary
Semantics. Blackwell, Oxford.
Lasnik, Howard. 1995a. ‘Verbal Morphology: Syntactic Structures Meets the Minimalist Program.’ In: H.
Campos & P. Kempchinsky (eds.), Evolution and Revolution in Linguistic Theory: Essays in Honor of Carlos
Otero. Georgetown Univ. Press, Washington, D.C., 251–275.
Lasnik, Howard. 1995b. ‘A Note on Pseudogapping.’ MIT Working Papers in Linguistics 27: 143–163.
Marantz, Alec. 1997. ‘No Escape form Syntax: Don’t Try Morphological Analysis in the Privacy of Yout Own
Lexicon.’ UPenn WPL. 4.2: 201–225.
May, Robert. 1985. Logical Form: Its Structure and Derivation. The MIT Press, Cambridge, MA.
Rebrus, Péter and Viktor Trón. 2000. ‘Constructional CV-phonology.’ Submitted to Bartos, H.
(ed.), Papers on the mental lexicon. Research Institute for Linguistics, Budapest.
Sag, Ivan A. 1976. Deletion and Logical Form. Doctoral diss., MIT.
Tomioka, Satoshi. 1997. Focusing Effects and NP-Interpretation in VP-Ellipsis. Doctoral diss., UMass,
Amherst.
Wilder, Chris. 1995. ‘Rightward Movement as Leftward Deletion.’ In: U. Lutz & J. Pafel (eds.), On Extraction
and Extraposition in German. John Benjamins, Amsterdam & Philadelphia. 273–309.
Wilder, Chris. 1997. ‘Some Properties of Ellipsis in Coordination.’ In: A. Alexiadou & T. A. Hall (eds.), Studies
on Universal Grammar and Typological Variation. John Benjamins,Amsterdam & Philadelphia. 59–107.
Williams, Edwin S. 1977. ‘Discourse and Logical Form.’ Linguistic Inquiry 8: 101–140.
Williams, Edwin. 1997. ‘Blocking and Anaphora.’ Linguistic Inquiry 28: 577–628.
Effect of verb complexity on agrammatic aphasic's sentence production
119
Effect of verb complexity on agrammatic aphasic's sentence production
Katalin Kiss
Introduction
In this paper we report on the investigation of those factors which have a role in the lexical
selection and activation of verbs in the sentence production of Hungarian agrammatic Broca's aphasic
speakers.
Verbs have a critical role in sentence planning processes since the lexical-semantic information
specified in the lexical entry of the verb determines the number and the relation of minimal
constituents of the sentence. Verbs are crucial to accounts of sentence production and syntactic
representation, because each verb carries syntactic well-formedness information which is represented
in the Mental Lexicon via the subcategorization frame and argument structure. Argument structure is
concerned with the number of participants described by the verb: e.g. transitive verbs select for two
arguments, they have two-place argument structure, other verbs entail three-place or one-place
argument structure. The arguments are those NPs, PPs to which thematic roles are assigned by the
verb. The thematic roles like Agent, Theme, Goal, Proposition are certain types of
semantic/conceptual information which define specific relationships between the verb and its
arguments (Jackendoff, 1990). The thematic roles must be assigned to the arguments which have to be
realized structurally in an appropriate syntactic position. Each verb selects its arguments and assigns
them thematic roles idiosyncratically. The argument structure pattern and argument types (externalinternal, direct-indirect) required by a verb are associated with a given construction type: the dstructure representation (Levin, 1993). During the lexical selection of a given verb the argument
structure representation, the d-structure representation and any other lexically selected material by the
verb, such as verb particles, obligatory adjunct and predicate adjectives, are also activated (Kegl,
1995).
The other type of syntactic well-formedness information that the verb carries is strict
subcategorization. Subcategorization determines the type of syntactic environment into which the
verb can enter. The syntactic categories of the complements (NP, S, PP) are assigned by the verb's
subcategorization frame.
The initial syntactic structure building processes see only the above-mentioned structurally
defined argument structure pattern, and manipulate only the structurally configured NP/DPs selected
by the argument structure and subcategorization pattern (Kegl, 1995). Assignment of thematic roles to
the arguments takes place only in a later processing stage.
Synactic realization of the arguments depends on the lexical properties of the verb, therefore
the choice of verb influences sentence production. Because strict subcategorization is relevant to
syntax (it imposes well-formedness conditions on a syntactic level) and semantic selection (argument
structure, thematic role information) is relevant both to semantics and syntax, the representational
complexity of a predicate has an effect on the generation of simple sentences in the performance of
agrammatic aphasics.
Agrammatic aphasics might select verbs of less complex representation structure more
succesfully because in this case, construction of phrase marker and monitoring of mapping of
arguments into the syntactic frame requires less computational load. Mapping procedures that are
responsible for transfer of information between different levels of representation (argument structure
pattern, d-structure representation and s-structure representation), can be disrupted if the verb has
more complex lexical-semantic structure such as
– more than one (external) argument
– some inner argument which can not be expressed syntactically or
120
Effect of verb complexity on agrammatic aphasic's sentence production
– alternating argument arrangement; and if the verb does not assign external argument (e.g.
unaccusative, passive verbs). In this case, the internal argument of the verb has to move from its
original d-structure position into the external argument position of the s-structure to get case (Kegl,
1995). Application of syntactic movement rule (NP movement type trarsformation) which displaces
an argument from its lexically specified position, increases the processing load of sentence formation
because assignment of thematic role for the displaced argument across syntactic chains, requires
greater memory capacity.
In the present study we will investigate the effect of verb complexity on the lexical selection of
predicates and on the semantic and syntactic mapping processes.
Agrammatic sentence production
Nonfluent Broca's aphasia is associated with damage to the left frontal lobe of the brain: the
lesion usually involves the Broca's area (a region in the third frontal convolution) but adjacent and
deeper areas of the anterior part of the brain are also affected (Naeser et al., 1989). Based on several
systematic studies of Broca's aphasic language deficit, it was recognized that Broca's aphasia involves
a range of linguistic impairments in both language production and comprehension. Verbal output of
Brocas's aphasics can be characterized by a variety of symptoms: articulatory disturbances, phonemic
and semantic paraphasia, nonfluent, reduced, effortful speech, "telegraphic" style. The spontaneous
speech is syntactically limited, deficient, "agrammatic". The most striking feature of agrammatism is
the dysfunction of the ability to construct syntactic structures. Agrammatic patients often produce
fragments: incomplete minimal phrases (DP, PP, VP) and sentences (IP/S) from which complements
(obligatory and implicit arguments), modifiers and often the verb itself are lacking. Assignment of
superficial sentence features is also disturbed: tense and agreement morphemes, case markers of
arguments (case suffixes, prepositions or determiners), verb inflexional endings and function words
(determiners, prefixes, prepositions) are often omitted or substituted. The patients often produce only
automatic, idiomatic phrases, speech panels with low productivity, isolate nouns or adverbs.
Agrammatic aphasics produce primarily simple SV, SVO structures. Complex sentences in which the
order of thematic roles (argument NPs) is noncanonical, e.g. passives, wh-questions, object relatives,
embedded clauses, are rarely used.
Patterns of verb production
Agrammatic aphasics show a marked deficit in naming actions; verbs are underrepresented in
their spontaneous speech, lexical verbs and auxiliaries are equally affected. Agrammatic Broca's
aphasics produce relatively small number of verbs in spontaneous speech (Saffran, et al., 1989).
Analysis of the spontaneous speech samples of Hungarian aphasics shows that the proportion of
lexical verbs used by the aphasic patients is much lower in the nonfluent, agrammatic group (17%)
than in the fluent Anomic (40%) and normal control group (43%) (Kiss, 1994 unpublished).
Bastiaanse and Jonkers (1998) found that Dutch agrammatic aphasics produce less inflected
verbs in their spontaneous speech than the normal control subjects (50% vs. 90%).
Friedmann and Grodzinsky (1997) found that in Hebrew inflection for Tense is more impaired
than inflection for Agreement and the production of copula is often difficult in sentence context.
Results from picture naming tasks demonstrate that verb retrieval is significantly more impaired
than noun retrieval in agrammatic aphasics, whereas Anomic aphasics show the opposite pattern
(Miceli, Silveri, Villa and Caramazza, 1984; Zingeser and Berndt, 1990; Marshall, Pring and Chiat,
1993). Some agrammatic Broca's aphasics show consistent selective deficit for verbs; verb-noun
dissociation can occur in their verbal output. Others did not found differencies between agrammatics
and anomics in action naming, both aphasic groups' performance was better in naming objects than in
naming actions (Williams and Canter, 1987).
The data indicates that noun and verb information may be stored separately in the Mental
Lexicon and is processed by distinct or different output systems. The reduced proportion of verbs in
121
Effect of verb complexity on agrammatic aphasic's sentence production
spontaneous speech of Broca's aphasics raises the question whether the difficulty of producing main
verbs and grammatical markers (closed class vocabulary: verb inflection, case markers of argument
DPs) is related or not? Does the tendency to omit main verbs imply lexical or rather a syntactic
processing deficit in agrammatism (Miceli et al., 1984)?
There is evidence in the literature which indicates that lexical selection of certain types of
predicates is more vulnerable than others, because the patients misselect a special type of information
represented in the conceptual-semantic structure of the verb. Jones (1984) found that understanding
and producing of sentences that contain directional motion verbs (follow, push) is more difficult for
the agrammatic patients than processing of sentences containing nondirectional motion verbs (throw,
visit) or nonmotion verbs (bandage, hold). Although the argument-structure complexity of these verbs
is the same (all verbs were reversible transitive verbs), directional motion verbs are more complex
because they encode an embedded preposition in their meaning representation which specifies the
spatial-directional relationship of the arguments. Selectional disorder of the inner preposition results
in argument misordering or accessing the wrong lexical entry: the converse of the target verb. Word
order deficit, argument reversing in a NP-V-NP clause involving directional motion verbs arises from
lexical-semantic, rather than syntactic, reasons. Jones (1984) claimed that the presence of an
embedded preposition increases the processing load of verb retrieval.
Processing load increases during the activation of those verbs as well, which allow more
subcategorization or argument-structure possibilities. Reaction time analyses, results of on-line cross
modal priming tasks indicate that the amount of representation information, subcategorization and
argument structure complexity of verbs, affects sentence processing (Shapiro, Zurif and Grimshaw,
1987). In the case of verbs that allow more semantic types (four argument-structures: e.g. remember
vs. two argument-structures: e.g. accept) or assign more arguments (transitive vs. dative verbs) in
their lexical entries, longer reaction times were observed during the activation of representation
information of verbs. Verbs allowing only one argument structure arrangement (transitive and
obligatory three-place verbs) did not differ significantly in contrast with verbs that allowed both a
two-place and an optional three-place argument structure (alternating and nonalternating dative). The
authors concluded that the relevant verb complexity metric for the sentence processing involves the
argument structure of the verb and not the syntactic subcategorization frame, and that the crucial
metric for sentence processing complexity is the number of possible argument-structure
arrangement and not the number of arguments within a given argument arrangement. Shapiro and
Levin (1990) found that neorologically healthy subjects and agrammatic aphasics showed the same
reaction time patterns in the above mentioned cross modal priming probe. Based on the results the
authors claim that the device that activates the verb and its structural properties operates normally
during sentence comprehension in Broca's aphasics. The difficulties in understanding and producing
complex structures (passive, relative clauses) thus arise from the dysfunction of the postactivation
processing which is responsible for the assignment of thematic roles to argument NPs.
Zurif et al. (1993) –using lexical priming probe–, found that Broca’s aphasics are sensitive to a
prime-target relation which means that they are able to access the full representation of the verb’s
argument structure. At the same time, the patients showed slower accessing rate than the normal
control group which means that the lexical activation of verbs are temporally protracted.
Some researchers assume that Broca’s aphasics are more successful in producing less complex
verbs: verbs with more, rather than fewer arguments are more difficult to activate. According to the
data of Byng (1988) English agrammatic aphasics used fewer complex verbs. Others found that the
number of arguments does not necessarily have an effect on the selection of verbs. Jonkers and
Bastiaanse (1996) found that their Dutch agrammatic aphasics named transitive verbs easier than
intransitive verbs.
Analysis of the production patterns of agrammatic spontaneous speech shows that agrammatic
patients tend to use complex verbs (which allow more argument structure arrangements or optional
argument) in their simplest form: if the verb allows sentential (S) and NP complements, the
122
Effect of verb complexity on agrammatic aphasic's sentence production
agrammatic aphasics prefer to build the NP complement into the sentence; if the verb selects an
implicit argument, the patients usually do not express it syntactically, they omit the non obligatory
argument from the syntactic structure (Thompson et al. 1995, Thompson et. al., 1997).
Bastiaanse and Jonkers (1998) found that Dutch Broca's aphasics who were relatively good at
verb retrieval in spontaneous speech, were relatively poor at producing finite clauses, the opposit
pattern was also found. The authors interpreted the data as a consequence of a 'trade-off' effect: the
patients produced finite clauses at the cost of lexical retrieval. According to their explanation the
phenomenon is the result of a syntactic disorder: the patients have difficulty to produce finite clauses.
To test this hypothesis, Bastiaanse and Van Zonneveld (1998) investigated the relation between
verb inflection and verb position. Analysis of the spontaneous speech samples of Dutch agrammatic
aphasics showed that in the matrix clauses the nonfinite verbs produced by the patients were always in
clause final position and the finite verbs in second position. In the embedded clauses (which occured
in very low proportion) the finite verbs also were in clause final position. On the basis of the results
the authors suggested that agrammatic Broca's aphasics are sensitive to the relation between finitness
and verb position, although the patients produce high number of nonfinite clauses. If they produce
finite verbs, the verb moves to the Verb Second position; if the verb is not inflected for Agreement
and Tense, the nonfinite verb remains in its base-generated (clause final) position. In a sentence
completion test an interesting pattern was found: the patients performed significantly better in
producing verb inflection if the inflected verb occured in clause final position. If the finite verb had to
be produced in Verb2 position in the matrix clause, the patients made significantly more inflectional
errors. Bastiaanse and Van Zonneveld concluded that "producing finite verbs that have been moved
from their base generated position is difficult for the agrammatic aphasics" (p.179.). The verb
inflection problem in production is not due to morphological disorder, but disturbance of verb
movement.
The above data indicates that the problem of verbs in agrammatic speech production can not be
explained only from one aspect. Complexity of the meaning representation (conceptual-semantic
description of the verb) and complexity of the argument-structure, subcategorization frame of the
predicate have an effect on successful verb retrieval. In the first case semantic selection deficit of
special information, the semantic feature of inner preposition, can be disturbed. The primary deficit is
lexico-semantic that causes a syntactic mapping disorder, reversing of arguments in the surface
structure. Preferring of the most simple argument structure of a given verb, at the same time, refers to
the restricted capacity to construct more complex sentence structure, and not to the lexical activation
problem of verbs. Agrammatic aphasics use shorter and less complex sentences than normal
speakers, they produce verb inflexion errors in their spontaneous speech, they use nonfinite verb [tense] forms –infinitive, participle–, instead of finite [+tense] verbs. These phenoma can be regarded
as a sign of adaptation to the limited memory capacity of the syntactic processsor. Agrammatic
patients avoid to use verbs assigning more complex argument structure in their lexical representation,
because of the primary disorder of syntactic structure building operations. Syntactic mapping of fewer
arguments into the sentence structure frame might be more successful when the capacity of
computational system is restricted.
The present study
The study based on the results of two experiments. Both experiments focus on the production of
verbs in Broca's aphasic speakers. We analysed the semantic and syntactic factors that influence the
lexical selection of verbs from the mental lexicon.
– In the Experiment I we investigated the role of representational complexity of predicates on
the lexical retrieval. As an elicitation method we used picture naming ('action description') test. We
compared the proportion of successfully selected verbs belonging to different types: the investigated
verbs differed in their argument structure complexities and in their morphological complexities.
Effect of verb complexity on agrammatic aphasic's sentence production
123
– In Experiment II we investigated the activation of complex predicates: [Verbal
modifier+Verb], in sentences with different structural complexites. Production pattern of neutral SVO
and non-neutral sentences were analysed. The question was whether the syntactic complexity of
sentences had an effect on accessing complex verbs or verbal modifiers.
Experiment I
The structure of the verbs used in Experiment I
Based on their argument structure complexities, the tested verbs formed three main groups:
one-place (I), two-place (II), three-place (III) predicates. Each main verb group is represented by
several subgroups.
Group I involves one-place intransitive predicates which take only one Agent or Experient
argument.
– Group I/A contains morphologically and semantically simple one-place verbs, e.g. alszik
(sleeps), ásít (yawns).
– Group I/B contains one-place reflexive verbs (R), e.g. borotválkozik (shaves oneself),
zuhanyozik (takes a shower), vakarózik (scratches oneself), nyújtózkodik (stretches oneself). These
types of reflexives in Hungarian, assign an 'inner' Patient (Theme) argument to which syntactic
function does not belong, it cannot be mapped into an overt object NP. The inner Patient argument is
identified by the semantic representation of the predicate, the Patient is identical with the Agent
(Komlósy 1994). e.g.:
1. fésülködik
semantic representation:x
combs oneself
fésüli
y-t
x=y
x-nom combs y-acc
Reflexives are morphologically complex because they are derived from a verb stem by
reflexive suffix. These verbs are semantically also more complex than the verbs in Group A, because
they assign inner semantic argument.
– Group I/C contains one-place verbs which are derived from a noun by a denominal
derivational suffix (N >V) e.g. teniszezik (plays tennis) síel (skies), csónakázik (boats)
The derived predicates in Group C contain an 'atomic predicate' in their semantic
representation, like 'MOVE/GO', 'PLAY', 'USE', which assign the original noun stems as argument.
Because of word formation/derivation these complements become 'inner' semantic arguments, they
are not expressed syntactically but are identified by the semantic representation of the derived
predicate (Jackendoff 1987; Komlósy, A. 1994). The derived verb 'gitározik' for example, has the
following semantic representation:
2. gitározik
' plays the guitar'
Morphological form:
gitár
N stem
– oz-ik
V derivational suffix-3sg. suffix
Semantic representation:
JÁTSZIK
a
gitár-on
PLAY-3sg
the
guitar-on
[atomic predicate]
[inner argument ]
`(he) plays the guitar`
Effect of verb complexity on agrammatic aphasic's sentence production
124
Based on their lexical representation, verbs belonging this type are semanticlly and
morphologically more complex than the verbs in Group A.
Group II involves two-place verbs, this group contained three subgroups :
– Group II/A contains reversible transitive predicates: the Theme argument is mapped to an
object NP in which the noun is specified as [+animate].
3.
megvigasztal (`comforts`) [NPnom., NPacc.]
A
férfi
meg|vigasztal-ja
a
lány-t.
the
man-nom.
prefix-comfort-3sg.def.
the
girl-acc.
`The man comforts the girl.`
– Group II/B contains irreversible transitive verbs: there is animacy contrast between the
Agent and Patient/Theme arguments, the Theme/Patient-object argument has [-animate] semantic
feature.
4.
hámoz
(`peels`)
[NPnom, NPacc]
A
fiú
meg|hámoz-za
a
banán-t.
the
boy-nom.
prefix-peel-3sg.def.
the
banana-acc
`The boy peels the banana.`
– Group II/C contains predicates that assign an Agent thematic role to the subject and
Goal/Source/Location thematic role to their locative arguments.
5.
átmegy (`cross`)
[NPnom, NPsuperess.]
A
férfi
át|megy-∅
the
man/nom
across (prefix)-go-3sg. the
az
úttest-en
road-on
`The man is crossing the road.`
Group III involves three-place predicates. Group III contains two subgroups:
– Group III/A contains verbs that assign Agent-Theme-Goal/Source/Location thematic roles
which have to be mapped into the subject, object and locative Nps.
6. beletesz (`put into`)
[NPnom, NPacc, NP illative]
A
férfi
be|tesz-i
a
the
man-nom.
into (prefix)-put-3sg.def. the
bôrönd-öt
az autó-ba.
suitcase-acc.
the car-into
`The man puts the suitcase into the car.`
– Group III/B contains verbs that assign Agent-Theme-Benefactive/Goal thematic roles,
requiring subject, object and dative complements.
7.
bemutat (`introduce`)
[NPnom., NPacc., NP dat.]
A
a
fiú
be|mutat-ja
lány-t
a
the boy-nom. into (prefix)-introduce-3sg.def. the
`The boy introduces the girl to his friend.`
barát-já-nak.
girl-acc
the friend-gen.3sg-dat
125
Effect of verb complexity on agrammatic aphasic's sentence production
Method
Subjects. Two agrammatic Broca`s aphasic patients were investigated. Both patients are native
speakers of Hungarian.
A.N. 55-yr-old, female, right-handed nurse (education: secondary school) was hospitalized
again 20 months post-onset when deterioration was observed in her condition, as a result of a second
cerebrovascular insult.
L.I. 37-yr-old, female, right-handed radiological assistant (education: specialized matriculation
examination) was investigated 36 month post-onset.
Both patients suffered left lateral cerebrovascular accident, both had right hemiparesis. CT scan
of A.N. showed previous parieto-occipital involvement and frontal hypodensity which was the sign of
a second, acute vascular lesion. L.I.`s CT scan showed large temporo-parieto-occipital hypodensity
which signalled a middle cerebral artery and posterior artery infarct. Classification of aphasia type
was made by the Hungarian Variant of the Western Aphasia Battery (Osman-Sági 1991). A.N.`s
WAB AQ:49; L.I.`s WAB AQ: 65,4.
A.N.`s spontaneous speech showed reduced fluency, telegraphic style and severe speech
initiation difficulty, it hardly contained intact propositions or sentences. Her speech consisted of
automatic, stereotyped phrases, isolated words and paraphasias. After a 3 months therapy period she
showed some improvement according to the Western test (WAB AQ: 54,6), but the character of her
spontaneous speech did not really change. She was tested with the 'action description test' at the
beginning of therapy and 3 months later. In the present study we analysed her answers given in both
of the 'action naming tests'.
L.I.`s spontaneous speech was also nonfluent, characterized by marked anomia and agrammatic
symptoms. She often used more automatic speech `panels` or expressions, e.g. "I knew that
something is wrong", "I am fed up", "It was evening when it happened", "I could not do anything".
She filled in the hesitation gaps caused by word finding difficulties with these grammatically wellformed phrases and with some adverbs. She was able to use active non-stereotype simple clauses but
she had difficulty producing complex sentences. She produced many incomplete phrases (VPs and
DPs as well) or sentences (with many often incomplete subordinate clauses). Semantic selection and
access problem of the 'lexical verbs' was an outstanding symptom in her speech; the type/token ratio
of verbs was relatively low in her spontaneous speech, she retrieved mostly modal verbs (e.g. must,
can, ought, might) and she often used the verb 'know'. Omission and morphological errors of verb
inflection and case marking of nouns were also observed.
Material. The test material contained 124 target verbs which belonged to 8 verb types (Group I
A-C, II A-C, III A-B, as described above). Our elicitation method was an 'action naming' /'picture
description' test. The pictures that represented the target verbs/actions were assembled from
Jacqueline Stark:"Everyday life Activities (photo series)" collection (Stark 1992) Each coloured
photocard of the series represents one particular action.
Procedure. The subjects were tested individually, the photocards were randomly presented one
after another. The patients were instructed to describe or tell us 'what happened in the picture', who
the actors were, what they were doing. (The subjects were asked to imagine desribing the picture to
someone is also in the room who does not see the picture, therefore they should try to give as precise a
desription as possible or as they can.) Responses were tape recorded and transcribed. General help
like `Could you tell me anything more?`, phonemic cues (whispering of first syllable), questions
refering to an argument noun or supplying of an argument noun were used only when the patients
asked for help. In the statistic analysis we used only the independent, spontaneous answers.
Effect of verb complexity on agrammatic aphasic's sentence production
126
Results
Table 1. shows the proportions of the verbs produced by the agrammatic aphasics according to
different types of predicates and the distribution of the complete, grammatically well-formed
sentences involving the target verbs.
We regarded an answer to be 'complete' if the patients were able to build the whole VP or
sentence. It means that the verb and its complements were lexically accessible, the argument NPs
were supplied with the appropriate overt case marker, noun-verb agreement was intact. If an
inflectional or a derivational ending or a determiner was omitted or any argument was missing we did
not accept the incomplete clause as a "correct" response even if the verb retrieval itself was successful
e.g. *Nem kártya. Fiúnak odaadta. (Not card-nom. Boy-dat. prefix-give-3sg.def.past) *Önteni önteni
a limonádét. (To pour to pour the lemonade-acc). *Doktornô, doktor megvizsgálja. (Doctoress-nom,
doctor-nom examine-3sg.def.).
Since Hungarian is a pro-drop language, the subject pronoun may be left unexpressed. If there
is no overt subject NP in the sentence the Agent thematic role is assigned to a phonetically empty
pronoun (pro). The verb inflection refers to the number and person of the subject (in the case of the
target verbs it was the 3. person singular form). We accepted those answers as 'complete clauses' in
which the aphasics did not assign an overt subject NP but used the right inflected verb form e.g.
'Felébreszti a fiút' ((pro) wake-3sg.def. up the boy-acc.).
Predicates 'precisely' described the action represented in the given picture constitute the target
verbs. Verb substitutions containe irrelevant responses or verbs which do not 'exactly' express the
given action although they are adequate to the situational context e.g.
target verb: The man goes upstairs. > substituted verb: "he is sitting"
target verb:The boy wipes the milk. > substituted verb: "he spills"
arget verb: The girl wakes up the boy. > substituted verb: "he is sleeping"
target verb: The man kisses the women. > substituted verb: "he loves her "
target verb: The man puts the suitcase into the car. > substituted verb: "mum packs up"
Effect of verb complexity on agrammatic aphasic's sentence production
127
Table 1. - Distribution of complete clauses and verbs (percentage values are enclosed
in parentheses)
Three-place
I/A
simple
One-place
Two-place
I/B
I/C
II/A
II/B
II/C
III/A
III/B
reflex.
N>V
[+anim]
[-anim]
locative
locative
dative
12
9
21
15
35
11
13
8
36
27
63
45
105
33
39
24
26
8
16
2
5
0
1
3
clauses
(72.2)
(29.6)
(25.3)
(4.4)
(4.7)
(0)
(2.5)
(12.5)
Mean score
(39.6)
Number of
target verbs
Total no. of
responses
Complete
(3.8)
(6.3)
of complete
clauses
26
8
16
11
24
0
4
4
(72.2)
(29.6)
(25.3)
(24.4)
(22.8)
(0)
(10.2)
(16.6)
Verb
substitution
5
7
6
29
41
25
20
15
Total verb
31
15
22
40
65
25
24
19
responses
(86.1)
(55.5)
(31.7)
(88.8)
(61.9)
(75.9)
(61.5)
(79.1)
Target verb
Mean of total
verb
(54.8)
(71.0)
(66.6)
responses
The analysis of the data shows that the mean scores of the complete clauses were much higher
in the 'one-place' verb group (39.6) than with the 'two-place' (3.8) or 'three-place' (6.3) predicates. In
the latter two groups the ratios of complete clauses were very low.
Not surprisingly, constructing a surface sentence which involved a one-place predicate was
easier for the agrammatic aphasics than constructing a syntactic structure which contained a two or
three-place verb. If the patient was able to access the one-place predicate she could also construct the
whole VP. In the case of the two and three place predicates however, retrieval of the verb`s phonetic
form did not mean a simultaneously successful syntactic structure building. The construction of the
surface sentence was perfect for roughly one quarter of those answers which contained the target verb.
This phenomenon shows a mapping disorder, the patients are limited in the 'monitoring' of the
assignment of multiple arguments into the appropriate slots of the syntactic frame.
Effect of verb complexity on agrammatic aphasic's sentence production
128
Comparing the distribution of the retrievable target verbs within each verb group we found the
following 'verb difficulty order':
simple one-place > morphologically complex one-place (R and N>V) =
transitive (two-place) > three-place (with locative and dative complement) >
two-place with locative complement
Access to the 'simple one-place' verbs was outstandingly succesful (72.2).
The lexical selection of the 'two-place verbs with locative complement' proved to be the most
difficult for the patients (they could not retrieve any verb in this group). These predicates were
directional motion verbs. The lexical representations of these verbs integrate mental knowledge
related to the cognitive representation of space or spatial relations. These verbs include such contents
as direction of the motion, place-coordinates, starting point and end point. This information is
encoded in the semantic representation and thematic roles of the predicate. Processing of this
information seemed to be more difficult for our patients, they produced marked selection disorder
when attempting to produce these verbs.
We did not find outstanding differences in the proportion of activated verbs among the 'oneplace reflexive' (29.6), 'one-place derived from noun' (25.3) and 'two-place transitive' (24.4; 22.8)
verb groups. Proportions of these verbs were lower in contrast to the 'simple one-place verb' group.
The ratios of the three-place verbs (10.2;16.6) were lower than the proportions of the one-place
and two-place verbs (except the '2-place locative' group).
The results show that the 'representational complexity' of the predicate has a direct effect on the
lexical accessibility of the verb for agrammatic aphasics. The argument-structure complexity of the
verb (number of obligatory arguments) plays an important role in verb retrieval but it is not the only
factor. The morphological and semantic representational complexity of the 'one-place derived verbs'
(I/B, I/C) and the semantic representational complexity of the 'two-place locative' verbs also had an
effect on the lexical-semantic selection of the predicates.
Effects of morphological complexity
In the Group I/C two types of answers were found which were related to the morphological
structure of the verb:
a.
noun stem (only)
b.
verbs:
– noun stem > target verb derivation
– target verb
– verb substitution
Table 2. - Proportion of answers belonging to the different types
N stem
22.22
N stem > targetV
30.15
target V*
44.44
substituted V
3,17
Effect of verb complexity on agrammatic aphasic's sentence production
129
(* number of the target verb answers is higher here than in Table 1. because those answers
which were not accepted as 'correct ' were also considered during `error analysis` that is answers
given by phonemic cue, infinitives, morphologically ill-formed forms)
As Table 2. indicates, the noun stem appeared in more than half of the answers, which means
that the noun stem was activated first in many cases and it was used as `access code` or `mediator` in
the retrieval process of the derived verb form, if the direct semantic access to the verb was not
possible. The patients used an indirect retrieval method for the predicate by activating morphological
processes/word formation rules. The noun stem was assigned first both in cases where the stem and
derived form were regular and in those where they were not, e.g. ló > lovagol (`horse > ride the
horse`) This strategy was found both in independent answers and in answers given by phonemic cue.
The data show that morphological complexity of the verb is a decisive representational factor
that has an effect on the accessibility of the verb.
Types of answers
Agrammatic aphasics gave different types of answers in the two- and three-place verb groups.
Three kinds of answer types can be distinguished.
Type A.
Isolated argument/s – activation of one argument or list of arguments (N or
DP, case marked or caseless forms)
Type B.
Argument/s assignment > followed by Verb selection / Clausal answer
(involving the target or substituted/paraphasia verbs which were one-place or two-place
predicates)
Type C.
Clausal answers (no previous argument N or DP assignment)
Type D.
Other: e.g. 'I dont know' answer, noun associations. (In the further analysis these
answers were not included.) These answers were rare 12/372 total answers.
130
Effect of verb complexity on agrammatic aphasic's sentence production
Table 3. - Distribution of answers in the 'A-C' types and distribution of target and
substituted verbs in clauses
II/A
II/B
II/C
III/A
III/B
2-place
[+animate]
2-place
2-place
3-place
3-place
[-animate]
(locative)
(locative)
(dative)
Total no. of answers
45
96
30
38
23
Total no. of clausal answers
40
65
25
24
19
Type A
11.1
32.2
16.6
36.8
17.3
Type B
28.8
39.5
43.3
52.6
43.4
Type C
60.0
28.1
40.0
10.5
39.1
Target verb
27.5
36.9
0
16.7
21.1
Substitut. for one-place V
40.0
24.6
76.0
29.1
42.1
Substitut. for two-place V
32.5
38.5
24.0
54.1
36.8
The data show that our subjects had a lexical selection problem in accessing target predicates;
they selected a high number of paraphasia verbs. Proportions of target verbs were lower than
substituted verbs in every verb group. This does not mean that the aphasic patients cannot obtain any
information from the feature structure of the verbs. The unsuccessful retrieval of the phonological
form of the target verb does not explain agrammatic verbal performance in itself. The argument
'enumerations' indicate that certain information concerning the argument-structure, thematic grid and
subcategorization list is available.
In Type A answers only nouns or noun phrases were produced, the verb was deleted, e.g.
Anyuka...virág.. A virág.. Férje.–mother-nom...flower-nom...The flower-nom..husband-gen 2sg.nom
(A férfi virágot ad a nőnek – The man gives flower to the women). In Type B answers the activation
of nominal elements of the subcategorization list preceded lexical access of the verb. During the
selection of argument nouns hesitation, pauses, semantic paraphasias, word initiation difficulties, self
corrections occurred. E.g. Busz..Fiú majd jön. – Bus-nom... Boy later comes. (A fiú felszáll a buszra.–
The boy gets on the bus). The subjects usually attempted to build the previously activated argument
Ns or DPs into a syntactic scheme. When the phonetic form of the target verb was inaccessible,
another predicate was selected (both one- or two-place verbs which were always adequate in the
situational context).
The "listed" Ns or DPs (in 'isolated argument' answer types) were always complements, never
randomly named nouns. (Instrument and locative adjunct NPs sometimes occurred.) The patients
never assigned "extra" arguments, only those which were required by the verb.
Type C (and clausal part of Type B) answers were either fragments/agrammatic or wellformed sentences. E.g. Önteni...Önteni a limonádét.– Pour-infinitive... To pour the lemonade-acc. (A
fiú kiönti a narancslevet a pohárba. – The boy pours out the juice into the glass)
131
Effect of verb complexity on agrammatic aphasic's sentence production
The data show that the subjects were able to get some specific information represented in the
lexical entry of the predicate: argument-structure and thematic information were accessible; argument
assignment was not accidental, the ratio of activated arguments with different thematic roles was
related to the thematic hierarchy.
Argument assignment and thematic hierarchy
According to the Projection Principle of GB Theory, lexical information is syntactically
represented. Argument structure is projected from the lexical semantic structure of the verb and the
base structure of the sentence is projected from the argument structure according to the parametric
characteristics of phrase structure of a given language (Grimshaw, 1990).
According to the 'thematic hierarchy' hypothesis the argument structure of the verb is not only a
set of arguments. It has its own internal structure which represents prominent relations that are
determined by the thematic information of the predicate (Grimshaw, 1990). Grimshaw suggested a
protoargument-structure which is a structured representation of arguments based on thematic
hierarchy:
(Agent (Experiencer(Goal/Source/Location(Theme))))
The hierarchy expresses which argument has more chance of getting into the subject position.
If the predicate assigns an Agent thematic role, this argument must be mapped into the syntactic
function of subject. If there is no Agent or Experience argument in the thematic grid of the verb, the
less prominent Goal/Source or Theme argument can get into the subject position.
As we described above, our agrammatic patients often gave answers in which they listed
argument Ns or DPs (Type A and argument enumeration part of Type B answers).
During the analysis of these kinds of answers, we found that the argument assignment was not
random. The selection of arguments was connected with their position in the thematic hierarchy
and the 'animacy' semantic feature, consequently the argument activation depended on the type of
target verb.
Table 4. shows the distribution of arguments produced first in the different verb subgroups.
(The numbers in parentheses show the number of total occurrences of a given argument: number of
occurrences of an argument produced first linearly in the list plus the number of the same argument
appearing as second or third element in another argument list).
Table 4. Distribution of arguments activated first (the total number of activated arguments
including first activated is indicated in parentheses)
Type of verb
Agent
2-place
12
[+anim.]
(13)
2-place
10
[-anim.]
(18)
Goal/
Goal/
Benefactive
Source
–
–
Theme
6
(12)
–
–
52
(60)
Effect of verb complexity on agrammatic aphasic's sentence production
132
2-place
7
[locative]
(8)
3-place
10
[locative]
–
11
–
(13)
–
(11)
3-place
0
4
[dative]
(5)
(6)
5
16
(10)
(26)
–
10
(13)
Based on the distribution of arguments the following thematic role preference was found in the
different predicate types:
Transitive verbs
Agent > Theme
([+animate] object constituent)
Transitive verbs
Theme > Agent
([-animate] object constituent)
2-place verbs
Goal > Agent
(locative constituent)
3-place verbs
Theme > Agent > Goal
(object+locative constituent)
3-place verbs
Theme > Goal/Benefactive > Agent
(object+dative constituent)
The subjects were able to produce every type of argument (Agent, Theme, Goal, Benefactive)
but a difference was found in the distribution of arguments activated first.
Activation of the arguments lower in the thematic hierarchy was more frequent than more
prominent arguments of a given predicate (e.g. Theme > Benefactive > Agent ; Theme > Agent; Goal
> Agent). Two exceptions were found: the Agent > Goal order in the 'three-place locative' group and
the Agent > Theme order in the 'transitive [+animate]' group.
Comparing the proportions of arguments, an outstanding contrast was found between the
activation of Agent and Theme arguments in the 'transitive [-animate]' and '3-place dative' verb
groups. In the case of the '3-place locative' group the Agent > < Theme contrast was not so sharp,
rather, the Goal/Source > < Theme and the Agent > < Goal/Source contrasts were considerable. The
contrast was also less sharp between the Agent and Goal arguments in the 'two-place locative' type of
verbs.
The data show that the less prominent Theme argument was activated faster than the other
arguments if the predicate assigned the thematic role of Theme mapped to an object NP specified as [-
133
Effect of verb complexity on agrammatic aphasic's sentence production
animate]. Activation of the Theme argument fell behind the Agent only if the verb was reversible (if
the Theme thematic role was mapped into an object specified as [+animate, +human]). Considering
the total numbers of activated arguments, only the 'Theme [-animate] > < other arguments' contrast
remained, the contrasts among the other arguments had been equalized. The Agent > < Theme
[+animate] contrast also 'disappeared' when considering the total number of arguments.
The Theme [-animate] argument seems to be a preferred argument. Among the elements of
the subcategorization list, mostly the Theme thematic role was assigned first. The Verb–Theme
argument relationship seems to be closer than the Verb–other argument connection.
Case assignment in isolated arguments
The argument Ns and DPs were mostly produced without a case marker but case marked forms
also occurred. Proportions of case marked forms in ' isolated argument assignment' answers were
slight compared to the caseless forms. [ Theme argument N with accusative: (4.5) ; Goal argument N
with locative case marker: (13.0); Benafactive argument with dative case marker: (16.6). Case marker
substitutions: Theme argument: (4.5); Agent argument: (3.6); Goal arguments: (8.6); Benefactive
argument: (0) ].
According to GB Case Theory case can be assigned under government. The verb governs its
complements and assigns them abstract cases. Hungarian has a morphologically rich case system,
cases are realized morphologically and marked by overt case markers (only the nominative case is
marked by a 'zero' morpheme). The verb idiosyncratically assigns case endings of their arguments,
this information is indicated in the subcategorization frame (as a case frame). Case assignment can be
realised in two ways:
– if we assume that root-form argument Ns are inserted into the base structure, further
morphological operations are needed to produce the appropriate case marked forms according to the
information specified in the subcategorization list. This process assumes intact lexical selection of the
given case maker from the 'grammatical marker lexicon' and intact morphosyntactic procedures that
assign the overt case markers to the argument nouns.
– we can also assume that the lexical entry of the noun involves the root form and all inflected
variants of the noun. In this case, complete case marked forms can be retrieved from the lexicon and
inserted into given syntactic slots. The verb assigns the case by checking the case markers of the NPs
in the argument positions.
Both procedures seem to be possible on the cognitive level and are supported by processing
data. The latter solution would correspond to a non-compositional holistic access regarding
morphology combined with a decompositional secondary checking.
The aphasic subjects' performance indicates that the root forms (nominative) are generally
accessible, our aphasic subjects rarely used a case marked form or direct holistic lexical access in their
isolated argument answers. The tendency to omit case markers indicates a dysfunction of the
syntactic structure building operations that can not create the minimally required syntactic domain
(VP) needed in the case assignment procedure. The slot of the case assigner (V) and the slot of the
case marked constituent (N or N+I) are not available at the same time, only one nominal category slot
is delivered and filled in by a lexical element.
The calusal answers (Type B and C)
After the activation of the available information from the lexical entry of the verb (argumentstructure and thematic information), the agrammatic patients tried to insert the activated arguments
into syntactic structure. The syntactic structure building mechanisms however, were also disturbed,
they generated both well-formed and agrammatic sentences.
(Note: These sentences involved target or paraphasia predicates)
Effect of verb complexity on agrammatic aphasic's sentence production
134
Table 5. - Proportion of agrammatic and well-formed sentences
II/A
II/B
II/C
III/A
III/B
2-place
2-place
2-place
3-place
3-place
[+anim]
[-anim]
(loc.)
(loc.)
(dat.)
Agrammatic S
47.5
58.4
44.0
58.3
42.1
Well-formed S
52.5
41.5
56.0
41.6
57.8
40
65
25
24
19
Total
Some agrammatic sentences were a. incomplete clause, e.g. VO*, SVO*, SLVO* etc., since
the complement NPs were deleted or the previously retrieved arguments were lost again when the
subjects wanted to frame them into the slots of the phrase structure (into the complement or subject/
Topic positions). Other agrammatic sentences were due to b. deletion of formatives. The complete
phrase structure was not built because of the lack of formatives (closed class items): case markers,
verb inflections (agreement of the verb inflectional ending with the definiteness or indefiniteness of
the object) and determiners/articles were deleted.
(* marks the deletion of the given constituent, D:dative complement, L:locative complement,
C:other complement, S': sentential complement)
Word order in the clausal answers
Our subjects produced different permutations of surface word orders in the clausal answers. In
Hungarian the verb's arguments can be freely topicalized but the sentence is unmarked if the Agent
NP or a constituent specified for [+human] or [+animate] semantic feature is preposed to the Topic
position. Our aphasic patients proved to be very sensitive to these features and to the prominent
relations in the argument hierarchy.
The word order variations in the different verb groups were the following:
(Note: – the verbs in these sentences were target or paraphasia verbs, –word orders with highest
occurrence are indicated by bold letters; word orders that occured only one or two times are indicated
in parentheses)
a./ 'Transitive [+animate]'/ reversible
V > SV > VO*> SVO*, VC , SOV > (OV, SVC, CVS, VS, VS', SVO, SCV)
b./ 'Transitive [-animate]'/irreversible
VO* > V > VO > VS,VC, SVO* > (SV, SVO*, VOC*, OV, SOV, OSV, SVC, CVS)
c./ '2-place verbs with locative complement'
V > SV > (VS, VO, LV, VL, SVL, VL* )
d./ '3-place with locative complement'
V > VO > VO*> (LVO, VOL*, VO*L*, SV, VDO*, LV, VSO*, CVS
e./ '3-place verbs with dative complement'
V > VO*, SDOV > (DVO*, SOVD, VDO, DSVO*, VO, VC, VSO)
Effect of verb complexity on agrammatic aphasic's sentence production
135
V, SV, VO (VO*) word orders were always higher than other word order variations, across all
permutations produced by the patients. When the predicate did not assign the Theme argument ( '2place locative') or there was no animacy contrast between the Theme and Agent arguments, the
number of V and SV structures was higher. In the case of those predicates that required Theme
arguments and in which an animacy contrast was found between the Agent and Theme arguments, the
V, VO, VO* answers occurred in the highest proportion.
Generating the types (a-e.) sentences, the aphasics used three main principles:
1. 'Insert the [+animate] [+human] argument into the subject or Topic position'.
In the transitive/reversible verb group the Theme argument was also specified as [+animate]
[+human], giving the possibility of 'perspective reversing': the 'original' Theme argument was mapped
into the subject position of a one-place verb as Agent. (Some SV answers belonged to this kind of
sentence type in this verb group.)
2. 'Keep the predicate and the Theme argument together'.
3. If other information is not available/accessible 'try to build the most simple structure
selecting a one-place predicate.'
These 'tendencies' are parallel to the strategies used by the subjects on the semantic mapping
level:
– the Agent argument was activated first in higher proportion in the case of the
'transitive/reversible' verb group.
– in the case of the other predicates the Theme arguments were activated first which can be
connected to the strategy: among the elements of the subcategorization list , 'map first the less
prominent argument into the syntactic frame. Construct the [V+O complement] structure first'.
Summary
We investigated the ability of Broca`s aphasic patients to produce simple active sentences
which involved verbs of different argument structures with varying morphological complexity.
'Task specificity', which is a characteristic feature of aphasic performance, appeared in our
investigation as well. Although our Broca patients omitted main verbs from their spontaneous speech,
omission of verbs was not characteristic of their performance in an 'action naming task' rather,
substitution of verbs occurred. Our subjects could retrieve one or two-place verbs (target and
paraphasia predicates) in a relatively high number (mean score of total verb answers in different verb
groups involving the substitutions): 54.8/ '1-place'; 71.0/ '2 -place'; 66.6/ '3-place'). The proportion of
the target verbs was lower than the ratio of the substituted predicates in the 2-, or 3-place verb types,
which showed a lexical selection disorder in verb retrieval.
We found that the representational complexity of the verbs had a direct effect on the
accessibility of the predicates. The 'morphologically simple one-place predicates' were produced in
the highest number. Much lower proportions were found in the 'morphologically complex one-place'
predicates and in the transitive verbs, and only some verbs were activated in the 3-place verb groups.
Production of the directional motion verbs proved to be the most difficult for the patients. This data
showed that the argument structure complexity of the verb is important but not the only factor in the
lexical selection of predicates. The semantic representational and morphological complexity of the
predicate is also relevant in the lexical-semantic selection of the verbs.
Although the phonological form of the target verb was often not accessible for the aphasics,
they could retrieve other specific information represented in the lexical entry of the verb. Argument
structure and thematic information were partly accessible. Activation of the arguments was not
random but related to the thematic hierarchy. The less prominent arguments were produced faster.
Effect of verb complexity on agrammatic aphasic's sentence production
136
The Theme argument specified as [-animate] was always the most preferred argument in the semantic
and syntactic mapping procedures.
When there was no animacy contrast between the Agent and Theme argument , more Agent
arguments were produced first than Theme arguments. The data show that the subjects used three
principles in the semantic and syntactic mapping processes: (1) Map the [+animate], [+human]
argument into the subject or Topic position, treating it as an Agent. (2) Construct first the verbobject complement structure, map the less prominent Theme argument specified as [-animate] first
into the object position of the syntactic frame. (3) Construct the most simple (S)V structure if more
information is not available from the semantic representation of the verb, selecting a one-place
predicate.
Dysfunction of the syntactic structure building mechanisms had a connection with the lexical
accessibility of formatives and nominal elements of the phrase structure. The reduced capacity to
preserve the previously activated argument Ns or NPs had a role in the unsuccessful structure building
operations. The previously activated argument nouns were often not preserved –they were not inserted
into the given positions of the constituent structure –, during the selection of the phonetic form of the
verb. This resulted in incomplete clauses or fragments. The formatives were not obtainable either,
they were not assigned to the appropriate slots of the case frame. This also resulted in agrammatic
sentences. The semantic and syntactic mechanisms rarely operated in a parallel way or in
coordination. The number of well-formed or complete clauses was very low in the case of predicates
which assigned more complex argument structure.
Based on the data, agrammatic performance can be interpreted by those asynchronic
mechanisms that cannot function simultaneously on/between the level of semantic operations
(activation of argument-structure and thematic information) and syntactic processing (procedures that
construct the syntactic phrase structure and map the arguments into the syntactic frame).
Experiment II
In Experiment II we investigated the production of 'complex predicates' belonging to different
types. Complex predicates contain a base verb (V) which selects a verbal modifier (VM) in its lexical
representation: verbal prefix (preverbs), predicate complement (resultative phrase) or unquantified
common noun (bare noun). Verbal modifiers are not arguments of the verb but modifier elements:
they modify, change the meaning of the verb, they have a role in expressing "aktionsart" (resultativity,
progressivity, instantaneity, accomplishment), they serve as aspectual markers in the sentence (e.g.
add a perfective meaning to the meaning of the verb).
In neutral sentences the canonical position of the verbal modifier is the immediate preverbal
position. The verbal modifier occurs to the left of the verb. Verbal modifiers specify [+verbal
modifier] feature in their lexical representation, according to which they must be mapped into the VM
position of the phrase structure (É.Kiss, 1998). É.Kiss (1998) claims that the verbal modifier is basegenerated in postverbal position and moves into the VM position in neutral sentences.
(1) Syntactic position of verbal modifiers according to the model of É.Kiss (1998)
VP
V`
XM
[+VM]
V
NP
Effect of verb complexity on agrammatic aphasic's sentence production
137
Komlósy's (1985) analysis implies that the "verbal modifier+verb" ([VM+V]) construction is a
lexical unit, base-generated in V0. Bródy (1990) also suggests that the [particle+V] complex (V+) sits
in V, it is base-generated under the V'.
(2) Syntactic representation of verbal particles (modifiers) in Bródy's (1990) model
VP
V`
V+
VM
NPSUBJ
NPOBJ
V
Syntactic complexity of the target sentences was varied: ‘neutral’ and ‘non-neutral’ sentences
were used to investigate the effect of sentence types on the lexical selection of verbal modifiers. We
assumed that the representational complexity of the syntactic structure had an effect on the selection
of complex predicates in the sentence production of agrammatic aphasics.
Neutral sentences contain topicalised argument(s). The topic serves as a logical or notional
subject, the VP serves as the logical predicate of the sentence (É.Kiss, 1987). In É.Kiss's (1987, 1995)
model all the arguments of the verb are generated VP internally in arbitrary order, all arguments are
sisters of the V. Topicalisation is a transformation rule which preposes an argument from the [XP, V']
position into the [Spec, TenseP] (=Topic) position. Horváth (1986) asssigns a configurational
structure to the Hungarian sentence, she claims that Hungarian is an SVO language, the subject is
base-generated in [Spec, IP], the object NP and other subcategorized arguments are base-generated in
a postverbal position.
In non-neutral sentences that contain focus or wh-operator the operator is preposed into the
Focus position. The focused element must be left-adjacent to the verb. In sentences with Focus the
VM is separated from the verb, the verbal modifier does not precede the verb but follows it.
According to É.Kiss (1995), the VM and the verb remain in their base-generated positions, and
only the focused constituent moves into the Focus ([Spec, VP]) position.
(3)
TenseP
spec
Tense`
Topick
Tense
VP
spec
V`
Focusi
V
AdvP
NP
NP
perfix
ti
tk
In Bródy`s (1995) model, the VM stays in its preverbal (VM) position while the verb moves to
the head position (F0) of FocusP across the AgrO, Tense and AgrS heads.
Effect of verb complexity on agrammatic aphasic's sentence production
138
(4)
FP
spec
F`
F
AgrSP
spec AgrS`
AgrS
TenseP
spec
Tense`
Tense AgrOP
spec
AgrO`
AgrO
VP
Test material
The test material contained simple neutral and non-neutral sentences. Neutral sentences
involved topicalized element(s) but not operators, like focus, wh-element, or negation word. Nonneutral sentences contained operators or a universal quantifier. The sentence types we used were as
follows:
A. neutral sentences containing topic (TopP)
B. wh-questions (WhP)
C. sentences containing focus-phrase:’only’-phrase (FP)
D. sentences containing negation (NegP), sentential or predicate negation
E. sentences containing a universal quantifier (QP)
F. neutral sentences modified by a sentential adverb (AdvP)
The complete test material contained 120 sentences (6 types X 20 sentences). There were 20
target sentences in each sentence type. The non-neutral sentences were alternations of the neutral
ones. For example:
Type A. neutral sentence (TopP)
A színésznő-∅
vörös-re
fest-ett-e
a
haj-á-t
the actress-nom
red-onto
paint-past-3SG/def
the
hair-her-acc
’The actress painted her hair red.’
Type B. wh-question (WhP)
Ki
festette
vörösre a
haját?
who-nom
painted
red-onto
the
hair-her-acc
’Who painted her hair red?’
Type C. focused sentence (FP)
Csakis a színésznő
festette vörösre a
Only the actress-nom painted red-onto
’Only the actress painted her hair red.’
Type D. Negation (NegP)
haját
the hair-her-acc
Effect of verb complexity on agrammatic aphasic's sentence production
139
Nem a
Not
színésznő
festette
the actress-nom painted
vörösre a
red-onto
the
haját
hair-her-acc
’It was not the actress that painted her hair red.’
Type E. universal quantifier (QP)
Minden
színésznő
vörösre festette a haját
Every
actress-nom
red-onto
painted the hair-her-acc
’Every actress painted her hair red.’
Type F. Neutral sentence modified by sentential adverb
Sajnos
a színésznő
vörösre festette a haját.
Unfortunately the actress-nom red-onto
painted the hair-her-acc
’Unfortunately the actress painted her hair red’.
Each sentence contained a complex predicate = [verbal modifier + verb]. Three types of verbal
modifier (VM) were used:
(a)
Prefix
e.g. ’ kirabol’ (robs)
(b)
KIVM
+
RABOLV
out
+
rob-3SG
Predicate complement
e.g. ’ laposra ver’ (beats flat)
(c)
LAPOS|RA
+
VERV
flat-into
+
beat-3SG
Unquantified common noun (bare noun)
e.g. ’csalódast okoz’ (causes disappointment)
CSALÓDÁS|T
+
OKOZ
disappointment-acc +
cause-3SG
Method
Subjects. For the purpose of Experiment II, two nonfluent Broca`s aphasics were selected. Both
aphasic patients showed typical agrammatic phenomena in spontaneous speech: telegraphic style,
omission of closed class elements (case suffixes, determiners), low proportion of verbs, reduced
phrase length, lack of the complex sentences, tendency to produce simple SV, pro-VO structures, and
isolate minimal phrases (NP, AdvP) and words (mostly nouns). Both patients displayed severe speech
initiation difficulty and moderate anomia.
N.Zs. 26 year-old female right-handed shop-assistant was investigated 5 years post-onset.
Sz.V. 42 year-old right-handed man, engineer was investigated 4 months post-onset.
The aphasia dignosis was made by the Hungarian variant of the Western Aphasia Battery: N.Zs
's WAB AQ: 71,8 ; Sz.V.'s WAB AQ: 64,2.
Effect of verb complexity on agrammatic aphasic's sentence production
140
CT-scan of N.Zs. (performed 5 years post-onset) showed left (middle cerebral artery)
hypodense lesion, involving the insula and basal ganglia.
CT-scan of Sz.V. revealed a large hypodense area in the left hemisphere, in the territory of the
middle cerebral artery. The lesion involved the frontal lobe, the insula, the temporal and the parietal
lobe.
Procedure. As elicitation methods we used Sentence Anagram Test and Sentence Repetition
Test. In both tasks the same sentence types and target sentences were used.
Anagram test is a widespread method to investigate the patients’ sensitivity to word-order rules
and the syntactic position of the sentence constituents, e.g. verbal modifier, verb, argument NPs,
adjunct, operators. The subjects are instructed to produce a sentence by rearranging written cards
which represent the sentential constituents. The argument DPs, the verb and the verbal modifier were
presented to the patients on separate cards, in random order. e.g.:
RABOLTA
A FÉRFI
KI
A BANKOT
rob-past-3SG.
the man-nom
out
the bank-acc
The constituents were supplied with proper inflexional endings: the verb had past-3SG suffix,
the argument DPs and the b, c type of verbal modifiers had the relevant case suffixes. Sentences
containing operators were elicited by using the story completion method, which introduced the special
context needed for production of these types of sentences. The task was administered without time
limits.
Sentence repetition test requires immediate repetition of sentences. The subject were asked to
repeat the target sentences presented in random order.
Reproduction of a sentence depends on the analysis of the syntactic structure of the sentence.
Grammatical function of the arguments can be assigned by the inflexional morphology of the
argument NPs. Case-frame, argument-structure and subcategorisation information of the verb has also
be activated. Sentential adverbs, topicalised constituents and operators (focus-phrase, wh-phrase,
negation-phrase and universal-quantifiers) can be identified via their lexically represented syntactic
features, their stress features and their position in the phrase marker. The relative order of the verb and
the verbal modifier also serves as a cue in processing of neutral and non-neutral sentences. All the
information on syntactic and lexical features of the arguments and the predicate has to be preserved
for the correct reproduction of the sentence with correct morphology.
We assumed that our agrammatic aphasic subjects strive to use active syntactic structure
building operations during the reproduction although other strategy can also be used in sentence
repetition, e.g. retrieval of words of the sentence (only) from the phonological memory.
Results of Experiment II.
Effect of the type of the verbal modifier
Table 6. 7. and 8. present the distribution of grammatically well-formed and ill-formed
sentences in the Anagram and Sentence repetition tests. In the statistical analysis Fisher's exact test
was used.
Effect of verb complexity on agrammatic aphasic's sentence production
141
Table 6. - Distribution of grammatically well-formed and ill-form sentences in the
different verbal modifier groups. - ANAGRAM test N.ZS.
type of verbal modifier
well-formed
ill-formed
n=
%
p= n=
%
a. PREFIX
35
0.83
–
7
0.17
b. PREDICATE
COMPLEMENT
31
0,42
0.74
11
0.26
c. BARE NOUN
24
0,11
0.67
12
0.33
Note: number of target sentences containing: prefix=42, predicate complement=42, bare noun=36. n = number
of answers % = percentage value * = significant (p < .01)
Table 7. - Distribution of grammatically well-formed and ill-form sentences in the
different verbal modifier groups. SENTENCE REPETITION test N.ZS.
verbal modifier
well-formed
n=
%
a. PREFIX
26
0.62
b.PREDICATE
COMPLEMENT
c. BARE NOUN
ill-formed
p=
n=
%
16
0.38
6
0.14
0,00001*
36
0.86
5
0.14
0,00002*
31
0.86
–
Table 8. - Distribution of grammatically well-formed and ill-form sentences in the
different verbal modifier groups. SENTENCE REPETITION test - Sz.V.
verbal modifier
well-formed
ill-formed
n=
%
p=
n=
a. PREFIX
5
0.12
–
37
0.88
b. PREDICATE
COMPLEMENT
0
0.0
0,055
42
1.00
c. BARE NOUN
0
0.0
0,058
36
1.00
b+c
0
0.0
0,0045*
78
1.00
%
In the Anagram test there is no difference in the distributions of correct responses in the
different verbal modifier groups.
Effect of verb complexity on agrammatic aphasic's sentence production
142
In the sentence repetion test however, frequency of the grammatically well-formed sentences is
significantly higher in the PREFIX group than in the predicate complemet (p= .00001) and in the bare
noun group (p= .00002). The same effect was found in the production pattern of Sz.V., although Sz.V.
produced complete, grammatically well-formed clauses in extremely low proportion in the sentence
repetition test.
Types of errors
The patients produced different types of errors in the sentence repetition test. Agreement errors
such as verb inflexion and case assignment errors, determiner omission, substitution/omission of
argument NP and lexical selection problem of complex verb were observed, which resulted in
agrammatic sentences. Incomplete sentences, sentence fragments were produced in higher proportion
in the non-neutral sentences.
N.Zs. produced verb and verbal modifier omission or substitution in higher rate than Sz.V.
N.Zs.
Sz.V.
AGReement error
25%
27%
Omisson or substitution error of
40%
17%
26%
8%
– bare verb
8%
3%
– complex verb
6%
5%
V-VM/VM-V inversion error
7%
2%
Determiner omission
5%
15%
10%
18%
5%
9%
complex predicate
– verbal modifier
Argument NP substitution
Argument NP deletion
In the case of N.Zs. 40% of the errors were related to the lexical selection of verbs, in the case
of Sz.V. this rate is only 17%. In the highest number verbal modifier omission and substitution errors
occured.
Analysis of the data showed that sentence complexity had an effect on the lexical selection of
verbal modifiers. In the neutral sentences (Type A.) omission of the complex verb or the verbal
modifier did not occur, substitution of the base verb occured but in very low proportion (N.Zs.: 2/20 ;
Sz.V. 1/20). Substitution of verbal modifiers occured mostly in 'non-neutral' sentences (B,C,D,E)
which involved operators such as wh-operator, focus operator, negation or universal quantifier and in
sentences modified by sentential adverbs (F). Agrammatic patients showed a tendency to replace the
predicate complement and the unquantified common noun type verbal modifiers to PREFIX in these
sentences.
Summary
The results of Experiment II show that syntactic complexity of the sentence and the lexical
representational complexity of the complex predicate equally had an effect on the lexical selection of
verbal modifiers. Our data indicates that the lexical selection problem of verbal modifiers during
sentence production can not be explained only by the disorder of the lexical accces to the complex
143
Effect of verb complexity on agrammatic aphasic's sentence production
predicate or the verbal modifier. Production of verbs and lexical material selected by the verb is
strongly related to the Broca's aphasics' ability to construct sentence structure.
In the Anagram test the agrammatic patient, N.Zs., assigned the verbal modifier position
correctly, in neutral sentences and in non-neutral sentences as well. She never handled the VM as an
argument of the verb, the VM never occupied argument position in the sentences constructed by the
patient. This means that the agrammatic patient is sensitive for the [+verbal modifier] semantic feature
of the VM; she was always able to make the distinction between the overt arguments and the modifier
element selected by the verb. This also means that the agrammatic patient can activate all the
information from the lexical entry of the verb: she retrived the argument-structure and
subcategorization information, as well as the semantic features of verb particles (prefixes) and
semantic features of the inner arguments (resultative phrases, unquantified common nouns) which
cannot be expressed syntactically as an argument only as a constituent which modifies the meaning of
the verb. Verbal Modifiers always occupied preverbal position in the phrase marker of neutral
sentences. In non-neutral sentences the VM remained in the appropriate position although verbmovement into the Focus position was disturbed.
Our agrammatic subjects were able to produce complex predicates (verbal modifier + verb)
correctly in sentence repetition task if the target sentence was an SVO neutral sentence. When
repeating non-neutral sentences, however, agrammatic patients produced omission and substitution of
verbal modifiers. In reproducing sentences containing operators, agrammatic patients showed a
tendency to replace the semantically more complex 'predicate complement' and 'bare noun' type of
verbal modifiers with prefixes that functioned as a perfectivity marker in the sentence. When they
used prefixes instead of the other type of verbal modifier, only the syntactically relevant [+verbal
modifier] and [+perfective] feature of the VM was preserved in the working memory, the other
lexico-semantic feature of the VM was neglected by the patients. With this strategy the agrammatic
subjects produced grammatically well-formed non-neutral sentences. Substitution of prefixes can be
regarded as an adaptation strategy that is suitable to decrease the processing overload. The patients
adapt themselves to the limited capacity of the sentence production system (slow activation of
syntactic information and limited memory time), keeping in the memory only those syntactic features
without which the syntactic representation can not be contsructed. This information remains available
only at the cost of "deleting" those lexico-semantic features of the VM which are not necessarily
involved in syntactic structure building.
It seems that the sentence type or sentence complexity also has a role in the lexical selection of
the verbal modifier. Non-neutral sentences involved operators: focused-phrase, wh-phrase or
negation-phrase. Syntactic representation of these sentences requires verb movement to the Focus
position and wh-movement to the operator position ([Spec, FP]). These kinds of operations seem to be
more demanding from a processing point of view than topicalisation, which moves an argument into
the Topic position.
Since in the Anagram task the agrammatic patient, N.Zs. proved to be very sensitive to the
semantic features of the verbal modifiers (she never mapped the verbal modifier into argument
position but into the VM position of the sentence structure) and because lexical selection of the verbal
modifiers was always preserved in neutral sentences, the selection problem of VMs cannot be
explained by a deficit of lexical access to the lexical representation of complex predicates.
Substitution of prefixes for resultative phrases and unquantified bare nouns in non-neutral
sentences but not in neutral sentences shows that there is a dependency relation between the sentence
complexity and lexical selection of verbal modifiers.
144
Effect of verb complexity on agrammatic aphasic's sentence production
References
Bastiaanse, R. and Jonkers, R. (1998) Verb retrieval in action naming and spontaneous speech in agrammatic
and anomic aphasia. Aphasiology, 12, 951-969.
Brody, M. (1990) Remarks on the Order of Elements in the Hungarian Focus Field. I. Kenesei (ed.) Approaches
to Hungarian, Vol. 3: Structures and Arguments, JATE, Szeged.
Byng, S. (1988) Sentence processing deficits:theory and therapy. Cognitive Neuropsychology, 5, 629-676.
É. Kiss, K. (1987) Configurationality in Hungarian. Reidel, Dordrecht.
É.Kiss, K. (1995) NP Movement, Operator Movement, and Scrambling in Hungarian. In: K. É.Kiss (ed.)
Discourse Configurational Languages, Oxford University Press.
É.Kiss, K. (1998) Mondattan. (Syntax) K.É.Kiss and F. Kiefer (eds.) Új magyar nyelvtan. Osiris Kiadó,
Budapest.
Friedmann, N. and Grodzinsky, Y. (1997) Tense and agreement in agrammatic production: Pruning the
syntactic tree. Brain and Language, 56, 397-425.
Grimshaw, J.B. (1990) Argument Structure. Cambridge, MA:MIT Press.
Jackendoff, R. (1987) The Status of Thematic Relations in Linguistic Theory. Linguistic Inquiry, 18, 369-411.
Jackendoff, R. (1990) Semantic structures. Cambridge, MA:MIT Press.
Jones, E.V. (1984) Word order processing in aphasia:Effect of verb semantics. Advances in Neurology, Vol.2,
159-181. In: F.C.Rose (ed.) Progress in aphasiology. Raven Press, New York.
Jonkers, R. and Bastiaanse, R. (1996) The influence of instrumentality and transitivity on action naming in
Broca’s and anomic aphasia. Brain and Language, 55, 37-39.
Kegl, J.(1995) Levels of representation and units of access relevant to agrammatism. Brain and Language, 50,
151-200.
Kiss, K. (1994) Analysis of aphasics' spontaneous speech (unpublished)
Komlósy, A. (1985) Predicate complementation. I. Kenesei (ed.) Approaches to Hungarian. Vol.1. JATE,
Szeged.
Komlósy, A. (1994) Complements and adjuncts. F. Kiefer and K. É.Kiss. (eds.) Syntax and Semantics. The
syntactic structure of Hungarian, Vol. 27., Academic Press.
Levin, B. (1993) English verb classes and alternations: A preliminary investigation. Chicago:University of
Chicago Press.
Marshall, J., T.Pring amd S. Chiat (1993) Sentence processing Therapy:Working at the level of the Event.
Aphasiology, Vol. 7, No.2, 177-199.
Miceli, G., M.C. Silveri, G. Villa, and A. Caramazza (1984) On the basis for the agrammatic's difficulty in
producing main verbs. Cortex, 20, 207-220
Miceli, G., M.C. Silveri, C.Romani, and A. Caramazza (1989) Variation in the Pattern of omissions and
substitutions of grammatical morphemes in the spontaneous speech of so-called agrammatic patients. Brain and
Language, 35, 24-65.
Naeser, M.A., C. Palumbo, N.Helm-Estabrooks, D. Stiassny-Eder, and M.L.Albert (1989) Severe nonfluency in
aphasia: Role of medial subcallosal fasciculus and other white matter pathways in recovery of spontaneous
speech. Brain, 112, 1-38.
Osman-Sági, J. (1991) Az afázia klasszifikációja és diagnosztikája [Diagnosis and classification of aphasia] In:
Ideggyógyászati Szemle, 44, 8. füzet, 339-362.
Saffran, E.M., R.S. Berndt, and M.F. Schwartz (1989) The quantitative analysis of agrammatic production:
Procedure and data. Brain and Language, 37, 440-479.
Shapiro, L. P., E.Zurif, and G.Grimshaw (1987) Sentence processing and the mental representation of verbs.
Cognition, 27, 219-246.
145
Effect of verb complexity on agrammatic aphasic's sentence production
Shapiro, L.P. and Levin, B. A. (1990) Verb processing during sentence comprehension in aphasia. Brain and
language, 38, 21-47.
Stark, J. (1992) Everyday Life Activites -photo series. Vienna, Printed in Austria by Druckerei Bosmüller.
Thompson, C.K., L.P. Shapiro, L. Li and L. Schendel (1995) Analysis of verbs and verb-argument structure: a
method for quantification of aphasic language production. Clinical Apahsiology, 23, 121-140.
Thompson,C.K., K.L. Lange, S.L. Schneider, and L.P.Shapiro (1997) Agrammatic and non-brain-damaged
subjects' verb and verb argument structure production. Aphasiology, 11, 473-490.
Zingeser, L.B. and Berndt, R.S. (1990) Retrieval of nouns and verbs in agrammatism and anomia. Brain and
Language, 39, 14-32.
Zurif, E., D.Swinney, P. Prather, J. Solomon and C. Bushell (1993) An on-line analysis of syntactic processing
in Broca’s and Wernicke’s aphasia. Brain and Language, 45, 448-464.
Williams, S.E. and Canter, G.J. (1987) Action.-naming performance in four syndromes of aphasia. Brain and
Language, 32, 124-136.
How to Cope with "Free Word Order": An Efficient Part-of-Speech Tagging
Method for Hungarian
146
How to Cope with "Free Word Order": An Efficient Part-of-Speech Tagging
Method for Hungarian
Csaba Oravecz, Péter Dienes
Abstract
There has been a growing need for language resources in the languages of Central and Eastern
Europe, for which such resources, bar a few exceptions, despite the efforts of some recent projects to
create them, still do not compare to those of languages such as English or German.
One reason for the significant difference in their availability might be attributed to the fact that
methods developed for western languages and used widespread to produce such resources cannot
readily be applied to Central and Eastern European languages, due to the large number of novel
phenomena exhibited in the syntax and morphology of these languages, which these methods have to
handle but have not been designed to cope with. The process of part-of-speech tagging when applied
to Hungarian data to produce corpora annotated at least at the morphosyntactic level is most indicative
of this problem: several of the algorithms (either rule-based or statistical) that have been used very
successfully in other domains performed disappointingly for Hungarian.
A possible way of improving performance offers itself handily from the recent discovery in
machine learning that the combination of multiple classifiers often significantly outperforms what can
be achieved with a single classifier. However, the basic assumption in trying to combine different
classifiers, in our case, for POS tagging, different taggers even of comparable accuracy, is that they do
not make identical errors, i.e. the errors made by the different taggers are strongly complementary.
The paper will present an experiment with several individual taggers, both rule-based and
statistical, and, combining the results from experiments in the same field by other researchers, will
show that the above assumption for Hungarian is not fully supported: there is a set of features relevant
for lexical disambiguation and manifest in a number of tagging instances which are not captured by
any of the algorithms and thereby cause all taggers to err. This finding seems to back up the
"suspicion" that since all of the different techniques are typically referring to features present in the
words and tags in a fixed window of tagging environment, they might well end up learning the same
information. However, owing to the unusual variability in the order of words and phrases in
Hungarian, this fixed domain of tagging environment may frequently miss some or all of the relevant
tagging clues in particular instances.
The paper will propose a possible method to remedy this situation with two basic objectives in
mind: (1) capture the features relevant for lexical disambiguation in a "flexible" tagging window,
independently of the number or order of words and tags in the given window; (2) reduce long distance
dependencies which are long only because of intervening items irrelevant for the particular tagging
instance. For (1), we will propose a learning procedure which specifies bounding tags that will define
a specific tagging domain on a sequence as a flexible tagging window within which it will be possible
to apply rules referring to any of the relevant features in that window to resolve ambiguities. Since
this procedure will construct a tagging window as containing a sequence of tags in between two
specific bounding tags irrespective of the order and number of tags belonging to this window, and
rules applicable in such a window will not depend on the distance between features used to resolve
ambiguities, crucial contextual information can be efficiently captured independent of its distance
from the tagging spot in the window. An algorithm to induce the rules applied in such tagging
windows from trainig corpus will also be discussed and presented.
For (2), the general idea is to eliminate all irrelevant items from the annotation temporarily and
thus to reduce the distance between the ambiguity and the point where the clue for disambiguation can
be found to a local context. A simple regular expression grammar to achive this aim will also be
discussed.
147
How to Cope with "Free Word Order": An Efficient Part-of-Speech Tagging
Method for Hungarian
Finally, we will present a tagging system built from component taggers to handle "local"
ambiguities stacked with the above methods to cope with phenomena specific to Hungarian, and will
show that this framework can significantly increase accuracy in the tagging process.
Keywords: combined classifiers, POS tagging, free word order languages
Aspect and Argument Structure
148
Aspect and Argument Structure
Introduction
In the semantic literature it has been recently proposed that event structure can be calculated on
the basis of the internal temporal constitution of the predicate, the referential type of the internal direct
argument in a compositional way (Verkuyl (1972, 1989,1993), Krifka (1989, 1992)). The aim of this
paper is twofold: The first is to propose a new classification of verbs different both from VendlerDowty's and from Verkuyl's classification that can account for the behaviour of achievement verbs
previous theories cannot account for.The second is to claim that on the basis of the internal temporal
constitution of a verb predictions can be made about the argument structure (external argument,
internal direct argument, internal indirect argument (Tenny (1995)) of the verb. This paper examines
the nature of the inner temporal constitution of an event on the basis of the interaction of the internal
temporal constitution of verb and the referential type of the internal direct arguments of the verb. and
We will argue that the internal temporal constitution of the predicate is just as relevent for the
calculation of internal aspect as the referential type of the internal direct argument.
On Aspect
In the semantic literature aspect deals with the internal temporal constitution of a proposition.
According to Verkuyl (1989, 1993) inner and outer aspect have to be distinguished depending on
whether we refer to the internal temporal structure of the VP or the internal temporal structure of the
sentence. (1)
In the traditional literature an event (2) is either telic or atelic. An event is telic if there is "a
built-in" endpoint in the event and it is atelic if the event lacks a "built-in" endpoint. (Comrie
(1985), Dahl (1985)).
1,a, János egy percen át/*egy perc alatt látott egy lovat.
J-nom for a minute/*in a minute saw a a horse-acc
John saw a horse for a minute/*in a minute.
b, Mari egy éven át/*egy év alatt beteg volt.
Mari-nom for a year/*in a year ill was
Mary was ill for a year/*in a year.
2, a, Lajos egy percen át/* egy perc alatt almát evett.
Lajos-nom for a minute/*in a minute apple-acc
Lajos ate apples for a minute/*in a minute
b, Péter egy órán át/*egy óra alatt aludt.
Peter-nom for an hour/*in an hour slept
Peter slept for an hour/*in an hour.
c, Aladár egy percen át/*egy perc alatt tolta a kocsit.
Aladar for a minute/*in a minute pushed the car.
Aladar pushed the car for a minute|/*in a minute.
d, Egyed egy órán át/*egy óra alatt vert egy darab vasat.
Egyed-nom for an hour/*in an hour hammered a piece of iron.
Aspect and Argument Structure
149
Egyed hammered a piece of iron for an hour/*in an hour.
3,a, Péter *egy even át/egy év alatt épitett egy házat.
P-nom *for a year/in a year built a house
Peter built a house *for a year/in a year.
b, Kazimir egy perc alatt/*egy percen át laposra vert egy darab vasat.
Kazimir in a minute/*for a minute flat hammered a piece of iron
Kazimir hammered a piece of iron flat in a minute/*for a minute
c, Egyed egy perc alatt/*egy percen át a garázsba tolt egy szekeret.
Egyed in a minute/*for a minute the garage-to pushed a cart
Egyed pushed the cart to the garage in a minute/*for a minute.
4,a, Egy perc alatt/*egy percen át vendég érkezett a városba
in a minute/*for a minute guest has arrived in the own
A guest has arrived in a minute.for a minute
b, Péter egy perc alatt/*egy percen át könyvet adott a barátjának.
Peter in a minute/*for a minute book gave his friend-to
Peter gave a book to his friend in a minute/*for a minute.
In 1 and 2, the verbal expression is atelic as its denotation has no set terminal point. In 3, and 4,
it is telic as it includes a terminal point. This well-known semantic distinction is supported by a
number of tests. In non-iterative reading atelic expressions allow for durative adverbs (e.g., for an
hour), but do not allow time-span adverbs (e.g., in an hour). In sentence 1, the inner aspect is stative,
in 2, it is process. In sentences 3, and 4, the natural temporal endpoint of the events are the existence
of the house, flat shape of the iron, the cart in the garage, the arrival of a guest and the new position
of the book.
Preliminaries
In the semantic literature aspectual properties of predicates have been investigated from various
perspectives. Various classifications of verbs have been proposed, the most influential one is the socalled Vendler-Dowty classification that derives inner aspect from the lexico-semantic poperties of
the verb. In this classification verbs are classified in four classes:state, process, accomplishment and
achievement.The problem with this type of classification that some verbs seem to move from one
class to the other.
5, a, Lajos *egy perc alatt/ egy percen át vizet ivott.
Lajos-nom *in a minute/for a minute water-acc drank
Lajos drank water *in a minute/for a minute.
b, János egy perc alatt/ ??egy percen át ivott egy pohár vizet.
Janos-nom in a minute/??for a minute drank a glass of wine-acc.
Aspect and Argument Structure
150
Janos drank a glass of wine in a minute/for a minute.
6 , a. János ??egy éven át/egy év alatt épitett egy házat.
Janos-nom ??for a year/in a year built a house-acc
Janos built a house ??for a year/in a year.
b, János egy éven át/ *egy év alatt házat épitett.
Janos for a year/*in a year house-acc built
Janos built houses for a year/*for a year.
In sentences 5 and 6 in the a, examples the verbs "drink" and "build" are process verbs while in
the b, examples they are acccomplishment verbs. In other words "drink" and "built" must be ambigous
between process and accomplishment. However, the difference clearly does not hold between the
verbs but rather between the reference type of internal direct argument, the patient. In the VendlerDowty classification the following regularities that have decisive roles in the calculation of the event
structure simply get lost:
7,
i, In every telic event there is an internal direct argument in the VP.
ii, In case of certain verbs the referential type of the internal direct argument defines the
interpretation of the event (process vs. accomplishment).
iii, There exists a group of process verbs whose quantized internal direct argument does not
delimit the event but the event can be delimited by introducing an internal indirect
argument.
iv, The referential type of the internal direct arguments of statives verbs do not have any role in
the calculation of the event, which is always atelic.
v, Though achievement verbs always have internal direct argument, their referential type do
not influence the event type. Achievement verbs always occur in telic events independently
of the referntial type of their internal direct argument.
These facts suggest that inner aspect is compositional, it is the combination of verbal and
nominal properties. This composition of nominal and verbal information constitutes the second trend
(Verkuyl (1972, 1989, 1993), Krifka (1992). In Verkuyl's theory inner aspect is composed of verbal
and nominal properties in the syntactic representation. According to Verkuyl VP (inner) aspect can be
reduced to verbal and nominal features. The temporal parameter for verbs is encoded by the verbal
feature (+/-add) where (+add) means roughly "progress in time" and (-add) means the lack of it. This
feature specification distinguishes stative verbs from dynamic verbs (process, accomplishment,
achievement). In the nominal domain the relevent feature is (+/-SQA), which means "specified
quantity of". The (+/-SQA) feature is the feature of the determiner system. A nominal construction is
(+SQA) if it has a determiner or it is quantized, if neither of them is present and the nominal
construction is a bare NP of existential reading, it has (-SQA) feature. Only if the properties of the
verb and the arguments meet what Verkuyl calls the Plus Principle (all verbal and nominal features
are specified for (+) value) is the VP telic.
Aspect and Argument Structure
151
8, a, János(+SQA) szeret(-ADD) néhány operát(+SQA).
Janos -nom
loves
some operas-acc
János loves some operas.
b, A fiúk(+SQA) házat(-SQA) épitenek(+ADD).
The boys-nom house-acc build
The boys are building houses.
c, A fiúk(+SQA) épitenek(+ADD) egy házat(+SQA).
the boys-nom build a house-acc
The boys are building a house.
d, János(+SQA) egy pénztárcát(+SQA) talált(+ADD).
János-nom a purse-acc found
János found a purse.
As we can see in 8, a, the verb has (-add) feature and the event is atelic, in 8,b, the internal
direct argument has (-SQA) feature and the event is atelic, in 8, c, and d, all the nominal and verbal
features are specified for (+), the events are telic.
Several semanticists have critized Verkuyl's theory because it has only descriptive but no
explanatory power. It correctly describes all the well-formed structures and excludes all the ill-formed
ones but does not explain why exactly that set of structures are the well-formed ones and why are the
rest excluded as ill-formed. It seems that there is a well-defined group of verbs namely, achievement
verbs that Verkuyl's theory cannot account for at all.
9, a, *Egy percen át/?egy perc alatt vendég érkezett.
*for a minute/in a minute guest-nom arrived
There have arrived guests in a minute/for a minute
b, *Egy percen át /?egy perc alatt János könyvet adott Péternek.
*for a minute/in a minute Janos-nom book-acc gave
John gave books in a minute/for a minute.
In 9, the internal direct argument is (-SQA) in Verkuyl's terms, both are bare existential NP. In
spite of that, the event interpretation of the VPs is telic. All achievement verbs can occur with bare
existential NPs and the result is telic. Achievement verbs can occur only in telic events.
Krifka (1992) analyzed the correspondence between the reference types of NPs and the
temporal constitutions of verbal predicates assuming an event semantics with lattice structures and
thematic roles as primitive relations between events and objects. The basic idea is that atelic
expressions are similar to mass nouns and bare existential plurals. They both refer cumulatively (3).
While telic expressions are similar to quantized nominal constructions. They have quantificational
reference. Krifka claims that there is a homomorphic mapping from object types to event types
provided the object is connected to the event throught the proper thematic role. If the object has
Aspect and Argument Structure
152
quantized reference, the complex verbal expression has quantized reference and event is telic. If the
object refers cumulatively the verbal expression will refer cumulatively, too. Then the event is atelic.
10, a, János *egy perc alatt/ egy percen át zajt halott a kertb l.
János-nom *in a minute/for a minute noise-acc heard from the garden.
János heard noise from the garden *in a minute/for a minute.
b, János *egy perc alatt/ egy percen át egy ember hangját hallotta a kertben.
János-nom *in a minute/for a minute a man's voice-acc heard in the garden
János heard a man's voice in the garden *in a minute/for a minute.
In 9 a, the direct argument refers cumulative, in 9, b, it has quantificational reference. In spite
of the difference in the reference type of the deep direct object, there is no difference in the aspectual
reading because stative verbs do not seem to assign the proper thematic role to their object. Krifka
distinguishes five different thematic roles verbs can assign to their direct internal argument:gradual
effected patient, gradual consumed patient, gradual patient, affected patient and stimulus. The
different patient relations are characterized by the following properties:Summativity, graduality and
uniqueness. If John saw a horse and Mary saw a horse then John and Mary saw two horses unless
they saw the same horse, (situation which is irrelevant for us now) then summativity is assumed for
the experiencer relation and the stimulus relation. Summativity relation is trivial for patient relations.
Graduality holds for events like read a book: the book is subjected to the event of reading in a gradual
manner, which is not true for the see a horse case. Uniqueness of objects chacterizes the patient
relation of effected and consumed patients that can be subjected to the event of eating or painting only
once. An object can delimit the event if the patient relation that relates the object to the event has the
properties of summativity, graduality and uniqueness and the object has quantized reference. Krifka's
theory has the same serious empirical problem that Verkuyl' s theory has. It cannot account for
achievement verbs. Achievement verbs have patient roles that are summative , some of them share the
uniqueness property, as well, but they are never gradual. In addition to these problems, they have to
be able to assign patient roles that are either effected or affected, they have to undergo some change,
but they do not need to be quantized so that a telic event could be obtained. Still we get telic events.
To define the relevant conditions for the event interpretation the following information is
needed :
11, i, information about the verb (roughly stative or dynamic)
ii, information about the reference type of the internal argument (bare existential or
quantized)
iii, information about the thematic roles that relate objects to event (patient relation)
(Smith (1996))
As we have seen earlier neither Verkuyl's nor Krifka's theory can account for the behaviour of
achievement verbs. The types of thematic roles that hold between achievement verbs and their
internal direct argument are those thematic roles that can delimit the event if they are quantized and
they are combined with dynamic verbs. Achievement verbs with their internal arguments always
form telic events independently of the reference type of the internal direct argument.
Aspect and Argument Structure
153
The theory
I will propose a compositional theory of events in the spirit of Krifka (1992), Verkuyl (1993). I
claim that event structures can be calculated on the basis of the referential type of the internal direct
argument, the thematic role that relates the internal argument to the event and the internal temporal
structure of the verbal predicate. I will further assume that not only events and nominal constuctions
have quantificational properties but verbs have such properties as well. The internal temporal
constitution of the verb can be specified for the feature (+/-quantized). I will claim that the
combination of the referential type of the nominal constructions, the verbs and the thematic roles
relating them would assign the event structures the proper event interpretation.
I assume the most relevant information for event interpretation is the thematic role of the
internal argument that relates the argument to the event. If the relevant thematic role is assigned, the
event can be delimited/telic or atelic/non-delimited, depending on the reference type of the internal
argument and the internal temporal constitution of the verb. There are two possibilities:
12, i, the internal temporal constitution of the verb is not delimited, the event can be delimited
by the quantized internal argument (process vs. accomplishment),
ii, the internal temporal constitution of the verb is delimited, the event can be temporally
delimited by the verb, it needs an object that undergoes some change of state but the
object does not need to be of quantized reference.(achievement).
Classification of verbs
Stative verbs
In the Vendler-Dowty classification of verbs the internal temporal constitution of statives verbs
do not consist of discrete atomic events.
13, a, János szereti Mári t.
János-nom loves Mária-acc
János loves Mária.
b, Ferenc féli Istent.
Ferenc-acc fears God-acc
Ferenc fears God.
c, Angéla látta a tömeget az utcán.
Angéla-nom saw the crowd-acc the street-on
Angéla saw the crowd in the street.
At any instant on the time span for which it is true that Janos loves Maria, Ferenc fears God,
and Angela saw the crowd in the street, the same statement is true. There is no development in the
event throught time. No change is encoded in the event. There is no way for any of the participants in
the situation to undergo some minimal change of state. The thematic roles assigned by stative verbs to
their internal argument is the least prototypical patient role (5), I will call this theta role stimulus
following Krifka's classification. Internal arguments that have stimulus theta role can never delimit an
event.
Aspect and Argument Structure
154
Process verbs
Process verbs are built up of an unlimited number of temporally ordered atomic events .
14, a, János (almát) evett.
János-nom (apple-acc) ate
János ate (apples).
b, János ((egy) traktort) vezett.
János-nom ((a )tractor-acc) drove
János drove a tractor.
c, János (egy darab) vasat vert.
János-nom (a piece of) iron-acc hammered
János hammered (a piece of) iron.
In sentence 14, a, each part of the optional effected object can be subjected to the temporally
ordered atomic events of eating only once. The relation between parts of the object and the atomic
events is unique. Verbs of consumption and creation assign the most prototypical thematic role, the
effected patient role to their internal argument. in 14, b, the whole object is subjected to the atomic
events of several unlimited number of pushing. What changes throught time is the position of the
object. In 14, c, the whole object is subjected to the event, again. In this case some property of the
object changes gradually in time.
Dynamic verbs differ from stative verbs with respect to their intern temporal constitution.
While stative verbs have a homogeneous internal temporal constitution, in process events there are
continuous changes encoded in the event. In processes for any atomic event for which the activity of
eating some part of an apple, pushing a tractor along some path or hammering some iron gradually
giving it some shape, will differ from the atomic event that preceeds it in time and from the atomic
event that follows it in time.
Those process verbs that have a unique relation to their direct internal argument assigns the
effected patient role to their object, those process verbs whose relation to their direct internal
argument is not unique assign the affected patient role to their internal direct argument. In 14 there is
no natural "built-in" end-point in the event. The verbs contain an unlimited number of atomic events,
in 14, a , the object has the effected theta role, but the reference type of the object is cummulative,
therefore the cummulative reference of the internal direct argument is carried over, the event has
cummulatve reference, that is the event is atelic. In 14, b, and 14, c, it seems that independently of the
reference type of the internal argument, the events are atelic. Affected arguments can undergo
changes imposed by the verb, but cannot delimit the event.
Accomplishment verbs
According to compositional theories accomplishment situations have a natural end point
provided by the quantized direct argument that is assigned the effected patient role by a process verb.
If the process verb can assigns affected patient role to its internal argument, the event can be delimited
by adding further internal arguments to the event (directional prepositional phrases, resultative
secondary predicates).
Aspect and Argument Structure
155
15, a, János házat épitett.
János-nom house-acc built
János built houses.
b, János épitett egy házat.
János-nom built a house-acc
János built a house.
16, a, Péter egy darab fát festett.
Peter-nom a piece of wood-acc painted
Peter painted a piece of wood.
b, Péter pirosra festett egy darab fát.
Peter-nom red-abl. painted a piece of wood.
Peter painted a piece of wood red.
17, a, Ferenc egy szekeret tolt.
Ferenc-nom a cart-acc pushed
Ferenc pushed the cart.
b, Ferenc a garázsba tolt egy szekeret.
Ferenc-nom the garage-into pushed the cart-acc
Ferenc pushed the cart into the garage.
According to Krifka (1992) direct arguments can carry over their referential type to the event if
they are connected to the event by the appropriate thematic roles. In 15,a, the amount of the house that
is involved in the building event is not well-defined at all. It can be one, more than one or even less
than one if Peter is involved only in the building of roof activity. In 16, b, the event of building a
house is completed when the house is ready. In 16, an existing object obtains a new property, in 17,
the cart as a whole is involved in the pushing event. As the verb "build" assigns effected patient role
to its internal direct argument, a quantized direct argument can delimit the event. In 16 a, 17 a, as
the verbs push and paint assign affected patient role to their arguments, they cannot delimit the event,
they need the help of other interal arguments. In 16,b, it is a directional prepositional phrase that
marks the endpoint of the event, while in 17, b, when the piece of wood is covered with red painting.
Process verbs can always be delimited, either by their internal direct argument. Process verbs can
always have a direct argument. If the direct argument do not have the effected argument role, it can
have an indirect argument, as too.
Achievement verbs
The internal tempora constitution of achievement verbs consists exactly two atomic events. As
oppossed to process verbs achievement verbs are delimited temporally.
18 a, János könyvet/ egy könyvet adott Máriának.
János-nom book-acc/a book-acc gave Mária-to
János gave a book/books to Mária.
Aspect and Argument Structure
156
b, Szinház/egy szinház alakult az egyetemen.
theatre/a theatre founded the university-at
There were theatres /there was a theatre founded at the university.
c, Vendég/egy vendég érkezett.
guest-nom/a guest-nom arrived
There has arrived a guest/there have arrived guests.
It seems that achievement verbs assign the same type of thematic roles (affected, effected
patient roles) that process verbs do. In 18 a, the position of the existing book changes, in 18, b, a new
theatre comes into existence, in 18, c, by being involved in the event of arriving the individual obtains
the property of becoming a guest. Achievement verbs do not need quantized internal arguments so
that they could occur in telic events. The referential type of the direct object cannot be carried over to
the event, with bare existential object achievemnt verbs form telic event. The internal temporal
constitution can temporally delimit the event.
Classification of Verbs
Verbs can be marked or unmarked with respect to the quantificational properties of their
internal temporal constitution. Stative verbs do not consists of atomic event,while process and
achievemnt verbs do. Process verbs is constructed of an unlimited number of atomic events, while
acievement verbs is temporally constituted of two atomic events exactly.Accomplishment situations
can be derived from process situations with the relevent type of internal argument. The claim is that
verbs have quantificational properties with respect to their internal temporal structure. Statuve verbs
are underspecified for (Q) feature as they do not have internal temporal structure, process verbs are (Q), as their internal temporal constitution is not limited. Achievement verbs consist a limited number
of atomic events, namely two. They are positively specified for the (Q) feature.
Telic events can be achieved if either the effected or affected patient role is assigned to the
internal direct object, which seems to be the case for all process and achievement verbs. Further on,
inner aspect can be calculated on the basis of the interaction of the quantificational features in the
verbal domain and the nominal domain I assume against Verkuyl (1993) that to get the telic reading it
is sufficient for either the verb or for the nominal to be quantificational.
Table 1
Classes of verbs
thematic roles
event type
V (Q)
stimulus
atelic
V (-Q)
effected arg. (-Q)
atelic
affected arg. (-Q)
atelic
effected arg. (+Q)
telic
affected arg. (+Q), XP
telic
effected arg. (+/-Q)
telic
affected arg. (+/-Q), XP
telic
V (-Q)
V(+Q)
In this theory stative verbs can never occur in telic events, because they are unable to assign
the proper thematic role to their internal direct due to their internal atemporal constitution. Process
157
Aspect and Argument Structure
and achievemnt verbs can always occur in telic event. They can assign the proper thematic roles to
their internal direct argument. Process verbs as they are inherently (-Q) can occur in telic events if
they are combined with quantized internal arguments. Achievement verbs are inherently telic as they
have (+Q) feature. The internal direct argument can be either quantized or bare existential.
External arguments and aspectual verb classes
It has been claimed that external arguments have no role in the calculation of inner aspectual
structure. External arguments are still sensitive to the internal temporal structure of the verb. They
can typically cooccur only with dynamic verbs, not with stative verbs (5).
Following Marantz (1983), Kratzer (1994) claims that external arguments are not arguments of
the verb. They are not in the VP in the initial syntactic structure, at all. External arguments are
generated in the specifier position of a functional projection she calls VoiceP which is headed by the
predicate CAUSE . Following Kratzer (1994) I claim that the external argument is not the argument of
the verb but the predicate CAUSE. That would explain, why external arguments do not participate in
VP aspect. They never occur in VP. I further assume that the predicate CAUSE has two arguments,
the external argument and the VP. The external argument can cooccur with only certain types of
verbs, namely dynamic verbs, because CAUSE selects for dynamic Vps.
Conclusion
A new classification of verbs is proposed on the basis of their internal temporal structure. This
new classification elimits the redundancy of the Vendler-Dowty classification, and gives the right
results with achievement verbs (problems that arise in the Verkuyl-Krifka theory) that heavily rely on
the referential type of the internal argument in the calculation of (a)telic event.
The theory further predicts the presence of external, direct internal and indirect internal
argument with minimal information about the verbal predicate.
Footnotes
*
I would like to thank Agnes Bende-Farkas, Grete Dalmi, Laszlo Kalman, Andras Komlosy, and Marta
Maleczki for discussion concerning the issues of this paper. Errors, of course, are all mine. This work has been
supported by the Research Support Scheme of the Open Sociaty Support Foundation, grant No.:320/1998.
1
, The inner aspect corresponds roughly to the composition between verbal features and referential features of
internal arguments. The outer aspect will correspond to the roles of adverbials, negation, progressive operators,
etc. They can modify inner aspect, but (a)telicity is specified for the VP once and for all..
2
I use the notion event for VP (inner) aspect.
3
The nominal construction refers cumulatively :whenever there are two entities to which that predicate applies,
this predicate applies to their collection, as well. A nominal construction has a quantificational reference
whenever there are two (different) entities to which that predicate applies, the predicate cannot apply to their
collection. Whenever there are two entities to which the predicate wine applies, this predicate applies to their
collection, as well, that is not the case with the predicate a glass of wine.
4
Dowty lists the properties which contribute to Proto-Agent and Proto-Patient role as (from Dowty (1991) p.
572):
i, Contributing Properties for the Agent Proto-Role:
a, volitional involvement in the event or state
b, sentience
c, causing an event or a change of state in another participant
d, movement (relative to the position of another participant)
(e, exists independently of the evnt named by the verb)
158
Aspect and Argument Structure
ii,Contributing Properties to the Proto-Patient Role:
a, undergoes change of state
b, incremental theme
c, causally effected by another participant
d, stationary relative to the movement of another participant
(e, does not exist independently of the event, or not at all)
5
Rizzi and Belletti (1988) claim, that psych verbs that are typically stative verbs do not have external
arguments, at all.
References
Bach, Emmon. (1986), The algebra of events, Linguistics and Philosophy, 9, 5-16.
Comrie, Bernard. (1987) Aspect. Cambridge:Cambridge University Press.
Diesing, Molly. (1992), Indefinites. Cambridge MA:MIT Press.
Dowty, David. (1979), Word meaning and Montague Grammar. The Semantics of Verbs and
Times in Generative Semantics and in Montague's PTQ. Reidel: Dordrecht.
Dowty, David. (1991), Thematic Proto-Roles and Argument Selection. Language 67.3:
547-619
de Hoop, Helen. ( 1992), Case configuration and NP interpretation. Doctoral Dissertation. U.
of Groeningen.
Johnson, Kyle. (1991), Object Positions. Natural Languge and Syntactic Theory 9. 577-636.
Kiefer, Ferenc. (1994), Aspect and syntactic structure, in F. Kiefer and K. E. Kiss (eds.) The
syntactic structure of Hungarian (Syntax and Semantics 27), Academic Press, New York.
Kratzer, Angelika, (1994), The event argument and the semantics of Voice. ms. University of
Massachusetts at Amhurst.
Krifka, Manfred. (1989), Nominal Reference, Temporal Constitution and Quantification in
Event Semantics. In R Bartsch, J. van Bethem, and P. van Emde Boas (eds.) Semantics
and Contextual Expression. Foris Dordrecht, 75-115.
Maleczki, Marta. (1992), Nominal reference, temporal constitution and quantification in event
semantics. In: Semantics and Contextual Expression (R. Bartsch, J. van Benthem and
van Emde boas, eds.), pp. 357-375. Institute for Logic, Language and Computation,
University of Amsterdam.
Pustejovsky, James (1989), The Geometry of Events. In C. Tenny (ed.) Generative Approaches
to Aspect. Cambridge, MA: MIT Lexicon Project.
Pustejovsky. James. (1991), The syntax of even structure. Cognition.
Szabolcsi, Anna. (1986), From the Definiteness Effect to lexical integrity. In: Topic, Focus and
Configurationality. (W. Abraham and S. de Meij, eds.), pp. 332-360. John Benjamins,
Amsterdam.
Tenny, Carol. (1994), Aspectual Roles and the syntax-semantics. Kluwer Academic
Publishers.Dortrecht, The Netherlands
Verkuy, Henk (1989) Aspectual Classes and Aspectual Composition. Linguistics and
Deconstructing syllable structure
159
Deconstructing syllable structure
Szigetvari Péter
By listing some empirical evidence and introducing theoretical considerations, this paper argues
for the idea that the phonological skeleton is made up of strictly alternating C and V positions. The
model advocated here claims that no two consonants and no two vowels are ever adjacent in the
phonological representation—if adjacency is defined at the level of the skeleton. This is rather
counterintuitive unless one accepts the possibility of empty skeletal positions. If so, the claim acquires
a new meaning: whenever adjacent consonants or adjacent, vowels (that is to say long vowels or
diphthongs) are encountered their representation will involve an intervening empty vocalic or
consonantal position, respectively. Accordingly, the first part of the paper shows why the possibility
of empty skeletal positions ought not to be rejected and, if looked at from a non-Indo-European
vantage point, is in fact the null hypothesis. The second part aims at demonstrating that the arguments
supporting the status of the syllabic constituent coda are rather weak, in fact, the traditional syllable
structure, incorporating an onset, a nucleus and a coda, can be dismantled in favour of a simpler
model involving only consonantal and vocalic skeletal positions.
Empty positions in the skeleton
One of the most important achievements of modern linguistics is the discovery of the use of
emptiness. Its relevance can be likened to that of the concept of zero, without which it is hard to
imagine the progress natural sciences have made in the last few centuries. The aim of the discussion
that follows is to convince the reader that empty positions in the phonological skeleton are not merely
a tricky device to ease the analysis, but rather a logical conclusion of various different lines of thought
pursued by theorists of modern phonology.
The skeleton-melody relationship
A not so recent advancement in phonological theory is the recognition of the necessity of
separating the quantitative and the qualitative aspects of segments. In this line of research, first called
prosodic, later reinvented as autosegmental phonology, the quantitative aspect is represented by a socalled skeletal tier, the qualitative by the melodic tier. The exact content of these two tiers is one of
the most important issues of current research. The skeletal tier primarily encodes the temporal
extension of the given stretch of the sound flow, while the melodic tiers contain melodic primes—
features, as they are standardly referred to—, which stand for the acoustic signal, on the one hand, and
the oral gymnastics; on the other, that are manifest in the period of time represented by the stretch of
the skeleton the given primes are associated with. The relationship between the elements of the two
tiers is negotiated by association lines.
With the advent of the autosegmental model, it becomes necessary to explore the consequences
of non-biunique relationships between the two. Having one batcli of melody defining primes
associated to two skeletal positions is the best-known and probably least controversial option,
standardly employed to represent some acoustic property stretching across multiple timing slots (la).
The realization of this configuration ranges from long vowels (a:) and some diphthongs (ei). through
genuine geminate consonants (t:). to partially identical clusters, like adjacent monomorphemic
homorganic consonants (mb). The complementary configuration — two pieces of melodic material
linked to the same skeletal position—is also a common thing, given that sounds are usually thought of
as composite entities (1b). (The Greek letters represent melodic primes.)
(1) a.
b.
Deconstructing syllable structure
160
Association of melodic material and skeletal slots includes not only one-to-two, but also one-tothrcc, one-to-four, etc., associations (2a). What is intriguing is that while such configurations
obviously exist — vowel harmony and tone phenomena very often exemplify unbounded spreading of
melodic material through longer skeletal strings—, three-long consonants (t::) or vowels (a::)
(allegedly present in, for example, Estonian) are standardly explained away, analysed in such a way
that does not involve a structure like the one in (2a) and supposed to be noncontrastive even if
phonetically existent. Phonological theory must find a way of rendering such structures impossible, or
at least highly marked.
(2) a.
b.
Two further options that deviate from the boring one-to-one relationship arc available in an
autosegmental model. One is melodic material without an associated skeletal slot. Such floating
segments are very useful in handling alternations where in what looks like the base form of a word
there is nothing to indicate the presence of melodic material surfacing in some other, oblique form.
This option is used, for example, by Kenstowicz & Rubach (1987) in their analysis of yers in
Slovakian. The phonetic identity of a realized yer is usually predictable in Slavic languages, but in
Slovakian the decisive factor, the palatalization of surrounding consonants: is lost, rendering the
quality of the surfacing yer unpredictable. Another alternation of this type is liaison, which is
especially intriguing when there exist other words with phonologically similar base forms which fail
to manifest the same alternation. Such is the case in, for example, the textbook RP19 grammar is
graema r iz vs. gamm∂ is gaema iz, where the base forms are graem∂ and gaem∂, respectively. The
presence of the r in the first but not in the second case is neatly explained by assuming that graem∂ is
lexically furnished with an r that lacks (or is unassociated to) a skeletal slot. while gaema has no r of
any kind, as shown in (3).
(3) a.
b.
Such an account avoids the use of brute force deletion, i.e.. maintains monotomcity (cf. Kálmán
1989), to explain the failure of the r to surface in case no vowel-initial string follows (e.g., grammar
book graem∂ buk). It also presupposes that phonetic interpretation proceeds on the skeleton, realizing
those and only those portions of the melody that are associated with the skeleton. If the mere presence
of melodic material in the representation were enough for its being phonetically interpreted the option
of unpronounced floating melody would not be viable.
The complementary situation is obviously a skeletal slot without any melodic content
associated to it. This configuration comes handy again in dealing with liaison phenomena:
for the floating liaison consonant to be interpreted it must be linked to a skeletal position. Since
such consonants are typically pronounced only if a vowel-initial word (or suffix) follows, all that need
be hypothesized is that such words carry a skeletal slot at their beginning which is not associated to
19
This dialect is sometimes claimed to be nonexistent outside prescriptively biased books on English
pronunciation (Harris 1994 : 293, note 5), though Jones, for example, claims he has had this dialect
(1967:xxvii). Whatever its reality, it illustrates the case in point.
161
Deconstructing syllable structure
any melodic material lexically, like at the beginning of is iz in (3). The floating melody thus has a
chance to associate and hence get interpreted.
Though this account appears elegant at first sight, there is some theoretical difficulty with it. If
the phonetic interpreter takes consecutive skeletal positions as its input and realizes whatever melody
is linked to each, one may wonder what should happen when it encounters a position to which no
melody is associated. There are two obvious possibilities, depending on the theoretical status of
skeletal slots, i.e., whether they represent a segment of the speech flow or a more abstract entity which
if empty is interpreted as silence. If we make the assumption that skeletal slots are segments of the
speech flow each slot must be interpreted, including empty ones. Without any explicit melodic
material the phonetic interpretation of a position is not trivial. It is only in a framework applying
exclusively unary features that this task is worth attempting: if features are binary or scalar phonetic
interpretation may only begin once all feature values arc present—some lexically given, others
supplied during the phonological derivation. As opposed to this, unary features model privative
oppositions, where a contrast is produced by one of the parties lacking some property the other
possesses. That is some segments are made up of less features than others, a property that also reflects
the relative unmarkedness of the former type as compared to the latter. The bottom extremity of
markedness is a segment containing no features at all, i.e., an empty skeletal position.
As we are going to see it also makes sense to assume that the phonetic interpretation of an
empty position is a function of other factors of the representation, that is, the same skeletal slot devoid
of any melodic content may under certain circumstances be interpreted as a very unmarked segment,
while in other cases remain silent. The advantage this assumption buys us that segment-zero
alternations will not have to involve the insertion or deletion of any phonological material, the former
option being arbitrary, the latter nonmonotonous, instead they will follow from the interpretative
conventions.
Empty skeletal positions and the null hypothesis
One way of classifying current phonological theories is by the criterion whether they allow
skeletal positions to be empty or not. The stance one adopts in this issue is of substantial relevance to
the whole of a given theory. There arc several questions that the existence or nonexistence of empty
skeletal positions bears upon. To mention but a few:
the association of segments in phonological strings to syllabic constituents will be seen
radically differently if empty positions may occur and cases of segments alternating with zero must
also be analysed differently if we are reluctant to accept that a skeletal position may be empty: the
destructive; non-monotonous device of resyllabification is very often called for if one wishes to have
only positions with melodic content on the skeleton.
Taking the first case, let us assume the conventional syllable structure comprising an onset, a
nucleus and a coda. In the standard textbook account all three constituents come with a practically
unbounded branching potential, i.e., the onset in English may contain 0-3; the nucleus 1-2 and the
coda 0-5 segments (e.g., Giegerich 1992:153; 167). Being empirically correct this analysis fares well
for a description but is unusable when searching for an explanation; the number of branches for each
constituent ranges within patently stipulative limits. One wonders why the onset may contain up to
three segments, what inhibits it from having, say, four. The tacit assumptions behind this analysis are
the axioms that syllable boundaries necessarily coincide with word boundaries20 and that segments are
fully integrated into the prosodic hierarchy, that is, each segment belongs to some syllabic constituent,
each syllabic constituent belongs to some syllable and so on. The unfoundedness of the first axiom
becomes apparent if we consider that on another level of the prosodic hierarchy, that of feet,
boundaries do not necessarily coincide; words may begin with a degenerate foot and may end with a
sole stressed syllable, which is not usually referred to as a degenerate foot, it still lacks a dependent
20
E.g., Blevins 1995 : 209: "In all languages, syllable edges correspond with word/utterance edges...''
162
Deconstructing syllable structure
second syllable. The second axiom, full integration of segments, has to be given up by theorists
following this line as soon as it is realized that word edges tolerate a wider range of phonotactic
freedom21, and to handle such phenomena the notion of extrasyllabicity has to be introduced22.
There is yet another reason why Giegerich's (or other analysts' similar) constraints are spurious
theoretically: while the two consonantal constituents, the onset and the coda may be empty, i.e., they
may contain zero segments, the same possibility is not available for the vocalic portion of the syllable,
the nucleus. The excuse that may be brought up to explain this discrepancy is the head status of
nuclei; as the head of the syllable they must not be empty. Again, if we move to other levels of the
prosodic hierarchy the situation is different: both headless feet and headless segments23 are possible.
As for segment-zero alternations, we have already seen a case where hypothesizing an empty
skeletal position facilitates the analysis: liaison phenomena arc neatly describable by positing an
empty consonantal position before vowel-initial words. To take another instance, this time a vowel
alternating with zero. consider the onset m of the unsyncopated faem∂li, which becomes a coda in the
bisyllabic faemli. A similar but converse situation often arises with morphological concatenation, e.g.,
the coda I of tell becomes an onset in telling. Both of these cases involve resyllabification in theories
that want to maintain that prevocalic consonants arc in an onset, but reject the possibility of having
empty skeletal positions. Resyllabification, however, subverts the result of core syllabification,
thereby representing a serious challenge to phonological parsing: if in a framework it is allowed that
the syllabic status of elements be freely changed during the derivation, the possibility of tracing back
the derivation, getting from the surface signal to the underlying representation, reduces radically.24
One could argue that resyllabification is necessary because a word-final or preconsonantal consonant
behaves differently from its prevocalic alternant. This, of course, is true, but one must also admit that
resyllabification is simply a way of representing this fact, nothing that would offer any explanation. In
such a framework we know a consonant is in coda position because it behaves like consonants in coda
position usually do. Since being in coda position is not an empirical issue, codas have no theoryexternal status; we have no independent evidence for the codahood of a consonant apart from the fact
that it behaves like other consonants that we believe to be in the coda. If one wants to avoid applying
resyllabification, the alternative analysis of segment-zero alternations and morphological
concatenations will involve empty skeletal positions.
What apparently justifies theories of the skeleton that reject the possibility of empty positions is
the assumption that this is the null hypothesis. That is, empty skeletal positions ought not to be posited
unless there is no other way to analyse phonological phenomena. While it is true that accepting
skeletal positions that fail to be interpreted phonetically does bring some abstractness into a theory, it
is controversial whether their rejection is the null hypothesis. The generative power of a theory having
syllables of an unlimited size may be just as excessive as that of one having empty skeletal positions,
what matters is whether there are adequate means of curtailing the possibilities.
I will now argue against this, aiming to show that the prevalent view rejecting empty positions
is somewhat accidental, relying on tradition. Let us imagine that modern phonological theory
happened to be developed by people who spoke languages like Desano or Zulu, in which
phonological domains are exclusively CVCV... CV on the surface. If they spoke about syllables at all.
these phonologists would surely claim that syllables have the structure CV—or onset-nucleus, for us.
21
If syllables in English could in fact begin with three consonants and end in five, we would expect eightconsonant-long intervocalic sequences within words, but this also turns out to be a disappointed expectation.
22
E.g., Goldsmith 1990:123: "prosodic licensing, which require[s] that all elements be a member of some
syllable, or else be marked as contingently extrasyllabic.''
23
The head-nonhead distinction in segments is not universally accepted, but cf. Anderson & Ewen 1987, Kaye
& al. 1985, Schafer 1995, among others.
24
It was for similar considerations the Chomskyan syntax has abandoned the device of MOVEMENT, replacing
it with the notion of CHAINS.
163
Deconstructing syllable structure
Indo-European phonologists. After encountering more complex languages like Luo or Krenak, which
allow consonants word-finally. or others like Japanese, which allow certain types of consonant
clusters word-medially, it is far from obvious that our hypothetical phonologists would extend their
syllable template by adding a further optional C position at the end. It is just as likely that they would
hypothesize an empty V position between the two consonants or after the word-final one. In fact
empirical evidence exists for the latter assumption: Harris & Gussmann (1998:141) claim the all
syllabic writing systems assign such offending consonants to an independent syllable with an
uninterpreted vowel (dummy syllables as Harris & Gussmann refers to them). Now one may argue
that this is so for reasons of economy: one needs much less new symbols for the vowelless syllables—
the number of all consonants in the worst case, but coda consonants are typically only a small subset
of all consonants in a given language—, while, again in the worst case, the number of all CV syllables
(maximally C x V", where C is the number of consonants and I" of vowels in the language) would be
multiplied by the number of coda consonants if CVC symbols were introduced, i.e., for each of the
hypothetical ta, sa and ma symbols we would need as many new symbols as there occur coda
consonants with them. Although this graphical economy is not a linguistic argument, one also has to
admit that economical considerations do feature in scientific theories, too, as noted by Occam quite a
while ago, furthermore, once dummy syllable symbols are used literate people will unavoidably
analyse their language as having empty vowels at certain points in phonological strings and
phonologist will follow this tradition and, more importantly, they will take it to be the null hypothesis.
I hope to have shown that while the acceptance or otherwise of empty skeletal positions appears
to be a matter of scholarly taste (analyses applying both approaches abound, after all), laying the
burden of proof on theories with empty positions thinking that we have the null hypothesis on our side
is not right after all. What the null hypothesis is in this issue is most probably a question of tradition.
Syllable structure
Many current theories of phonological representation assume one or more levels between feet
and the skeleton in the prosodic hierarchy. These are occupied by so-called syllabic constituents
which organize skeletal positions and other syllabic constituents into syllables. Syllabic constituents
gain theoretical relevance when they prove to be indispensable in—or at least result in a substantial
simplification of—the formulation of phonological generalizations.
Syllables, on the other hand, are not uncontroversial entities. The notion has been abandoned
several times in the history of phonological theory, the best known case is probably that of the SPE
(Chomsky & Halle 1968). From the 1970s mainstream phonology has gradually returned to applying
this traditional concept, but interestingly in most cases25 it is not the syllable constituent itself that is
necessary for the analyses, but its subconstituents, the onset, the nucleus and the coda.
Why have syllable structure?
It has been noticed—e.g., by Kahn (1976)—that certain consonantal processes favour the
phonological environment depicted in (4).
(4)
If syllables have a theoretical status, the environment in (4) can simply be referred to as the end
of the syllable, i.e., its coda. There are two problems with this formula: first, it is not true that all
preconsonantal consonants exhibit coda-like behaviour, for example, we find glottalization in an
25 7
Reduplication may appear be an exception, though here again it is often not a syllable that is repeated, but
the head of the first onset and the following nucleus (Brockhaus 1995:215ff).
164
Deconstructing syllable structure
English word like A[?t]lantic but aspiration in a[?t]ractive. although the t is preconsonantal in both
cases. Thus it seems that syntagmatic relationships in the string of segments are not in themselves
enough to properly capture phonological environments. Second, even if they were so, the formula in
(4) makes use of an unnatural disjunction: there is nothing more common in the word boundary and
consonants than in. say, the word boundary and vowels.
As we have seen. the two contexts, — C and _#, can be unified by assigning both types of
consonants to a coda constituent. The relevant phonological rules can now be formulated by the
structure in (5).
(5)
In the case of contrasts like A[?t]lantic vs. a[?t]ractive all there is to do is to assign one of the
t's to the coda and the other elsewhere—obviously to the following onset. In many cases such
distinctions can be justified by independent evidence, in this one, for example, we can note that one of
the clusters in question, tl, does not occur word-initially, the other, tr. does.
One cannot, however, be satisfied with this much. While a significant degree of descriptive
adequacy is reached by the formulation in (5), explanatory adequacy is still wanting. For example,
lenition, a phenomenon typically associated with the coda position, manifested as glottalization in the
previous example, may be adequately captured by the generalization that coda consonants lenite. there
is, nevertheless, no reason why it should be the coda position of all that triggers weakening. One
promising initiative to an explanation is made by Itô (1986) and Goldsmith (1990), who claim that
codas have a weaker prosodic license than other domains of the syllable, therefore coda consonants
are more prone to lenition. There is still ground for insisting on the question why it is codas that have
a weaker prosodic license. An answer couched in the Government Phonology framework is provided
by Harris (1997), who posits a so-called licensing path in phono-logical domains ranging from the
most prominent nucleus through least prominent ones to the onsets of these nuclei. The claim is that
the further away a position is from the prime licensor, the more prone it is to lenition.
Problems with the standard view
In (6) I give a diagram that shows my interpretation of the syllable tree most widespread in the
literature (e.g., Lass 1984 : 252, Durand 1990 : 204, Giegerich 1992 :138, Carr 1993 : 196.
Kenstowicz 1994:253, Roca 1994:141, Blevins 1995:213).
(6)
The solid lines in (6) represent obligatory associations, the dashed lines are optional. i.e., one
nuclear segment is obligatory for any syllable, all the others—another nuclear segment and practically
any number of onset and coda segments—may or may not be added to complete a syllable.
Deconstructing syllable structure
165
Given this syllable template syllabifying strings is still not a trivial issue: the length of both
onsets and codas is rather flexible. Nuclei can be found applying the SONORITY SEQUENCING
PRINCIPLE, one possible wording of which is quoted in (7).
(7) The Sonority Sequencing Principle (SSP)
Within a syllabic sonority rises from the onset towards the nucleus and falls from the nucleus
towards the coda.
That is, the sonority peaks of a certain string, away from which sonority falls in both directions,
can be identified with the syllabic nuclei. Even if nuclei are spotted easily, the consonantal interlude
streching between two sonority peaks must be properly distributed among the coda and the onset. To
be able to do this in a principled way the ONSET MAXIMIZATION PRINCIPLE26 is formulated to the
effect of (8).
(8) The Onset Maximization Principle (OMP)
If a consonant can be assigned both to a coda and the following onset, assign it to the onset.
Equipped with this principle, consonantal interludes can be unambiguously divided: in a
VC4C3C2C1V string C1 always goes with the second vowel, then one has to test whether C2C1 is a valid
onset, if yes it goes with the second vowel, else the syllable boundary-is between C2 and C1, and so on.
One difficulty comes with deciding whether a given consonant cluster is a valid onset or not. The
assumption that the set of word-initial clusters is coextensive with that of valid onsets—and likewise
that of word-final clusters with that of valid codas—is often accepted (cf. footnote 2) but rarely if ever
supported by any evidence. In fact, what can be supported by empirical evidence is the falsity of this
hypothesis, as, for example, the CLOSED SYLLABLE ADJUSTMENT rule of French shows. According to
this rule e and ∂ surface as å in closed syllables, and althought sC clusters do occur word-initially,
they also close a syllable (Lowenstamm 1981:598f). If sC clusters are heterosyllabic within a word,
then it cannot be concluded that the set of well-formed onsets is that of word-initial clusters. On the
other hand, in most—perhaps all—languages single consonants that can turn up before a vowel may
also turn up word-initially27. On the other hand; it is not true that in all languages single consonants
that can turn up before a consonant may also turn up word-finally—this is most evident in the case of
languages that have word-internal codas, but lack word-final consonants, like Italian. Also word-final
consonants can very often not stand before a consonant word-medially—the distribution of English ð
and the affricates could exemplify this situation28. Therefore, we may conclude that the only inference
that can be drawn is the following:
whatever is an onset may turn up at the beginning of a word. To schematize:
(9) The relationship of consonant (s) at word and syllable margins
NAIVE VIEW
word-initial consonant(s) ⇔ syllable-initial consonant (s)
word-final consonant(s) ⇔ syllable-final consonant(s)
EVIDENCED VIEW
26
An alternative, negative name of the principle could be the ''coda minimalization principle." Both names
convey the superiority of onsets over codas. In Optimality Theory the same idea is manifest in the ONSET and
NoCoDA constraints.
27
Counterexamples include r and in English, as Péter Siptár (voce) points out. To explain them away. the first
is a variant of t or d, thus its status is not obvious, the special status of the second is copiously documented, see
Gussmann 1998 for a recent discussion.
The only counterexamples are rhythmic and logarithmic for ð — both have forms; rhythm and logarithm, in
which the ð and the m are not adjacent—, and some other syncope created clusters like in natural 'naet r∂lfor
the affricates.
28
Deconstructing syllable structure
166
word-initial consonant (s) ⇒
⇐ syllable-initial consonant (s)
word-final consonant (s) ⇒
⇒ syllable-final consonant (s)
Another method that may be of use in determining the end of the coda and the beginning of the
onset, i.e.. the syllable boundary; is provided by the SONORITY DISPERSION PRINCIPLE proposed by
Clements (1990). quoted in (10).
(10) The Sonority Dispersion Principle (SDP)
a. The preferred initial demisyllable maximizes the dispersion in sonority. b. The preferred final
demisyllable minimizes the dispersion in sonority.
An initial demisyllable is the first half of the syllable up to and including the vowel — with
certain language specific differences in the case of long vowels and diphthongs—, a final
demisyllable is the second half from and including the vowel; i.e.. the onset with the (first half of the)
nucleus and the (second half of the) nucleus with the coda. respectively. Sonority dispersion is
maximized if the individual members of the demisyllable are evenly distributed on the sonority scale:
in an initial demisyllable the first member being the least sonorous (an obstruent29), the last the most
sonorous (a vowel) and if there is a further member between them then that should be a liquid. In the
final demisyllable, sonority dispersion is minimized, that is, the best case is not to have a coda at all.
or at least have very sonorous segments in it. The OMP is a derivate of the SDP: it is not only
preferable net to have a coda. but also to have an unset and thereby a large—or at least some—
sonority distance in the onset-nucleus sequence.
In the case of a string like atla both the SDP and the OMP prefers the syllabification a.tla, yet
in many languages, including English or French, at.la is the accepted division, since tl is not
encountered word-initially and—as already noted—the t behaves differently before I and r. The third
logical possibility, atl.a, is the worst, it even violates the SSP, introduced in (7). What we end up with
arc both principles, the SDP and the OMP, only partially satisfied. One way out of this situation is to
abandon the apparently self-evident hypothesis that superficial adjacency is evidence of adjacency at
all levels. Syntacticians have long noticed this fact30, for phonologists it still is not always obvious.
Accepting the—let's call it—adjacency hypothesis makes it seem trivial to determine syllabic
structure simply by looking at the string of segments constituting the word. The price to pay is that we
have to content ourselves with dispreferred syllable structures and contacts, on the one hand. and the
unbelievable complexity and number that syllable types will exhibit, on the other. If we are not
willing to pay this price, we have to allow some degree of abstraction—although it is controversial
whether this is indeed an abstraction after all, as shown in section 1.2—, dispensing with the view that
adjacent segments are necessarily adjacent underlyingly. In this way, syllable structure can be
radically simplified.
Empty nuclei in the skeleton
In this section I am going to introduce a train of thought that allows skeletal positions to remain
empty, abandoning the adjacency hypothesis. Government Phonology (GP), especially Kaye & al.
1990, Kaye 1990 and Charette 1991; is one theory that uses empty vocalic positions, but is not unique
29
Clements assumes a five-step sonority scale: obstruents < nasals < liquids < glides < vowels. He claims that
the algorithm he gives for measuring sonority dispersion also works for more refined scales, but argues that such
scales lose cross-linguistic generalizations and become too language specific.
30
For example, current syntax posits an empty category in the string the man I want 0 to go but not in I want to
go in order to explain, among other things, the impossibility of wanna-contraction in the first.
Deconstructing syllable structure
167
in this respect, cf., for example, Anderson 1982, Spencer 1986, Burzio 1994, Siptár & Törkenczy
forthcoming31.
One motivation for Kaye & al. (1990) to assume empty nuclei bears close resemblance to the
impasse situation encountered above, the syllabification of atla. The claim is that any two consonants
that are indeed adjacent are in a governing relationship with each other, i.e., one of them governs the
other. The governing potential of specific consonants is determined by their melodic content32: some
consonants are typically governors, others typically governees. As a result, if a consonant cluster xy is
established as a coda x followed by an onset y—in which then y governs x—, the opposite; yx, will
definitely not be the same type of cluster, coda-onset in this case, since that would require the
previous governing relationship to be swapped, the governor y to now be governed by the governee x.
This is deemed impossible, because codas must always be governed by the following onset — the
theory claims.
Translated to our case, if alta is syllabified al.ta—and there is good reason to do that: having a
small sonority distance in the nucleus-coda sequence and a great one in the onset-nucleus sequence, it
perfectly matches the requirements of the SDP—, atla cannot be analysed as a coda-onset cluster too.
i.e., *at.la.33 If we arc also unable to squeeze both consonants into the onset (*a.tla) or the coda
(*atl.a), there is no possible syllabification in a model that accepts the adjacency hypothesis. It would
be desirable to say that the t of atla is an onset and the l a coda, since—as the SDP suggests—t is an
ideal onset consonant, and l is okay for a coda. This would unfortunately lead to a violation of the
constraint banning crossing lines as shown in (11), where
denotes the syllable node. 0, N and C
should be obvious.
(11)
Allowing melodically empty skeletal positions into our theory offers a solution to this problem:
we are now able to say that the two consonants are not adjacent underlyingly, there is an empty
vocalic34 position (Ø) between them. Thus we can have both consonants in separate onsets {a.tØ.la),
in an onset and a coda {a.t0l.a, this is a possible manifestation of the idea in (11)) or in separate codas
(ai. Øl.a), though the second option is a bit strange, the last one rather perverse and neither is favoured
by the SDP. The two-onset representation is the most plausible, (12) shows this option syllabified
with an empty skeletal position. The skeletal tier is now included since once we have empty positions
on it the alphabetic symbols abbreviating melody cannot simultaneously represent skeletal positions
anymore.
31
It is interesting to note that hypothesizing empty consonantal positions is more obvious—and chronologically
earlier (e.g., Selkirk & Vergnaud 1973, Clements & Keyser 1983)—than empty vocalic positions: the silence of
the former is more straightforward than that of the latter.
32
In some versions of the theory governing potential is a function of the charm value of the given segment, but
then charm is dependent on melodic content.
33
Note that Clements's (1990) theory would allow this option, albeit as a highly marked and un-preferable
syllable contact. By doing so, Clements is paving the way towards Optimality Theory, where "anything goes,"
constraints are more or less preferably violable.
34
Of course, one might hypothesize an empty consonantal position between the two consonants but that would
not bring him any closer to a viable analysis: hosting the extra C position is vet another pain in the neck.
Deconstructing syllable structure
168
(12)
It is an interesting question to ask how the SDP would react to the syllabification a.tØ.la. The
sonority of an unpronounced segment is undefined, therefore the sonority rise in the syllable t0 is
indeterminable. Nonetheless, the absence of codas is of merit in the eyes of the SDP; onset
maximization is fully performed.
There seems to be a difficulty with this solution. As we have seen in section 1.1. the phonetic
interpretation of inelodically empty skeletal positions is not. obvious: it may be the most unmarked
vocalic segment (∂, ø , i or something similar) if dominated by a nuclear position, or the most
unmarked consonantal segment (the identity of which is debatable and indeed debated in the
literature) if dominated by a nonnuclear position. i.e., the onset or the coda. This means that the
phonetic interpretation of the representation in (12) should be atl∂a or atøla, a pronunciation that
would cause no debate in phonologist circles as regards its syllabification. If we are to maintain the
results of section 1.1 and posit unpronounced empty positions simultaneously we have to claim that
some melodically empty skeletal positions arc pronounced, others arc not. The theory must provide
some means to predict the pronunciation or nonpronunciation of a skeletal position in each case. GP's
solution is the formulation of the phonological EMPTY CATEGORY PRINCIPLE, of which I will here
mention but one clause: "a melodically empty skeletal position remains unpronounced if properly
governed [... ]" (Kaye & al. 1990:
219). I am not going to present all the details of proper government at this point. Let it suffice
that a vocalic position is properly governed if followed by one consonant and a pronounced vowel. It
is in fact this vowel that is said to govern the one that precedes it; i.e., in the configuration V1CV2 Vg
properly govern V1.
To conclude the discussion of empty positions, we may say that by positing empty nuclear
positions in the skeleton the theory reduces the cases where consonants are syllabified into the coda
position. This tendency is in line with the generally accepted view that onsets are to be preferred over
codas in syllabification. One salient feature of GP is its affinity to turn generalizations that other
theories look at as universal preference statements into unviolable constraints. This property
distinguishes the approach quite radically from Optimality Theory, where any constraint is violable.
In the case discussed above, the fact that an obstruent-liquid cluster is a dispreferred coda-onset
cluster is tightened to the claim that it is never a coda-onset cluster. If one dares take this thought to its
conclusion, the next question to ask is if codas exist at all; after all the optimal final demisyllable is
one without a coda. We are going to proceed in this direction.
Does the coda exist?
What we have to examine is the arguments supporting the existence of the coda position. As it
was already noted there is a sharp asymmetry between the two margins of the syllable, the onset and
the coda. The most unmarked syllable type, available in all languages, is CV, i.e., one that contains an
onset but no coda. Furthermore, while in the unmarked case the onset is obligatory, it is the marked
case to have a coda.
One of the reasons why codas are posited in the first place is the assumption that syllable
boundaries and word boundaries coincide. If consonants arc found at the right margin of words then
169
Deconstructing syllable structure
they obviously occupy the right margin of a syllable. But. as we have seen, there is also phonological
evidence which indicates that word-final consonants are not uncontroversially codas.
Codas also have explanational value in the formalization of stress rules. In languages with
unfixed stress, rules are often sensitive to syllable weight. The standard case is that syllables with only
a short vowel count as light (therefore usually unstressable), while syllables more fleshy than that—
either closed by a consonant or containing a long vowel— are heavy (and attract stress). Positing a
constituent, the rhyme, dominating the nucleus and the coda facilitates the definition of heaviness:
syllables with rhymes containing one segment are light, those with multisegmental rhymes are heavy.
Unfortunately, neither the branching of the rhyme, nor that of the nucleus may be held to be
responsible for heaviness, all we can say is that one of the two must branch. Another problematic
aspect of this approach to syllable weight is the fact that onsets (apart from very few and therefore
suspect cases) do not contribute to it. One cither stipulates that only the size of the rhyme is relevant
or offers some theory that assigns weight, standardly referred to as mora. to the appropriate segments.
However, even the latter option does no more than formalizing the observation that coda consonants
do; while onset consonants do not influence the weight of a syllable, without explaining why this and
not the opposite should be the case. The alternative to be discussed below fares better in both respects:
it explains why both closed and long-vowelled syllables are heavy and why onsets do not count.
The minimal word phenomenon; that constrains the size of lexical words in a number of
languages as diverse as English, Hungarian, Beijing Mandarin, Khalkha Mongolian and Turkish, also
depends on a plausible formulation of heavy syllables. The observation is that in these languages a
lexical word cannot be a single light syllable, it must either be a heavy syllable or two light syllables.
In monosyllables the necessary weight is provided either by the length of the vowel or a final,
allegedly coda, consonant.
One of the standard arguments for constituenthood in the subsyllabic domain is the existence of
phonotactic constraints. For instance, the almost very strict restrictions holding between the two
members of a branching onset—disregarding sC clusters now— may be seen as evidence that such
consonants form a constituent. Similarly, in nuclei the types of attested vowel clusters, i.e.,
diphthongs and long vowels, is restricted to a small subset of all the possibilities. As opposed to this,
very few qualitative phonotactic constraints apply to VC clusters, that is, within the rhyme. Where we
do encounter phonotactic constraints between consonants is in intervocalic and word-final clusters.
Intervocalic clusters of the type nt, mp are rather unanimously analysed as heterosyllabic, coda-onset
clusters. Yet, it is not usual to consider these clusters as members of the same syllabic constituent.
Therefore, we may conclude that the existence of some phonotactic constraint between two segments
does not necessarily imply that they share their host constituent.
Recall that different syllabifications were suggested for al.ta and a.tØ.la, as shown here. If we
accept that some intervocalic consonant clusters are coda-onset clusters, while others are onset-onset
clusters containing an empty nucleus between them, our theory becomes indeterminate. Nothing
excludes the syllabification a.lØ.ta: there will be no way of knowing whether a cluster that satisfies
the criteria for coda-onset clusters is to be analysed as such or as an onset-onset cluster that
accidentally happens to contain consonants which would also make a coda-onset cluster35.
To summarize: the theoretical status of the coda is strongly challenged. It is an outcast in
markedness universals: onsets may even be obligatory but are never impossible in languages, codas
arc never obligatory and may even be impossible. Though positing a coda position seems to help in
distinguishing heavy and light syllables, there are serious problems with the formulation. Finally, the
possibility of analysing some clusters both as coda-onset and as onset-onset clusters loosens the
theoretical tightness of the framework.
35 17
Of course, phenomena like closed syllable shortening or heaviness for stress assignment may tilt the balance
in this or that direction, but only in case C.C and C0.C are treated differently in the analysis of these
phenomena.
Deconstructing syllable structure
170
Without codas
Making a constraint out of the preference of the Sonority Dispersion Principle, one may claim
that all syllables have an onset and none have a coda (cf. Lowenstamm 1996).36 Setting aside for the
time being the possibility of having more than one consonant in a single onset constituent, this means
that whenever we find a consonant that is not followed by a vowel it must be followed by an empty
nucleus — to make it, at least theoretically, an onset.
It is important to bear in mind that the question whether something is in coda position or not is
not an empirical one; this property does not in itself have any physical correlate. The rationale of
positing a coda position is to unify the contexts that pattern together in certain phonological
phenomena. If these contexts may be unified by other means there is no strong argument for keeping
codas in the theoretical vocabulary, unless one needs it for descriptive purposes, as a dated but useful
term, similarly to the way a syntactician would refer to S(entence)s even after showing that they are
I(nflection)P(hrase)s or C(omplementizer)P(hrase)s. This is the sense the word coda will be used
hereafter. Actually, if codas do not have a theoretical status then it does not make much sense to talk
about onsets either, even if—what is kept in benign ignorance—they are imagined to be potentially
branching; the onset constituent becomes the consonantal domain, as opposed to the nucleus, which
is, and always was, the vocalic domain. What are thus left of syllabic constituents is a consonantal and
a vocalic constituent.
Having stripped syllabic constituency so brutally, one might as well take the last move and
claim that neither the consonantal, nor the vocalic constituent ever branches, that is, the skeleton
contains a strict alternation of consonantal and vocalic positions; this is exactly what Lowenstamm
(1996) does. Arguments for this final step do not readily offer themselves, some motivations will,
nevertheless, be pointed out in the next section. Even without explicitly arguing against branching
nuclei and branching onsets, formal simplicity is a criterion that opts for nonbranching constituents.
Recall (from section 1.2), if current phonological theory were initiated by phonologists whose native
language was Zulu a CVCV skeleton would surely have been the starting point.
In this section we are going to see the way the CVCV framework handles some coda-related
phenomena discussed in section 4.
Heavy versus light syllables
In a theory comprising only CV pairs to represent syllabic structure; a light syllable will be
made up of one such pair. while a heavy syllable will contain two of them as shown in (13). where the
Greek letters stand for any, potentially identical, melodic material (if identical, the two symbols arc
merged in (13b)):
(13) a.
b.
c..
The advantages of the representations in (13) are the following: (i) the formulation of what
constitutes a heavy syllable is much less clumsy than if we were using the coda; all that has to be
distinguished is one vs. two CV pairs, as opposed to statements like ''either the nucleus or the rhyme
36
Note that GP theorists regularly argue that the coda consituent is nonexistent in their theory. There still is a
coda position in GP, since rhymes may branch, what the right branch dominates is the coda as opposed to the
other two consonantal positions that are in the onset (which may also be branching), i.e., the term coda is a
shorthand for the "postnuclear rhymal complement''. My aim above, however, was to show that as regards their
skeletal status all consonantal positions are equal, the only difference is whether a consonantal position is
followed by an interpreted vocalic position or not.
Deconstructing syllable structure
171
is branching." (ii) We get an explanation of why onsets do not contribute to syllable weight:
paradoxically rhymes do not contribute either, the question itself loses its significance. All we need
for a heavy syllable is two pronounced CV pairs, that is two CV pairs both containing some melodic
material37. The onset of such a syllable is the C of the first pair but whether it is filled or not is
immaterial, since its V will be filled, that is why it is taken to be a syllable in the traditional approach.
In a sense then a CV slice of the skeleton is the equivalent of the mora in frameworks that measure
syllable weight by that means, but unlike moraic frameworks we get a nonstipulative account for the
lack of onset weight. The CVCV approach, however, still owes an explanation for why word-final
consonants often fail to contribute to syllable weight.
Note also that in languages like Latin or English, where stress rules typically take the form "if
the penult is heavy stress it, if it is light stress the antepenult38,there is room for a simplified
formulation: e.g., stress the third last CV pair, boxed in the Latin words illustrating the rule in (14):20
(14)
a.
b.
c.
It is rather complicated to capture the minimal word constraint, which limits the size of content
words to two moras at least, in the traditional GP framework. Since word-final consonants are claimed
to be onsets followed by an empty nuclear position, one has to say that either the nucleus of the only
syllable of the minimal word must branch or the word must contain two onset-rhyme sequences39. The
CVCV formulation is trivial:
the minimal word contains two CV pairs (perhaps in order to be stressable).
Compensatory lengthening
Compensatory lengthening is another phenomenon that appears to call for coda positions in
representations. After the total lenition of a consonant in a weak prosodic position the loss is made up
37
In a subset of the languages distinguishing heavy and light syllables only (C)VV, but not (C)VC counts as
heavy. In such languages it is apparently the pronunciation of the V part of the CV unit that is taken into
account. Crucially, no language takes (C)VC to be heavy to the exclusion of (C)VV. This falls out neatly in the
CV model: in such a language the interpretation of the V should matter in the first, but that of the C in the
second CV pair. With rhymes and nuclei it is not so evident why there exist no languages where the branching
of the rhyme would make a syllable heavy, that of the nucleus would not.
38
The situation is not as neat as depicted here. Difficulties arise in the following cases: the third last CV pair
may contain an empty V position, stress in this case appearing on the fourth (fórmula 'rule'), word-final long
vowels count as if short {fáciõ 'make') and word-final consonants do not count (ácidus 'sour').
39
An alternative, slightly less disjunctive but no more plausible formulation is the following: a minimal word
must contain two slots dominated by a nuclear node.
Deconstructing syllable structure
172
for by the propagation of either the preceding vocalic or the following consonantal material, for
example, the reconstructed Greek form *esmi is realized in Classical Attic as e:mi 'I am', while Aeolic
has em:i. The latter event, where the place of a consonant is taken up by another consonant, is rather
easy to handle for both theories. Vowel lengthening on the other hand happens again in violation of
structure preservation in the coda approach: what used to be a consonantal position, coda. is lost and a
vocalic, nuclear position appears instead. The model offered by the CVCV approach does not face
such problems: the vacation of the C position by the loss of s either opens the way for the following C
position to occupy it (15c) or removes the obstacle that has prevented the preceding V from taking it
(lob). Which of the two strategies is applied can be predicted on a language—here dialect—specific
basis: it looks very much like a parameter.
(15)
a.
b.
c.
Against constituency
All three syllabic constituents, the onset; the rhyme and the nucleus, arc imagined to be
potentially branching by GP theorists and most mainstream researchers alike (for the latter, even the
coda is potentially branching). In the former; more restrictive, framework a maximal syllabic has one
of the structures depicted in (16).
(16)
In a GP-like framework the nonexistence of codas amounts to the claim that the rhyme
constituent does not branch, and if it does not branch it is not a syllabic constituent—it shares the fate
the coda has suffered earlier. It is in fact a felicitous development of the theory to have got rid of the
rhyme constituent, which is a nuisance in more than one respects. For one thing, the rhyme is the only
syllabic constituent that does not dominate exclusively skeletal slots but also another syllabic
constituent, the nucleus. This fact has led to uncertainty about whether and why a branching nucleus
may occur in a branching rhyme. In one view (that of, e.g.. Kaye & al. 1990) it cannot, because in
such a constituent—shown in (17) for those with a visual disposition—no head can be assigned. The
Deconstructing syllable structure
173
two constraints that head and dependent must be adjacent and that their relationship is unidirectional
destroys the hopes of all three possible candidates: the first is not adjacent to the third, the second
would have one dependend on the left, one on the right, the third is not adjacent to the first.
(17)
However, when forced to accept the structure in (17), as Harris (1994 :68f, 76f, 82f) is in order
to cater for words like dainty, easier. b[a:}sket. saint, post, wild40 etc., one may seek refuge in the idea
that the head of the rhyme is not on the skeleton, but it is the nuclear node itself. It is not unreasonable
to look for the head of a constituent among its daughters, after all. If the rhyme should no more exist,
the dilemma also perishes.41
If syllable heaviness is not (merely) a function of the number of skeletal positions in the rhyme,
representing long vowels and diphthongs by brandling nuclei becomes much less obvious. The wish
to keep syllables together as onset-rhyme sequences is also in vain if codas arc let loose. The
'"phonetic unity" of long vowels—whatever that should mean— is not a strong argument: a long
vowel is just as much a unit as a long consonant, the latter is. nevertheless, a coda-onset cluster, thus
not one constituent, in most frameworks. (Not to mention the fact that without codas long consonants
hopelessly become CØC clusters.)
The claim that "all feet are minimally binary and that the word in many languages must consist
minimally of a foot" (made by McCarthy & Prince (1986) and quoted by Harris (1997)) suggests that
just as tata and tat (the latter obviously tat. Ø) are binary feet—hence qualify for minimal words in the
languages concerned—, ta: must also somehow make a binary foot. The number of vocalic positions
involved in the string is undisputedly two, but. the immediate constituens of foot nodes arc usually
either syllable nodes or. in their absence, nuclei. Only by analysing the a: as two nuclei, i.e., NØN, do
we obtain a binary foot, thus satisfying the minimal limit on word size. Note that the same argument
was already brought up in section 5.1 cast in a slightly different form.
Kaye (1985:290f) and Lowenstamm & Kaye (1985-1986:99f) claim that there is an
implicational relationship between branching rhymes and branching onsets. The observation, called
the RHYME-DOMINANT PRINCIPLE, is that languages having branching onsets invariably have
branching rhymes (i.e., closed syllables), while the opposite is not true, languages with branching
rhymes may or may not have branching onsets. To put it in other words, branching onsets are more
marked than closed syllables42. Whether this calls for the abandonment of the hypothesis that onsets,
or rather, the consonantal constituent, may branch is not fully obvious. The question basically boils
down to the markedness of branching constituents and that of empty skeletal positions.43Theoretical
uniformity requires either the retention of constituency throughout the whole range of syllabic
constituents or their total abandonment, which means positing a CØC structure to branching onsets as
well.
40
Though Harris does allow type (17) superheavy rhymes (1994 :G9, 83), he also has to strictly limit their
occurrence to ones with coronal and very few other consonant clusters.
41
The problem of superheavy rhymes unfortunately does not disappear with this move. 17
42
Lowenstamm & Kaye (1985-1986: 111) also claim that long vowels are more marked than closed syllables,
that is, there exist no languages with long vowels and/or heavy diphthongs and only open syllables. If one
accepts the proposal suggested here, this is a further argument for the VØV representation of long vowels.
43
There is a third possibility, branching onsets could be considered to be contour segments (cf. Rennison 1998).
This idea includes large scale reshuffling of segmental representations, space limitations inhibit further
discussion here.
174
Deconstructing syllable structure
One last consideration that is relevant for the total rejection of syllabic constituency is that if
the skeleton contains strictly alternating C and V positions—no adjacent Cs and no adjacent Vs—then
it is trivial to parse a phonological string, provided the listener can distinguish consonants and vowels:
whenever he encounters two instances of the same category an empty position of the opposite type
must be inserted between them, while two different categories will be adjacent.44 This advantage is not
available in a system where at some points one may assume two adjacent Cs or Vs, at another they
will be separated by an empty category. Consequently, allowing empty skeletal positions into
phonological representations concludes to the hypothesis that the phonological skeleton must be made
up of strictly alternating Cs and Vs.
Conclusion
In this paper I collect evidence for a rather impoverished model of prosodic structure, one that
involves strictly alternating consonantal and vocalic positions. These two skeletal primes arc not
incorporated in further structures, thus the traditional syllable and syllabic constituents are dispensed
with. Whether their functions can be taken over by the simplistic organization proposed is beyond the
scope of the present paper, but indications of a positive answer arc suggested by recent work in the
area (e.g., Harris 1997. Ségéral & Scheer 1999, Dienes & Szigetvari 1999).
References
Anderson. Stephen R. 1982. The analysis of French schwa: or. how to get something for nothing. Language
58.3:534-573.
Anderson, John M. and Colin J. Ewen. 1987. Principles of Dependency Phonology. Cambridge: Cambridge
University Press.
Blevins, Juliette. 1995. The syllable in phonological theory In Goldsmith 1995:206-244.
Brockhaus, Wiebke. 1995. Skeletal and suprasegmental structure within Government Phonology. In Jacques
Durand and Francis Katamba (eds.). Frontiers of Phonology: Atoms; Structures. Derivations. Harlow: Longman.
180-221.
Burzio; Luigi. 1994. Principles of English Stress. Cambridge: Cambridge University Press. Carr. Philip. 1993.
Phonology. Houndmills & London: Macmillan.
Charette, Monik. 1991. Conditions on Phonological Government. Cambridge: Cambridge University Press.
Chomsky, Noam and Morris Halle. 1968. The Sound Pattern of English. New York: Harper & Row. Clements;
George N. 1990. The role of the sonority cycle in core syllabification. In John Kingston and
Mary E. Beckman (eds.) Papers in Laboratory Phonology 1: Between the Grammar and the Physics
of Speech. Cambridge: Cambridge University Press. 283-333.
Clements. George N. and S. Jay Keyser. 1983. CV Phonology: A Generative Theory of the Syllable.
Cambridge, Mass.: The MIT Press. Dienes, Peter and Peter Szigetvari. 1999. Repartitioning the skeleton: VC
Phonology. Ms., Eötvös
Loránd University.
Durand, Jacques. 1990. Generative and Non-Linear Phonology. Harlow: Longman. Giegerich,Heinz J.1992.
English Phonology: An Introduction. Cambridge: Cambridge University Press.
Goldsmith, John A. 1990. Autosegmental and Metrical Phonology. Oxford & Cambridge, Mass.: Black-well.
Goldsmith; John A. (ed.) 1995. The Handbook of Phonological Theory. Cambridge, Mass. & Oxford:
Blackwell.
44 )
This is only true if two adjacent empty positions are not allowed, two instances of the opposing categories may or may
not be adjacent (CøøV or VøøC).
175
Deconstructing syllable structure
Gussmann, Edmund. 1998. Domains, relations, and the English agma. In Cyran 1998:101-126. Harris, John.
1994. English Sound Structure. Oxford & Cambridge, Mass.: Blackwell.
Harris, John. 1997. Licensing Inheritance: an integrated theory of neutralisation. Phonology 14: 315-370.
Harris, John and Edmund Gussmann. 1998. Final codas: why the west was wrong. In Cyran 1998: 139-162.
Itö, Junko. 1986. Syllabic Theory in Prosodic Phonology. Doctoral dissertation, University of Massachusetts.
(Published by Garland Press, New York. 1988.)
Jones, Daniel. 1967. Everyman's English Pronouncing Dictionary (13th ed.). London: Dent.
Kalin. Daniel. 1976. Syllable-based generalizations in English phonology. Doctoral dissertation, Massachusetts
Institute of Technology, Cambridge, Mass.
Kalman. Laszlo. 1989. Monotonicity in phonology. Acta Linguistira Hungarica 39:133 147.
Kaye. Jonathan. 1985. On the syllable structure of certain West African languages. In Didier Goyvaerts (ed.).
African Linguistics: Essays in Memory of M. W. K. Semikenke. Amsterdam: John Benjamins. 285 308.
Kaye, Jonathan. 1990. 'Coda' Licensing. Phonology 7:301 330.
Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1985. The internal structure of phono-logical
elements: a theory of charm and government. Phonology Yearbook 2 :305 328.
Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1990. Constituent structure and government in
phonology. Phonology 7:193 232.
Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Cambridge, Mass. & Oxford: Blackwell.
Kenstowicz, Michael and Jerzy Rubach. 1987. The phonology of syllabic nuclei in Slovak. Language 63.3 :
463-497.
Lass, Roger. 1984. Phonology: An Introduction to Basic Concepts. Cambridge: Cambridge University Press.
Lowenstamm, Jean. 1981. On the maximal cluster approach to syllable structure. Linguistic Inquiry 12 : 575604.
Lowenstamm, Jean. 1996. CV as the only syllable type. Ms., Université Paris 7. Appeared in Jacques Durand
and Bernard Laks (eds.). Current Trends in Phonology: Models and Methods. European Studies Research
Institute, University of Salford Publications. 419-442.
Lowenstamm, Jean and Jonathan Kaye. 1985-1986. Compensatory lengthening in Tiberian Hebrew. In Leo
Wetzels and Engin Sezer (eds.) Studies in Compensatory Lengthening. Dordrecht: Foris. 97-132.
McCarthy, John J. and Alan Prince. 1986. Prosodic morphology. Ms., University of Massachusetts, Amherst
and Brandeis University.
Nádasdy, Adam and Peter Siptar. 1989. Issues in Hungarian phonology: preliminary queries to a new-project.
Acta Linguistica Hungarica 39 : 3-27.
Rennison. John R. 1998, Branching onsets are contour segments. Paper presented at the 31st Poznan Linguistic
Meeting, Government Phonology Workshop, 2 May 1998.
Roca, Iggy. 1994. Generative Phonology. London & New York: Routledge.
Schafer, Robin. 1995. Headedness in the representation of affricates. The Linguistic Review 12:61-87.
Ségéral, Philippe and Tobias Scheer. 1999. The Coda mirror. Ms., Universite de Paris 7 and Universite de Nice.
Selkirk. Elisabeth 0. and Jean-Roger Vergnaud. 1973. How abstract is French phonology? Foundations of
Language 10:249-254.
Siptar, Peter and Miklós Törkenczy. forthcoming. The Phonology of Hungarian. Oxford: Clarendon
Press/Oxford University Press.
Spencer, Andrew. 1986. A non-linear analysis of vowel-zero alternations in Polish. Journal of Linguistics
22:249-280.
Phonotactic grammaticality and the lexicon
176
Phonotactic grammaticality and the lexicon
Miklós Törkenczy
Introduction
Native speakers have the ability to pass judgements on the acceptability (well-formedness) of
words, i.e. they can judge whether (or perhaps to what extent) a given string of segments may be a
word in their language. The well-formedness or ill-formedness of a word is attributable to an interplay
of several factors some of which are unquestionably nonphonological. For instance, the hypothetical
string */ba+rda+• / can be seen as an ill-formed Hungarian word, but only if analysed as a
combination of the noun bárd ‘hatchet’ and the deverbal suffix -ás/-és ‘-ness’: bárdN+ás, which is
morphologically ill-formed since this suffix cannot be attached to nouns. Phonologically, unattested
*/ba+rda+• / is just as acceptable as the attested word /± a+rda+• / csárdás ‘a type of Hungarian
dance’, i.e. the string of segments is a phonologically well-formed sequence: there is no restriction on
the combination of segments into strings in Hungarian that this sequence violates.1 In other words, it is
phonotactically grammatical.
This paper explores phonotactic grammaticality, specifically, the properties phonotactic
grammars (must) have. A phonotactic grammar (a subcomponent of the phonology) is the set of
constraints (or the mechanism) which specifies the (degree of) well-formedness of strings of
segments.
In what follows first I will discuss some general issues pertaining to (the interpretation of)
phonotactic grammaticality, then I will review and comment on three existing phonotactic models2
two of which work with more than two degrees of phonotactic well-formedness (multi-levelled
models), and in the last section I will discuss the factors that can influence phonotactic grammaticality
and could/should be incorporated into a phonotactic grammar. I will mainly use English examples
because the models to be discussed were mainly designed for English, but will point out differences
that would arise if a given model/assumption were applied to Hungarian.
Let us assume (as is usual) that the strings whose phonotactic grammaticality is to be
established ultimately are words. Given an inventory of segments (the phonological alphabet of L1)
and an upper limit on word-size, the set of logically possible word-strings can be computed.
Henceforward, I shall refer to this set as the ‘hypothetical set (Set H)’. A phonotactic grammar
partitions Set H into non-overlapping subsets which represent different degrees of grammaticality.
The number of these subsets depends on our view of phonotactic grammaticality. We may assume
that any string of segments is either fully well-formed (grammatical) or fully ill-formed
(ungrammatical) - in this case the number of subsets is two -, or we may assume that there are several
intermediate degrees of well-formedness between the fully grammatical and the fully ungrammatical in this case the number of subsets is more than two. Trivially, the number of subsets cannot exceed
the number of strings Set H contains. It is also reasonable to suppose that the number of subsets is low
although no reliable evidence is available as to how many degrees of well-formedness native speakers
can distinguish. Furthermore, the divisions should be established in a principled way, and not be
simply listing the members of the different subsets.
Naturally, only a subset of the strings Set H contains occurs as attested words in any given
language. I shall call this subset the ‘occurring set (Set O)’. It has been long recognised (and is
universally accepted in the literature; e.g. Fischer-J rgensen 1952, Vogt 1954, Halle 1962, Chomsky
1964) that there is no simple correspondence between the partitioning of Set H into subsets that differ
in grammaticality and the division of Set H into Set O and its complementary subset. Whether a
specific string occurs or not, does not (uniquely) determine its phonotactic grammaticality: nonoccurring strings may be grammatical (accidental gaps) and occurring strings may be ungrammatical
(unsystematic occurrences or irregular strings). Most phonotactic analyses (explicitly or implicitly)
assume a fourfold division of strings:
Phonotactic grammaticality and the lexicon
177
(1)
occurring
non-occurring
well-formed
regular occurrence
accidental gap
ill-formed
irregular occurrence
systematic gap
The classic examples from English are brick (regular occurrence), /blwk/ (accidental gap) and
/bnwk/ (systematic gap) (cf. Halle 1962). Schweppes, tsetse could be added as examples of irregular
occurrences. It is easy to find example from other languages, e.g. Hungarian: tÅgy ‘udder’ (regular
occurrence), bÅgy (accidental gap), akn (systematic gap), tromf (irregular occurrence).
Typically, two degrees of phonotactic grammaticality are assumed in phonotactic analyses and
discussions of phonological phenomena related to phonotactics. The reason is that - like elsewhere in
phonology - only two degrees of well-formedness (the contrast between the ill-formed and the wellformed) are needed in order to account for phenomena related to or motivated by phonotactics (like
epenthesis, deletion, syncope, etc). It is also usually taken for granted that native speaker judgements
also reflect this binary division of strings into well-formed and ill-formed. It has to be pointed out,
however, that this is not necessarily true, and there exists some experimental evidence that native
speakers can distinguish more than two degrees of phonotactic grammaticality ( Ohala 1984, 1986,
Scholes 1966, Greenberg and Jenkins 1964).3 It is perfectly possible then that a phonotactic grammar
should be able to divide Set H into more than one subsets ranked according their well-formedness.
Before we go on to review some of these models, some remarks are in order about the fourfold
division charted in (1) and the nature of the strings evaluated by a phonotactic grammar. It has to be
noted that commitment to a division of strings into four classes as in (1) does not necessarily mean
commitment two a two-level model of phonotactic grammaticality (fully well-formed vs. fully illformed). The reason lies in the meanings of the terms ‘gap’ and ‘occurrence’. These terms are used in
a dual meaning in the literature. Let us call the object whose phonotactic grammaticality is to be
ultimately determined a structural base. We have assumed that these are words.4 On the one hand,
occurrence and gap can be used to refer to the occurrence or non-occurrence of a sequence of
segments which is a substring of a structural base, but does not realise a structural base in itself. This
is the substring interpretation of the terms. For instance, it is often claimed that /stw/, which is
unattested in word/morpheme-initial position in English, is an accidental gap.5 This clearly is a
substring interpretation of the term ‘gap’. On the other hand occurrence and gap can refer to the
occurrence or non-occurrence of a sequence of segments which realises a structural base in itself (a
base-string). This is the base-string interpretation of the terms gap and occurrence. Under the latter
interpretation ‘gap’ does not mean ‘gap in the pattern of clusters’ but a gap in the lexicon: an item is
missing from the lexicon6. Under this interpretation, occurrences are items found in the lexicon. The
structural base and the substring senses of the terms ‘gap’ and ‘occurrence’ are obviously related. The
grammaticality of a particular structural base depends on the grammaticality of the substrings making
up the string realising the structural base. This means that the phonotactic grammaticality of a
structural base is derived of that of the substrings it contains. We shall see that it does not follow that
the algorithm that assigns structural bases to degrees of phonotactic grammaticality necessarily has to
be indirect, i.e. refer to the pre-established phonotactic grammaticality of the substrings. But if the
algorithm is indirect, it is perfectly possible for the substrings to be assigned to just two degrees of
grammaticality (well-formed and ill-formed) while the ultimate grammaticality ratings of the
structural bases themselves are multi-level (for instance, a base-string may contain more than one illformed substring and the algorithm assessing the grammaticality of base-strings is sensitive to the
number of ill-formed substrings a base sting contains).
178
Phonotactic grammaticality and the lexicon
The central question of phonotactic grammaticality is how to distinguish the different degrees
(the fully well-formed from the fully ill-formed, or the different degrees from one another) in a
principled way. All treatments agree in the central intuition that while unlistedness in the lexicon does
not necessarily entail ungrammaticality, the more similar an item is to most of the listed items, the
more grammatical it is (or in the case of two-level models: the more likely it is to be grammatical).
The problem is how to formalise this intuition into an algorithm.
The SPE-a algorithm
The only explicit early generative approach to phonotactic grammaticality (the first measure
proposed in SPE, cf. pp. 380-381) incorporates the pre-generative claim that strings excluded by a
general statement are structural gaps, and those excluded by a non-general statement are accidental
gaps.7 The basic idea is the following: ‘[...] when a rule specifying coefficients of features in certain
configurations is added to the lexicon, the predicted values can be left unspecified in lexical entries.
We might propose that if the number of predicted coefficients is greater than the number of specified
features in the rule in question, then the addition of the rule represents a true generalisation. Once
added to the grammar, this rule excludes certain unattested configurations that would be inconsistent
with it. On the other hand, when all such rules are added, there will still be many unattested
configurations consistent with this “simplest set” of rules; these, then, would be the “accidental gaps”,
the admissible, unrealized matrices. Thus, we can draw the distinction between admissible and
inadmissible configurations in terms of a rather natural extension of the method of evaluation to the
lexicon’ (SPE: pp. 381-382). Thus, SPE-a claims that if we have a rule R (an MSC in SPE) whose
cost (i.e. the number of features needed to state the rule) exceeds the number of features that could be
left unspecified in the lexicon if we added the rule to the lexicon, then R does not represent a ‘true
generalisation’, consequently, must not be added to the grammar. The non-occurring strings that
would be excluded by R are accidental gaps. On the other hand, a rule represents a true generalisation,
and must be added to the grammar if its cost is smaller than the saving its addition would effect in the
lexicon. The non-occurring strings excluded by such a rule are systematic gaps, and the non-occurring
strings permitted by such a rule (and not excluded by other such rules) are accidental gaps. This
measure applies to the classic examples brick, /blwk/, /bnwk/ in the following way. No rule can be
formulated to exclude brick because it is listed in the lexicon. The rule excluding unlisted /bnwk/ can
be formulated with reference to the initial consonant cluster (nasals cannot occur in the second place
in initial clusters English). It is a fairly simple rule in terms of the number of features needed to state
it, and would result in a ‘saving’ of several features for every lexical item beginning with two
consonants in the lexicon. Therefore, the rule excluding it is a ‘true’ phonotactic rule, and /bnwk/ is a
systematic gap. By contrast, the rule formulated to exclude /blwk/ cannot be formulated with reference
to any of its substrings, but would have to refer to the whole string as a base-string, i.e. it would have
/blwk/ as a lexical item. Thus, it would be a highly specific rule which would save very few features.
Consequently, the putative rule formulated to exclude /blwk/ is not a phonotactic rule of English and
/blwk/ is an accidental gap. It is clear that the assumption underlying the measure is the following: if
we can only formulate a statement excluding a string realising a structural base in such a way that
must refer to the whole base-string and not to any of its substrings, then the non-occurrence of the
string in question is accidental. The view of the lexicon, lexical representations, underspecification,
and the notion of MSC-s as conceived by SPE are outdated, but this is beside the point: the measure
could be made to conform to more recent views (e.g. instead of counting the features that are left
unspecified, one could count the features that could be unspecified, etc.). SPE-a is a measure of only
two degrees of phonotactic grammaticality (well-formed vs. ill-formed). It can only distinguish
accidental gaps from systematic gaps, but cannot distinguish regular occurrences from irregular ones
(i.e. it assumes that the items listed in the lexicon are uniformly regular/well-formed).8 It measures the
grammaticality of base-strings directly and cannot measure the grammaticality of substrings, i.e. any
unlisted base-string that can be excluded with reference to a substring will be evaluated as a
systematic gap: unlisted *stwig will be just as ungrammatical as *spwig - even though it can be argued
that */stw/ is and accidental gap while */spw/ is a systematic gap in English (see note 5).
Phonotactic grammaticality and the lexicon
179
The SPE-b algorithm
In SPE another phonotactic measure is put forward (I shall call it SPE-b). SPE-b is clearly
meant to supersede the first measure. SPE-b, as opposed to SPE-a is a multi-level phonotactic
algorithm, i.e. it assigns strings to more than two degrees of phonotactic grammaticality. SPE-b
claims that ‘Items that do not occur in the lexicon differ strikingly in their “degree of admissibility”.
Hence a real solution to the problem of admissibility will not simply define a tripartite categorization
of occurring, accidental gap and inadmissible, but will define the “degree of admissibility” of each
lexical matrix [...]’ (SPE: pp. 416-417.). The degree of admissibility is measured as a distance of a
string from the lexicon. The distance of a potential lexical item (a matrix of features) : from the
lexicon (L) is defined as follows: a rule R ‘distinguishes : from L if [R] does not change any member
[<] of L (i.e. given < 0 L, either [R] is inapplicable to < or it leaves < unaltered) but [R] does change :;
and furthermore [R] is minimal in that any other rule meeting these conditions contains at least as
many features specified [+F] or [-F] as does [R]’ (SPE: p. 417.). The distance of a potential lexical
item from the lexicon is 1/n where n is the number of specified features in the rule distinguishing the
string in question from the lexicon. The distance is taken to be zero if there is no rule that would
distinguish the item from the lexicon. ‘[...] to determine the distance of a matrix from L, we find the
simplest rule which is “true of L”, in the obvious sense, but not true of :, and we take the distance of :
from L to be inversely related to the complexity of this rule’ (SPE: p. 417.).
Thus, the distance of brick from the English lexicon (Le) is zero because there is no rule that
would distinguish it from Le (i.e. it is listed). According to SPE the distance of /blwk/ from Le is 1/17
because the minimal rule that distinguishes it from Le has 17 specified features (the rule states in the
SPE feature system that a consonant must be non-lateral in the context /b_wk/ initially in a basestring). The distance of /bnwk/ from Le is 1/5 because the minimal rule that distinguishes it from Le
has 5 specified features (the rule states that a consonant must be non-nasal following an initial noncontinuant in a base-string). The distance of /bnzk/ from Le is 1/4 because the minimal rule that
distinguishes it from Le has 4 specified features (the rule states that a segment following a base-initial
two-term cluster whose second consonant is a nasal must be a vowel). Thus, according to SPE-b the
above strings would be rated on a scale of decreasing phonotactic grammaticality as show in (2):
(2)
brick > /blwk/ > /bnwk/ > /bnzk/
These assignments seem correct intuitively. Note that SPE-b is an extension of SPE-a. The
main difference between the two algorithms is that SPE-b allows for grammaticality assignments
where the number of degrees of phonotactic grammaticality exceeds two. Otherwise, the two models
are very similar. Both are based on the assumption that the more general the (negative) phonotactic
rule is, the fewer features are needed to state it, and it is the generality of the rule (whether it is
actually part of the grammar or an extragrammatical measure) that is the core of phonotactic
grammaticality: the more general the rule is, the less grammatical the measured string will be. Like
SPE-a, SPE-b also measures (i) base-strings only (and cannot measure the grammaticality of
substrings), (ii) unlisted base-strings only (and cannot distinguish between regular and irregular
occurrences) because (iii) it assumes that all listed base-strings are fully grammatical. We have seen
that (iii) is incorrect. There are two additional notable features of SPE-b: (iv) given an unlisted basestring whose grammaticality is to be established, its rating will only depend on a single deviation from
the lexicon, the one that can be excluded by the most general rule (thus, an unlisted base-string such
as /bnwNp/ that contains two ill-formed substrings /bn-/ and /-np/ is judged just as deviant as /bnwk/,
which only contains just one ); (v) there is no upper limit on the number of degrees of phonotactic
grammaticality the measure allows for. These assumptions make predictions about the
behaviour/intuitions of native speakers, and there is experimental evidence that (iv) is untrue (Ohala
1984, 1986).
Phonotactic grammaticality and the lexicon
180
The Greenberg and Jenkins algorithm
As opposed to SPE-b, the Greenberg and Jenkins algorithm, henceforward GJ (Greenberg and
Jenkins 1964) assumes that number of violations in a base-string influences its phonotactic
grammaticality, i.e. that native speakers are sensitive to the number of violations within a base-string.
G&J is a segment substitution algorithm. ‘If we have a sequence of length n, we can substitute
from zero up to n at a time and each such number of substitution can be done n over i ways where i
takes on values from 0 to n (Greenberg and Jenkins 1964: 158). The total number of substitutions is
2n. Zero substitution means leaving the word as it is (=no substitution). Thus, in a CVC string the total
number of substitutions is 23=8. Greenberg and Jenkins claim that the phonotactic grammaticality of a
word depends on the number of successful substitutions, i.e. those substitutions that yield an existing
word. They propose the following measure of phonotactic grammaticality:
(3)
G = (Sp + 1) _ Ss
where Sp is the number of logically possible substitutions and Ss is the number of successful
substitutions. If G=1, the string is fully grammatical. The greater G is, the less grammatical the string
is. Figure (4), which is based on Ohala (1984), shows the grammaticality ratings of three candidate
words /kr k/, /kl b/ and /kleb/ in English (the substituted segments are emboldened in the CV
skeletons):
(4)
/kr k/
/kl b/
/kleb/
CCVC
crack
!
!
CCVC
track
slab
!
CCVC
clack
crab
!
CCVC
creek
club
club
CCVC
crass
clam
Clem
CCVC
smack
stab
!
CCVC
click
crib
crib
CCVC
cream
clip
clip
CCVC
shriek
slob
slob
CCVC
clash
slam
phlegm
CCVC
trash
cram
kress
CCVC
slick
grub
grub
CCVC
clean
cream
cream
Phonotactic grammaticality and the lexicon
181
CCVC
trim
flip
flip
CCVC
blab
spat
bread
CCVC
gleam
gruff
gruff
Ss =
16
15
12
G=
1
2
5
(Sp=24=16)
Thus, these base-strings would be rated on a scale of decreasing phonotactic grammaticality as
show in (5):
(5)
crack >/kl b/ >/kleb/
Greenberg and Jenkins impose a number of restrictions on the way substitutions can be made:
(6)
i. consonants may only be substituted for consonants, vowels for vowels;
ii. substitution by zero (subtraction) is not permitted;
iii. additions are not permitted.
An important feature of the model is that Greenberg and Jenkins permit identity substitutions in
strings attested as words. This means that whenever substitution is not successful in a given place or
places in an existing word, they permit that the word itself should qualify as a successful substitution.
Thus, for example, in English the string /str]+/ (which realises an attested word straw), the
substitution of /p/ or /k/ (the only two consonants other than /t/ that can occur after /s/ in an initial
CCC-cluster) for /t/ does not produce an existing word. Nevertheless, since identity substitution
permitted, substitution in the second place (i.e. the substitution of /t/ for /t/) is taken to be successful.
‘Thus, any existent word automatically receives all possible substitutions’(Greenberg and Jenkins
1964: p. 159). Allowing identity substitution is equivalent to the assumption that all listed items are
grammatical. This position is taken because otherwise existing words might receive different ratings,
in fact, they might end up being just as ungrammatical, or even less grammatical than some nonexisting words. Figure (7) shows the ratings that we would get for the English words free, stay, mew
and unattested /• tew/ if identity substitutions were not permitted:
Phonotactic grammaticality and the lexicon
182
(7)
/fri+/
/mju+/
/stew/
/• tew/
CCV
free
mew
stay
!
CCV
tree
new9
-
stay
CCV
flee
-
slay
-
CCV
fry
-
sty
-
CCV
glee
clue
clay
clay
CCV
fly
-
spy
shrew
CCV
try
cure
-
sty
CCV
clue
stay
clue
clue
Ss =
8
5
6
5
G=
1
4
3
4
(Sp=23=8)
The ratings for these base strings would be the following without identity substitutions:
(8)
free >stay > mew >/• tew/
As can be seen, only one of the base-strings would be fully grammatical (free). It would be
more grammatical than the other two lexically listed base-strings (stay and mew) of which the former
would be more grammatical than the latter. Furthermore, one of the listed base-strings (mew) would
turn out to be just as (un)grammatical as the unattested base-string */• tew/. This complexity of ratings
is counterintuitive (mew, a listed base-string that contains no deviant substrings10 would be judged just
as (un)grammatical as */• tew/, an unattested base string that contains the deviant initial substring
*/• t/. Thus, a simple modification of GJ by disallowing identity substitutions produces undesirable
results. Unfortunately, however, permitting identity substitution causes problems as well. Consider the
ratings for two words prey and schwa when identity substitutions are permitted:
Phonotactic grammaticality and the lexicon
183
(9)
/prew/
/• wY+/
CCV
pray
schwa
CCV
tray
schwa
CCV
play
schwa
CCV
pry
schwa
CCV
clay
spa
CCV
ply
shrew
CCV
try
sway
CCV
fly
fly
Ss =
8
8
1
1
G=
3
(Sp=2 =8)
Identity substitutions make both (listed) base-strings in (9) equally (fully) grammatical. It is
notable, however, that (not counting zero substitution) for pray none of the substitutions are identity
substitutions whereas for schwa three of the substitutions are. This fact reflects the intuitive
judgement that the latter string - although it is an attested word - is not as grammatical as e.g. pray.
GJ does not allow for such distinctions.
Another serious problem is that GJ does not make it possible to compare the grammaticality of
base-strings of different length. In its original form the algorithm favours shorter words: the longer the
measured string is, the less likely it is that a word can be produced by substitution. This is especially
problematic in languages (such as Hungarian) where the length of words is typically longer than in
English. Also there is a wide range within which words/morphemes can differ from one another
(length, number of vowels, etc.). It would be desirable to measure the grammaticality of a unit that
shows a limited range of variety. An obvious choice would be the syllable. If the phonotactic
grammaticality of syllables were measured with a substitution algorithm, the a component would have
to be added that calculates the grammaticality of words with reference to the grammaticality of
syllables.
GJ also does not take into consideration the number of successful substitutions in a given place:
it makes no difference whether only one successful substitution is possible or several successful
substitutions are possible in a given place. E.g. given hypothetical /pju+/ only one successful
Phonotactic grammaticality and the lexicon
184
substitution is possible in the third place in British English: pure (assuming that it ends in the vowel
• c). By contrast, several successful substitutions are possible in the same place in hypothetical /tr]+/:
e.g. tray, try, tree, true, etc.
Greenberg and Jenkins tested their measure and found that there is a good agreement between
their grammaticality ratings and the test subjects’ reactions. Ohala (1984, 1986) also tested GJ and
found that it is correct on two counts: (i) native speakers react to both small (particular) and large
(general) deviations from the native pattern (ii) native speakers are sensitive to the number of
violations in a hypothetical word. Note these experimental result do not mean that GJ is correct, only
that it is correct on these two counts (Ohala 1984, 1986).
Summary
In the final section of this paper I will give a brief summary of the most important issues
discussed formulate questions for further research and identify the factors that can possibly influence
the grammaticality of a string and should be built into a phonotactic grammar.
(i)Similarity to the lexicon. A phonotactic grammar measures the similarity of a string to the
listed strings in the lexicon. Unlistedness does not necessarily entail (full) ungrammaticality, but the
more similar a string is to most of the listed strings, the more grammatical it is (or the more likely it is
to be grammatical).
(ii) Two-degrees or more than two degrees of phonotactic grammaticality? There is some
experimental evidence (Greenberg and Jenkins 1964, Ohala 1986, 1986) that native speakers can
distinguish more than two degrees of grammaticality. This suggests that a phonotactic grammar that is
to model native speaker intuition/behaviour manifested in well-formedness judgements has to be
multi-level.
There are, however, two problems related to (ii):
(iii) The discrepancy between phonotactic grammaticality manifested in well-formedness
judgements and in phonotactics-driven phonological processes/phenomena. While more than two
degrees seem to be required to model native speaker well-formedness judgements, only two degrees
of phonotactic well-formedness are only ever needed in the phonology to account for phonotacticsdriven phenomena such as epenthesis, syncope, deletion, etc. This discrepancy is usually ignored in
the literature and an agreement between grammaticality judgements and phonological role is assumed.
This relationship should be explored by future research.
(iv) The number of degrees of phonotactic grammaticality. The multi-level models we have
discussed do not impose an upper limit on the number of degrees of phonotactic well-formedness.
Taken to the extreme, this makes the prediction that native speakers are capable of distinguishing an
infinite number of degrees of grammaticality. Such a claim is certainly untrue (and is reminiscent of
the SPE position on the number of stress levels). A realistic multi-level model must have an upper
limit on the number of degrees.11 Preferably, the number of degrees should be low. It is unclear how
such an upper limit can be determined as experimental evidence is lacking.
(v) Base-strings and substrings. The algorithms discussed measure the well-formedness of
base-strings directly, i.e. without reference to the possible differences between the phonotactic
grammaticality of substrings. While the grammaticality of substrings is often referred to in the
literature, there is no evidence available about the role it plays. There are many questions one could
ask, for instance: Are there just two degrees of substring well-formedness or more than two?; Are
native speakers sensitive to the grammaticality of substrings independently of base-strings?, etc. It
would be possible to construct phonotactic grammars that are indirect in the sense that they measure
the grammaticality of base-strings with reference to a pre-established grammaticality ratings of
substrings.
Phonotactic grammaticality and the lexicon
185
(vi) The number of violations in a base_string. There is evidence that native speakers are
sensitive to the number of deviant substrings in a base-string (Ohala 1986, 1986). The higher the
number is the less grammatical the string is. This favours GJ over SPE-b.
(vii) Uniformity of strings measured The algorithm should be one that allows the comparison of
the grammaticality of any two strings whose grammaticality is measured. This disfavours algorithms
that measure the grammaticality of morphemes/words directly since they may considerably differ
from one another in size and structure. A more uniform unit, such as the syllable, is easier to measure.
This favours indirect models if the ultimate task is to measure the grammaticality of words.
(viii) The role of (un)listedness. It is not clear precisely what role (un)listedness in the lexicon
plays in determining phonotactic grammaticality. In two-level models unlistedness plays no role at all
since unlisted items may be fully grammatical (accidental gaps). It is clearly untenable to consider
listed items always fully grammatical, since there are irregular occurrences. The models we have
reviewed nevertheless make this assumption as an (over)simplification, which makes it easier to
approach the problem. A more satisfactory phonotactic grammar must suppose that the lexicon
contains items that are not (fully) grammatical. Even if listedness does not entail full grammaticality,
it is possible that (assuming a multi-level model) it contributes to the grammaticality of an item. Note
that this also suggests that different speakers (who may have different lexicons) may have different
grammaticality judgements.
(ix) Sublexicons. There are languages that are known to have sublexicons that are identifiable
phonotactically in that they have (partly) different phonotactic regularities (Itô and Mester 1995;
arguably the Hungarian verb has different phonotactics from the noun). It is a nontrivial problem how
grammaticality ratings are influenced by a more intricate structuring of the lexicon. Several scenarios
seem possible (e.g. each sublexicon has its own phonotactic subgrammar, or there is a designated
default sublexicon that determines the grammaticality ratings of unlisted strings, etc).
(x) Frequency. Type and token frequency may also have a effect on phonotactic
grammaticality. It may be hypothesized that more frequent items are felt more grammatical than
infrequent ones.
This report on phonotactic grammaticality has to end on a rather desperate note: it seems that
everything we know about phonotactic grammaticality is an oversimplification. All the existing
models contain features that are definitely incorrect. GJ seems to be the most promising of the models
discussed. Future research must make modifications (along the dimensions discussed here) and test
the modified model experimentally.
Notes
1 There are aspects of phonological well-formedness that are not phonotactic in nature, for example, the wellformedness of words related to alternation patterns. A word may be phonologically ill-formed even though it
is phonotactically well-formed. For instance, *sárat ‘mud’ (accusative) is ill-formed because the vowel of
the stem sár ‘mud’ shortens before suffixes like the accusative (sarat). However, there is nothing anomalous
about the combination of segments in *sárat - it is just as well-formed phonotactically as attested várat
‘castle’ (accusative), járat ‘passageway’. We abstract away from this phenomenon in this study.
2 There are some other approaches to phonotactic grammaticality I do not discuss here in detail, e.g. Scholes
(1966), Clements and Keyser (1983) - both of which are multi-levelled models. I will not to review them
because they are either not formalised at all, or are based on highly outdated assumptions, or do not present
an algorithm in sufficient detail.
3 It is an interesting problem how a linguistic ability of native speakers that appears not to be grammatical is to
be evaluated/interpreted. This, however, is beyond the scope of this paper.
4 In some treatments they are morphemes, cf. Chomsky and Halle (1968).
5 The reason is the following: there is a correspondence between initial CC and CCC clusters in English. The
first element of a CCC cluster must be /s/. The second consonant must be a voiceless plosive, and the third
one a sonorant. In addition to these restrictions, the combination of the second and third consonants in a
186
Phonotactic grammaticality and the lexicon
CCC cluster seems to be governed by the same constraints that govern the combination of the first and the
second consonants in CC clusters: if the CC cluster is attested the CCC cluster is attested, if the CC cluster is
excluded, the CCC cluster is excluded too: /pl/ : /spl/ (play, splash); */pw/ : */spw/. This correspondence
breaks down in the case of /stw/: /tw/ - */stw/ (twig, -). If /stw/ is regarded an accidental gap the
correspondence can be maintained.
6 This could mean morpheme or word, depending on our view of the lexicon. As the question of precisely
what kind of units are listed in the lexicon is not crucial in the problem in focus, I will disregard it in this
paper. For the sake of simplicity, I will assume that attested word forms of any kind are listed. This is clearly
an oversimplification because there are suffix boundaries that are opaque to phonotactic interaction and no
phonotactic constraints apply across them (word-level/analytic/Level 2 suffixes; cf Harris 1994).
7 The treatment in SPE is a summary of identical approaches to the problem by Halle (1959, 1962) Chomsky
(1964), and Chomsky and Halle (1956). Interestingly SPE never refers to the pre-generative precursors of
the approach, such as Fischer-J rgensen (1952) and Vogt (1954).
8 Later in SPE Chomsky and Halle note that ‘in any real grammar, the lexicon will actually contain items that
are “inadmissible”’ (p.416), but they never incorporate this realisation into the algorithms they actually
propose.
9 Throughout the paper Standard Southern British pronunciation is assumed. Furthermore, for the sake of
simplicity, substitution is taken to apply to a near surface representation, e.g. there is no r in non-prevocalic
position (cure is three segments). Naturally, different representations would give slightly different ratings.
10 GJ probably more-or-less correctly measures the (dis)similarity of mew to other items in the lexicon. mew is
unusual (and gets a low grammaticality rating) because) after a cluster Cj the choice of vowels in English is
very limited: the vowel can only be [u+, •c, • ] (fjord is an isolated counterexample) . No such restrictions
apply after any other cluster.
11 The only algorithm known to me that is multi-level and has an upper limit is Clements and Keyser (1983).
They have three levels of phonotactic well-formedness (fully grammatical, fully ungrammatical and
intermediate). However, Clements and Keyser do not go into details and propose this measure without any
argumentation/experimental evidence.
Bibliography
Chomsky, Noam; Halle, Morris (1968) The Sound Pattern of English New York: Harper and Row
Chomsky, Noam; Halle, Morris (1956) ‘Some controversial questions in phonological theory’, Journal of
Linguistics 1.2. pp. 97-138.
Chomsky, Noam (1964) Current Issues in Linguistic Theory The Hague: Mouton.
Clements, George N. and Samuel J. Keyser (1983) CV Phonology: A Generative Theory of the Syllable,
Cambridge, Massachusetts: MIT Press
Fischer-J rgensen, Eli (1952) ‘On the definition of phoneme categories on a distributional basis’, in E. P. Hamp,
F. W. Householder and Robert Austerlitz (eds.) Readings in Linguistics II., Chicago, 1966. pp. 229-321.
Greenberg, J, H; Jenkins J. J. (1964) ‘Studies in the Psychological Correlates of the Sound System of American
English’ Word 20.2 pp. 157-177
Halle, Morris (1959) The Sound Pattern of Russian The Hague: Mouton
Halle, Morris (1962) ‘Phonology in generative grammar’, in J. Katz and J. Fodor (eds.) The Structure of
Language, Englewood Cliffs, N. J. Prentice Hall, pp. 334-352.
Harris, John (1994) English Sound Structure Oxford: Blackwell
Ohala, John (1986) ‘Consumer's Guide to Evidence in Phonology’, Phonology Yearbook 3. pp. 3-26
Ohala, John (1984) ‘Consumer's Guide to Evidence in Phonology’, Berkeley Cognitive Science Report 17.
Scholes, Robert J. (1966) Phonotactic Grammaticality The Hague: Mouton
Vogt, Hans (1954) ‘Phoneme Classes and Phoneme Classification’, Word 10. pp. 28-34.
Aspect and Argument Structure
187
Aspect and Argument Structure
Introduction
In the semantic literature it has been recently proposed that event structure can be calculated on
the basis of the internal temporal constitution of the predicate, the referential type of the internal direct
argument in a compositional way (Verkuyl (1972, 1989,1993), Krifka (1989, 1992)). The aim of this
paper is twofold: The first is to propose a new classification of verbs different both from VendlerDowty's and from Verkuyl's classification that can account for the behaviour of achievement verbs
previous theories cannot account for.The second is to claim that on the basis of the internal temporal
constitution of a verb predictions can be made about the argument structure (external argument,
internal direct argument, internal indirect argument (Tenny (1995)) of the verb. This paper examines
the nature of the inner temporal constitution of an event on the basis of the interaction of the internal
temporal constitution of verb and the referential type of the internal direct arguments of the verb. and
We will argue that the internal temporal constitution of the predicate is just as relevent for the
calculation of internal aspect as the referential type of the internal direct argument.
On Aspect
In the semantic literature aspect deals with the internal temporal constitution of a proposition.
According to Verkuyl (1989, 1993) inner and outer aspect have to be distinguished depending on
whether we refer to the internal temporal structure of the VP or the internal temporal structure of the
sentence. (1)
In the traditional literature an event (2) is either telic or atelic. An event is telic if there is "a
built-in" endpoint in the event and it is atelic if the event lacks a "built-in" endpoint. (Comrie
(1985), Dahl (1985)).
1,a, János egy percen át/*egy perc alatt látott egy lovat.
J-nom for a minute/*in a minute saw a a horse-acc
John saw a horse for a minute/*in a minute.
b, Mari egy éven át/*egy év alatt beteg volt.
Mari-nom for a year/*in a year ill was
Mary was ill for a year/*in a year.
2, a, Lajos egy percen át/* egy perc alatt almát evett.
Lajos-nom for a minute/*in a minute apple-acc
Lajos ate apples for a minute/*in a minute
b, Péter egy órán át/*egy óra alatt aludt.
Peter-nom for an hour/*in an hour slept
Peter slept for an hour/*in an hour.
c, Aladár egy percen át/*egy perc alatt tolta a kocsit.
Aladar for a minute/*in a minute pushed the car.
Aladar pushed the car for a minute|/*in a minute.
d, Egyed egy órán át/*egy óra alatt vert egy darab vasat.
Egyed-nom for an hour/*in an hour hammered a piece of iron.
Aspect and Argument Structure
188
Egyed hammered a piece of iron for an hour/*in an hour.
3,a, Péter *egy even át/egy év alatt épitett egy házat.
P-nom *for a year/in a year built a house
Peter built a house *for a year/in a year.
b, Kazimir egy perc alatt/*egy percen át laposra vert egy darab vasat.
Kazimir in a minute/*for a minute flat hammered a piece of iron
Kazimir hammered a piece of iron flat in a minute/*for a minute
c, Egyed egy perc alatt/*egy percen át a garázsba tolt egy szekeret.
Egyed in a minute/*for a minute the garage-to pushed a cart
Egyed pushed the cart to the garage in a minute/*for a minute.
4,a, Egy perc alatt/*egy percen át vendég érkezett a városba
in a minute/*for a minute guest has arrived in the own
A guest has arrived in a minute.for a minute
b, Péter egy perc alatt/*egy percen át könyvet adott a barátjának.
Peter in a minute/*for a minute book gave his friend-to
Peter gave a book to his friend in a minute/*for a minute.
In 1 and 2, the verbal expression is atelic as its denotation has no set terminal point. In 3, and 4,
it is telic as it includes a terminal point. This well-known semantic distinction is supported by a
number of tests. In non-iterative reading atelic expressions allow for durative adverbs (e.g., for an
hour), but do not allow time-span adverbs (e.g., in an hour). In sentence 1, the inner aspect is stative,
in 2, it is process. In sentences 3, and 4, the natural temporal endpoint of the events are the existence
of the house, flat shape of the iron, the cart in the garage, the arrival of a guest and the new position
of the book.
Preliminaries
In the semantic literature aspectual properties of predicates have been investigated from various
perspectives. Various classifications of verbs have been proposed, the most influential one is the socalled Vendler-Dowty classification that derives inner aspect from the lexico-semantic poperties of
the verb. In this classification verbs are classified in four classes:state, process, accomplishment and
achievement.The problem with this type of classification that some verbs seem to move from one
class to the other.
5, a, Lajos *egy perc alatt/ egy percen át vizet ivott.
Lajos-nom *in a minute/for a minute water-acc drank
Lajos drank water *in a minute/for a minute.
b, János egy perc alatt/ ??egy percen át ivott egy pohár vizet.
Janos-nom in a minute/??for a minute drank a glass of wine-acc.
Janos drank a glass of wine in a minute/for a minute.
Aspect and Argument Structure
189
6 , a. János ??egy éven át/egy év alatt épitett egy házat.
Janos-nom ??for a year/in a year built a house-acc
Janos built a house ??for a year/in a year.
b, János egy éven át/ *egy év alatt házat épitett.
Janos for a year/*in a year house-acc built
Janos built houses for a year/*for a year.
In sentences 5 and 6 in the a, examples the verbs "drink" and "build" are process verbs while in
the b, examples they are acccomplishment verbs. In other words "drink" and "built" must be ambigous
between process and accomplishment. However, the difference clearly does not hold between the
verbs but rather between the reference type of internal direct argument, the patient. In the VendlerDowty classification the following regularities that have decisive roles in the calculation of the event
structure simply get lost:
7,
i, In every telic event there is an internal direct argument in the VP.
ii, In case of certain verbs the referential type of the internal direct argument defines the
interpretation of the event (process vs. accomplishment).
iii, There exists a group of process verbs whose quantized internal direct argument does not
delimit the event but the event can be delimited by introducing an internal indirect
argument.
iv, The referential type of the internal direct arguments of statives verbs do not have any role in
the calculation of the event, which is always atelic.
v, Though achievement verbs always have internal direct argument, their referential type do
not influence the event type. Achievement verbs always occur in telic events independently
of the referntial type of their internal direct argument.
These facts suggest that inner aspect is compositional, it is the combination of verbal and
nominal properties. This composition of nominal and verbal information constitutes the second trend
(Verkuyl (1972, 1989, 1993), Krifka (1992). In Verkuyl's theory inner aspect is composed of verbal
and nominal properties in the syntactic representation. According to Verkuyl VP (inner) aspect can be
reduced to verbal and nominal features. The temporal parameter for verbs is encoded by the verbal
feature (+/-add) where (+add) means roughly "progress in time" and (-add) means the lack of it. This
feature specification distinguishes stative verbs from dynamic verbs (process, accomplishment,
achievement). In the nominal domain the relevent feature is (+/-SQA), which means "specified
quantity of". The (+/-SQA) feature is the feature of the determiner system. A nominal construction is
(+SQA) if it has a determiner or it is quantized, if neither of them is present and the nominal
construction is a bare NP of existential reading, it has (-SQA) feature. Only if the properties of the
verb and the arguments meet what Verkuyl calls the Plus Principle (all verbal and nominal features
are specified for (+) value) is the VP telic.
8, a, János(+SQA) szeret(-ADD) néhány operát(+SQA).
Janos -nom
loves
János loves some operas.
some operas-acc
Aspect and Argument Structure
190
b, A fiúk(+SQA) házat(-SQA) épitenek(+ADD).
The boys-nom house-acc build
The boys are building houses.
c, A fiúk(+SQA) épitenek(+ADD) egy házat(+SQA).
the boys-nom build a house-acc
The boys are building a house.
d, János(+SQA) egy pénztárcát(+SQA) talált(+ADD).
János-nom a purse-acc found
János found a purse.
As we can see in 8, a, the verb has (-add) feature and the event is atelic, in 8,b, the internal
direct argument has (-SQA) feature and the event is atelic, in 8, c, and d, all the nominal and verbal
features are specified for (+), the events are telic.
Several semanticists have critized Verkuyl's theory because it has only descriptive but no
explanatory power. It correctly describes all the well-formed structures and excludes all the ill-formed
ones but does not explain why exactly that set of structures are the well-formed ones and why are the
rest excluded as ill-formed. It seems that there is a well-defined group of verbs namely, achievement
verbs that Verkuyl's theory cannot account for at all.
9, a, *Egy percen át/?egy perc alatt vendég érkezett.
*for a minute/in a minute guest-nom arrived
There have arrived guests in a minute/for a minute
b, *Egy percen át /?egy perc alatt János könyvet adott Péternek.
*for a minute/in a minute Janos-nom book-acc gave
John gave books in a minute/for a minute.
In 9, the internal direct argument is (-SQA) in Verkuyl's terms, both are bare existential NP. In
spite of that, the event interpretation of the VPs is telic. All achievement verbs can occur with bare
existential NPs and the result is telic. Achievement verbs can occur only in telic events.
Krifka (1992) analyzed the correspondence between the reference types of NPs and the
temporal constitutions of verbal predicates assuming an event semantics with lattice structures and
thematic roles as primitive relations between events and objects. The basic idea is that atelic
expressions are similar to mass nouns and bare existential plurals. They both refer cumulatively (3).
While telic expressions are similar to quantized nominal constructions. They have quantificational
reference. Krifka claims that there is a homomorphic mapping from object types to event types
provided the object is connected to the event throught the proper thematic role. If the object has
quantized reference, the complex verbal expression has quantized reference and event is telic. If the
object refers cumulatively the verbal expression will refer cumulatively, too. Then the event is atelic.
Aspect and Argument Structure
191
10, a, János *egy perc alatt/ egy percen át zajt halott a kertb l.
János-nom *in a minute/for a minute noise-acc heard from the garden.
János heard noise from the garden *in a minute/for a minute.
b, János *egy perc alatt/ egy percen át egy ember hangját hallotta a kertben.
János-nom *in a minute/for a minute a man's voice-acc heard in the garden
János heard a man's voice in the garden *in a minute/for a minute.
In 9 a, the direct argument refers cumulative, in 9, b, it has quantificational reference. In spite
of the difference in the reference type of the deep direct object, there is no difference in the aspectual
reading because stative verbs do not seem to assign the proper thematic role to their object. Krifka
distinguishes five different thematic roles verbs can assign to their direct internal argument:gradual
effected patient, gradual consumed patient, gradual patient, affected patient and stimulus. The
different patient relations are characterized by the following properties:Summativity, graduality and
uniqueness. If John saw a horse and Mary saw a horse then John and Mary saw two horses unless
they saw the same horse, (situation which is irrelevant for us now) then summativity is assumed for
the experiencer relation and the stimulus relation. Summativity relation is trivial for patient relations.
Graduality holds for events like read a book: the book is subjected to the event of reading in a gradual
manner, which is not true for the see a horse case. Uniqueness of objects chacterizes the patient
relation of effected and consumed patients that can be subjected to the event of eating or painting only
once. An object can delimit the event if the patient relation that relates the object to the event has the
properties of summativity, graduality and uniqueness and the object has quantized reference. Krifka's
theory has the same serious empirical problem that Verkuyl' s theory has. It cannot account for
achievement verbs. Achievement verbs have patient roles that are summative , some of them share the
uniqueness property, as well, but they are never gradual. In addition to these problems, they have to
be able to assign patient roles that are either effected or affected, they have to undergo some change,
but they do not need to be quantized so that a telic event could be obtained. Still we get telic events.
To define the relevant conditions for the event interpretation the following information is
needed :
11, i, information about the verb (roughly stative or dynamic)
ii, information about the reference type of the internal argument (bare existential or
quantized)
iii, information about the thematic roles that relate objects to event (patient relation)
(Smith (1996))
As we have seen earlier neither Verkuyl's nor Krifka's theory can account for the behaviour of
achievement verbs. The types of thematic roles that hold between achievement verbs and their
internal direct argument are those thematic roles that can delimit the event if they are quantized and
they are combined with dynamic verbs. Achievement verbs with their internal arguments always
form telic events independently of the reference type of the internal direct argument.
The theory
I will propose a compositional theory of events in the spirit of Krifka (1992), Verkuyl (1993). I
claim that event structures can be calculated on the basis of the referential type of the internal direct
argument, the thematic role that relates the internal argument to the event and the internal temporal
Aspect and Argument Structure
192
structure of the verbal predicate. I will further assume that not only events and nominal constuctions
have quantificational properties but verbs have such properties as well. The internal temporal
constitution of the verb can be specified for the feature (+/-quantized). I will claim that the
combination of the referential type of the nominal constructions, the verbs and the thematic roles
relating them would assign the event structures the proper event interpretation.
I assume the most relevant information for event interpretation is the thematic role of the
internal argument that relates the argument to the event. If the relevant thematic role is assigned, the
event can be delimited/telic or atelic/non-delimited, depending on the reference type of the internal
argument and the internal temporal constitution of the verb. There are two possibilities:
12, i, the internal temporal constitution of the verb is not delimited, the event can be delimited
by the quantized internal argument (process vs. accomplishment),
ii, the internal temporal constitution of the verb is delimited, the event can be temporally
delimited by the verb, it needs an object that undergoes some change of state but the
object does not need to be of quantized reference.(achievement).
Classification of verbs
Stative verbs
In the Vendler-Dowty classification of verbs the internal temporal constitution of statives verbs
do not consist of discrete atomic events.
13, a, János szereti Mári t.
János-nom loves Mária-acc
János loves Mária.
b, Ferenc féli Istent.
Ferenc-acc fears God-acc
Ferenc fears God.
c, Angéla látta a tömeget az utcán.
Angéla-nom saw the crowd-acc the street-on
Angéla saw the crowd in the street.
At any instant on the time span for which it is true that Janos loves Maria, Ferenc fears God,
and Angela saw the crowd in the street, the same statement is true. There is no development in the
event throught time. No change is encoded in the event. There is no way for any of the participants in
the situation to undergo some minimal change of state. The thematic roles assigned by stative verbs to
their internal argument is the least prototypical patient role (5), I will call this theta role stimulus
following Krifka's classification. Internal arguments that have stimulus theta role can never delimit an
event.
Aspect and Argument Structure
193
Process verbs
Process verbs are built up of an unlimited number of temporally ordered atomic events .
14, a, János (almát) evett.
János-nom (apple-acc) ate
János ate (apples).
b, János ((egy) traktort) vezett.
János-nom ((a )tractor-acc) drove
János drove a tractor.
c, János (egy darab) vasat vert.
János-nom (a piece of) iron-acc hammered
János hammered (a piece of) iron.
In sentence 14, a, each part of the optional effected object can be subjected to the temporally
ordered atomic events of eating only once. The relation between parts of the object and the atomic
events is unique. Verbs of consumption and creation assign the most prototypical thematic role, the
effected patient role to their internal argument. in 14, b, the whole object is subjected to the atomic
events of several unlimited number of pushing. What changes throught time is the position of the
object. In 14, c, the whole object is subjected to the event, again. In this case some property of the
object changes gradually in time.
Dynamic verbs differ from stative verbs with respect to their intern temporal constitution.
While stative verbs have a homogeneous internal temporal constitution, in process events there are
continuous changes encoded in the event. In processes for any atomic event for which the activity of
eating some part of an apple, pushing a tractor along some path or hammering some iron gradually
giving it some shape, will differ from the atomic event that preceeds it in time and from the atomic
event that follows it in time.
Those process verbs that have a unique relation to their direct internal argument assigns the
effected patient role to their object, those process verbs whose relation to their direct internal
argument is not unique assign the affected patient role to their internal direct argument. In 14 there is
no natural "built-in" end-point in the event. The verbs contain an unlimited number of atomic events,
in 14, a , the object has the effected theta role, but the reference type of the object is cummulative,
therefore the cummulative reference of the internal direct argument is carried over, the event has
cummulatve reference, that is the event is atelic. In 14, b, and 14, c, it seems that independently of the
reference type of the internal argument, the events are atelic. Affected arguments can undergo
changes imposed by the verb, but cannot delimit the event.
Accomplishment verbs
According to compositional theories accomplishment situations have a natural end point
provided by the quantized direct argument that is assigned the effected patient role by a process verb.
If the process verb can assigns affected patient role to its internal argument, the event can be delimited
by adding further internal arguments to the event (directional prepositional phrases, resultative
secondary predicates).
Aspect and Argument Structure
194
15, a, János házat épitett.
János-nom house-acc built
János built houses.
b, János épitett egy házat.
János-nom built a house-acc
János built a house.
16, a, Péter egy darab fát festett.
Peter-nom a piece of wood-acc painted
Peter painted a piece of wood.
b, Péter pirosra festett egy darab fát.
Peter-nom red-abl. painted a piece of wood.
Peter painted a piece of wood red.
17, a, Ferenc egy szekeret tolt.
Ferenc-nom a cart-acc pushed
Ferenc pushed the cart.
b, Ferenc a garázsba tolt egy szekeret.
Ferenc-nom the garage-into pushed the cart-acc
Ferenc pushed the cart into the garage.
According to Krifka (1992) direct arguments can carry over their referential type to the event if
they are connected to the event by the appropriate thematic roles. In 15,a, the amount of the house that
is involved in the building event is not well-defined at all. It can be one, more than one or even less
than one if Peter is involved only in the building of roof activity. In 16, b, the event of building a
house is completed when the house is ready. In 16, an existing object obtains a new property, in 17,
the cart as a whole is involved in the pushing event. As the verb "build" assigns effected patient role
to its internal direct argument, a quantized direct argument can delimit the event. In 16 a, 17 a, as
the verbs push and paint assign affected patient role to their arguments, they cannot delimit the event,
they need the help of other interal arguments. In 16,b, it is a directional prepositional phrase that
marks the endpoint of the event, while in 17, b, when the piece of wood is covered with red painting.
Process verbs can always be delimited, either by their internal direct argument. Process verbs can
always have a direct argument. If the direct argument do not have the effected argument role, it can
have an indirect argument, as too.
Achievement verbs
The internal tempora constitution of achievement verbs consists exactly two atomic events. As
oppossed to process verbs achievement verbs are delimited temporally.
18 a, János könyvet/ egy könyvet adott Máriának.
János-nom book-acc/a book-acc gave Mária-to
János gave a book/books to Mária.
Aspect and Argument Structure
195
b, Szinház/egy szinház alakult az egyetemen.
theatre/a theatre founded the university-at
There were theatres /there was a theatre founded at the university.
c, Vendég/egy vendég érkezett.
guest-nom/a guest-nom arrived
There has arrived a guest/there have arrived guests.
It seems that achievement verbs assign the same type of thematic roles (affected, effected
patient roles) that process verbs do. In 18 a, the position of the existing book changes, in 18, b, a new
theatre comes into existence, in 18, c, by being involved in the event of arriving the individual obtains
the property of becoming a guest. Achievement verbs do not need quantized internal arguments so
that they could occur in telic events. The referential type of the direct object cannot be carried over to
the event, with bare existential object achievemnt verbs form telic event. The internal temporal
constitution can temporally delimit the event.
Classification of Verbs
Verbs can be marked or unmarked with respect to the quantificational properties of their
internal temporal constitution. Stative verbs do not consists of atomic event,while process and
achievemnt verbs do. Process verbs is constructed of an unlimited number of atomic events, while
acievement verbs is temporally constituted of two atomic events exactly.Accomplishment situations
can be derived from process situations with the relevent type of internal argument. The claim is that
verbs have quantificational properties with respect to their internal temporal structure. Statuve verbs
are underspecified for (Q) feature as they do not have internal temporal structure, process verbs are (Q), as their internal temporal constitution is not limited. Achievement verbs consist a limited number
of atomic events, namely two. They are positively specified for the (Q) feature.
Telic events can be achieved if either the effected or affected patient role is assigned to the
internal direct object, which seems to be the case for all process and achievement verbs. Further on,
inner aspect can be calculated on the basis of the interaction of the quantificational features in the
verbal domain and the nominal domain I assume against Verkuyl (1993) that to get the telic reading it
is sufficient for either the verb or for the nominal to be quantificational.
Table 1
Classes of verbs
thematic roles
event type
V (Q)
stimulus
atelic
V (-Q)
effected arg. (-Q)
atelic
affected arg. (-Q)
atelic
effected arg. (+Q)
telic
affected arg. (+Q), XP
telic
effected arg. (+/-Q)
telic
affected arg. (+/-Q), XP
telic
V (-Q)
V(+Q)
In this theory stative verbs can never occur in telic events, because they are unable to assign
the proper thematic role to their internal direct due to their internal atemporal constitution. Process
196
Aspect and Argument Structure
and achievemnt verbs can always occur in telic event. They can assign the proper thematic roles to
their internal direct argument. Process verbs as they are inherently (-Q) can occur in telic events if
they are combined with quantized internal arguments. Achievement verbs are inherently telic as they
have (+Q) feature. The internal direct argument can be either quantized or bare existential.
External arguments and aspectual verb classes
It has been claimed that external arguments have no role in the calculation of inner aspectual
structure. External arguments are still sensitive to the internal temporal structure of the verb. They
can typically cooccur only with dynamic verbs, not with stative verbs (5).
Following Marantz (1983), Kratzer (1994) claims that external arguments are not arguments of
the verb. They are not in the VP in the initial syntactic structure, at all. External arguments are
generated in the specifier position of a functional projection she calls VoiceP which is headed by the
predicate CAUSE . Following Kratzer (1994) I claim that the external argument is not the argument of
the verb but the predicate CAUSE. That would explain, why external arguments do not participate in
VP aspect. They never occur in VP. I further assume that the predicate CAUSE has two arguments,
the external argument and the VP. The external argument can cooccur with only certain types of
verbs, namely dynamic verbs, because CAUSE selects for dynamic Vps.
Conclusion
A new classification of verbs is proposed on the basis of their internal temporal structure. This
new classification elimits the redundancy of the Vendler-Dowty classification, and gives the right
results with achievement verbs (problems that arise in the Verkuyl-Krifka theory) that heavily rely on
the referential type of the internal argument in the calculation of (a)telic event.
The theory further predicts the presence of external, direct internal and indirect internal
argument with minimal information about the verbal predicate.
Footnotes
*I
would like to thank Agnes Bende-Farkas, Grete Dalmi, Laszlo Kalman, Andras Komlosy, and Marta
Maleczki for discussion concerning the issues of this paper. Errors, of course, are all mine. This work has been
supported by the Research Support Scheme of the Open Sociaty Support Foundation, grant No.:320/1998.
1
, The inner aspect corresponds roughly to the composition between verbal features and referential features of
internal arguments. The outer aspect will correspond to the roles of adverbials, negation, progressive operators,
etc. They can modify inner aspect, but (a)telicity is specified for the VP once and for all..
2
I use the notion event for VP (inner) aspect.
3
The nominal construction refers cumulatively :whenever there are two entities to which that predicate applies,
this predicate applies to their collection, as well. A nominal construction has a quantificational reference
whenever there are two (different) entities to which that predicate applies, the predicate cannot apply to their
collection. Whenever there are two entities to which the predicate wine applies, this predicate applies to their
collection, as well, that is not the case with the predicate a glass of wine.
4
Dowty lists the properties which contribute to Proto-Agent and Proto-Patient role as (from Dowty (1991) p.
572):
i, Contributing Properties for the Agent Proto-Role:
a, volitional involvement in the event or state
b, sentience
c, causing an event or a change of state in another participant
d, movement (relative to the position of another participant)
(e, exists independently of the evnt named by the verb)
ii,Contributing Properties to the Proto-Patient Role:
a, undergoes change of state
197
Aspect and Argument Structure
b, incremental theme
c, causally effected by another participant
d, stationary relative to the movement of another participant
(e, does not exist independently of the event, or not at all)
5
Rizzi and Belletti (1988) claim, that psych verbs that are typically stative verbs do not have external
arguments, at all.
References
Bach, Emmon. (1986), The algebra of events, Linguistics and Philosophy, 9, 5-16.
Comrie, Bernard. (1987) Aspect. Cambridge:Cambridge University Press.
Diesing, Molly. (1992), Indefinites. Cambridge MA:MIT Press.
Dowty, David. (1979), Word meaning and Montague Grammar. The Semantics of Verbs and
Times in Generative Semantics and in Montague's PTQ. Reidel: Dordrecht.
Dowty, David. (1991), Thematic Proto-Roles and Argument Selection. Language 67.3:
547-619
de Hoop, Helen. ( 1992), Case configuration and NP interpretation. Doctoral Dissertation. U.
of Groeningen.
Johnson, Kyle. (1991), Object Positions. Natural Languge and Syntactic Theory 9. 577-636.
Kiefer, Ferenc. (1994), Aspect and syntactic structure, in F. Kiefer and K. E. Kiss (eds.) The
syntactic structure of Hungarian (Syntax and Semantics 27), Academic Press, New York.
Kratzer, Angelika, (1994), The event argument and the semantics of Voice. ms. University of
Massachusetts at Amhurst.
Krifka, Manfred. (1989), Nominal Reference, Temporal Constitution and Quantification in
Event Semantics. In R Bartsch, J. van Bethem, and P. van Emde Boas (eds.) Semantics
and Contextual Expression. Foris Dordrecht, 75-115.
Maleczki, Marta. (1992), Nominal reference, temporal constitution and quantification in event
semantics. In: Semantics and Contextual Expression (R. Bartsch, J. van Benthem and
van Emde boas, eds.), pp. 357-375. Institute for Logic, Language and Computation,
University of Amsterdam.
Pustejovsky, James (1989), The Geometry of Events. In C. Tenny (ed.) Generative Approaches
to Aspect. Cambridge, MA: MIT Lexicon Project.
Pustejovsky. James. (1991), The syntax of even structure. Cognition.
Szabolcsi, Anna. (1986), From the Definiteness Effect to lexical integrity. In: Topic, Focus and
Configurationality. (W. Abraham and S. de Meij, eds.), pp. 332-360. John Benjamins,
Amsterdam.
Tenny, Carol. (1994), Aspectual Roles and the syntax-semantics. Kluwer Academic
Publishers.Dortrecht, The Netherlands
Verkuy, Henk (1989) Aspectual Classes and Aspectual Composition. Linguistics and
Philosophy 12, 39-95.
198
Aspect and Argument Structure
Verkuyl, Henk (1993) A Theory of Aspectuality. Cambridge: Cambridge University Press.
Ertékek azonossága-e az egyeztetés?
199
Ertékek azonossága-e az egyeztetés?
Kálmán László és Trón Viktor
Bevezetés
Ennek a cikknek kettős célja van. Egyrészt megpróbáljuk megmutatni hogy a morfo-szintaktikai információnak a megszorítás-alapú nyelvtanokban használatos kezelése nem meg-felelő.
Látni fogjuk, hogy néhány egészen általános jelenséget - így a morfológiailag semle-ges alakok
viselkedését többszörös egyeztetés esetén, valamint különböző kategóriájú kife-jezesések
koordinációját - absztrakt jegyek használatával, értékeik unifikációjával és típu-sos jegy-érték párok
hierarchiába rendezésével nehéz vagy egyenesen lehetetlen elemezni. Másrészt azt állítjuk, hogy az
említettjelenségek megfelelő módon és formálisan elegánsan kezelhetőek akkor, ha szakítunk az
absztrakt morfológiai jegyekkel. Javaslatunk lényege, hogy a tulajdonságot kifejező jegy-érték
párokat az öröklődési hálózatba felvett tulajdon-ságokkal mmt típusokkal helyettesítjük. Másrészt az
egyeztetést és a hasonló viszonyokat jegy-értékek azonossága helyett kifejezések közötti relációkkal
modellezzük. Ehhez ki kell terjesztenünk az öröklődési hálózatok fogalmát, a bennük szereplő típusok
közé relációs típusokat is bevéve. Végül érvelünk a grammatikai viszonyok szigorú lokalitása ellen, és
megmutatjuk/ hogy az általunk javasolt elemzés hogyan használható bizonyos nem-lokális
összefüggések kezelésére. Megközelítésünk lehetővé teszi, hogy a nyelvtant tisztán dek-laratív módon
modellezzük, megőrizve a típusrezolúció mint egyetlen grammatikalitási feltétel meghatározó
szerepét.
Az unifikáció hiányosságai
Az első két alfejezetben két jól ismert adathalmazt ismertetünk. Az első az egyeztetéssel, a
második pedig a szubkategorizációval kapcsolatban mutat rá az unifíkáció- és jegy-alapú
megközelítések elégtelenségére. A fejezet végén bemutatunk néhány lehetséges kiutat a vizsgált
problémákból.
Egyeztetés
Tekintsük a következő mondatokat:
(1) Alanyi-tárgyas egyeztetés a magyarban
a. Afát látom/*látok.
b. Egy fát látok/*látom.
Ezekben a példákban a következőt figyelhetjük meg: véges mondatokban az ige határozottságban egyezik a tárggyal. Ez azt jelenti, hogy a ragozott ige határozott tárgy esetében tárgyas,
egyébként alanyi ragozású. Vannak azonban olyan esetek, amikor az alanyi és a tárgyas ragozású alak
egybeesik (pl. láttam, látnánk). Az ilyen alakokat semleges alakoknak fogjuk nevezni. Ezek mind
határozott, mind határozatlan tárggyal előfordulhatnak:
(2) Semleges alakok viselkedése
A fát/egy fát láttam.
Érdekes módon a semleges alak a többi alakkal szemben olyan koordinált NP-vel is előfordulhat, melyben a tagok különböző határozottságúak:
(3) Alanyi-tárgyas egyeztetés koordinált NP-vel
a. *Látom/*Látok a fát és egy madarat.
b. Láttam afát és egy madarat.
Altalánosságban a következő megfigyelést tehetjük:
Ertékek azonossága-e az egyeztetés?
200
(4) Vegyes egyeztetés
A határozottság szempontjából semleges alakok határozott és határozatlan tárggyal is
egyezhetnek, akár egyidejűleg is.
Kézenfekvőnek tűnhetne az a feltételezés, hogy a határozottság tekintetében semleges ala-kok
szótári tételében a határozottsági jegy értéke nincs specifikálva. Ha így lenne, akkor valóban unifkálni
lehetne akár a határozott, akár a határozatlan tárgy megfelelő jegyének értékével. Így lehetne elemezni
a fenti (2) példát. Láttuk azonban azt is, hogy a semleges alakok egyszerre két eltérő határozottságú
tárggyal is elő tudnak fordulni, ha azok koor-dinált NP-t alkotnak. Ez azonban eleve feltételezi, hogy
vannak „vegyes határozottságú" NP-k, ezt azonban a mellérendelés szokásos kezelése kizárja, hiszen
a tagok határozottsági jegyének az értékei nem unifikálhatóak (erre először Ingria (1990) mutatott rá).
Felhívjuk a figyelmet arra, hogy morfológiailag semleges alakok a világ legkülönbo-zőbb
nyelveiben előfordulnak, és - talán meglepő módon - mindenütt a fenti mintát köve-tik, azaz vegyes
egyeztetési viszonyba tudnak lépni:
(5) Ingria általánosítása
Azok a formák, amelyek egy bizonyos jegy szempontjából semlegesek/ ezen jegy
szempontjából külön-külön vagy akár egyszerre is egyeztetési viszonyba léphetnek tetszőleges
formákkal (tehát attól függetlenül, hogy azokban az illető jegynek mi az értéke).
Alljon itt két angol példa, amelyek azt mutatják, hogy a jelenség nem korlátozódik a szigorú
értelemben vett egyeztetésre. (Ingria (1990) további példákat is idéz.)
(6) Számbeli egyezés az angolban
a. Why did
John leave and the others stay? miért AUX-PL/SG János elmegy és a többiekmarad
'Miért ment el János és maradtak a többiek?'
b. *Why does/do
John leave and the others stay? AUX-SG/AUX-PL
(7) Esetkonfliktus többszörös űröknél (Grover, 1996)
a. Who did
you say John's criticism of
_i(ACC) would ki/kit MÚLT-AUX te mond Jánosnak
a kritikája BIRT-PREP
FELT-AUX make us think _i(NOM) was stupid? csinál minket gondol
volt ostoba?
'Mit mondtál, hogy ki az akiről azt gondoltuk, hogy ostoba/ miután János kriti-zálta?'
b. Which
Caesar did Brutus imply [_i(NOM) was no good] while melyik/melyiket Ceasar AUX
Brutus sugall
volt nem jó mialatt ostensibly praising _i(Acc)?
látszólag dicsőítve
'Melyik Caesar volt az, akiről Brutus az mondta/ hogy nem jó, miközben látszó-lag dicsőítette?'
Szubkategorizáció
Egy másik jól ismert jelenség, amely problematikus az unifikációs nyelvtan számára, a
szubkategorizációval és a mellérendeléssel kapcsolatos. A mellérendelés önmagában nem követeli
meg, hogy a tagok kategória-jegyeinek értékei azonosak legyenek: az olyan pre-dikátumok/
amelyeknek elég lazák a szubkategorizációs kötöttségei, azok vegyes szófajú ko-ordinált
szerkezetekkel is előfordulhatnak:
(8) Vegyes kategóriájú mellérendelés az angolban (Sag és mtsai., 1985)
a. That was [NP a rude remark] and [pp in
a very bad taste]. Ez volt egy durva megjegyzés
és BAN-PREP egy nagyon rossz ízlés 'Ez egy durva megjegyzés volt, mégpedig ízléstelenül
előadva'
Ertékek azonossága-e az egyeztetés?
201
b. Pam is [NP a republican] and [AP proud ofit]. Pam van egy republikánus és
republikánus/ és büszke is rá'
c. Pam is [AP healthy] and [pp of sound mind]. Pam van egészséges és
Tam egészséges és értelmes.`
büszke rá 'Pam
PREP józan értelem
Hangsúlyozzuk, hogy ez a jelenség igen általános mindenféle fej-módosító és fejkomplementum viszonyban a világ nyelveiben (pl. német: Zaenen és Karttunen (1984); francia:
Pullum és Zwicky (1986)).
A következő megtigyelést tehetjük:
(9) A grammatikai funkciók kifejezése
Azok a fejek, amelyek mellett valamelyik grammatikai funkciót több különböző ka-tegóriájú
kifejezés is betölthet, megengedik ezek egyidejű megjelenését is, melléren-delő szerkezet formájában.
Világos, hogy ez a jelenség hasonló problémát okoz a mellérendelő szerkezetek unifikáci-óval
való kezelésében, mint a vegyes egyeztetés (Id. (4).
Általánosítás és típusrezolúció
Láttuk, hogy a mellérendelésnek az a fajta kezelése, amely szerint a tagok morfo-szintaktikai
jegyeinek értékeit unifikálni kell, nem összeegyeztethető a vegyes koordinált szerkezetek létezésével.
Van azonban egy másik lehetőség is: mondhatjuk, hogy a mellérendelő szerke-zet kategóriáját nem
unifikációval, hanem általánosítással kell kiszámolni.45 Ezt elsőként Sag és mtsai. (1985) javasolták,
majd Pollard és Sag (1994:203) a HPSG keretein belül az ún. koordinációs elv részévé tette:
(10) Koordinációs elv
A mellérendelő szerkezetekben a tagok CATEGORY és a NONLOCAL jegyeinek érté-kei
specifikusabbak, mint az anyacsomópont megfelelő jegyeinek értékei.
Természetesen ez az elv feltételezi, hogy az illető jegyek értékeit specifikusság szerint rendezzük. Például Jackendoff (1972) szermt a négy fő szófajt (főnév, melléknév, ige, elöljáró/
névutó/esetrag) kétjegy különböző értékeivel jellemezhetjük:
(11) Jackendoff-féle kategória-definíció
A jegy-értékek specifikusság szerinti rendezését úgy is elérhetjük, hogy az értékeket ún. típushierarchiában helyezzük el. Például a határozottság esetében:
(12) A ha tározo ttsági értékek hierarchiája
45
Valójában elég annyit megkövetelni, hogy a komplex szerkezet kategóriája általánosabb legyen, mint bármely
tagjának kategóriája, Id. Shieber (1992:132-134).
Ertékek azonossága-e az egyeztetés?
202
Ha ezt a megközelítést választanánk, akkor a vegyes határozottságú koordinált NP-k határozottsági jegye a defíniteness értéket kapná.
Csakhogy a fenti, látszólag tetszetős elemzés nem fér meg a típus-hierarchiákkal dol-gozó
elméletekben általában feltételezett típusrezolúcióval. A típusrezolúció azt jelenti, hogy a nyelvi
objektumokat kimerítően típusozott j egy-struktúrákkal modellalljuk: ha egy nyelvi objektumra
egy jegy értelmezve van, akkor annak a jegynek egy teljesen spe-cifikus érteket meg is kell kapnia.
Alulspecifikáció tehát csak a nyelvtani leírás nyelvében van, de azok az objektumok, amelyekről a
leírások szólnak, tehát a modell entitásai nem lehetnek parciálisak.46
Ezért a koordinációs elvben szereplő megkötés a mellérendelő szerkezet kategóriájá-nak
kiszámítására a teljes típusrezolúcióval együtt kizárja a vegyes koordinált szerkeze-teket, mert az
ilyen szerkezeteknek parciális objektumoknak kellene lenniük. Csak úgy tudnánk a vegyes koordinált
szerkezeteket jellemezni, ha az illető tulajdonságot három-értékű jeggyel kódolnánk (pl. a
határozottságnál: határozott, határozatlan, vegyes). Ekkor azonban az egyeztetés nem lenne
kifejezhető unifikáció segítségével, hiszen a semleges alakok mindhárom értékű kifejezéssel
egyezhetnek.
Ha azonban elvetnénk a szigorú típusrezolúció elvét, akkor valamilyen más eszközt kellene
találnunk arra, hogy az agrammatikus képződményeket kizárjuk. Az egyetlen szisz-tematikus módja
ennek, amely a szakirodalomban előfordul, a korábban alkalmazott újra-író szabályrendszer,
amelyet azonban az elmúlt évtizedekben már számos szempontból meggyőzően bíráltak. Márészt ha a
típusrezolúciót bizonyos konkrét esetekben akarnánk felfüggeszteni (például a határozottság
feloldását éppen mellérendelt szerkezetek eseté-ben), ezt csak önkényes meta-szintű megszorítások
formájában tehetnénk meg. Mi több, a morfo-szintaktikai jegyek értékeinek alulspecifikálása a
koordinált NP-kben önmagában nem is lenne elegendő a teljes jelenségkör leírásához. Amint Ingria
(1990) kimutatja, az a törvényszerűség, amelyet a fenti (5) alatt megfogalmaztunk, messze túlmutat a
melléren-delés tartományán:
(13) Német vonatkozó mellékmondatok
a. *Wer/*wen du mir vergestellt hast, ging nach Hause. aki/akit te nekem bemutattál menthaza
'Az, akit bemutattál nekem, hazament' b. Was
ami/amit te nekem adtál
du mir gegeben hast, ist prachtig.
nagyszerű
'Az, amit adtál nekem, nagyszerű.`
Itt a második mondatban a főmondat állítmánya (ist prachtig nagyszerű') alanyesetű NP-t vár,
míg a vonatkozó mellékmondat (du mir gegeben hast nekem adtál') tárgyesetűt. Ezeket a
követelményeket a was egyszerre képes kielégíteni, mert eset szempontjából morfológia-ilag
semleges forma (alany- vagy tárgyesetű). Azonban a főmondat állítmánya és a vonat-kozó
mellékmondat semmiféle mondattani viszonyban nem állnak egymással, ezért nem lehetséges
olyanfajta mondattani összetevő feltételezése, melyben valamely jegy értéke ál-talánosítás
segítségével számolható ki úgy, ahogy azt a koordináció esetében lehetséges megoldásként láttuk.
Kálmán és Goldberg (1992) aztjavasolják, hogy Ingria általánosítása (Id. (5)) úgy ragad-ható
meg, ha az unifikáció művelete helyett ilyenkor az unifikálhatóság relációját használ-juk. Ez a
javaslat megoldja ugyan a nem mellérendelésben előforduló vegyes egyeztetés problémáját is,
azonban szintén döntően támaszkodik a parciális objektumokra, hiszen a morfológiailag semleges
alakokat nekik is ilyeneknek kell tekinteniük. Összefoglalásként a következő megállapítást tehetjük:
(14) A fejezet fő következtetése
46
Részletesen Id. Johnson (1988), Shieber (1992), King (1989,1994).
203
Ertékek azonossága-e az egyeztetés?
A morfo-szintaktikai osztályok közötti kongmencia-relációk nem ragadhatók meg az unifikáció
vagy az unifikálhatóság segítségével.
Jegy-érték párok mint tulajdonságok
Az atomi jegy-értékek unifikációjának mindig is főleg az egyeztetésszerű viszonyok leírá-sában
volt szerepe. (Az unifikáció egyéb előfordulásai valójában a frázisokra vonatkozó megszorítások
egyidejű érvényesülését írják elő, tehát komplex struktúrák unifikációját je-lentik.) Viszont láttuk,
hogy az unifikáció művelete erre nem megfelelő eszköz. Felmerül tehát a kérdés: szükségesek-e
egyáltalán a „[DEF = +]" és „[DEF = -]" jellegű jegy-érték párok. Ha ehelyett a határozottságot és a
határozatlanságot egyszerű, elemzetlen tulajdon-ságokként képzeljük el, akkor a magyar alanyitárgyas egyeztetéssel kapcsolatban említett szóosztályokat a következőképpen jellemezhetjük:
(15) Határozottsági osztályok a magyarban
A határozottak;
B határozatlanok;
C semlegesek = A ∩ B.
Ez egyszerű magyarázatot adna arra, hogy a semleges alakú ige miért fordulhat elő ha-tározott
és határozatlan NP-kkel egyaránt: mivel mindkét határozottsági osztálynak tagja. Tehát a semleges
igealakok nem parciális obj'ektumok, kettős természetük nem alulspeci-fikáltságból fakad, hanem
abból, hogy egyszerre két különböző osztályba tartoznak.
Ez a megközelítés nyilvánvalóan egyszerűbb, mint többértékű jegyek használata. A
fonológiában már korábban széles körben elfogadták a többértékű jegyekről az ún. uná-ris jegyekre
való áttérést. Formálisan a fenti osztálycímkék tulajdonságoknak (objektum-osztályoknak) felelnek
meg. Vagyis az itt javasolt osztályokat úgy kell érteni, mmt bizo-nyos magyar szótári egységek
egymástól független tulajdonságait. Igaz, hogy a két tulaj-donság egymással kapcsolatban áll, hiszen
ugyanazon dimenzió mentén jellemzik a kife-jezéseket. Ezt a tényt a jegy-érték struktúrákban úgy
jelölik, hogy egyetlen jegyet használ-nak, míg itt ezt az fejezi ki, hogy a fenti osztályok egy nagyobb
osztály alosztályai (nevez-zük a nagyobb osztályt DEF-CLASS-nak). Ez a nagyobb osztály azoknak
az objektumoknak az osztálya, amelyekre a határozottság egyáltalán értelmezve van. Magát az
osztályozást a típus-hierarchia részeként képzeljük el, tehát az osztályok, amelyeket fent bevezettünk,
valójában típusok.
Természetesen nincs elvi különbség az objektumok tulajdonságai és az objektumok-ból álló nesek tulajdonságai, vagyis a relációk között. Így az egyeztetésszerű viszonyokat nyelvi
objektumpárok tulajdonságaként (osztályaként), azaz bináris relációkként ábrázol-hatjuk. Például a
magyar alanyi-tárgyas egyeztetés legegyszerűbb esetei a következők:
(16) Az alanyi-tárgyas egyeztetés alapesetei
a. határozott NP - tárgyas ragozású igealak
b. határozatlan NP - alanyi ragozású igealak 6
A semleges igealakokat itt nem is kell említenünk, hiszen az a képességük, hogy mindkét-fajta
NP-vel előfordulhatnak, következik abból, hogy mindkét osztályba beletartoznak.
Mivel a relációk maguk is tulajdonságok, ők is szerepelhetnek a specifikusság szerinti
rendezésben, azaz a típus-hierarchiában: a fenti két alapeset valójában alosztálya egy álta-lánosabb
relációnak, amit DEF-AGR-REL-nek (határozottság szerinti egyezésnek) nevezhe-tünk.
Grammatikai viszonyok és mellérendelő szerkezetek
Amint fent megmutattuk/ a határozottság szerinti egyezés a magyarban nem kezelhető jegyértékek unifikációjának segítségével, amennyiben a típusrezolúció szerepét meg akar-juk tartani. Az
Ertékek azonossága-e az egyeztetés?
204
alábbiakban amellett fogunk érvelni, hogy a grammatikai viszonyokat for-málisan is relációknak lehet
megfeleltetni, sőt, erre szükség is van, ha a mögöttes ontológiát a lehető legszűkebbre kívánjuk
szabni. Ez azt jelenti, hogy teljes egészében mellőzhetjük az olyan morfo-szintaktikai jegyeket, mint a
határozottság (és remélhetőleg a többi hason-lót is), és a nyelvtani kongruencia-relációkat (mint pl. az
egyeztetés és a szubkategorizáció) annak tekintsük, amik: kifejezés-osztályok közötti relációknak.
Ezért, bár a következőkben a határozottsági egyeztetés, közelebbről pedig a magyar alanyi-tárgyas
egyeztetést hasz-náljuk példáinkban, az elemzés elvei és eszközei egyszerűen alkalmazhatók más
gramma-tikai viszonyokra is.
Az első alpontban a magyar alanyi-tárgyas egyeztetésre összpontosítva rámutatunk arra, hogy
milyen közvetlen előnyökkel jár az egyeztetés osztály-alapú elemzése. A má-sodik alpontban azt
vizsgáljuk, hogy e szerint az elemzés szerint hogyan kell jellemezni a mellérendelő szerkezeteket a
határozottság szempontjából. Látni fogjuk, hogy problé-mákba ütközünk, ha a jegy-alapú
megközelítést próbáljuk alkalmazni. Az utolsó alpont-ban erre kínálunk egy egyszerű megoldást.
Az egyeztetés osztály-alapú elemzése
A magyar igealakok és NP-k természetes felosztása határozottság szempontjából könnyen
kódolható típus-hierarchia segítségével47
(17) Természetes határozottsági osztályok
Mint már említettük, a semleges alakokat úgy jellemezhetjük, mint több osztályba egy-szerre
tartozó alosztályokat. Ezt a típus-hierarchiák nyelvén többszörös öröklésnek ne-vezzük:
(18) Semleges alakok és többszörös öröklés (igéknél)
Az egyeztetési relációk természetes osztályait hasonlóan egyszerű formában ábrázolhat-juk:
47
A határozottság és hatarározatlanság szemantikai velejáróit természetesen teljesen figyelmen kívül hagyjuk.
Lehet, hogy van ezeknek a fogalmaknak szemandkai alapjuk, de nem csodálkoznánk, ha a formai osztályozás
nem illeszkedne a szemantikaira.
205
Ertékek azonossága-e az egyeztetés?
(19) Alany-tárgy egyeztetés alapesetei
Ha egy szerkezet egyazon elemre vonatkozóan két egyeztetési relációt is előír, akkor a típusrezolúció értelmében a két reláció feloldható az egyeztetés két különböző aleseteként. Például a
vegyes egyeztetésnek nevezett jelenségnél a semleges alak egy-egy egyeztési relációban áll két
különböző osztályba tartozó kifejezéssel. Az unifikáció-alapú megkö-zelítéssel szemben ez azért nem
vezet konfliktushoz, mert a reláció fennállása nem okoz „változást" a relációban álló objektumok
belső szerkezetében. Ennek következtében az itt ismertetett megoldás közvetlenül megmagyarázza
Ingria (1990) általánosítását (Id. (5)). Azonban adósok vagyunk még annak a megvilágításával, hogy
hogyan áll elő a fenti álta-lános szituáció koordinált szerkezetek esetében.
Határozottság koordinált NP-kben
Az unifikációs nyelvtanokban a morfo-szintaktikai jegyek és az értékeiken végzett unifikáció
ill. általánosítás művelete igen fontos ahhoz, hogy a komplex kifejezések tulajdonsá-gait a részeik
tulajdonságaiból számoljuk ki (ennek legegyszerűbb esete a jegy-értékek ún. perkolációja/
„felszivárgása"). Ez azt a célt szolgálja, hogy a grammatikai relációk szigo-rúan lokálisan
kifejezhetőek legyenek. Ezért az ilyen elméleteket szigorúan lokalistának nevezhetjük. Ennek
következtében azonban ezek a nyelvtanok kénytelenek a koordinált szerkezetekhez mint
kifejezésekhez ugyanolyan jellemzéseket rendelni (pl. a határozottság tekintetében), mint az
egyszerűekhez. Ekkor azonban, mint már említettük, az unifikáció nem lenne használható az
egyeztetés leírásához.
Az itt javasolt keretben (jegyek nélkül) természetesen le tudnánk írni azokat a bonyo-lult
teltételeket, amelyek a különböző típusu (homogén és vegyes) koordmált szerkezetek egyeztetését
megszabják a különböző (jelölt és semleges) alakokkal, hiszen az eddig el-mondottak alapján
bármilyen relációt fel tudunk írni a kifejezések osztályai között. Csak-hogy a háromféle koordinált
szerkezet (pl. határozott, határozatlan, vegyes) lehetséges fe-lépítései (tehát az, hogy hogyan
határozzák meg a különböző koordinált szerkezetek tulaj-donságait a tagjaik tulajdonságai) eleve elég
bonyolult leíráshoz vezetnek:
(20) Az NP-k határozottsági osztályai szigorú lokalista szellemben
Továbba, mint említettük, ebben a megközelítésben az egyeztetést nem lehet egyszerű unifikáció segítségével leírni. Ehelyett az NP-k osztályai és az egyező alakok (pl. tárgyas, alanyi,
semleges igealak) közötti viszonyokat az egyeztetési reláció aleseteiként explicit mó-don meg kell
adni:
Ertékek azonossága-e az egyeztetés?
206
(21) Az egyeztetési reláció alesetei szigorú lokalista szellemben
A fenti megoldási javaslat nyilvánvaló bonyolultsága felveti azt a kérdést, hogy mit nyerünk a
morfo-szintaktikai jegyek elvetésével, ha maga az elemzés nem egyszerűsödik. A válaszunk az, hogy
sokat, feltéve, ha a szigorú lokalitás elvét is elvetjük. Erről lesz szó a következő alpontban.
Grammatikai viszonyok rekurzív definíciója
Ebben az alpontban megmutatjuk, hogy a szigorú lokalitás elvetésével úgy elemezhetjük a
vegyes egyeztetés jelenségét, hogy elkerüljük a különböző alesetek elburjánzását az osz-tályozásban
(Id. az előző alpontot). Ezt például az alanyi-tárgyas egyeztetetés esetében úgy érjük el, hogy a
határozottsági osztályokat csak a nem-komplex NP-k tartományára értelmezzük, a koordinált NP-kkel
való egyeztetést visszavezetjük a tagjaikkal való egyez-tetésre. Az ötlet a következő: az egyeztetést
úgy definiáljuk/ hogy annak egy alesete a koor-dinált NP-kkel való egyeztetés legyen. Ebben az
alesetben kimondjuk, hogy a reláció akkor áll fenn a koordinált NP és az igealak között, ha az
egyeztetési reláció fennáll az igealak és a koordmació tagjai között. Az a tény, hogy az egyeztetésnek
van egy olyan alesete, amely hivatkozik a koordinált NP-k összetevőire, ellentmond a szigorú
lokalitás elvének.
Matematikai értelemben ez azt jelenti, hogy az egyeztetést rekurzívan definiáljuk. A rekurzív
lépés az, hogy a koordinált NP egyeztetését visszavezetjük az egyszerűbb esetre, a tagjaival való
egyeztetésre. Maga a definíció a következő:
(22) Az alanyi-tárgyas egyeztés rekurzív definíciója
def-agr-rel(x,y) - def
(i) def-NP(x) ∧ def-V(y);
(ii) indef-NP(x) ∧ indef-V(y); vagy
(iii) def-agr-rel(CONJ1(x),y) ∧ def-agr-rel(CONJ2(x),y).
Ezt a definíciót a típus-hierarchiában a következőképpen kódolhatjuk:
(23) Az egyeztetési reláció a típus-hierarchiában
Ez az elemzés tehát csak a természetes határozottsági osztályokra támaszkodik, nem feltételez
mesterséges, „vegyes" típusú kifejezéseket. A koordinált kifejezések sajátos vi-selkedéséért kizárólag
a nekik megfelelő aleset (amely a rekurzív lépést tartalmazza) a felelős. Ezért a mellérendelő
Ertékek azonossága-e az egyeztetés?
207
szerkezetek felépítésének leírásában nem kell külön szólnunk arról, hogy hogyan viselkednek
egyeztetés szempontjából.
A koordinált szerkezetekkel való egyeztetés így egy elem többszörös egyeztetésére redukálódik, tehát automatikusan érvényes rá mindaz, amit a fenti 3.1 alpontban a vegyes egyeztetésről
elmondtunk. Így nem kell további kikötéseket tennünk ahhoz, hogy az elem-zés helyesen jósolja meg,
hogy az Ingria által megfigyeltek erre az esetre is teljesülnek.
Összefoglalás és további lehetőségek
Az alábbiakban először tömören, pontokba szedve ismertetjük a cikk legfontosabb állítá-sait,
majd röviden kitérünk arra, hogy milyen kiterjesztései lehetnek a javasolt elemzésnek.
A cikk fontosabb állításai
1. A kongruencia-osztályokat csak a nem-koordinált NP-k tartományában értelmez-zük.
2. A kongruencia-osztályok a kifejezéseket alcsoportokra osztják, de a semleges ill. ve-gyes
kifejezések nem alkotnak külön alosztályokat.
3. Az egyeztetés leírásához nincs szükség olyan jegy-érték párokra, amelyekben az ér-ték
atomi. Ehelyett csak tulajdonságok (osztályok) ábrázolására van szükség, s ehhez az amúgy is
feltételezhető típus-hierarchia elegendő.
4. A semleges alakok egyidejűleg több alosztályba beletartoznak, közös alesetei több
természetes osztálynak.
5. A mellérendelt szerkezeteket nem kell az osztályok valamelyikébe besorolni.
6. A koordinált NP-vel való egyeztetés visszavezethető a tagokkal való egyeztetésre. Ez az
egyeztetésnek egy sajátos alesete, így az egyeztetés a koordinációra vonatkozó rekurzióval
definiálható.
7. Az egyeztetés relációs felfogásához csupán a típus-hierarchia viszonylag egyszerű
kiterjesztésére van szükség: meg kell engednünk rendezett n-esek tulajdonságainak, azaz n-áris
relációknak megfelelő típusokat is a típus-hierarchiában.
További lehetőségek
Érdekes kérdés, hogy a magyar alanyi-tárgyas egyeztetéssel illusztrált elemzés kiterjeszthető-e
az alany és az igealak szám szerinti egyeztetésére. A fontosabb adatok a következők:
(24) Számbeli egyeztetés a magyarban a. János és Kati elmentek moziba.
b. János és Kati elment moziba.
c. Te és te kimehettek.
d. ?Te és te kimehetsz.
A magyar nyelv különleges sajátossága, hogy az egyes számú főnévi csoportból álló koordinált alanyok egyes számú igealakkal is állhatnak. Ennek a jelenségnek a leírásában is
használhatjuk a koordinált NP-vel való egyeztetésnek azt a rekurzív fajtáját, amelyet a határozottsági
egyeztetésnél láttunk. Azokban a nyelvekben azonban, amelyeknél nincs ilyen lehetőség (és
benyomásunk szerint ezek vannak többségben) a koordinált NP-k jel-lemezhetők grammatikai szám
vonatkozásában (ti. mindig többes számúak), talán azért, mert ez a tulajdonság igen szoros
összetüggésben van a jelentéssel. Az a tény, hogy a ma-gyarban kétféle egyeztetési minta is előfordul,
azzal függhet össze, hogy a magyarban az egyes számnak általában is kevesebb jelentéstani
következménye van (Id. három alma, mossa a kezét, ceruzát vettem). Kétféle elemzést lehet
elképzelni: vagy azt mondjuk, hogy a magyar koordinált NP-k is jellemezhetők szám tekintetében
Ertékek azonossága-e az egyeztetés?
208
(többes számúak), ezért töb-bes számú NP-ként az egyeztetés alapesete is alkalmazható rájuk, de a
rekurzív aleset is, s innen a kétféle lehetőség, vagy pedig nem jellemezzük a koordinált NP-ket
grammati-kai szám szempontjából, hanem az alany-állítmány egyeztetés egyik sajátos aleseteként
értelmezzük azt, amikor többes számú igealakkal állnak.
A szigorú lokalitás elvének feladásával az általunk javasolt eszközök olyan jelenségek
leírásában is használhatók lehetnek, amelyek nem kapcsolódnak sem a vegyes egyeztetéshez, sem a
koordinációhoz. Ilyen például a szigorúl lokalista elméletek számára hagyomá-nyosan fejtörést okozó
„emelésnek" nevezett jelenség:
(25) Emelés a magyarban
a. Pista meg akarja látogatni a nagymamáját.
b. Pista meg akar látogatni egy lányt.
Annak ellenére, hogy közvetlen szubkategorizációs viszony szempontjából a nagimamáját
illetve az egy lányt a meglátogatni ige tárgya, és a meg igekötő lexikailag a meglátogat ige része, a
fenti mondatokban mind az igekötő helye (prozódialiag a meg a meg akarja része), mmd a tárgyi
egyeztetés (a tárgy az akar főigével egyezik) arra utalnak, hogy a főige az infinitívuszi vonzatának
bővítményeit sajátjaként kezeli, „felemeli". Ezt úgy szeretnénk kezelni, hogy az igekötő-ige, ige-tárgy
stb. viszonyok (relációk) osztályának van olyan sajátos alosztályuk, amelyre az jellemző, hogy az ige
„emelő", a bővítmény pedig az emelő igéhez tartozó infinitívusszal áll a megfelelő relációban:
(26) Az ige-bővítmény kapcsolat rekurzív definíciója comp-rel(x,y) - def
(i) verb(x) AREL(x,y);
(ii) raising-verb(x) A comp-rel(lNF(x),y).
Végül további nyitott kérdés, hogy milyen természetes kikötések érvényesek arra, hogy milyen
kifejezések között állhat fenn grammatikai viszony. A szigorú lokalitás elvetése ugyanis szándékunk
szerint nem jelenti azt, hogy minden korlátozás nélkül bármely két kifejezés között grammatikai
viszony állhat fenn. Például természetes megkötésnek látszik, hogy A és B között csak akkor állhat
fenn egy bizonyos viszony, ha van olyan C, amelynek B része, és A és C között lokálisan áll fenn az
illető viszony. A kérdés pusztán az, hogy mit kell pontosan a „része" viszonyon érteni a fenti
megfogalmazásban.
Hivatkozások
Goldberg, Jeff és Kálmán László. 1992. Unification or unifiability? [Unifikáció vagy unifi-kálhatóság?] In
Proceedings of CLIN.
Grover, Claire. 1996. Parasitic gaps and coordination m HPSG. [Parazita űrök és koordi-náció a HPSG-ben.] In
Claire Grover and Enric Vallduví, editors, Edinburgh Working Papers in Cognitive Science, Vol. 12: Studies in
HPSG. Centre for Cognitive Science, Uni-versity of Edinburgh, Scotland, May, chapter 2, pages 33-69.
Ingria, Robert J.P. 1990. The limits of unification. [Az unifikáció korlátai.] In 28th Annual meeting ofACL:
Proceedings ofthe Conference. ACL, Morristown, NJ. Pp. 194-204.
Jackendoff, Ray. 1972 Semantic Interpretation m Generative Grammar. MIT Press, Camb-ridge MA.
Johnson, Mark. 1988. Attríbute-value logic and the theory of grammar. [Attribútum-érték logika és a nyelvtan
elmélete.] CSLI Lecture Notes. CSLI Publications.
King, Paul. 1989. A Logical Formalismfor Head-Driven Phrase Structure Grammar. [A HPSG egy logikai
formalizmusa.] Ph.D. thesis, University of Manchester.
Kmg, Paul. 1994. An expanded logical formalism for Head-Driven Phrase Structure Gram-mar. [A HPSG egy
kiterjesztett logikai formalizmusa.] Arbeitspapiere des sfb 340, Uni-versity of Tübingen.
Pollard, Carl J. and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press,
Chicago.
209
Ertékek azonossága-e az egyeztetés?
Pullum, Geoffrey K. and Arnold M. Zwicky. 1986. Phonological resolution of syntactic fe-ature conflict. [A
mondattani jegy-konfliktus fonológiai feloldása.] Language, 62(4):751-773, December.
Sag, Ivan A., Gerald Gazdar, Thomas Wasow, and Steven Weisler. 1985. Coordination and how to distinguish
categories. [A koordinációról, és arról/ hogy hogyan különböztessük meg a kategóriákat.] Natural Language
and linguistic Theory, 3(2):34-43,16-18, June.
Shieber, Stuart M. 1992. An introduction to Unification-based Approaches to Grammar. [Beve-zetés a nyelvtan
unifikációs alapú megközelítéseibe.] CSLI Lecture notes.
Zaenen, Annie and Lauri Karttunen. 1984. Morphological non-distinctiveness of coor-dination. [A koordináció
nem-megkülünböztető morfológiai jellege.] In Eastern States Conference on Linguistics 1, Columbia, Ohio. The
Ohio State University.
A Magyar Igekoto Egyeztetese
210
A Magyar Igekoto Egyeztetese
Kálmán László És TRÓN Vlktor
Bevezetés
A magyarban némelyik igekötő igen gyakran fordul elő olyan főnévi csoporttal, amelynek
határozóragja vele etimológiailag azonos:
(1) Igekötős ige és ragos főnévi csoport együttjárása
a. Belelépett a sárba.
b. Ránézett a lányra.
c. Benne maradt a dobozban.
d. Rajta felejtette a szemét afiún.
Ezt a jelenséget hagyományosan nem tekintik egyeztetésnek, pontosab-ban nem is foglalkoznak
vele, egyszerűen egy igekötős ige és egy vonzat kapcsolatának tekintik (ilyen elemzés például E. Kiss
(1998) is). Ezzel az a probléma, hogy nem magyarázza meg, hogy miért nincs olyan ige, amely bele,
rá, hozzá, neki, benn(e) ill. rajta igekötős, mégsem a megfelelő -ba/-be, -ra/-re, -hoz/-hez/-höz, -nak/nek, -ban/-ben ill. -on/-en/-ön/-n ragos vonzattal
48
jár.
Az egyeztetéses elemzésnek is vannak nehézségei, hiszen általában azt szoktuk egyeztetésnek
nevezni, amikor egy szónak különböző, jelentésük-ben azonos paradigmatikus alakjai vannak,
márpedig ezek a ragos főne-vek a maguk jogán is megjelenhetnek az ige mellett, amikor az nem igekötős. Ezekben a fentiekhez igen közel álló szerkezetekben nyilván nem tekinthető az esetrag
egyeztetési morfémának49
(2) Főnévi csoport igevivői helyzetben
a. A sárba lépett.
b. A lányra nézett.
c. A dobozban maradt.
d. A fiún felejtette a szemét.
Ez a fajta eloszlás sokkal inkább az expletívumos szerkezetekre emlékez-tet. Az expletívumos
elemzés szerint az ige előtti kitüntetett pozíciót, az ún. igevivő pozícióját valaminek el kell foglalnia
(hogy miért, az most közömbös, de világos, hogy ha nem lenne kitöltve, akkor valamilyen je-lölt,
például progresszív olvasatú mondatot kapnánk). Ha ezt a pozíciót a ragos főnévi csoport nem tölti ki
(Id. a (1) mondatokat), akkor valami másnak kell kitöltenie. Expletívumnak az ilyen helykitöltő
elemet szokás nevezni (ebben az esetben tehát az igekötő lenne az), amelynek más funk-ciója nincs is,
mint hogy elfoglaljon egy pozíciót, a jelentéshez legfeljebb ennyiben járul hozzá. Gyakori, hogy
alakilag rokona olyan tartalmas ele-meknek, amelyek abban a pozícióban előfordulhatnak.
48
Az is kérdés, hogyan lehetne egy ilyen elemzés esetén egyazon szótári tételből eredeztetni a fenti mondatokat
és névmási vonzatos változataikat is: Belelépett, Ránézett, Benne maradt, Rajta felejtette a szemét.
49
Cikkünkben nem foglalkozunk azokkal az esetekkel, amikor az igének névutós fő-névi csoport a vonzata, és a
névutó az igevivő helyét is elfoglalhatja (ment a hídon át és átment a hídon). Ez olyan névutókkal fordul elő,
amelyek előtt ragos főnévnek kell állnia (vmin át, vmin keresztül vmivel szemben). Alapvető különbség azonban
az általunk vizs-gált szerkezetek és ezek között, hogy ezekben mind a névutő, mind pedig a ragos főnévi csoport
minden szórendi változatban kötelezően jelen van. É. Kiss (1998) cikkével szem-ben nem tartjuk
elfogadhatónak az ilyen szerkezeteket ragtalan főnévvel álló névutók esetében (*alatta futott a fáknak, *mellette
úszott a korlátnak, *utánukfutott a lányoknak).
A Magyar Igekoto Egyeztetese
211
Ennek ellenére amellett fogunk érvelni, hogy az expletívumos elemzés nem tartható ebben az
esetben. Az adatok némileg zavarba ejtőek, de arra is szükség van az elemzéshez, hogy egyeztetési
viszonyt tételezzünk fel az igekötő és a vonzat között. Ugy gondoljuk, hogy érvelésünk a konstrukciós elemzés felsőbbrendűségét mutatja a fejközpontú megközelítéssel szemben.
Alapfogalmak
Először is vizsgáljuk meg, hogy mit is jelentenek azok az alapfogalmak, amelyeknek a fenti
jelenségek leírásában esetleg hasznát vehetjük. Minde-nekelőtt magának az egyeztetésnek a fogalmát
próbáljuk meghatározni, hogy lássuk, vajon alkalmazható-e itt ez a kategória:
(3) Az egyeztetés fogalma
Egyeztetésnek azt nevezzük, amikor egy bizonyos szintaktikai vi-szonyban álló elemeknek egy
bizonyos szempontból azonos (mor-fológiai) osztályba kell tartozniuk.
Szándékosan nem köteleztük el magunkat abból a szempontból, hogy az egyik elemnek
kitüntetettnek kell-e lennie (ti. annak, amelyikkel a má-sik „egyeztetve van"); ez a megszorítás
például rákényszerítene, hogy az alany-állítmány viszonyban (amely alapvetően exocentrikus) az
egyik ele-met kitüntessük - ennek a következőkben nem lesz jelentősége.
Az egyeztetéssel szemben azokat a morfológiai
aszimmetrikusak, szubkategorizációnak szokás nevezni:
viszonyokat,
amelyek
valóban
(4) A szubkategorizáció fogalma
Szubkategorizációnak azt nevezzük, amikor egy bizonyos szintakti-kai viszony a benne
szereplő elemek mibenlététől függően más- és másképpen lehet jelölve az elemeken.
Ez a meghatározás olyan általános, hogy az egyeztetés akár az alesetének is tekinthető, de azt
gondoljuk, hogy valóban nagyon sokféle esetben hasz-nálják, s ezek lefedéséhez nem fogabnazhatunk
ennél korlátozóbb módon.
Expletívum-e az igekötő?
Az expletívumos elemzés legnagyobb és legnyilvánvalóbb problémája, hogy az igekötő
megjelenésének - tapasztalatunk szerint - majdnem min-dig van jelentésbeli következménye, ha csak
árnyalatnyi is. Az eddigi pél-dáinkat úgy választottuk meg, hogy ez ne legyen nagyon feltűnő, de tekintsük a következőket:
(5) A kétféle változatnak különböző a jelentése
a. A doktornő belenézett a szemébe.
b. A doktornő a szemébe nézett.
c. Pista a heverőrefeküdt.
d. Pista ráfeküdt a heverőre.
Nem is kell talán magyaráznunk, milyen távol áll a fenti mondatpárok tag-jainakjelentése
egymástól. Az is azonnal látható, hogy ajelentéskülönbség nem rendszeres. Azt ugyan mondhatjuk,
hogy a szemébe néz valójában idi-óma, de a második mondatpár egyik tagja sem tartalmaz idiómát,
mégis jelentéskülönbség figyelhető meg (a második esetben a heverőnek érintett-nek kell lennie
azáltal, hogy Pista ráfeküdt, például le kellett szakadnia, vagy össze kellett gyűrődnie a lepedőnek).
Viszont az sem igaz, hogy idi-ómát nem tartalmazó mondatpároknál mindig a bővítménynek
megfelelő dolog érintettségében van a különbség:
(6) A kétféle változatnem a bővítmény érintettségében különbözik
a. A macska a dobozban van.
b. A macska benne van a dobozban.
212
A Magyar Igekoto Egyeztetese
Itt a második mondatból az következik, hogy a doboz a macska várható vagy szokásos helye, a
doboz érintettségéről pedig szó sincs. Ha az exple-tívumos elemzés helyes lenne, akkor semmilyen
jelentéskülönbséget sem várnánk a mondatpárokon belül, de legalábbis rendszeresnek kellene len-nie
a jelentéskülönbségnek. Mivel egyik sem teljesül, az expletívumos elem-zést el kell vetnünk.
Egy másik érvünk is van az expletívumos elemzés ellen: az igekötő je-lenléte ugyanis nemcsak
akkor hordozza azt a bizonyos (számunkra egy-előre megjósolhatatlan) jelentéskülönbséget, amelyről
fent szóltunk, ami-kor az ige előtti („kötelezően kitöltendő") pozícióban áll:
(7) A jelentéskülönbséghez az igekötőnek nem kell igevivőnek lennie
a. Miattad maradt a dobozban.
b. Miattad maradt benne a dobozban.
c. Tegnap nézett a doktornő a szemébe.
d. Tegnap nézett bele a doktornő a szemébe.
Nem lehet igazolni, hogy ezeknek a pároknak a második mondatai exp-letív igekötőt
tartalmaznak, hiszen nincsen semmiféle egyéb jele annak, hogy a ragozott ige után is lenne kitöltendő
hely a magyarban.
Milyen mondattani viszonyokról van szó?
Az igekötőnek és a vele „társult" főnévi csoportnak mindenképpen vala-milyen mondattani
viszonyban kell állniuk egymással ahhoz, hogy akár egyeztetési, akár valamilyen általánosabb
szubkategorizációs kapcsolatot feltételezzünk közöttük. De vajon miben áll ez a viszony? Sajnos
minden jel arra utal, hogy a szokásos értelemben nem tudjuk őket egy összetevő-nek tekinteni, hiszen
sosem fordulnak elő együtt olyan helyzetben, amely-ben egyébként összetevők szoktak:
(8) Az igekötőt és a főnévi csoportot nem lehet együtt íókuszba tenni
a. *NEKI A FALNAK/*A FALNAK NEKI ment az autóval.
b. *HOZZÁ A FALHOZ/*A FALHOZ HOZZÁ vágta a könyvet.
(9) Az igekötőt és a főnévi csoportot nem lehet együtt topikalizálni
a. *NEKI A FALNAK/*A FALNAK NEKI nem ment a kocsival.
b. *HOZZÁ A FALHOZ/*A FALHOZ HOZZÁ sose vágta a könyvet.
Igaz, hogy néha előfordulnak ilyen szerkezetek (pl. bele a dobozba), de az, hogy általában nincs
mód arra, hogy egy összetevőnek tekintsük az igekö-tőt és a főnévi csoportot, mindenképpen ellene
szól ennek a típusú elem-zésnek.
A másik ellenjavallat az, hogy nem találunk olyan idiomatikus kapcso-latokat, amelyek
igekötőből és ragos főnévi csoportból állnának, találunk viszont olyanokat, amelyek az igekötőt és az
igét tartalmazzák, márpedig általánosan elfogadott nézet, hogy idiómákká csak összetevők
válhatnak50? Például a rájön, ráér igék nyilvánvalóan ilyenek, és bár -ra/-re ragos vonza-tuk van, az
igekötő - mivel az idióma része - sohasem maradhat el, nincs *A megoldásra jöttem, csak Rájöttem a
megoldásra.
Az egyetlen elfogadható elemzés tehát az, hogy az igekötő az igével alkot összetevőt, de
mindig olyat, amelynek a megfelelő ragos főnévi cso-port a vonzata, pontosabban: amelynek vonzata
egyezik az igekötővel. Például a rak igének amúgy is van 'valahová' értelmű vonzata, és úgy is
felfoghatjuk - sőt, ez az egyszerűbb -, hogy ezt a vonzatát a rárak alakban is megtartja, csak éppen az
50
A magyarban van ugyan néhány látszólagos ellenpélda (pl. kítör vkit afrász, megjön vkinek az esze stb.), de
nem vagyunk benne biztosak, hogy a magyarban az alany mindig egy állítmányi összetevővel kapcsolódik
össze, tehát ezekben a példákban igenis feltéte-lezhető, hogy az alany (a frász ill. az esze) az igével összetevőt
alkot.
A Magyar Igekoto Egyeztetese
213
igekötő kikényszeríti a vonzat egyezését (a -ra/-re ragot). Akármilyen keretet válasszunk is tehát a
jelenség leírására, a vizsgált szerkezetek elemzésénél mindenképpen hivatkoznunk kell az igekötő és a
ragos főnévi csoport közötti egyeztetés-szerű viszonyra.
Különös, számunkra egyelőre megmagyarázhatatlan jelenséget talál-tunk, amely ezt
alátámasztja. Az igekötős igék általában nem tiltják, hogy bármelyik bővítményük is személyes
névmás legyen, az általunk vizsgált osztály azonban igen:51
(10) Az igekötős ige bővítménye általában lehet személyes névmás a. Keresztülment rajta.
b. Meghívta 'ót.
(11) A ragos főnévi csoport nem lehet személyes névmás
a.
b.
c.
d.
*Ráunt rám.
*Beleszeretett belém.
*Ráismert rám.
*Ránéztem rájuk.
Az oda + -nak/nek ill. ott + vhol vonzatkeretű igéket többek között azért nem tekintjük cikkünk
tárgyának, mert nem így viselkednek: Odaadtam neki, Ottjártam nála. Másrészt azért nem
foglalkozunk velük, mert az ott, oda igekötővel álló igéknek sokféle vonzatuk lehet, vagyis nem
rendelkez-nek az egyeztetés tulajdonságával. Ennek a jelenségnek a fényében meg kell
különböztetnünk az általunk vizsgált szerkezeteket a közönséges ige-kötős ige-vonzat kapcsolattól. A
magyar nyelvtanban egyetlen ehhez ha-sonló jelenséget találtunk: a -nak/-nek ragos birtokost
tartalmazó birtokos szerkezetben (legalábbis akkor, amikor ezek elszakítatlanul állnak) nem lehet a
birtokost személyes névmással kifejezni (*NEKEM A KALAPOM veszett el, *NEKTEK A HÁZATOK
égett le). Mind a vizsgált igekötős igék, mind abirtokos szerkezetek esetébennémelyestjavítja a
grammatikalitást, ha a személyes névmás hosszabb alakját (enrám, terád, őbelé stb.) használ-juk,
ebben is hasonlít a két szerkezet. Akárhogy is áll a dolog, ez a j'elenség amellett szól, hogy a vizsgált
szerkezetet ne egyszerűen igekötős ige és bő-vítmény kapcsolataként elemezzük, hanem a leírásnál
mind az igekötőt, mmd az igét, mind pedig a bővítményt közvetlenül a szerkezet részeként fogjuk fel.
A személy- és számegyeztetés hiánya
A legtöbb adatközlő szerint vannak a (11) példákhoz hasonló mondatok, amelyek közel vagy
teljesen elfogadhatóak:
(12) Személyes névmás ismétlése
?%NEKEM jött nekem.
Ezt a mondatot azonban nem lehet igekötős ige és személyes névmás kap-csolatának tekinteni,
hiszen az igekötő személyben és számban nem egye-zik a hozzá tartozó bővítménnyel:
(13) Nincs személy- és számbeli egyezés
a. *Rájuknéztem a fiúkra.
b. *Beléjükszeretett a lányokba.
Ez a tény azonban rejtélyessé teszi a következő adatokat:
(14) Ragozott igekötő? Személyes névmás mint igevivő?
a. Rámunt. (Vö. *A fiúra unt.)
51
A többi mellett ez is nehézséget okozna akkor, ha a kérdéses szerkezeteket egysze-rűen egy szótári igekötős
ige és egy ragos vonzat kapcsolatának tekintenénk, hiszen eze-ken kívül nmcsenek olyan szótári igék,
amelyeknek nem lehet a vonzatuk személyes név mas.
A Magyar Igekoto Egyeztetese
214
b. Belémszeretett. (Vö. *A fiúkba szeretett.)
c. Belédvágta a kést. (Vö. OKA fiúkba vágta a kést.)
Az (a-b) esetekben mondhatnánk azt, hogy az igekötő mégiscsak ragozva van (éppen akkor,
amikor a neki megfelelő bővítmény nincsen jelen). Eb-ben az esetben azt kellene mondanunk, hogy
az igekötő személyragja név-másként működik, s ez kedvező lenne, mert megmagyarázná, hogy miért
nincs ilyenkor lehetőség arra, hogy még egy személyes névmás is megje-lenjen. Hátránya viszont
ennek a leírásnak, hogy nem jósolja meg, miért nem jelenhet meg a személyes névmás az egyéb nem
ragozható igekötők mellett (pl. nincs *rájukfázom, de *ráfázom rájuk sem).
A másik lehetőség, hogy azt állítjuk, ezekben az igekötőtlen ige szerepel, a személyes névmás
pedig bővítmény, amely az igevivő szerepét játssza. Ennek a lehetőségnek azonban több nehézsége is
van. Egyrészt azt kell megmagyarázni, hogy miért csak személyes névmás játszhatja ezt az igevivői
szerepet (amint ezt a zárójelbe tett ellenpéldák mutatják), és miért nem ez a helyzet a (c) mondat
esetében, valamint így is „kilógnak" azok az igék, amelyek nem engedik meg ezt a lehetőséget
(*rájukfázom). A nagyobbik baj az, hogy sokszor az ige magában nem is engedné meg az illető
bővítményt: az az igekötő megjelenéséhez van kötve (pl. nincs *szeret vkibe).
Ezzel kapcsolatban még egy jelenségre kell felhívnunk a figyelmet:
(15) A névmás nem mindig lehet fókuszban
a. Pista BELÉD vágta a kést.
b. *Pista BELÉD szeretett.
Mi az oka ennek a kontrasztnak? Nyilvánvaló, hogy az igekötő amúgy lehet fókuszos a
magyarban (FELvitte, nem pedig LE), tehát meg kell ma-gyaráznunk, hogy a (b) példa miért rossz.
Ezt semmiképpen sem tudnánk megmagyarázni, ha a második elemzési lehetőséget választanánk,
misze-rint a ragozott névmási elem valójában bővítmény, és ilyenkor nincs igekötő, mert akkor semmi
különbség nem lehetne az (a) és a (b) eset között. Csak az lehet a magyarázat alapja, hogy az első
mondat esetében lehet a névmás bővítmény (mert az igének a maga jogán, igekötő nélkül is lehet -ba/be ragos bővítménye), míg a (b) mondatban kell lennie igekötőnek. Miért nem lehet az igekötő itt
fókuszban? A választ abban kell keresnünk, hogy a fókusz csak egyetlen szemantikai mozzanat
előtérbe állítását en-gedi meg, márpedig a ragozott igekötő - ha elfogadjuk ezt az elemzést jelentéstanilag tartalmazza az igekötőt és a személyes névmást is, ezért összeférhetetlen a fókusszal.
Az (a) mondatban viszont, mint ezt már ko-rábban említettük, csak egyedül a személyes névmás
ragozott alakja van fókuszban.
Leírási kísérlet
Bár az adatok egy része továbbra is rejtélyesnek mutatkozik, a jelenség leírásának bizonyos
lehetőségeit sikerült kizárnunk, így tehát megkísérel-hetjük a fennmaradó lehetőségek szellemében
legalább részlegesen leími a tényeket.
Az elemzés legfontosabb eleme, hogy sajátos lexikai osztályt alkotnak azok az igék, amelyek
igekötőszerű névmási elemmel és azzal egyező ra-gos főnévi csoporttal állhatnak. Azok a szerkezetek
pedig, amelyekbe ezek beilleszthetők, sajátos konstrukciókkal írhatók le. Ezek közül a két legfontosabb az igevivőt és főnévi csoportot egyaránt tartalmazó, valamint a csak főnévi csoportot
tartalmazó. Ezt mind jelentéstani, mmd mondattani sajátosságaik indokolják. Az ezekben a
konstrukciókban szereplő igeosz-tályok nem feltétlenül diszjunktak, innen ered, hogy néhány ige
azonos jelentésben is többféle típusú mondattani szerkezetben is előfordul (pl.:
ránézett a lányra, a lányra nézett). A kétféle konstrukcióban előforduló igék jelentése között
rendszeres kapcsolatot legfeljebb néhány altípusuk eseté-ben tételezhetünk fel (pl. az igevivős
változat a bővítmény érintettségét fejezi ki a ráül, ráfekszik stb. esetében).
A szóbanforgó konstrukciók mondattani viselkedésének legegyszerűbb magyarázata az, hogy
bennük az igekötőszerű névmási elem és az ige ugyanolyan viszonyban van, mint általában az igekötő
A Magyar Igekoto Egyeztetese
215
és az ige (tehát eb-ből a szempontból ilyenkor is igekötős igéről beszélhetünk), a kettő együtt pedig
olyan viszonyban van a bővítménnyel, mint általában az igekötős igék a bővítményeikkel.
Példának álljon itt az a konstrukció, amelyben mind az igekötőszerű névmási elem, mind a
ragozott főnévi csoport szerepel:
(16) Az igevivő egyeztesését mutató konstmkció
igevivő-egyeztetés(x) <=>def
vonzatos-ige (x)
VONZAT(x) = fncs
főnévi-csoport(fncs)
IGE(x) = ige igekötős-ige(ige)
IGEKÖTŐ(ige) = ik
névmási-igevivő (ik) IGETŐ(ige) = tő
igevivő-egyezteséses-ige (tő)
rag-egyeztetés(ik, fncs)
Itt a félkövérrel szedett szimbólumok csomópontok nevei a nyelvtant leíró általános öröklési
hálóban. Az egész leírás a beleszeret-vkibe típusú konst-rukció leírása, definíciója. Az x változó a
nyelvészeti objektumok felett fut, a definíció ezek közül jelöli ki azokat, amelyek a szerkezet
követelmé-nyeinek megfelelnek. Az „ik"/ a „tő" és az „fncs", a szerkezet építőköveinek, a névmási
igevivőnek, az igének és a főnévi bővítménynek felelnek meg. Ezekről kimondja a definíció, hogy
rendre tartozzanak bele a meg-felelő szóosztályokba (pl. a bele névmási igevivő, a szeret pedig
igevivő-egyeztetéses ige). Az „ige" nevű változó az igekötőből és az igéből álló „virtuális" igekötős
igének felel meg (azért virtuális, mert ezek nem alkot-nak a szokásos értelemben vett összetevőt).
Feltételezzük, hogy az igekötős-ige csomópont felelős az igekötős igék mondattani viselkedéséért
általá-ban. Az igéből és vonzatából álló „virtuális" összetevő viselkedését jel-lemzi a vonzatos-ige
nevű csomópont, végül pedig a rag-egyeztetés cso-mópont leírja, hogy mely igekötőszerű névmás
milyen ragozott főnévi csoporttal áll egyeztetési viszonyban. Feltételezzük, hogy a névmási ige-vivő
és a közönséges igekötők osztálya diszjunkt, ezért pl. a bele csak akkor lehet igevivő, ha egyezik az
egyik vonzattal.
Külön gondoskodnunk kell annak a lehetőségnek a leírásáról, hogy az igekötő és a bővítmény
összeolvadhat, ha a bővítmény személyes névmás. Fel kell tételeznünk, hogy ha morfológiailag
lehetséges az összeolvadás, akkor meg is kell történnie. (Az egyelőre nem világos számunkra, hogy ez
következik-e valamilyen univerzális nyelvtani elvből, vagy külön stipu-lálnunk kell, és ha az utóbbi
áll fenn, akkor hogyan kellene ezt megtenni.) Kivételt képez az az eset, amikor fókuszba kellene
helyeznünk a névmást vagy az igekötőt, mert - mmt említettük - az összeolvadt egység fókuszba
helyezése összeférhetetlen lenne a fókusz jelentéstanával.
Külön figyelmet érdemelnek azok az igekötős igék, amelyekben az ige-kötő mindig ragtalan.
Ezek mind idiomatikus jelentésűek, mint például a rájön vmire:
(17) A rájön vmire szótári tétele
rájön-vmire(x) <=>def
vonzatos-ige (x) VONZAT(x) = fncs
főnévi-csoport(fncs) IGE(x) = ige
A Magyar Igekoto Egyeztetese
216
igekötős-ige(ige) IGEKÖTŐ(ige) = ik rá(ik)
IGETŐ(ige) = tő jön(tő)
rag-egyeztetés(ik, fncs)
ALAK(ik) = „rá"
Ennek a leírásnak - a benne szereplő lexikai elemektől eltekintve - csak az utolsó sora újdonság
a fenti konstrukcióhoz képest, tehát a rájön vmire le-xikai tétele annak alesete. Az utolsó sor pedig azt
fejezi ki, hogy a névmási igevivő csak rá alakú lehet. Ilyen módon az igevivő és a névmási vonzat
összeolvadása csak akkor következhet be, ha az illető névmás egyes szám harmadik személyű: Rájött,
de nem jó: *Rájukjött,
Hivatkozások
É. Kiss K. 1998. „Verbal prefixes or postpositions? Postpositional aspectu-alizers m Hungarian" [Igekötők vagy
névutók? Névutói aspektusjelö-lők a magyarban]. In: C. de Groot és Kenesei I. (szerk.), Approaches to
Hungarían 6: Papers from the Amsterdam Conference. JATE Press, Szeged.
Representation of Linguistic Knowledge in GIN 99
217
Representation of Linguistic Knowledge in GIN 99
László Kálmán, Vlktor TRÓN
Introduction
GIN (Generalized Inheritance Networks) is a framework for linguistic knowledge
representation based on typed feature structures. It differs from systems like HPSG (Pollard and Sag
(1987), Pollard and Sag (1994), King (1989), King (1994)) in its ability to deal with relational
information, which makes it easier to represent semantic information and so-called correspondence
rules (Jack-endoff (1997)). We will present the system in two steps. First, we will argue for the
enrichment of the so-called attribute/value formalism in order to represent relational information,
and present the enriched formal language that GIN relies on (section 2). Second, we will explain how
the type resolution algorithm simulates the performance processes that operate on the knowledge
base embodied in the type hierarchy (section 3).
The GIN language
Motivation
Our main motivation in creating the GIN framework was to provide a formal model for
linguistic theories like Construction Grammar (Fillmore and Kay (1993), Goldberg (1995)) and
Jackendoff's (Jackendoff (1997)) cognitive grammar. Such theories crucially rely on so-called
correspondence rules at the various types of linguistic interfaces (such as the syntax/semantics
interface), which are difficult to represent in the traditional attribute/value structures (AVS)
formalism. We will first present the AVS formalism, then point to its shortcomings from the point of
view of representing relational information.
Attribute/value structures
The common format of linguistic representations is the so-called attribute/value structure or
AVS. An AVS is a set of ordered pairs consisting of an attribute and a value. The way in which
AVSs describe linguistic objects is the following. Properties are conceived of as 'having a certain
value for a certain attribute'. For example, 'being a noun' can be represented as 'having the value noun
for the attribute CATEGORY'. This allows a finer-grained representation of properties than primitive
predicate symbols. In particular, the value of an attribute may be either atomic (a variable or a
constant, like the constant noun above) or an AVS itself. The latter type of values cannot be represented with primitive predicate symbols. For example, if the linguistic object to be represented *
The research forming the basis of this paper was part of the COMPLIN Project. The second author
was supported by Group Research Support Scheme, Grant No. 320/1998. of the Open Society
Foundation.
Is a sentence, then its attribute SUBJECT may have an AVS as a value, which specifies, say,
the category and the agreement properties of the subject. This allows the representation of
underspecified linguistic objects, e.g., the subject of the sentence in this example does not correspond
to a unique linguistic object, but to the entire class of nouns having certain agreement properties.
Clearly, every AVS may contain at most one pair with a given attribute. For example, the category of
an object is either not specified at all or it is characterized with a given value, but not with several
values.
Formally speaking, the AVS language is a fragment of first-order predicate logic with the
AVSs corresponding to the terms of the language. The only sentential connective is conjunction, and
the only type of quantification is existential. I will refer to such fragments as positive fragments in
the following. Furthermore, the language contains equality as a term connective, and unary function
symbols. Each AVS can be written down as an equation system with equations of the form ' f(t1) = t2,
Representation of Linguistic Knowledge in GIN 99
218
where x and y are variables or constants, and f is a unary function symbol, plus one of the terms
occurring in the equation system is distinguished as the root node of the AVS. For example, the AVS
(1)
can be translated into the equation systém
(2)
CATEGORY(r) = noun
AGREEMENT(r) = 3Sg
with r as the root node. The AVS
(3)
can be translated into the equation systém
(4)
with r as the root node. The term root node originates from the graph representation of AVSs.
An AVS can be represented as a directed graph the edges of which are labelled with function
symbols, and the nodes with terms. The AVS graph has a distinguished root node from which every
other node is can be reached through some directed path. AVSs are connected in this special sense.
(5) AVSs
The set of AVSs over F (a set of functors) is the set of ordered pairs A = <q, E> such that q
(the root of A) is a term (a variable or constant), and E is a set of equations such that, for every e ∈ E,
e has the form f(t) = v, where f ∈ F is a functor, and t, r are terms. We say that t, v ∈ Nodes(A) in this
case. Furthermore, we require that Q ∈ Nodes(A), and that every node a ∈ Nodes(A) \ {q} be
accessible to the root node Q; accessibility is defined below.
(6) Accessibility in AVSs
If A = <Q, E> is an AVS then, for every t ∈ Nodes(A), ACC(t, E) or the set of terms accessible to t in
E is defined as the smallest set satisfying the following criteria:
Representation of Linguistic Knowledge in GIN 99
219
(i) if f(t) = v ∈ E for some f, v, then v ∈ ACC(t, E);
(ii) for every w,v ∈ Nodes(A)/ if v ∈ ACC(t,E) and w ∈ ACC(v, E), then w ∈ ACC(t,E).
That is, the special kind of connectedness mentioned above means that n 6 ACC(g,E) for every
n ∈ Nodes(A) \ {q}. The AVS language has the advantage that the complexity of the decision of the
term equivalence problem in it has a relatively low complexity, which is important for practical
reasons (and other operations on AVSs are similarly complex). Here is the definition of the
equivalence of two AVSs:
(7) Equivalence of AVSs
The AVSs A1 = <qI, E1> and A2 = (q2, E2) are equivalent (written A1≡A2) iff there is a bijection
β : Nodes(A1) -> Nodes(A2) such that
(i) β (q1) = q2;
(ii) For every equation f(n1)= n′1 ∈ E1 and f(n2)= n′2 ∈ E2, if β(n1)= n2, then β(n′1)=n′2
(iii) If C1 is the set of constants in Nodes(A1), and C2 is the set of constants in Nodes(A2),
then β is an identity bijection w.r.t. C1 and C2.
The algorithm of deciding the term identity problem aims at constructing the relation R, which
is initially R0 = {<Q1, Q2>}, then we have to examine all the equations of the form f(q1) = n1 ∈ E1 and
f(q2) = n2 ∈ E2 and (a) stop (fail) if n1 and n2. are different constants, or proceed with R1 =
R0∪{<n1,n2>}. Thus, in the worst case, we have to compare |Nodes (A1)| • |Nodes(A2)| nodes. The
same considerations apply to the operation of unification, which consists of the union of the two
equation systems (plus identifying the root nodes and whatever nodes have to be identified in
addition). Here is its definition:
(8) Unification of AVSs
The unification of the AVSs A1 = <q1, E1> and , A2 = <q2, E2> (written , A1 ∪ A2) is an AVS
A= <q,E> such that there are mappings µ1 : Nodes(A1) —> Nodes(A) and µ2: Nodes(A2) —>
Nodes(A) satisfying the following conditions:
(i) µ1(q1) = µ2(q2) = q;
(ii) if µ1(n1) = µ2(n2) = n, and there are equations f(n1) = n′1 ∈ E1 and f(n2) = n′2 ∈ E2
then
(i) µ1(n′1) = µ2(n′2) = n′, and
(ii) f(n) = n′ ∈ E;
(iii) If C1 is the set of constants in Nodes(A1) and C2 is the set of constants in Nodes (A2),
then µ1 and µ2 are identity mappings w.r.t. C1 and C2.
(iv) The only nodes and equations of A are the ones defined by (i-iii) above
= ⊥).
If the mappings µ1, µ2 do not exist, the unification is said to fail (yield a failure, written A1 ∪ A2
(It is easy to prove that all AVSs A satisfying this definition are equivalent to each other.) The
unification of two AVSs corresponds to conjunction, and the interpretation of A1 ∪ A2 is the
intersection of the set of objects in the interpretation of A1 with the set of objects in the interpretation
of A2. The interpretation of ⊥ is the empty set.
Representation of Linguistic Knowledge in GIN 99
220
AVSs and relations
The drawback of the AVS language, on the other hand, is that it cannot express relational
information. For example, we cannot express agreement as a relation between AVS nodes, because
the language has no relation symbols. Consequently, if the model contains relations between entities,
then those relations cannot be expressed in the AVS language in the general case. In this section, we
will present pieces of motivation for the need of relational information in linguistic representations.
Consider the following English sentences:
(9)
a. Why did Joe leave and the others stay?
b. *Why does Joe leave and the others stay?
c. *Why do Joe leave and the others stay?
Here did, which is unmarked for the NUMBER feature, can agree both with the singular noun
phrase Joe and the plural noun phrase the others at the same time, whereas do and does, which are
marked with respect to number, cannot occur in such a context. The same applies to first person
singular past-tense Hungarian verb forms which are unmarked for the DEF-OBJ feature (definiteness
of the third-person object), whereas the same form is marked for DEF-OBJ in the present.
(10) a. Minden fiút és a két lányt láttam. every boy-ACC and the two girl-ACC saw-I I saw
every boy [indefinite] and the two girls [definite]'
b. *Minden fiút és a két lányt látok.
see-I-INDEF-OBJ
c. *Minden fiút és a két lányt látom.
see-I-DEF-OBJ
The problem with this type of phenomena is that, if one was to treat agreement in terms of
structure sharing, then unification would fail in the cases when an unmarked form is present in just the
same way as in the ungrammatical sentences above. For example, if the phrase structure rule
generating transitive structures prescribed that the DEF-OBJ feature of the verb must have the same
value as the DEF feature of its direct object, then either we should say that co-ordinated noun phrases
with both a definite and an indefinite conjunct are unmarked for definiteness (in which case we would
wrongly predict the (a) and (b) examples to be grammatical) or we should stipulate that the verb has
to agree with every conjunct in definiteness, in which case we would predict the (a) sentence to be
agrammatical, too, since the two definiteness values of the conjuncts should be unified because of
structure sharing.
The problem lies in the fact that structure sharing is a very special type of relation, namely,
term identity, and this is the only type of relational information available in the AVS formalism. It
seems that the phenomena in question require a type of agreement similar to structure sharing in terms
of its being a relation, though not the identity relation (in particular, this is a relation that holds
between identical values or a marked value and an unmarked value). This type of cases could be
successfully dealt with using a relation-based account.
Another, much more general, motivation for relational information comes from the essentially
cumulative character of semantic representations. For example, it would be unnatural to assign
hierarchical semantic representations to co-ordinated structures and multiple modifier structures. For
example, take the following sentence:
(11) Joe is reading the book and the -paper.
In the semantic representation of this assertion, either there is one theme, a set consisting of the
book and the paper, or there are two themes (a book and a paper), or there are two predicates (read the
221
Representation of Linguistic Knowledge in GIN 99
book and read the paper), or there are two events described (Joe reading the book and Joe reading the
paper). In all these cases, two entities play the same role in one and the same (sub-)structure. Thus,
technically speaking, those roles cannot be characterized as functions.
2.4 Representing relations in an AVS format
The only way of creating the illusion that the AVS refers to a relation is to use an attribute or an
attribute/value pair representing the relation symbol, and some distinguished attributes (such as ARGl,
ARG2 etc.) the values of which correspond to the arguments of the relation:
(12)
The reason why we qualify this method as creating an illusion is the following. If we attributed
a special semantics to attributes like RELNAME, ARGl and ARG2 in order to force an AVS like the
above to denote a genuine relation, the resulting language would fail to be complete. For example,
assume that, for some reason, we want to express the agreement of subjects and predicates in a
relational format, then we could modify our previous example in either one of the following ways:
(13)
(14)
Under the intended relational interpretation, these two AVSs should have the same denotation
without their being syntactically equivalent, which means is that the language is not complete.
However, if we were to modify the definition of AVS equivalence in such a way that AVSs like the
above should come out equivalent, we would end up with a very complex computational procedure.
222
Representation of Linguistic Knowledge in GIN 99
Note, in addition, that the above encoding allows only one relational statement per AVS, since
RELNAME is an attribute, so it can have at most one value. We could use the relation symbol itself as
an attribute and put the arguments into its value:
(15)
This would make it possible to state more than one relation in one AVS, but at most one with
each relation symbol. To overcome this, one could embed each relational statement under a different
dummy relation name:
(16)
However, if one intended to maintain the genuinely relational interpretation of such formulae,
then one should ensure that the attributes AGR-A and AGR-B are syntactically indistinguishable,
which is an additional source of incompleteness.
As a matter of course, AVSs can be enriched with relational formulae:
(17)
(We just improvised this AVS for the sake of illustration: we know of no language in which the
topic and the comment stand in an agreement relation.) This would yield a language equivalent to the
positive fragment of first-order predicate logic. This can be proved easily by just giving a translation
algorithm from 'positive first-order' formulae to AVSs with relational sub-formulae. In this algorithm,
existential quantification corresponds to underspecified values for the arguments in question (and
structure sharing by co-indexing if one and the same existentially bound variable occurs in several
relational statements). The complexity of the term equivalence problem in positive first order logic,
however, is of an extremely high complexity.
This type of enriched language has been used by various authors (Pollard and Sag (1994),
Copes take, Flickinger, and Sag (1996?)) under the disguise of set-valued attributes:
223
Representation of Linguistic Knowledge in GIN 99
(18)
Here the value of the RELS attribute is meant to represent all the relational statements in one
set. In the literature, usually several dummy attributes (e.g., SEMANTICS, PHONOLOGY) are used,
but this is not meant to have formal consequences, they are used only for mnemotechnic reasons.
Alternatively, one could have the relation symbols themselves having set values:
(19)
These enriched AVS languages exhibit the same incompleteness problem as the ones
mentioned earlier, which can be overcome by modifying the definitions of AVS operations. The
complexity characteristic for 'positive first-order logic' manifests itself as the problem of unifying and
comparing set-values. To put it in simple terms, the problem is what to unify with what. Assume that
the sets to be unified contain statements analogous to the first-order formulae R(x, y), P(x, y), R(z, u),
Q(u) and P(w,u), where x,y,u,w and z are existentially bound. Depending on what the rest of the
formula says about these variables, we have to decide whether to unify u with y and/or x with z.
What is at stakes here is whether we can find a fragment of first-order logic that is more expressive than the traditional AVS formalism (relational information can be expressed in it), and yet it is
computationally tractable, i.e., the complexity of the operations on formulae is relatively low, which
means that it must be less expressive than 'positive first-order logic'. We propose to achieve this by
allowing genuine relational information and, at the same time, strictly controlling what to unify with
what. To ensure this, we will impose limitations on the use of existential quantification (see 2.5).
Yet another fragment
In this section, I will put forward a fragment of first-order logic reminiscent of the AVS
formalism that has less expressive power than 'positive first-order logic', yet allowing relational
statements.
The way in which we will enrich the traditional AVS language in order to deal with relations is
to introduce n-ary functors into the equation sets. That is, we will allow equations of the form
f=(t1,…,tn) in the equation sets. Then, for example, we can say that both AGR(t1,t2)=1 and
Representation of Linguistic Knowledge in GIN 99
224
AGR(t3,t4)=1 are in our equation system, which corresponds to the first-order formula
AGR(t1,t2)∧AGR(t3,t4). We can do this because the uniqueness of features in an AVS corresponds to
the fact that f(t) = v1 and f(t) = v2. entails v1=v2, but no such entailment is present if the arguments of
the functor / are different in two equations. I will call such enriched AVSs multi-AVSs.
(20) Multi-AVSs
A multi-AVS over the set F of functors (which contains n-ary functors for every natural number
n) is an ordered pair M = <q, E> such that q (the root of M) is a term, and E is a set of equations such
that, for every e ∈ E, e has the form f(t1,..., t2) = v, where / e F is an n-argument functor, and t1,..., tn, v
are terms. We say that t1,..., t2, v ∈ Nodes(M) in this case. Furthermore, we require that q ∈ Nodes(M),
and that every node n ∈ Nodes(M) \ {q} be accessible to at least one root node in q, which means
this:
(21) Accessibility in multi-AVSs
If M = <q, E> is a multi-AVS then, for every t ∈ Nodes(M), ACCn(t, E) or the set of terms
accessible to t in E is defined as the smallest set satisfying the following criteria:
(i) if f(t1,..., t2) = v ∈ E for some f, t1,..., t2, v (where t = ti, for some 1 ≤ i ≤ n), then
v∈ACCn(t,E);
(ii) for every w,v ∈ Nodes(M), if v ∈ ACCn(t, E) and w ∈ ACCn(v, E), then w ∈ ACCn(t,E).
This is a straightforward extension of the concept of accessibility and connectedness defined
for AVSs.
As can be seen, multi-AVSs constitute an extension of the original AVS language, yet they can
be used for expressing relational statements by using functions with Boolean values. The question is
whether the expressive power of the multi-AVS language is indeed weaker than that of 'positive firstorder logic'. To answer this question, we have to see whether an ordinary first-order formula like
AGR(t1, t2,) ∧ AGR(t3, t4) can be encoded in the multi-AVS formalism. We have seen that we can
produce two equations that do the job, but the connectedness requirement imposes a serious limitation
on this, namely, the terms t1,..., t4 have to be accessible to the root node in the sense defined above.
This, however, means that this formula cannot be translated into the multi-AVS language (unless t1 =
... = t4 all represent the root node). No analogous limitation is imposed by the 'positive' fragment of
first-order logic, of course. Thus, multi-AVSs correspond to a fragment of first order logic in which
the atomic formulae are not valid formulae of the language at all: all variables in a formula have to be
either identical to the single existentially bound variable that corresponds to the root node, or else be
explicitly identified with the value of some n-ary function symbol applied to n arguments that are
independently licensed in this way. The only connective of the fragment is conjunction, and there is
exactly one existential quantification per formula.
The concepts of equivalence and unification of multi-AVSs are analogous to the definitions
seen in 2.2 (see the final version in (24)).
It is easy to see that the complexity of the unification operation (and, similarly, of other operations) for multi-AVSs is of the same order of magnitude as for traditional AVSs. At any rate, it is
significantly smaller than in 'positive first-order logic'.
The multi-AVS formalism allows us to formulate the relational analysis of Hungarian definite
object agreement we mentioned earlier. As we said, a special type of agreement (called DEF-AGR),
different from the structure sharing of definiteness values, is at work in such structures: the unmarked
verb can agree with any type of noun phrase (including an unmarked one), whereas marked verb
forms must strictly agree with the noun phrase. Moreover, the conjuncts in a coordinated noun phrase
stand in the same relation with the definiteness of the entire noun phrase.
225
Representation of Linguistic Knowledge in GIN 99
So both the transitive verb phrase rule and the NP co-ordination rule must make reference to the
DEF-AGR relation. (In the formulae below, we use the AVS formalism enriched with relations for the
sake of readability.)
(22)
(23) NP co-ordination rule
Multi-AVSs and types
Multi-AVS type hierarchies
The aim of GIN is to try to reconstruct whether a given formula can be seen as subsumed by the
unification of a set of pre-defined concepts (or constructions, when we refer to specifically linguistic
concepts). GIN is based on an inheritance network of concepts, i.e., a set of types forming a type
hierarchy. Each type is itself represented with a multi-AVS (so that we can perform AVS operations
on concepts). The sub-types of a type t all subsume the multi-AVS that characterizes t.
As in all knowledge representation frameworks, we assume that the concepts are related to each
other by more complex relations than subsumption: an attribute in a concept may be a relational
concept itself, and the arguments and values in a multi-AVS may be restricted to certain types of
entities, i.e., we use typed multi-AVSs. As far as we can see, typing does not crucially affect the
complexity of AVS operations.
Typing basically means that we enrich the multi-AVS representation with an extra component
which enables us to have type statements. A type statement can be thought of as conveying the
assertion that a certain n-tuple of terms is an instance of a type and is expressed by assigning typenames to n-tuples of terms. These types (concepts) are charaterized by multi-AVSs themselves. So the
226
Representation of Linguistic Knowledge in GIN 99
set Type of types is in fact a set of names of multi-AVSs. Obviously typing is reasonable only in the
context of a set of multi-AVSs, which by way of referring to each other in the form of type statements
and at the same time acting as types themselves constitute a network. Therefore we can refer to
multi-AVSs representing a concept in a network as network nodes. Moreover, in order to allow
relational concepts (n-ary types), we have to modify the definition of multi-AVSs so that they can
have multiple roots. Now we give the final formal definition of multi-AVSs with multiple roots.
(24) Multi-AVSs with multiple roots
The structure M = <R, E> is a multi-AVS with multiple roots if and only if
(i) R = <q11,...,q12> are terms (the roots of M); {q1, ...,q2} ∈ Nodes(M); (the roots are nodes of the
multi-AVS);
(ii) E is a set of equations such that, if e ∈ E, then e has the form f(t1,..., tn) = v, where f is an n–
argument function (f ∈ F), and t1,..., tn, v are terms. We say that t1,...,tn v ∈ Nodes(M) in this case;
(iii) every node n ∈ Nodes(M) \ {q1,….,qn} is accessible to a root node qi (1 ≤ i ≤ n) in the sense of
(21).
A multi-AVS of this sort is said an n-ary multi-AVS, which means that it has n root nodes.
(25) Equivalence of multi-AVSs with multiple roots
The multi-AVSs M1 = <R1, E1> and M2 = <R2,E2> are equivalent (written M1 ≡ A1) iff R1 =
<q11,….q1n> and R2 = <q21,….,q2n> (the two multi-AVSs have the same number of root nodes), and
there is a bijection β : Nodes(M1) -> Nodes(M2) such that
(i) For every 1 ≤ i ≤ a, β(q1i) = q2i;
(ii) For every equation f(n11,…n1m) = n′1 ∈ E1 and f(n21,…n2m) = n′2 ∈ E2 , if β(n11) = n21,...,β(n1m) =
n2m ,…, β(n1m) then β(n′1) = n′2
(iii) If C1 is the set of constants in Nodes(M1), and C2 is the set of constants in Nodes(M2),
then β is an identity bijection w.r.t. C1 and C2.
(26) Unification of multi-AVSs with multiple roots
The unification of the AVSs M1 = <R1,E1> and M2. = <R2,E2> (written M1 ∪ A2) is an AVS M =
<R,E> such that there are mappings µ1 : Nodes(M1) -> Nodes(M) and µ2 :
Nodes(M2) -> Nodes(M) satisfying the following conditions:
(i) R1 = <q11 ,…, q1n> R2 = <q21,…,q2n>, and R = <q1,…,qn> and (i.e., the number of the
root nodes of the three multi-AVSs is the same);
(ii) For every 1 ≤ i ≤ n, µ1(q1i) = µ2(q2i) = qi ;
(iii) if µ1(n11) = µ2(n21) = n1,…, µ1(n1m) = µ2(q2i) = qi and there are equations f(n11,…,n1m) = n′1 ∈
E1 and f(n21,…,n2m) = n′2 ∈ E2, then
(i) µ1(n′1) = µ2(n′2) = n′, and
(ii) f(n1,…,nm) = n′∈ E;
(iv) If C1 is the set of constants in Nodes(M1), and C2 is the set of constants in Nodes(M2), then µ1 and
µ2 are identity mappings w.r.t. C1 and C2.
Representation of Linguistic Knowledge in GIN 99
227
(v) The only nodes and equations of M are the ones defined by (i-iv) above. If the mappings µ1, µ2 do
not exist, or A1 and A2, have a different number of root nodes, then the unification is said to fail
(yield a failure, written M1 ∪ M2 = ⊥).
Finally, typing a multi-AVS consists in associating types (i.e., multi-AVSs) with n-tuples of
nodes in the AVS. By a type hierarchy we mean a set of multi-AVSs which are typed and may
function as types themselves. Of course, only those AVSs can be used for typing that are members of
the hierarchy. The set Type of names of multiple-rooted multi-AVSs is ordered with a relation that is
a subset of the subsumption relation. We refer to this order as domination.
(27) Type hierarchy of multi-AVSs
A type hierarchy M is an ordered pair <M,T>, where M is a set of multi-AVSs, and T is the typing
relation, which means that, if <<n1M,...,nnM>>, Tn> ∈ T, then {n1M,…,nnM} ⊆
Nodes(M) for some M ∈ M. and Tn ∈ Type is the name of an n-ary multi-AVS Mn ∈ M.
For example/ we can use a sub-hierarchy in a type hierarchy to encode the DEF-AGR relation,
mentioned earlier, then use the typing relation in the network instead of relational statements in order
to guarantee definiteness agreement:
(28) Sub-hierarchy for DEF-AGR
This is a flat hierarchy with five multi-AVSs, which contain no equations (the second element
of the ordered pair is the emptyset), and their root nodes are constants (namely, those that qualify as
agreeing values in terms of DEF-AGR). These are to be the sub-types of DEF-AGR for this analysis
to work. Then we could repeat the multi-AVSs in (22) and (23) without the relational statements
there, and stipulate (using the typing relation of the type hierarchy) that the relevant pairs of nodes
belong to the type DEF-AGR.
Answering a query in GIN is effectively a type resolution problem since each concept is a
type. Then the question is whether the input formula can be seen as (the partial description of) an
instance of a set of types.
The type resolution process
The type hierarchy is interpreted in such a way that every concept refers to the disjunction of
the concepts that it dominates. Proving that a representation belongs to a certain type implies proving
that it belongs to some sub-type of it. This proof consists in unifying the root nodes of the concept in
question with the respective arguments of a relational type statement.
During parsing, given the normal procedure of type resolution above, the relational statement
referring to the binary relational concept DEF-AGR in our earlier example should be regarded as a
step of type-proving.
Observe the following sentence:
(29) a könyv, amelyiket ?? olvastál
/ ?? olvastad
/ the book which read-2SG-INDEF-OBJ / read-2SG-DEF-OBJ /
OK olvastam
228
Representation of Linguistic Knowledge in GIN 99
read-1SG-UNMARKED-DEF-OBJ
'The book that you/I read'
Amelyik 'which' is a relative pronoun which, when in the accusative case, can agree with verb
forms unmarked for definiteness agreement. In some dialects it is ungrammatical with either marked
definite or marked indefinite verb forms. This can be captured if we assign it the 0 ('unmarked') value
for definiteness. As it refers to a relational concept, the agreement statement induces a step in the
resolution process: a subtype has to be instantiated. Given the sub-hierarchy of DEF-AGR, however,
there is no appropriate subtype to instantiate if the verb is definite or indefinite, i.e., <<0, +>, Ø> and
<<0, ->, Ø> are not instances of the DEF-AGR relation.
In actual fact, it is not always necessary to go all the way in such a type proof. In particular, if
no contradiction may arise during the proof, then the proof can be considered successful. A relational
statement may lead to contradiction if and only if the relation-name in question refers to a concept and
all the arguments of the relation are variable leaves of the multi-AVS. We give the formal definition
below:
(30) Let M = <<q1 ,..., qn> ,E>bean n-rooted multi-AVS. Then a. Leaves
Leaves(M) is the set of leaves of M iff
Leaves(M) = def {t ∈ Nodes(M)|Acc(t, E) = 0}
b. Potential contradiction
Let R(x1,... ,xn) ∈ {T, q, E} be a relational statement in a multi-AVS (M), where R refers to a concept
(T(M) = R for some M). This statement may NOT lead to a contradiction iff
(i) {x1,..., xn} ⊆ Leaves (M) and
(ii) {x1,,..., xn} ∩ Constants (M) = 0.
The following example illustrates the point. As said above there is some hesitation between the
definite and indefinite agreement on the verb the object is the specific relative pronoun amelyiket. For
some speakers, in these cases either the definite or the indefinite agreement marking on the verb is
well-formed.
If we choose to represent the lexical entries for olvastam ('I read') as well as amelyiket ('which')
with underspecified values for definiteness, then the following representation arise for the VP (we
think of the relative pronoun as a normal object here):
(31)
229
Representation of Linguistic Knowledge in GIN 99
The relational type expression in the AVS would normally induce a step in the type proof,
which comprises decsending to some most specific subtype of the relational type DEF-AGR. Given
the above definition of lazy evaluation, however, this step can be omitted due to the fact that both of
the arguments of relation DEF-AGR are variables and leaves of the multi-AVS and therefore
instantiation of DEF-AGR can never yield a contradiction.
References
Copestake, Ann, Dan Flickinger, and Ivan A. Sag. 1996? Minimal recursion semantics — an introduction.
CSLI, Stanford University, Stanford, CA 94305, USA.
Fillmore, Ch.J. and P. Kay. 1993. Construction Grammar Coursebook. Reading materials for Linguistics X20,
University of California, Berkeley CA.
Goldberg, A.E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. The
University of Chicago Press, Chicago.
Jackendoff, Ray. 1997. The architecture of the language faculty. Linguistic Inquiry monographs. MIT Press.
King,Paul. 1989. A Logical Formalism for Head-Driven Phrase Structure Grammar. Ph.D. thesis, University of
Manchester.
King, Paul. 1994. An expanded logical formalism for Head-Driven Phrase Structure Grammar. Arbeitspapiere
des sfb 340, University of Tubingen.
Pollard, Carl J. and Ivan A. Sag. 1987. Information-based Syntax and Semantics, Vol. 1. Number 13 in CSLI
Lecture Notes. CSLI Publications, Stanford University. Distributed by University of Chicago Press.
Pollard, Carl J. and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press,
Chicago.
Is Agreement Value Sharing?
230
Is Agreement Value Sharing?
László Kálmán and Viktor Tron
Abstract
The goal of this paper is twofold. On the one hand, we aim at providing evidence that
traditional approaches to representing morphosyntactic information in constraint-based grammars is
on the wrong track. Focussing on very general phenomena such as multiple agreement with
morphologically neutral elements as well as non-category co-ordination, we propose that the use of
abstract features, unification and type resolution on sorted attribute/value structures makes it rather
difficult if not impossible to handle these problems. On the other hand, we hope to illustrate that, by
abandoning attribute/value pairs in the representation of morphosyntactic information, such phenomena can be given an adequate and also formally elegant treatment. The essence of our proposal is
to replace attribute/value pairs with properties represented by types in the type hierarchy. As a
consequence, agreement and similar congruence relations must be modelled by relations over classes
of expressions instead of value sharing. To do this, we have to extend the concept of type hierarchies
with relational types. Finally, we will argue against the strict locality of grammatical relations and
show how the type of analysis that we propose can be applied to the description of certain non-local
dependencies. Our approach makes it possible to keep the grammar purely declarative while retaining
the central role of type resolution as the only constraint on grammati-cality.
Problems with unification
In the first two subsections, we will outline two well-known arrays of data, one related to
agreement, and the other to subcategorization. Following Ingria (1990), we will use these data to point
at the inadequacy of a treatment based on the unification of feature values. In the end of this section,
we will sketch certain possible solutions to the problems presented.
Agreement
Consider the following Hungarian examples:
(1) Definiteness agreement in Hungarian
a. A fát
*látok/ok látom.
the tree-ACC see-lSG-lNDEFOBJ/see-lSc-DEFOBJ 'I can see the tree.'
b. Egy fát
ok látok/ *látom.
a tree-ACC see-lSG-lNDEFOBJ/see-lSG-DEFOBJ 'I can see a tree'
As can be seen, the finite verb agrees with the direct object in definiteness.52 There are some
cases, however, when the definite-object form and the indefinite-object form coincide, namely, in the
first person singular form of the past and the first person plural form of the conditional. We will call
such forms neutral:
(2) Neutral forms in Hungarian a. láttam
see-PAST-lSG-DEFOBJ/lNDEFOBJ
'I saw'
b. látnánk
52
ln sentences without a direct object, the INDEFOBJ forms are used unless there is an implicit definite direct
object singular pronoun, which can be dropped in Hungarian.
Is Agreement Value Sharing?
231
See-COND-1pl-defobj/ indefobj
'we would see'
Such neutral forms can co-occur with either definite or indefinite direct objects:
(3) Behaviour of neutral forms in Hungarian
A / egy fát
láttam the/a tree-ACC see-PAST-lSG-DEFOBJ/lNDEFOBJ
'I saw the/a tree'
Interestingly, the neutral forms - but not the others - can co-occur with co-ordinated direct
objects in which the definiteness of the members is not uniform:
(4) Definiteness agreement with co-ordinated direct objects in Hungarian
a. *Látom /*Látok
a fát
és egy madarat. see-lSc-DEFOBJ/see-lSG-lNDEFOBJ the treeACC and a bird-ACC Intended reading: 'I see the tree and a bird'
b. ok Láttam
ACC and a bird-ACC
a fát
és egy madarat. see-PAST-lSG-DEFOBJ/lNDEFOBJ the tree-
'I saw the tree and a bird' In general, we can make the following observation:
(5) Mixed agreement
Items that are morphologically neutral with respect to definiteness can (possibly simultaneously) agree with either definite or indefinite direct objects.
It would seem natural to assume that lexical entries of items that are neutral in terms of
definiteness do not specify the value of their definiteness attribute, i.e., they do not contain either
'[DEFINITE +]' or '[DEFINITE -]' in their representations. The underspecification of such feature values
would make it possible for the agreement values to unify with either definite or indefinite values. This
would explain why neutral verbs can agree with both definite and indefinite NPs (cf. the example (3)
above).
As we have seen, however, items that are morphologically neutral with respect to def-initeness,
can agree with items of potentially conflicting definiteness simultaneously if those items constitute a
co-ordinate structure. If this was an accurate description of the situtation, then we would have to
recognize the existence of direct objects with 'mixed definiteness'. Usual treatments of co-ordination,
however, rule out such heterogeneous co-ordinate structures, given that the conflicting definiteness
values of the conjuncts (+ and -, respectively) would block unification (this was first pointed out by
Ingria (1990)).
We note here that morphologically neutral forms abound in the world's languages and - perhaps
surprisingly - universally conform to the pattern shown above, i.e., neutral elements happily enter into
mixed agreement relationships:
(6) Ingria's generalization
Forms that are neutral with respect to a certain feature enter into congruence relations, either
separately or simultaneously, with entities of arbitrary values for that feature.
We give here two examples from English to show that the phenomenon is not limited to
agreement proper. We refer to Ingria (1990) for further examples.
(7) Number agreement in English
a. Why did John leave and the others stay?
b. *Why does/do John leave and the others stay?
Is Agreement Value Sharing?
232
(8) Case mismatch in multiple gaps in English (Grover, 1996)
a. Who did you say John's criticism of___í(ACC) would make us think____Í(NOM) was stupid?
b. Which Caesar did Brutus imply [___í(NOM) was no good] while ostensibly praising____i(ACC)7
Subcategorization
Another well-known phenomenon that is problematic for unification-based treatments is related
to subcategorization and co-ordination. Co-ordination in itself does not require the identity of the
values of categorial features. Predicates imposing sufficiently under-specified subcategorization
restrictions on their complements can occur with categorially heterogeneous co-ordinate structures:
(9) Co-ordination of mixed categories (Sag et al./1985)
a. That was [NP a rude remark] and [PP in a very bad taste].
b. Pam is [NP a republican] and [AP proud of it].
c. Pam is [AP healthy] and [PP of sound mind].
We emphasize that this phenomenon is very general in all sorts of head-modifier and headcompelment relations universally in the world's languages (e.g., German (Zaenen and Karttunen,
1984) and French (Pullum and Zwicky, 1986)).
We can make the following empirical observation:
(10) Expression of grammatical functions
If a grammatical function related to a head can be fulfilled by expressions of various categories,
then that head will allow their simultaneous appearance in a co-ordinate structure.
Clearly, this phenomenon raises a problem for the unification-based treatment of co-ordinate
structures: the problem is analogous to that raised by mixed agreement (cf. (5)), namely, the values of
the category attributes in the 'non-category co-ordinations' predicted by the observation above cannot
be unified with each other.
Generalization and type resolution
We have mentioned above that the usual treatments of co-ordination (in which the morphosyntactic features of the conjuncts are unified) is incompatible with the existence of heterogeneous
co-ordinate structures. There is, however, another possibility, namely, assuming that the value of the
categorial attributes of co-ordinated phrases are to be calculated using the generalisation rather than
the unification of those of their conjuncts53? This was first proposed by Sag et al. (1985); then Pollard
and Sag (1994, pp. 203) made it part of (the weak version of) the so-called Co-ordination Principle
within HPSG:
(11) Co-ordination Principle
In a co-ordinate structure, the CATEGORY and NONLOCAL value of each conjunct
daughter is subsumed by that of the mother.
As a matter of course, this presupposes that the values of the attributes in question (i.e.,
and NONLOCAL) are partially ordered with respect to specificity (subsump-tion). This can
be done, for example, by refining the internal structure of the main types of part of speech in the way
proposed by Jackendoff (1972), i.e., by decomposing category types into the features N and V:
CATEGORY
53
As a matter of fact, it is sufficient to require the value of the co-ordinate structure to subsume (be more
general than) the respective values of the conjuncts. See also Shieber (1992, pp. 132-134) for a discussion.
233
Is Agreement Value Sharing?
(12) Jackendoff's category definitions
adjective
noun
verb
preposition
Then one can represent the generalisation of certain categories using underspecification. For
example, the class of expressions containing just nouns and adjectives can be categorized as '[N +]/;
the class containing noun phrases and prepositional phrases is represented by '[v —]'. Consequently,
the subcategorization requirements of predicates can be under-specified in the same way.
The specificity ordering of the values of attributes can also be achieved by putting them into a
type hierarchy. For example, in the case of definiteness:
(13) Hierarchy of definiteness values
If we were to opt for this approach, then the value of the definiteness attribute of heterogeneous
co-ordinate noun phrases would be definiteness.
However, this simple and seemingly appealing solution turns out to be problematic from
another point of view: it is incompatible with the role that type resolution plays in those theories
assuming type hierarchies. Type resolution means that linguistic entities are to be modelled using
strictly typed feature structures: if a certain attribute is defined for a linguistic entity, that attribute
must be assigned a fully specified value for the entity in question. Underspecification only exists in
the language of linguistic description; the entities described, however, cannot be partial54. The
subsumption requirement in the Co-ordination Principle together with total type resolution exclude
heterogeneous co-ordinate structures, since the co-ordinated phrase should be unspecified for certain
attributes (e.g., if the definiteness of a co-ordinated NP is the generalization of '[DEFINITE +]' and
'[DEFINITE -]', then it must be unspecified for the attribute DEFINITE, i.e., it should be a 'partial object').
The only alternative (using the example of definiteness agreement again) is to use a three-valued
attribute, and represent mixed co-ordinate structures with the third value (i.e., neither + nor -). But
54
For a discussion we refer the reader to Johnson (1988), Shieber (1992) and King (1994; 1989
Is Agreement Value Sharing?
234
then agreement cannot be characterized in terms of unification, because neutral forms can agree with
expressions of all three sorts (definite, indefinite and mixed), but no fully specified value can possibly
be unified with all three values.
If, on the other hand, we were to abandon strict type resolution, we should find another means
to exclude agrammatical structures. The only systematic way of doing that in the literature is to use a
rewrite rule system which, however, has been convincingly criticized over the last few decades. Yet
another possibility would be to lift the constraint of strict type resolution in certain particular cases
(e.g., for the resolution of the value of the attribute DEFINITE in the case of co-ordinate structures); this
could only take the form of arbitrary meta-level constraints, which would be undesirable.55
Moreover, the underspecification of the morphosyntactic features of co-ordinate NPs would not
in itself account for the whole range of phenomena. As pointed out by Ingria (1990), the seemingly
universal phenomenon mentioned in (6) extends well beyond the domain of co-ordination. Consider:
(14) German headed/headless relatives
a. *Wer/*wen
du mir vergestellt hast, ging nach Hause.
who-NOM/who-ACC you to-me introduced went home
Intended reading: 'The person whom you introduced to me went home'
b. Was
du mir gegeben hast, ist prächtig.
what-NOM/ACC you to-me gave
is splendid
'What you gave me is splendid.'
Here the main predicate of the second sentence (ist prächtig 'is splendid') expects an NP in the
nominative, whereas the relative clause (du mir gegeben hast 'you gave me') expects one in the
accusative. Was 'what-NOM/ACC' meets these requirements at the same time because it is
morphologic ally neutral in terms of case. However, the predicate of the main clause and the relative
clause do not stand in any syntactic relationship whatsoever with each other. Therefore, there is no
way to postulate a syntactic constituent in which some feature values could be calculated using
generalization.
Goldberg and Kálmán (1992) argue that the generalization in (6) can be captured by using the
relation of unifiability instead of the operation of unification. Even though this proposal solves the
problem of mixed agreement even in cases not involving co-ordination, it crucially relies on the
underspecified nature of certain linguistic entities, namely, neutralized forms, therefore, it also runs
counter total type resolution.
Let us conclude by stating the most important point made in this section:
(15) Main point of this section
The congruence relations between morpho-syntactic classes cannot be captured in terms of
unification or unifiability.
Attribute/value pairs as properties
Agreement-like relations have been and still are the principal motivation for the unification of
atomic feature values. (All other instances of unification express the simultaneous satisfaction of
constraints on phrase types, i.e., they prescribe the unification of complex structures.) However, as we
have seen, the unification operation is not an appropriate device for the description of agreement
phenomena in general. Therefore, the question arises whether attribute/value descriptions such as
'[DEFINITE +]' or '[DEFINITE —]' are necessary at all. If we considered definiteness and indefiniteness
55
Pollard and Sag (1994, p. 203, fn.39.) actually admit this deficiency
235
Is Agreement Value Sharing?
as simple (unanalysed) properties instead, then the phrase classes mentioned in connection with the
Hungarian agreement phenomena could be characterized as follows:
(16) Hungarian definiteness classes
A definites;
B indefinites;
C neutral = A ∩ B.
This would explain in a straightforward manner why a neutral verb can agree with definites and
indefinites alike: it is a member of both classes. Thus, instead of underspecifi-cation, the double
nature of neutral verbs is a result of their multiple class membership. Otherwise, they are first-class
citizens just like definites and indefinites.
This formulation is clearly simpler than using multi-valued features. Our move is similar to the
one made by many phonologists who have abandoned multi-valued attributes in favour of so-called
unary features (the first such moves were made as early as Done-gan (1973) and Schane (1984)). In
formal terms, such class labels correspond to properties (i.e., classes of entities). So, in our analysis,
'definite' and 'indefinite' will be understood as two independent properties of some linguistic entities
in Hungarian. True, these two concepts are somehow related (they characterize words and phrases in
the same 'dimension'), which the attribute/value notation makes explicit, whereas our class labels do
not in themselves. However, it is easy to express the relatedness of the two classes by making them
subclasses of the same large class which we might call 'DEF-CLASS' (representing the dimension itself,
i.e., the class objects for which definiteness is defined). That is, if we assume a framework using a
type hierarchy, the properties can be readily represented as types.
As a matter of course, there is no difference in principle between properties of entities and
properties of n-tuples of entities, i.e., relations. Thus, agreement-like relations can be seen as
properties (i.e., classes) of pairs of linguistic objects. For example, the simplest cases of definiteness
agreement in Hungarian are the following:
(17) Base cases of definiteness agreement in Hungarian
a. definite NP - definite-object form of the verb
b. indefinite NP - indefinite-object form of the verb
There is no need to mention 'neutral' verbs here, since their ability to agree with definites and
indefinites alike follows from their double class membership.
Since relations are properties themselves, they can be included in the specificity ordering, i.e.,
in the type hierarchy. For example, the two base cases above form sub-classes of a more general
relation in the type hierarchy, which we will call 'DEF-AGR-REL' (definiteness agreement relation) in
what follows.
Grammatical relations and co-ordinate structures
As we have shown above, the definiteness agreement of Hungarian verbs with their direct
objects cannot be treated using the unification of the values of the attribute DEFINITE if we are to
retain the role of type resolution. In what follows, we will argue that grammatical relations in general
can be represented as relations in formal terms. Moreover, it is even necessary to do so if we want to
keep our underlying ontology as concise as possible. This means that we can dispense with morphosyntactic features like definiteness in general, and consider grammatical congruence relations what
they actually are: relations between classes of expressions. That is, we assume that the principles and
tools used in our analysis can be extended to other grammatical relations, also we keep using the
example of Hungarian definiteness agreement.
Is Agreement Value Sharing?
236
In the first subsection, we point at the immediate advantages of the class-based analysis of
agreement. In the second subsection, we will examine how co-ordinate structures must be
characterized in terms of definiteness under the analysis proposed. We will see that we encounter
certain problems if we just reproduce an attribute/value-based analysis. In the last subsection, we offer
a simple solution to them.
Class-based analysis of agreement
The natural division of Hungarian verb forms and noun phrases in terms of definiteness can
easily be encoded in a type hierarchy:56
(18) Natural definiteness classes in Hungarian
As we have mentioned earlier, the neutral forms can be characterized as sub-classes that belong
to several class at the same time. This is called multiple inheritance in the language of type
hierarchies:
(19) Neutral forms and multiple inheritance (for verb forms)
The natural classes of agreement relations can be given in a similarly simple form:
(20) Base cases of definiteness agreement in Hungarian
If a certain structure requires two agreement relations for one and the same expression, then
they can be resolved as two different sub-cases of agreement in terms of type resolution. For example,
in the case of mixed agreement, the neutral form stands in different instances of the agreement
relation with two expressions belonging to two different classes. As opposed to unification-based
approaches, this does not lead to conflicts because the fact that a relation holds between two entities
does not bring about a 'mutation' in the internal structure of the entities in question. As a consequence,
the solution sketched here immediately explains Ingria's (1990) generalization (see (6)). However, it
56
As a matter of course, we have to neglect here the semantic correlates of definiteness and indefiniteness.
Maybe these concepts are somehow semantically grounded, but we would not be surprised if their formal
classification would not fit their semantic behaviour.
237
Is Agreement Value Sharing?
remains to be explained how the general situation described as mixed agreement arises in the case of
co-ordinate structures.
3.2 Definiteness in co-ordinate NPs
In unification-based grammars, morpho-syntactic attributes and the unification of their values
plays a very important role for calculating the properties of complex expressions from the properties
of their parts. (The simplest case is the so-called percolation of attribute values.) This mechanism
serves the purpose of expressing grammatical relations in a strictly local manner. Therefore, theories
that admit can be called strictly localist. Such theories are bound to associate co-ordinate structures
with the same kinds of characterizations as simple expressions (e.g., in terms of definiteness). This,
however, as we have demonstrated, is incompatible with using unification to account for agreement.
In the framework proposed here, it would be possible to describe the complex conditions that
determine the agreement properties of the various types of co-ordinate structures (homogeneous and
heterogeneous) since, as should be clear from the above, practically any relation between expression
classes can be expressed in our formalism. However, consider the three types of co-ordinate NPs that
we have seen in the definiteness agreement phenomenon examined (i.e., definite, indefinite and
mixed). Already the ways in which these can arise (i.e., in what ways their members determine their
properties) yield a rather complex description:
(21) Definiteness classes of Hungarian NPs in a strictly localist spirit
Moreover, as we have mentioned earlier, agreement cannot be described in terms of simple
unification in this framework. Instead, we have to explicitly list the instances of relations between NP
classes and the verb forms that agree with them as sub-classes of the agreement relation:
(22) Sub-cases of agreement in a strictly localist spirit
The obvious complexity of the above solution clearly raises the question whether we gain
anything from abandoning morpho-syntactic features if the analysis itself does not become simpler.
Our answer to this is yes, we gain a lot, provided that we also abandon strict localism. This is what the
next subsection is about.
Recursive definition of grammatical relations
In this subsection, we show that the phenomenon of mixed agreement can be analysed without
the proliferation of sub-cases in our classifications that we have seen in the previous subsection by
simply abandoning strict localism. For example, in the case of Hungarian definiteness agreement, we
achieve this by defining the definiteness classes for simplex NPs only, then reducing the agreement
Is Agreement Value Sharing?
238
properties of co-ordinate NPs to those basic cases through postulating agreement with their members.
The idea is the following: we define agreement in such a way that one of its sub-cases is agreement
with co-ordinate NPs. What we say about this sub-case is that the relation of agreement holds between
the co-ordinate NP and the verb form if it holds between the members of the co-ordination and the
verb separately. The fact that agreement has a sub-case that makes direct reference to the constituents
of the co-ordinate NP contradicts strict localism.
In formal terms this means that we define agreement recursively. The recursive step is that the
agreement of a co-ordinate NP is derived from the simpler case, namely, agreement with its members.
The definition itself could be written as follows:
(23) Recursive definition of Hungarian definiteness agreement def-agr-rel (x.y) - def
(i) def-NP(x) A def-V(y);
(ii) indef-NP(x) A indef-V (y); vagy
(iii) def-agr-rel(CONJ1 (x), y) A def-agr-rel(coNJ2(x), y).
Here the values of the attributes CONJ1 and CONJ2 are the first and the second
conjunct, respectively.
This definition can be encoded in the type hierarchy as follows:
(24) Hungarian definiteness agreement in the type hierarchy
The above analysis relies on natural definiteness classes only, it does not postulate artificial,
'mixed' type expressions. The particular behaviour of co-ordinate structures is solely due to the special
sub-case that they correspond to (which contains the recursive step). That is, we do not have expand
the description of co-ordinate structures themselves with information on how they behave in terms of
agreement.
In this way, agreeing with a co-ordinate structure is reduced to the multiple agreement of one
entity (with the members of the co-ordinate structure). As a consequence, whatever was said about
mixed agreement (cf. Section 3.1) is automatically true for it. So the analysis correctly predicts that
Ingria's observation applies to this case without any further stipulation.
Summary and further perspectives
In what follows, we first summarize the most important statements made in this paper, then
briefly introduce certain possible extensions of the analysis proposed here to slightly different
phenomena.
Main statements
1. Congruence classes are to be defined for simplex (not co-ordinate) structures only.
Is Agreement Value Sharing?
239
2. Congruence classes divide the class of expressions into sub-classes, but the neutral and
mixed expressions do not constitute separate sub-classes.
3. There is no need for attribute/value pairs in which the value is atomic for describing
agreement. Instead, we just need a representation of properties (classes), which is given in the
form of the type hierarchy that we assume anyway.
4. Neutral forms can be characterized as belonging to several classes at the same time;
they are shared sub-classes of several natural classes.
5. Co-ordinate structures do not belong to any natural class.
6. Agreement with a co-ordinate NP can be reduced to agreement with its members. This is a
particular sub-type of agreement, so the agreement relation can be defined through recursion on
co-ordination (although this contradicts strict localism).
7. In order to implement this relational view of agreement, we only need to extend the concept
of the type hierarchy to properties (classes) of n-tuples of entities, i.e., n-ary relations. Those
classes are to be nodes in the type hierarchy just like unary properties are.
Further perspectives
It is an interesting issue whether the analysis that we illustrated with Hungarian definite
agreement can be extended to number agreement facts. The most important data are the following:
(25) Number agreement in Hungarian
a. János és Kati elmentek
moziba.
John and Kate AWAY-go-PAST-3PL movie-lNTO
'John and Kate went to the movies'
b. János és Kati elment
moziba.
AWAY-go-PAST-3SG
‘Id.’
c. ?Te és te kimehettek.
you and you OUT-go-MAY-2PL
'You and you may go out'
d. Te és te kimehetsz.
you and you OUT-go-MAY-2Sc
‘Id.’
It is a particularity of Hungarian that co-ordinated subjects consisting of singular NPs can cooccur with singular verb forms. To describe this phenomenon, we can use the recursive type of
agreement that we have seen in connection with definiteness agreement. In those languages, however,
which do not exhibit this phenomenon (and we believe most languages are like that) co-ordinate NPs
can be characterized in terms of number (namely, they are plural NPs), probably because this property
is very closely related to semantics. The fact that Hungarian exhibits two distinct patterns of
agreement may be related to the fact that the morphological singular in general has less semantic
motivation, consider:
Is Agreement Value Sharing?
240
(26) Morphologically singular - semantically plural
a. három alma
b.
mossa
a kezét
three apple
wash-3SG-DEFOBJ the hand-His/HER-ACC
'three apples'
'[(s)he] is washing his hands'
c. ceruzát
vettem
pencil-ACC buy-PAST-1SG-lNDEFOBJ
'I bought some (one or more) pencils'
We can think of two types of analyses: The first is to say that Hungarian co-ordinate NPs, too,
can be characterized in terms of number (namely they are plural), therefore, the basic case of plural
agreement can be applied to them, but the recursive case applies as well, so the verb form may agree
with the members separately. The second option is that coordinate NPs are not characterized in terms
of grammatical number, but there is a special sub-case of subject-predicate agreement that specifies
that they may co-occur with a plural verb form. The latter option would be corroborated by the fact
that the order of the subject and the predicate matters:
(27) Order dependency of Hungarian number agreement
a. *Elmentek
János és Kati moziba.
AWAY-go-PAST-3PL John and Kate movie-lNTO
Intended reading: 'John and Kate went to the movies'
b. OK Elment
János és Kati moziba.
AWAY-go-PAST-3SG
'Id.'
Since we have abandoned strict localism, the analysis proposed here can be used to describe
phenomena that are not related to either mixed agreement or co-ordination. To quote one, consider the
phenomenon called 'raising':
(28) Raising in Hungarian
a. Pista meg akarja
látogatni a nagymamáját.
Steve PREF want-3SG-DEFOBJ visit-INF the grandma-HIS/HER-ACC
'Steve wants to visit his grandma'
b. Pista meg akar
látogatni egy lányt.
want-3SG-lNDEFOBJ
a girl-ACC
'Steve wants to visit a girl'
Note that the direct object NPs a nagymamáját 'his grandma' and egy lányt 'a girl' are directly
subcategorized by the verb meglátogatni 'visit', and the verbal prefix meg is lexically part of the verb
meglátogatni 'visit'. Nevertheless, both the position of the verbal prefix (it forms a prosodic
constituent with the main verb: meg akarja is pronounced as a single prefixed verb) and the
definiteness agreement (which appears morphologically on the verb akar 'want') show that the main
verb treats the complements of its infinitival argument as if they were its own: it 'raises' them. One
way of treating this is to assume that the class of relations 'verbal prefix-verb', 'verb-direct object' etc.
241
Is Agreement Value Sharing?
have special sub-classes characterized by the following: the verb is a 'raising verb', and the
complement stands in the given relation with the infinitive argument of the raising verb:
(29) Recursive definition of verb-complement relations comp-rel(x,y) — def
(i) verb(x)AREL(x,y);
(ii) raising-verb(x) ∧ comp-rel(lNF(x),y).
Finally, it is an open question what natural restrictions apply to the expressions which
potentially stand in a grammatical relation. In our view, abandoning strict localism should not mean
that any two expressions whatsoever can stand in any grammatical relation. For example, it seems
natural to assume that the expressions A and B can be grammatically related only if there is an
expression C such that B is a part of C, and the given relation holds locally between A and C. The
question that remains to be answered, then, is what exactly we should mean by 'a part of in this
formulation.
References
Donegan, P. J. 1973. Bleaching and coloring. CLS, 9:386-397.
Goldberg, J. and L. Kálmán. 1992. Unification or unifiability? In Proceedings of CLIN '92, Amsterdam. CLIN.
Ingria, R. J. P. 1990. The limits of unification. In 28th Annual Meeting of ACL: Proceedings of the Conference.
ACL, Morristown, NJ, pages 194-204.
Jackendoff, R. 1972. Semantic Interpretation in Generative Grammar. MIT Press, Cambridge MA.
Johnson, M. 1988. Attribute-Value Logic and the Theory of Grammar. CSLI, Stanford CA.
King, P. 1989. A logical formalism for Head-Driven Phrase Structure Grammar. Ph.D. thesis, University of
Manchester, Manchester.
King, P. 1994. An expanded logical formalism for Head-Driven Phrase Structure Grammar. Arbeitspapiere des
sfb, 340, Universitát Tübingen, Tübingen.
Pollard, C. J. and I. A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press,
Chicago IL.
Pullum, G. K. and A. M. Zwicky. 1986. Phonological resolution of syntactic feature conflict. Language, 62:751773.
Sag, I. A., G. Gazdar, T. Wasow, and S. Weisler. 1985. Coordination and how to distinguish categories. Natural
Language and Linguistic Theory, 3:34-43.
Schane, S. A. 1984. The fundamentals of particle phonology. Phonology Yearbook, 1:129155.
Shieber, S. 1992. An Introduction to Unification-based Approaches to Grammar. CSLI, Stanford CA.
Zaenen, A. and L. Karttunen. 1984. Morphological non-distinctiveness of coordination. In Eastern States
Conference on Linguistics 1, Columbia, Ohio. Ohio State University.
Constructional CV phonology
242
Constructional CV phonology
Péter Rebrus and Viktor Trón
Abstract
The present study is an attempt to develop a framework for the analysis of (morpho)phonological phenomena, especially in Hungarian. First we will show in what ways theoretical
assumptions of Government Phonology and (strict) CV Phonology can be reformulated and integrated
into a constraint-based lexicalist conception of grammar. The sufficiently abstract but simple notation
that encode phonological representations in CV phonology lends itself to an appealing formalization
in a strictly declarative constraint-based setting in the form of what we call phonological
constructions. We demonstrate how our framework called Constructional CV Phonology fits into the
general framework of constraint-based grammars advocating a hierarchical lexicon. In this spirit we
provide a novel account of a number of phenomena of Hungarian morphophonology. It is assumed
that particular lexical classes can arbitrarily restrict the range of licensing constructions available for
parsing phonological structures. The main claim of the paper is that selective reference to certain
constructions and their preference over others on the part of a lexical class can be responsible for
morphology-driven but phonologically relevant regularities. The novelty of the approach is that the
above conception allows for a unified treatment of phonotactic generalizations, suffixation patterns,
irregularities in the paradigmatic behaviour and exceptional stem alternations alike.
Keywords:
constructions
morphophonology,
Hungarian,
CV-phonology,
constraint-based
grammar,
Construction phonology
CV Phonology in a constraint-based setting
Strict CV Phonology (Lowenstamm, 1996), as opposed to other prosodic theories, utilizes flat
phonological representations without evoking constituency. In this theory prosodic representations
contain exclusively CV sequences. The phonological stucture involves licensing and government.
These are strictly local and directed relations constrained by several conditions. The Licensing
Principle states that every position in a phonological domain should be licensed. Licensing
Inheritence allows a position to transmit its licensing potentials to another position. (The concepts of
licensing and government come from the Government Phonology tradition, see Kaye et al., 1990;
Harris, 1990, among others).
Phonological processes are restricted in the sense of not allowing destructive (non-monotonic)
operations. Phonological alternations can only be expressed by the interpretational difference of
representations, i.e., there are no phonological derivations in the traditional sense. The Projection
Principle says that licensing relations (which correspond to 'syllable structure') are given in the
lexicon, and remain constant (this conception is to be slightly modified here).
Attempts to adapt CV Phonology to Hungarian data are found in several papers (e.g., Pol-gárdi,
1999; Szigetvári, 2000). Rebrus (to appear) reduces different licensing (and government) relations to
four. This latter approach argues that licensing is defined by the domain it is applied to, i.e., each type
of licensing can be substituted for by a licensing domain. This conception is explored and refined in
this paper inasmuch as licensing domains are considered to be typed phonological constructions.
The well-formedness of a structure depends on the (possibly overlapping) constructions defined on it.
In this paper we assume that possibly underspecified phonological representations in the
lexicon are 'parsed' by some phonological constructions. These constructions are thought as
controlling licensing relationships between prosodic positions and their segmental content. These
constructions, imposing constraints on the representation on various levels, filter out ungrammatical
structures.
243
Constructional CV phonology
The CV skeleton of the phonological domain is required to be parsed by possibly overlapping
constructions. The constructions more or less stand for various types of licensing configurations. The
boundaries of constructions, which are thought of as the 'constituents' of our representations, reflect
the natural boundaries of individual licensing domains. Constraints might enforce lexically
underspecified segments to be interpreted, i.e., for some segmental content to be supplied. We will
call this situation epenthesis. If in a language only interpreted vowels are allowed to license their
preceding onset, then the language is a strict CV language. We will call this most unmarked
licensing configuration the canonical CV construction.
(1) Sequence parsed by canonical CV constructions
alibi 'alibi'
In the following subsections we go through some special configurations, such as the ones resulting in consonant clusters57, and discuss how the assumptions of Strict CV Phonology and
Government Phonology about the nature of licensing and government can be reinterpreted and
reformulated in the language of constructions.
Domain-final empty nuclei
In addition to the least marked situation, consonants can appear at the end of words in a great
deal of languages. These would be ruled out if only the overt nucleus of the CV unit could locally
license a consonant to its left. Therefore, the licensing of domain final consonants, for instance, is to
be accounted for by the introduction of some extra device in addition to the canonical construction.
The standard solution in CV Phonology and Government Phonology is the explicit introduction of
domain final licensing (Kaye et al., 1990; Harris, 1990). As there is an obvious interaction between
the licensing of empty vowels and the licensing of consonants, it is standard to assume that the
licensing relationship always holds between the vowel and the consonant of the same CV item and
licensing by non-canonical (i.e., empty) vowels is actually equivalent to the licensing of an empty
vowel. The usual interpretation of domain final licensing then is the following: the domain final
nucleus is allowed to have no interpreted segmental content. For this situation we might come up with
a new construction, the domain final licensing construction:
(2) Domain final licensing
motor engine:
Domain final licensing will turn out to have a special status in our account of Hungarian
phonology in that it will have a restricted domain of application (cf. section 2.4). In a great deal of
cases, however, the relevant configurations are made possible by a licensing construction subsuming
domain final licensing: this is the topic of the next subsection.
57
Sybilants behave in a strange way in consonant clusters. We ignore them throughout the whole paper.
244
Constructional CV phonology
Intervocalic consonant clusters
In the previous section we pointed out how one can extend the inventory of licensing constructions in order to account for domain final consonants. However, the surface configurations
covered so far only feature sequences of strictly alternating C and V elements possibly ending in any
of the two. In a great deal of languages, however, phonotactic restrictions are not this severe: in
addition to strictly alternating consonant and vowel units, they allow certain domain internal
consonant clusters. Now we focus our attention on clusters constituted by exactly two consonants.
However, we cannot generalize the device of domain final licensing to word internal contexts,
i.e., the licensing of a consonant by any empty V position, which in turn is licensed exceptionally. If
we were to do that, we could not block the occurrence of arbitrary consonants with an empty nucleus
between them. It is generally true that the range of clusters that can appear intervocally is wider than
that of domain final clusters. These clusters do not necessarily exhibit any segmental interaction. In
other words, there is virtually no limitation concerning the segmental content of intervocalic clusters.
This generalization naturally lends itself to an account which does not involve any licensing
relationship between the consonants of the cluster, giving instead a leading role to the surrounding
vowels.
The second consonant is licensed by way of the canonical case of onset licensing by the
interpreted vowel in the right context. But what licenses the empty cluster internal V position and the
first consonant of the cluster in the lack of an overt nucleus? The standard solution to the above
problem is to introduce a new device, proper government, i.e., the right-to-left licensing of an empty
V position by a full vowel (Kaye et al., 1990; Harris, 1990).
Such an account, however, makes no reference to the vowel preceding the cluster, therefore, it
wrongly legitimates any bogus cluster word initially. The distribution of bogus clusters is bound to a
left-context vowel, therefore it is imperative to base our account on the interpreted left-context V
position. Therefore we choose to replace right-to-left proper government with a left-to-right licensing
construction, called trochaic licensing. To encode this we introduce a new construction called the
trochee construction.
(3) Trochaic licensing
a. The trochee construction
b. Parsing of bogus clusters
hagyma 'onion'
Van der Hulst and Rowicka (1997) and Rowicka (1999) argue that phenomena explained with
the help of proper government can easily be accounted for by trochaic licensing. In our analysis of
Hungarian morphophonology the former can be totally disposed of in favor of the latter. The trochaic
construction (called 'trochee') licenses an (empty) vowel after an interpreted one.
Whichever way one proceeds, what remains to be explained is the licensing of the cluster-initial
consonant. One is supposed to allow a right-to-left coda-licensing or to generalize onset-licensing to
245
Constructional CV phonology
empty V positions. We are going to argue for a solution in which the former can be reduced to the
latter. In keeping with the above discussion on domain-final licensing we say that licensing of an
empty V-position is sufficient to license the consonant of the same CV unit. By this we can maintain
that licensing is always possible between the nucleus and onset no matter the nucleus is interpreted or
empty.
The above analysis gains further support from a peculiar phonotactic restriction of Hungarian
monomorphemic stems. In Hungarian, clusters after long vowels are only allowed if the vowel is /á/
or /é/.58
(4) The *lómpa constraint
No long vowels, except /á/ and /é/ can be followed by a consonant cluster. No hypothetical
monomorphemic stem of the form *lómpa exists.
With trochaic licensing at hand, this generalization falls out. It is assumed for independent
reasons that no segmental content is normally associated with the second V position of long vowels in
the lexicon. The long vowels that do license bogus (or other) C-clusters happen to be the ones which
are to be represented as lexically present in both V positions for independent reasons (cf. Rebrus in
print). If the second V position is filled, canonical trochaic licensing can be made responsible for the
licensing of the CC-cluster at the lexical level (see below for a detailed discussion of such 'levels').
(5) VVCC sequences in the lexicon
a. vowel lexically represented as short *lómpa:
b. vowel lexically represented as long lámpa 'lamp':
The 'strength' of licensing — which is in proportion to the complexity of the inheritance path
of local licensing relations — can be naturally thought to have an effect on the segmental content of
licensees (Harris 1990, 1994). In the case of consonant clusters, licensing proceeds from the
interpreted left-context vowel to the empty nucleus, which in turn licenses its onset, which is to say
that licensing is inherited (as opposed to inherent or local). As this inheritance path is longer in a
trochee, one might naturally assume that the strength of trochaic licensing is weaker than in the local
case. This can readily account for the well-known fact that the coda position is generally a lenition
site (cf. section 1.5).
(6) Trochaic construction and licensing inheritance
58
Most exceptions from this generalization are either German borrowings with a peculiar -li/ni suffix-like
ending (e.g., ródli 'sleigh', műzli 'muesli', csúzli 'slingshot') or nontransparent derived forms (e.g., tozsde 'stockmarket')
Constructional CV phonology
246
The phonetic realization of the phonemes /v/ and /h/ in Hungarian is also a straightforward
example of the weaker licensing potential of the trochaic pattern. These phonemes appear as
obstruents before consonants — and as a consequence can take part in voice assimilation — , whereas
in prevocalic position they are realized as approximants. Obstruentization, then, could also be
explained by the lesser extent of licensing supplied with the trochaic pattern both in a bogus cluster
and at the end of a domain.
Trochaic licensing can do the job for domain final licensing, since it is not restricted to contexts
where the consonant is followed by another CV segment. This obviates the need for a domain final
licensing construction for languages where bogus clusters are attested. The availability of full-strength
domain final licensing together with full-strength trochaic licensing would allow for word final bogus
clusters, which do not generally occur in Hungarian monomor-phemic stems. Therefore we posit that
the two devices are not freely available constructions in the phonological representation of
Hungarian words (see the detailed account below).
(7)
Trochee at the end of parsed structures motor engine
Domain-final consonant clusters
In Hungarian clusters are possible at the end of a phonological domain, too. The consonants in
these clusters comply with rather strict requirements on sonority. As opposed to the general case of
intervocalic clusters, the consonants in these clusters also show a certain extent of segmental
interaction. They comprise the following types:
(8) Possible domain-final clusters
a. geminates,
b. nasal + stop clusters (henceforth partial geminates),
c. liquid + C clusters.
Real clusters
If we do not take this interaction into consideration, the intervocalic occurrence of these clusters is a sub-case of the bogus clusters discussed in the previous section. This is to say that the V
position within the cluster is licensed by trochaic licensing. As for the additional relationship, we
stipulate, in agreement with traditional analyses, that these consonant clusters can form a special unit,
hence called real cluster. Licensing relations are very special in these constructions as the first
consonant position — licensed by the same trochaic pattern as in bogus clusters — can license the
second when it is at the end of a domain.
(9) Licensing in the real-cluster construction
247
Constructional CV phonology
(10) Parsing domain final real clusters leng'(it) swings' (cf. (15))
The question arises whether we have to treat all occurrences of possible domain final clusters as
real clusters. First we are going to argue that real clusters are not only relevant at the end of a domain,
but also domain internally. This is apparent in our account of three-consonant clusters. The first two
consonants in a CCC-cluster can always occur domain finally. Just like in the case of a single
consonant which can equally appear in domain final as well as pre-consonantal (bogus) position, the
trochaic pattern proves to offer a unified solution we wish to give an account for these clusters. The
real-cluster analysis is already at hand, which, in a way analogous to closed syllable (coda-position)
consonants, can account for the distribution of these clusters with one and the same device. We only
need to allow real clusters (together with trochaic licensing) to occur inside a domain. (See the CCC
clusters in (20-c).) This is clearly superior to an analysis where the licensing of domain final clusters
is related to domain final licensing.
On the other hand, it is to be established whether it is desirable to allow an analysis of potential
real clusters as actually bogus in some cases or not, i.e., force a real cluster to be 'formed' each time
the two candidate consonants are adjacent. Our answer to this is on the positive though we will strive
to erect rather severe criteria on the individual constructions in order to reduce spurious ambiguities to
the minimum.
Mandatory real clusters
As argued above, the licensing of a consonant is compulsory in an analysed domain, therefore
the vowel following the first consonant in a CCC cluster can only remain empty if there it is licensed
by a real-cluster construction. Therefore we need to focus only on intervocalic occurrences of our
candidate clusters as potentially ambiguous. We follow the tradition to treat all instances of geminates
and nasal + stop sequences as real clusters:
(11) Mandatory real clusters
(Partial) geminates are always real clusters.
Our reasons for this are manifold. On the one hand, requiring real clusters to form, we can
simplify the account of a number of phonological phenomena. Degemination in Hungarian shows no
lexical exceptions. Though the conditions on degemination are rather complex (cf. Siptar 1992), it can
be analyzed with the help of real-cluster processes, therefore a unified representation of geminates is
desirable.
On the other hand, there is a very general constraint on bogus clusters: the consonants involved
cannot have the same place of articulation. This could be accounted for by the ubiquitous Obligatory
Contour Principle (henceforth OCP), which requires the nonidentity of adjacent segments in terms
of their place of articulation. The only exceptions to this generalization are geminates and nasal + stop
sequences. If all these were unambiguously analysed as real clusters, and the sharing (spreading) of
the place of articulation was seen as a result of a cluster-triggered agreement process (Rebrus 1999),
then we would obtain a situation when the OCP could be made a hard-wired inviolable constraint (as
opposed to soft phonotactic tendencies), i.e., a requirement on representations:
248
Constructional CV phonology
(12) The Obligatory Contour Principle of CV Phonology
Adjacent consonants have to have different places of articulation.59
Geminates We assume that geminates are always analysed as clusters on the surface with
identical segmental contents, while they are represented with the first C being lexically empty in the
lexicon (i.e., at least when the geminates are within one morpheme). If OCP is a constraint on
lexically specified segmental content, this stipulation becomes a necessity.
(13) Lexical representation of geminate /p/ kappa 'kappa' a. desired representation
b. representation that violates the OCP
There is another very strong argument in favor of this decision: the lack of Hungarian
monomor-phemic stems with a long vowel followed by a geminate. (ViViCiCi sequences). We have
already seen that no clusters are possible after vowels that are lexically represented as short but (can)
appear long on the word level. Apart from these contexts, however, the restriction appears to extend to
vowels which can normally be represented as long in the lexicon, i.e., a, é:60
(14) The *láppa constraint
In a monomorphemic stem a long vowel is never followed by a true geminate. No hypothetical
monomorphemic stem of the form *láppa exists.
The above lexical-level constraint can only be explained if we assume a unified representation
for all true geminates. This is what we do instead of letting them be represented as bogus in the
lexicon.
Partial geminates OCP and phonological processes operating on partial geminates are similar
to the ones referred to in the case of true geminates61 These make it desirable to represent nasals
without any place and stop specification in the lexicon.
59
It is a common assumption in theories using unary (place) features (Government Phonology and CV
Phonology) that alveolar is the default place of articulation, therefore alveolar sounds are marked with the
absence of any place specification. This can readily account for the fact that alveolars are the only consonants
occuring next to one another in bogus clusters. If the OCP is formulated within these theories, then alveolars,
having no feature marking their place of articulation, are naturally exempt from the constraint.
60
The only exception to this generalization is geminate 1 which behaves in a strange way in other respects, too.
We find stems like száll (verb) 'fly', váll 'shoulder', but here, too, they are mainly morpheme-final. One reason to
think that the lack of such stems is not an accidental gap can be their comparison with geminate + long vowel
sequences, which, though rare, are attested in monomorphemic stems (with gemmate / we find many). On the
other hand, we might compare their distribution to á/é'+ nasal + stop sequences (a very common pattern), which
proportion should reflect the general distributional rate of geminates and partial geminates.
61
One such process is the labial assimilation of nasals, also postlexically, as well as the nasalization of vowels
(appearance of the nasal element on the vowel) before a continuant C (e.g., vonz [voz] 'attract', vonsz [vos] 'you
drag').
Constructional CV phonology
249
(15) Lexical representation of partial geminates leng'(it) swings'
Optional clusters
General considerations about the compulsory segmental interaction as well as the phonological
processes appearing in consonant clusters have led us to the assumption that all kinds of geminates are
to be analysed as real clusters. The same arguments, however, do not hold for other possible stem
final consonant clusters still conforming to standard sonority restrictions, namely liquid + C clusters
(lk, jd, rm etc.). Therefore we have decided not to represent them necessarily as real clusters. This
potential ambiguity will turn out to reflect the dual behaviour of this sort of cluster correctly inasmuch
as it will prove useful in the representation of some lexical contrasts.
The generalizations are the following:
1. Possible domain final clusters must obey severe restrictions.
2. The segmental interactions between the consonants of a real cluster and the licensing
relation between them are mutually dependent, i.e., they are regarded as the effects of one
and the same construction.
3. All geminates are represented alike whichever position they occupy. No element with
consonantal features is lexically associated to the first C position. A nasal coda is required to
surface as part of a real cluster or otherwise they nasalize the nucleus if followed by a
continuant.
4. Not all possible domain final clusters are always analysed as real clusters. Liquid + stop
clusters, in which no segmental interaction is observed, are ambiguous: a real cluster is
assumed only when necessary, i.e., when licensing conditions force it.
5. Real clusters can occur as the first pair of CCC clusters, in which the second two consonants
constitute a possible word initial cluster.
It is natural to assume that trochaic licensing is responsible for the licensing of the first consonant as well as the empty vowel, since real clusters always stand after vowels. As we argued above,
we do not want to use domain final licensing in the general case, because that (together with trochaic
licensing) would also allow for bogus clusters in word final position. This requirement, together with
the fact that there are segmental processes going on within the cluster, make it desirable to analyse the
second consonant as one licensed by the first. As the first consonant, in turn, is licensed through the
trochaic pattern, this VCC is one complex licensing domain. The licensing relation between
consonants can be made responsible for the phonological processes always applying in these clusters.
The forming of real clusters then can be forced by two different principles. On the one hand,
lexically empty or floating segments in consonantal positions must receive an interpretation. This
process presupposes the specification of some of the consonants articulation features. Putting aside
the case of nasalized vowels before continuants, this specification is made possible by the 'agreement
relation' holding between the members of a real cluster. Therefore the requirement of word-level
interpretability coerces a lexical coda into a real cluster with the subsequent C having lexically
associated segmental content. This constraint obviously does not affect liquids in a coda.
250
Constructional CV phonology
On the other hand, parsing two consonants as a real cluster is necessary if this is the only way
to license the second consonant, i.e., in CCC clusters or domain finally (see section 1.7 for
exceptional licensing) .62
Here we present the lexical and analysed (parsed) representation of real clusters, the latter as a
complex licensing domain embodying a combination of the trochee and the real-cluster licensing
construction:
(16) Complex licensing domain with real clusters a. Real geminates galopp 'galopp'
a. Real geminastes
galopp 'galopp'
b. Nasal + stop clustres
kolomp 'bell'
c. Liquid + C clusters
kudarc 'failure'
Long vowels
It is evidenced by a great deal of phenomena that long vowels and trochaic constructions are
very intimately related. Nasalization (a nasalized long vowel appears before a continuant),
supplementary lengthening (optional drop of liquids (mainly /1/) in coda position) take place in
canonical real-cluster constructions. This suggests that the construction responsinble for the licensing
of long vowels is akin to a trochaic pattern. Though in this paper we will neglect a detailed discussion
of long vowels, we argue that in a hierarchical network of constructions one can generalize these
constructions as the complex syllable construction.
By the hierarchy sketched below we propose a generalization network of some of our phonological constructions. The generalization yields a subsumption order on the construction types which
62
We have to note here that trochaic licensing is not thought to be a more marked construction than domainfinal licensing. This is evidenced by the fact that languages exist where certain intervocalic real clusters are
allowed, whereas there are no word final consonants (e.g., Italian). This can be accounted for if we suppose that
the trochaic licensing of a consonant in these languages is restricted only to those with minimal segmental
features lexically specified. The interpretation of the trochaically licensed consonant, however, is strongly
influenced by the inventory of the phonemes of the language. At the end of the domain they cannot acquire their
articulation features in the lack of a real cluster. In this situation, either they cannot appear (Italian) or they are
realized without any consonantal features (Japanese). That partial geminates are still less marked than true
geminates is a consequence of another restriction, disfavouring lexically empty C positions.
Constructional CV phonology
251
reflects the complexity (information content) of licensing relations.63 An intermediate construction
might not have any interesting theoretical status besides being the generalization of some fully
specified structures (i.e., its sub-types). We believe that a classification of these constructions can
actually take place in more than one dimension (class of constructions consisting of two CV units at
most, constructions with some vowel-vowel relation, etc.), and classes generalized across these
dimensions actually count as special subsets available for or disfavoured by specific sorts of
phonological domains or morphological classes.
(17) Typology of complex constructions
We should regard this figure as generalizing over every construction involving a complex
licensing domain, i.e., a licensing sort of relation between two CV units, hence the name complex
syllable. One of the two sub-types of this complex construction is one in which the vowel in the
modifier unit is interpreted. In this case, similarly to a real-cluster construction it is assumed that there
is some relation between two adjacent V positions. This is called long-vowel relation and is thought to
be constrained by the set of phonemes in the given language.
Below we give the types of long-vowel constructions.
(18) Typology of long-vowel constructions
In the other subcase the vowel in the modifier construction is not interpreted, it is licensed to
remain empty.
63
The particular typing in the figure is meant to be tentative only to be further refined if the analysis makes it
necessary.
Constructional CV phonology
252
(19)
Typology of trochaic constructions
The intuitive typology sketched here becomes relevant when one wants to relate arbitrary sets
(classes) of constructions to certain lexical classes (cf. section 1.8) and are relevant especially in the
case of synthetic suffixation (cf. sections 2.5 and 2.6).
Domain initial clusters
For an account of domain initial clusters we will introduce a special construction. Clearly, we
cannot give an account for them in terms of any licensing condition discussed so far. At first sight,
domain initial clusters may seem parallel to domain final consonant relations inasmuch as similarly
strict requirements are imposed on them in terms of sonority. However, the segmental content of these
clusters is more constrained than that of domain final sequences. When they occur intervocalically,
they are subsumed by the general bogus cluster construction with trochaic licensing. The question
arises whether we should introduce a special licensing construction similar to the domain final case.
With such a solution one would predict that these clusters could appear as the second and third
consonants of a CCC sequence in which the first two Cs do not constitute a cluster. These examples
are typically not attested.64
There is no symmetry with respect to place assimilation, either. Actually, just like in the case of
any non-interacting consonants, the OCP has the effect that the places of articulation within domain
initial clusters are obligatorily different (no stems starting in e.g. pv, bv). The fact that in domain
initial clusters, the effect of OCP extends to alveolars (no stems starting in *tl, dl exist, either) also
indicates that this is a highly marked construction.65
There are, however, three-consonant clusters in case the first and second Cs constitute a domain
final cluster, and the second and third constitute a possible domain initial cluster (e.g., templom
'church').66
64
Monomorphemic exceptions (e.g., oktrojál 'force') to this can only be treated as phonological compounds,
which is motivated on independent grounds.
65
We note here that the analysis totally excludes four-consonant clusters: whatever combination of constructions
license the consonants and the surrounding vowels, the V position in the middle is not licensed, i.e., cannot
remain empty.
66
Again, there are marginal exceptions like arktikus 'arctic'. More relevant is the tendency that there are clusters
which are not allowed domain initially, though they appear in CCC, e.g., pantlika 'ribbon' or partner 'partner',
though tl and tn are forbidded domain initially.
Constructional CV phonology
253
(20) Domain initial clusters
a. domain initially krimi 'thriller'
b. in intervocalic position dupla 'double'
c. as part of a three-conconant cluster szimpla 'simple'
Exceptional domain final licensing
As alluded to before, contrary to the fact that trochaic licensing subsumes domain final licensing, we will demonstrate that a special restricted version of domain final licensing is nevertheless
useful and necessary. Its special status motivates the term exceptional licensing construction (cf.
Rebrus, to appear). We emphasise here again that one has to take care of domain final licensing in the
general case since, when preceded by a trochaic construction, it may result in stem-final bogus
clusters. There is a special class of monomorphemic nominal stems in Hungarian, however, that
makes it necessary to introduce of some special device for domain final licensing. The stems in
question all end in a peculiar bogus cluster (Rebrus to appear):
(21)
Exceptional nominal stems ending in a bogus cluster
a. bogus-final nominal stems ending in /v/
(i) szarv horn
(ii) kedv mood
(iii) könyv 'book'
b. bogus-final nominal stems ending in /j/
(i) szomj 'thirst'
(ii) alj 'bottom'
(iii) furj 'quail'
c. bogus-final nominal stems ending in /h/
?enyh 'relief
254
Constructional CV phonology
These are all consonant + fricative clusters ending in a consonant alluded to above (cf. section
1.3). This phonological generalization can serve as a basis for the characterization of this special class
of consonants, which we will call exceptional consonants. These consonants appear unlicensed
domain finally, therefore we are bound to introduce a special type of domain final licensing
construction restricted to exceptional consonants.
The need to apply exceptional licensing appears with some suffixed forms ending in a bogus
cluster, however. This occurs only in the case of two verbal analytic suffixes made up of one
consonant. These are the imperative inflectional suffixes -d, and -j, which attach to verbal stems
without an epenthetic vowel and yield a bogus cluster which may never be attested elsewhere:
(22) Imperative forms ending in bogus
a. hagyd (hagy + IMP-2ND-SG-DEF.OBJ) 'leave (that)'
b. lengj (leng+ IMP-2ND-SG-INDEF.OBJ) 'swing'
The real-cluster status of these consonant pairs is out of the question, so on the word level the
domain final consonant has to be exceptionally licensed:
(23) Exceptional licensing kedv 'mood'
Constructions and the hierarchical lexicon
Morphologically distinct classes (lexically classified by their part-of-speech status and the
complexity or type of the suffixation) can have their own characteristic choice of constructions
selected out of the general inventory as available for phonological parsing of the representation.
It is clear that the class of monomorphemic stems represent a stratum of language that has its
own grammaticality criteria. Now it seems natural to assume that the phonological representations of
the members of this class are parsed according to these specific constraints. In the following we will
refer to this level as the lexical or monomorphemic level.67 The more restricted nature of
monomorphemic patterns is due to the availability of a smaller set of licensing constructions (and a
more restricted inventory of phonemes available) on the lexical level.
The level of interpretable phonological representations represent another such level, which is
(conforming to standard terminology) called word level in general. However, the fact that the
representational language of these two levels is considered to be the same, and the constraints
controlling the grammaticality of structures are of the same sort, discredits the postulation of two
distinct representation levels. Also, as most (or at least a large part) of the correspondence between
lexical and word level representations is taken to be monotonic, we do not need to conceive of the
levels as constraining different informational parts of our linguistic entities.68 This is why we will take
67
This term is used to avoid confusion with different levels of representation, i.e., the morphemic and word
levels used in Cognitive Phonology (Lakoff 1995). Our lexical level comprises monomorphemic stems only.
68
Unfortunately, such a solution could not even receive an appropriate formalization in a restricted declarative
framework, such as Generalized Inheritance Networks (GIN) (Kálmán and Trón 1999). This is because the
intuitive idea of carrying over information from one level to another can only be expressed by structure sharing
which makes contradicting constraints on distinct levels useless anyway. In Lakoff's (1995) three-level phonology the derivational cycles of lexical phonology are replaced by a fixed number of representational levels. It is
assumed there that constructions establish correspondences between the levels, or, as a default, information is
255
Constructional CV phonology
a daring step and say that our phonological representation comprises only one representational level.
The actual strata of language imposing various (maybe simultaneous) constraints on parsing a
structure are thought to restrict this single phonological representation:
(24) Different levels constraining phonological representations
If a structure belongs to more than one such stratum, then it is constrained by the restrictions on
each of them simultaneously (e.g., a monomorphemic stem in its base form as a word). We believe
that the seemingly contradicting constraints imposed by the different 'levels' should be handled by
way of underspecification and intricate subsumption relations holding between the constructions,
some of which has already been alluded to (cf. 1.5).69
Additionally, it will turn out that the phonological behavior of various morphologically exceptional subclasses of stems and suffixed forms can readily be captured if we conceive of our
previously mentioned strata as further divided into smaller classes. These classes are supposed to have
their own phonological regularities expressed by the inventory of available constructions. This
basically means that our levels are far from being homogeneous, rather they form an intricate
hierarchy of sublexicons.
(25) The organization of the morphological sublexicons
We emphasize that reference to sublexicons (more or less arbitrary lexical classes) is natural
and necessary in any account of class-specific phonological regularities or category-sensitive
processes. Therefore we rely only on what is needed, anyway.
We do not want to postulate a clear-cut division between phonological regularities and exceptional behaviour. Instead, we conceive of them as categories exhibiting graduality. We will express
the degree of markedness in terms of the number and complexity of constructions a certain
morphological or lexical class selects. This is reflected by the hierarchical structure of the lexicon,
i.e., information content defines a partial order of subsumption. Phonotactic grammaticality, thought
to be gradual independently as a result of the markedness order of sounds, is now interpretable only
relative to a given form-class (Torkenczy 1999).
shared by two levels. However, there are well-formedness constraints on the actual levels which, in case they
are incompatible, might have unwanted interaction with monotonicity.
69
Both devices are readily available in a declarative framework of typed relational inheritance network (Kálmán
and Trón 1999, Rádai 2000). The formal details of the encoding and the implementation are not at issue in this
paper.
256
Constructional CV phonology
Phonotactics and morphophonology
Types of suffocation
The usual division between analytic and non-analytic suffixation is also relevant in the case of
Hungarian. Analytic suffixation is never accompanied by stem alternations (except for stem final
vowel lengthening, which is ignored here), such as shortening, stem internal vowel deletion etc. This
empirical categorization is often captured by the difference in the domain in which the phonological
rules are applied.
(26) Types of word-formation
a. synthetic suffixation: [AB] b. analytic suffixation: [[A]B] c. compounding: [[A][B]]
This suggestive notation is an attempt to capture the intuition that in the case of synthetic
suffixation the lexical representations of the morphemes are simply concatenated, thereby constituting
one phonological domain within which the rules apply. On the other hand, compounding is a case
where the concatenation takes place after the rules have applied.
To formulate this in an analogous way within our treatment, we can say that a synthetically
suffixed form that is the concatenation of the morphemes constitutes one parsing domain. As we
assume that monomorphemic stems are also parsed in the lexicon, the only difference between
synthetic and analytic suffixation will be that in the former case the stem is parsed on the
monomorphemic level, whereas in the latter, on the word level.70
(27) Two types of suffixation
a. synthetic suffixation: [[A]mB]w b. analytic suffixation: [[A]wB]w
The behavior of synthetic suffixation can be handled in a simple manner if we assume that the
word level constraints restrict the suffixed form together with its suffix as one phonological domain.
The account for analytic suffixation, in turn, relies on the distinctness of the two domains. In other
words, it makes reference to the word level base form of the stem to which the analytic suffixes are
attached.
As a result, analytic suffixation, though it does not trigger any stem alternations, is parsed as
one word domain. Therefore all the postlexical assimilation processes and degemination affect the
domain. As opposed to compounding, however, word internal phonological rules, such as vowel
harmony and complete palatal assimilation do apply here.
(28) Types of suffixation
However, there are constraints which typically distinguish synthetic domains from analytic
ones. Support for this comes from the fact that contrary to the generalizations about stems these
suffixed forms can violate phonotactic constraints on consonant clusters. Some extreme examples are
worth mentioning:
70
We do not yet have an account of sequences of suffixes in words, therefore we focus only on monomorphemic
stems and suffixed versions of these. Multiple suffixation is discussed only marginally in this paper.
Constructional CV phonology
257
(29) Analytically suffixed forms violate phonotactic constraints on stems a. fajlja (fajl + POSS)
'his/her file' b. gombnak (gomb + DATIVE) 'for button'
As is apparent, analytically suffixed forms can also violate the OCP, i.e., the adjacent consonants on the suffix boundary can have the same place of articulation:
(30) Analytic suffixed forms violating the OCP
a. dobva dob + va 'thrown/throwing' b. savban sav + ban 'in the acid'
This fact is instructive that we should handle the OCP as a domain internal constraint restricting
only synthetic domains, therefore a distinction between the two types of domains may be desirable.
In what follows we will discuss the particulars of an analysis of some morphophonological
phenomena in Hungarian. Our account is not meant to be exhaustive in any sense, rather we wish to
illustrate what we think motivates a reinterpretation of CV phonology within a radically lexicalist
declarative framework.
General constraints on monomorphemic stems
To repeat what has already been discussed we give the relevant generalizations referring to all
monomorphemic stems:
(31) Constraints on monomorphemic stems
a. The *lómpa constraint
No V*CC occurs in a monomorphemic stem.
b. The *láppa constraint
No VVC1C1 occurs in a monomorphemic stem.
There is ample evidence that this constraint is relevant only for monomorphemic stems, and
cannot be accounted for without assuming special phonotactics for the stems.
(32) Violation of the *lómpa constraint in suffixed forms VV + CC
a. hajónk hajó + -nk{POSS- 1ST-PL)
'our ship'
suffixed noun
hajóstul hajó + stul(DERIV. SUFFIX) 'with ship'
hajónként hajó + nként(DERIV. SUFFIX) 'by ship'
b. lottek, lo + (t)t + (e)k(PAST-3RD-PLUR)
suffixed verb 'he shot'
ródd, ró + (d)d(lMP-2ND-SIND-DEF) 'tell off
(33) Violation of the *lómpa constraint in suffixed forms VVC + C
a. bunt bun + t(ACC)
accusative nouns
'sin' b. súgta súg+ ta(PAST-3RD-SING-DEF)
past tense verbs
'he whispered it' c. óvnak óv + nak(3RD-PLUR)
pseudo-analytic suffixation
'they protect (me)'
Constructional CV phonology
258
(34)
a. védd véd + d(IMP-2ND-SING-DEF)
‘defend it‘
analytic suffixes
b. vé[tt]e véd + te (PAST-3RD-SING-DEF)
‘he defended it’
past verbs VVC+C
c. lábbol láb + ból(ELATIVE)
‘from leg’
suffixed noun VVC+C
As a first example, we will focus our attention on the phonotactic generalization in (4). For
independent reasons we do not want to represent long vowels. As it has been alluded to, an intuitive
way to explain the constraint would be to say that the empty second vowel of the VV in the lexicon
cannot head a trochaic construction, therefore no cluster is possible lexically after the long vowel. In
other words, no matter the V between the two Cs is specified as empty in the lexicon, already on the
monomorphemic level an epenthesis takes place in order to license the consonant in the absence of a
preceding vowel. This would yield the base form *lómopa on the word level.
The problem, however, arises with stem final Cs, which occur frequently after any type of long
vowel. This is a serious problem for any analysis that attempts to handle coda consonants alike.
Clearly, we cannot extend our notion of domain final exceptional licensing to any consonant, and we
want no special device to distinguish domain final and domain internal trochees. Instead of trying to
put the burden on the different licensing constructions, we use a global device to solve the problem:
extrametricality. In particular, in the case of stems (or maybe all lexemes), we allow the final CV
unit of the lexical representation to remain un-parsed. The last syllable is called extrametrical since
lexical-level parsing constructions may ignore it.
(35) Principle of extrametricality at the lexical level
Lexical-level parsing constructions may ignore the last syllable (CV unit) of the lexical
representation of monomorphemic stems.
Owing to this global device, the last C of consonant-final stems is exempt from the obligation
of being lexically licensed. This means that their distribution is not restricted by the interpretation of
the preceding vowel. They might appear irrespective of the left context.
(36) Representation of canonical VVC final stems hód'beaver'
In the following subsection we give further evidence supporting extrametricality on the lexical
level.
Epenthetic stems
There is a great deal of stems — nominal as well as verbal — in Hungarian which show
vowel/zero alternation in their second V position in suffixed forms. These are called epenthetic stems
(Torkenczy, 1992; Rebrus, in print):
Our usual methodology makes it necessary to represent these stems without segmental content
associated to the epenthetic vowel position. Fortunately this is feasible since the quality of the
Constructional CV phonology
259
epenthetic vowel is predictable: it is always a middle vowel altering according to usual regularities of
(palatal and labial) vowel harmony.71
With no vowel present in the lexical representation, the stem final consonant remains unlicensed. The first consonant in the lexical cluster might remain ambiguous with respect to whether it is
licensed by a trochaic or a canonical licensing construction. If the last consonant of these stems was
always parsed by some construction, the last two consonants would either have to form a real cluster
— if that is possible —, or a trochaic construction would require the epenthetic vowel position to be
interpreted, therefore vowel/zero alternation would be banned. In keeping with our previous
assumptions, the very fact that epenthetic stems exist makes extrametricality indispensable.
(38) Lexical representation of epenthetic stems bokor 'bush'
We note that for the vowel/zero alternation to take place one option for lexical parsing should
be trochaic construction involving the epenthetic position. This presupposes a vowel lexi-cally present
before the lexical bogus cluster within the stem. We correctly predict then that epenthetic stems with a
long first vowel (not a, e) are exceptional.
As argued above, the first consonant of (partial) geminates can only be interpreted if the real
cluster is formed, whereby any alternation is blocked. Because of the OCP, on the other hand, no
specified representation of these geminate clusters is allowed in the lexicon. Thus, we correctly
predict that no epenthetic stems exist where the cluster is a (partial) geminate.
(39) Hypothetical stems that could not be epenthetic
(We give the base form, and we underline the epenthetic vowel.)
a. konop
interpretation of N
b. lomop
OCP
c. kopop
OCP
d. kodop
voice assimilation
With the exclusion of geminates and nasal + stop clusters, however, we are still left with a
specific cluster type, which are possible real clusters and which can nevertheless occur as consonants
of epenthetic stems, namely liquid + C clusters. This strongly supports our previous decision that
liquid + C clusters are not necessarily represented as real clusters in the lexicon, but are ambiguous
(cf. 1.2).
As the epenthetic V position is empty within the cluster in the lexical representation of the
stem, it should somehow be stipulated that, in cases like this vowel, epenthesis should take place
when the stem ends in a closed syllable, i.e., we have to forbid *to[rk] instead of torok as the free
base-form of 'throat' (cf. (37)). In other words, we have to make sure that a construction involving the
formation of a real-cluster is dispreferred (to trochaic licensing) even at the cost of vowel epenthesis.
A straightforward solution to this is the assumption that epenthetic stems are characterized by their
71 15
There are a handful of exceptions (all nouns) where the epenthetic vowel is not middle but low (e.g., ajak
'lip' - ajkam 'my lip'). Further two exceptions (e.g., bajusz 'mustache' - bajszom 'my mustache') in the case of
which the epenthetic vowel is not middle and not low, either. In none of these cases does the epenthetic vowel
violate vowel harmony.
260
Constructional CV phonology
not allowing any consonant-relations to form. This would immediately explain why there are no
epenthetic stems in which (some member of) the paradigm would presuppose consonant relations.
As the consonant of the stem should not be involved in any real cluster (since a real cluster
relies on a consonant relation), we predict that no (consonantal) synthetic suffix morpheme can attach
to the stem without a preceding epenthetic vowel. This consequence is borne out: epenthetic stems
always parse by two trochaic constructions in the case of consonantal suffixes (see (a) below) and also
disallow real clusters in the stem domain (see (c) below). The contrast with other stems is relevant
only in the case of the accusative form where non-epenthetic stems allow the accusative case suffix -t
to attach directly to the stem final consonant if a legitimate real cluster is formed (see (b) below):
(40)
a. bokor ‘bush’ – bokrok (PLUR) – bokrot (ACC)
epenthetic
b. motor ‘engine’ – motorok (PLUR) – motort (ACC)
non-epenthetic
c. mokor – mokrok (PLUR) – mokort (ACC)
non-existent bypothetical epenthetic
Below we give the word level representations of the forms of an epenthetic stem. Apart from
the fact that it contains an empty V position internally, there is nothing special about this
representation, actually there is not any specified construction lexically given. This fact is in
agreement with the relatively unmarked status of these stems within exceptional (here: nonproductive)
stem classes. Below we give the parsed word level representations of epenthetic stems:
(41) Parsed word level realizations of epenthetic s
a. free (base) form bokor 'bush'
Lowering and exceptional licensing
In this section we focus on an exceptional subclass of nouns called lowering stems. A stem is
called lowering if the epenthetic vowel between the stem and its suffixes is low (a, e), as opposed to
the canonical middle (o, e, o). Lowering stems constitute an apparently exceptional closed class. By
definition, then, a lowering stem ends in a consonant (otherwise there is no vowel epenthesis).
At first sight, lowering cannot be given a direct phonological explanation: as the existence of
straightforward minimal pairs, such as dal 'song' (non-lowering: dalok) and hal 'fish' (lowering: halak)
suggests, lowering cannot always be predicted from the surface form of the stem. However, the
regular correlation of lowering to other phonological phenomena as well as a number of phonotactic
generalizations about the form of these stems cast a doubt on the independence of this seemingly
morphological dimension from phonology proper. If we were to analyze lowering simply as a lexical
property in terms of an indication of an arbitrarily chosen special paradigm to which the stem belongs,
we would fail to capture a great deal of generalizations. Therefore, no matter how arbitrary the
property of lowering turns out to be in some cases, we choose to give it a specific phonological
representation, whereby the explanation of some interesting entailments concerning the phonology of
lowering stems will naturally fall out.
Stems ending in a bogus cluster (bogus-final stems, see (21)) are classified as lowering. This
phenomenon urges one to presume some connection between lowering and phonological licensing
constructions. There is another rather important across-the-board generalization on lowering stems.
Constructional CV phonology
261
Normally, the accusative -t (the only synthetic case suffix) can attach to the stem-final consonant
without epenthesis provided it yields a proper real cluster. In the case of lowering stems, however,
epenthesis is mandatory also in the accusative form:
(42) Epenthesis in the accusative
a. halat (hal + ACCUSATIVE) 'fish' (cf. halak)
b. dalt (dal + ACCUSATIVE) 'song' (cf. dalok)
lowering
non lowering
It is imperative, then, to connect the behavior of lowering stems with licensing conditions at the
end of the stem. This virtually amounts to saying that however idiosyncratic the property of lowering
is, it should be (lexically) encoded in the phonological representation instead of simply indicating an
arbitrary paradigmatic choice. With this solution at hand, we might hope to explain the positive as
well as the negative implications of lowering simply as the natural consequence of the inconsistence
of some licensing constructions at some level of representation.
Positive implications
Below we give two positive entailments concerning lowering:
(43) Properties implying lowering
a. A shortening stem is lowering.
No hypothetical singular—plural pair of the form sam-samok exists.
b. A bogus-final stem is lowering.
No hypothetical singular—plural pair of the form *samj-samjok exists.
It seems tempting to revert to a proper-government analysis saying that the quality of the vowel
in non-lowering epenthesis is unable to properly govern the preceding V, therefore it cannot remain
empty (cf. Rebrus, in print). Something in the same vein can in fact be said under the present
treatment: non-lowering is allowed only in case it is weakly licensed from the left. Since the
constructions constituting the two examples above have a licensing-domain boundary before the
epenthesis site, we could say that no mid vowel is allowed in these configurations. However, there are
serious problems left to solve.
On the one hand, the representation of bokor-type epenthetic stems (cf. section 2.3) should
receive the same word-level representation as bogus-final stems, since there, too, epenthesis takes
place after a bogus cluster. The fact that mid-vowel epenthesis can take place in these cases discredits
any analysis involving the condition that the mid vowel is licensed by an interpreted vowel from the
left. (We note that a quality-sensitive proper-government analysis faces the same problem.) On the
other hand, on the morphemic level most long vowels are represented as short (cf. (36)), therefore
their underlying form is the same as that in most lowering stems. As a consequence, reference to such
restrictions on non-lowering either on the word or the monomorphemic level cannot explain the facts.
Shortening stems It was argued on independent grounds that long vowels (just like real
clusters, which are not represented with their domain on the lexical level) are formed whenever
possible. With an appeal to monotonicity, our decision is to represent the vowel without its domain on
the stem level (though, as argued above, long á and é are the only unmarked long vowels with their
domain possibly present in the monomorphemic representation). But then the data about shortening
stems contradicts our previous assumptions, since in the synthetically suffixed forms of nyár 'summer'
(nyarak, nyaral etc.) contain no long vowels.
Constructional CV phonology
262
(44) Lexical representation of shortening stems nyár 'summer'
Whatever is responsible for the fact that the non-analytic suffixed forms of these words contain
a short vowel (cf. (52)), the V position is always empty before the stem final consonant in the lexical
representation. Since even on the word level no trochaic licensing can then také place between an
empty vowel and the one after the consonant, any subsequent structure that presupposes trochaic
licensing by this empty position is ruled out. As a consequence, we obtain an immediate explanation
for three generalizations. As for any kind of consonant cluster within the non-analytic domain,
trochaic licensing is needed, it falls out without any further stipulation that the consonant after the
shortening vowel has to be followed by a vowel (cf. section 2.2). This requirement rules out
shortening stems with their (shortening) vowel followed by a consonant cluster as part of the stem
itself (see (a) below) and consonant clusters formed as a result of suffixation (see (b) below):72
(45) Hyphotetical shortening stems ruled out
a. * sánt (Noun, stem) – santak/santok (plural)
skálár (Noun, stem) – saklarak/saklarok (plural)
stem final real cluster
any intervocalic cluster
b. * sán (Noun, stem) – sant (accusative)
süt (Verb, stem) – sütte (past tense)
lack of epenthesis in accusative
lack of epenthetic vowel in past tense
Interestingly, shortening in the stem vowel is not independent from lowering: all shortening
stems are lowering.73 Note that a simple postulation of this entailment as an intended explanation is
not sufficient to account for some other limitations on shortening stems, such as forbidden word final
CC clusters, since some of these can occur word finally in lowering stems, e.g., szorny 'monster',
targy 'object'. The phonological explanation of exceptional licensing as discussed in lowering stems,
however, can immediately be connected to the above finding.
The canonical epenthetic vowel of nominal stems involves some floating U element (the one
that is responsible for inducing the middle vowel of the epenthesis). Therefore, throughout the
paradigm of a shortening stem the floating U would associate to the empty V position of the stem.
This would result in an ill-formed diphthong (e.g., au, eu, iu) or a persistent long vowel with a 'U'
element (e.g., u, u, o, o), thereby rendering potential lowering ineffective: 74
(46) Shortening stems are lowering
*nyarok (from nyár 'summer'; OKnyarak 'summer, PLURAL' )
72
Final low vowels are either always long as a result of compulsory lengthening or already represented as long
in the case of á and é. Therefore we can rule out hyphotetical pairs with shortening stem-final vowels of the
form masé-maset on independent grounds.
73 17
The reverse statement is not true as evidenced by pairs such as vár 'castle', várak 'castle, PLURAL'.
74
This might gain further support from the distribution of vowels in epenthetic stems. Whatever is responsible for the quality
of vowels in stem internal epenthesis (cf. section 2.3), stems with exceptionally low stem internal epenthesis (cf. footnote
15) are all lowering (i.e., an a followed by an o in two potential epenthetic sites is not allowed). If the mid-vowel epenthesis
of suffixed nouns is explained by the presence of a lexical floating U, this generalization falls out.
263
Constructional CV phonology
However arbitrary the class of lowering stems may be (cf. section (52)), in order to avoid
serious detectivity through the entire synthetic paradigm its members have to be marked as lowering.
Therefore, we conjecture that, provided we need a special lexical class marked for shortening stems, it
has to be subsumed by the class of lowering stems.
Bogus final stems Next we consider the second problem concerning the contrast between
bogus final and epenthetic stems. Now suppose that there is no difference between bogus final and
epenthetic stems in the lexicon with respect to the licensing of the penultimate CV unit. If lowering
for bogus final stems was not specified in the lexicon, either, then the lack of epenthesis in the base
case (nominative) — and similarly lowering in other suffixed forms — should be taken as a
consequence of the availability of exceptional licensing of the stem final unit on the word level. The
different behavior of epenthetic stems and exceptional bogus final stems would hinge exclusively on
the particular range of consonants for which exceptional domain-final licensing is an option.
Unfortunately, this treatment would predict that there is no lexical contrast between epenthetic and
bogus final nouns with the same final consonant, which is falsified by the following pair:
(47) Epenthesis contrasting stems ending in lexically exceptional clusters
a. bagoly - baglyok' owl'
b. szomj - szomjak 'thirst'
The whole truth is that there is no epenthetic stem ending in /v/ or /h/ at all. Epenthetic stems
with /v/ as the first C of the cluster are also missing and the ones with /h/ are meta-thetic (e.g., teherterhek 'burden'). We take this to be a consequence of the constraint that stem internal exceptional
consonants have to be specified already in the lexicon. If their realization depends on the licensing
construction they are involved in, they cannot remain ambiguous, hence the paradigmatic integrity of
these vowels.
If there is a difference with respect to the compulsory presence of trochaic licensing in the lexical representation of bogus final stems, then we can explain why the stem has to be lowering. On the
other hand, however, lowering should also be represented in the lexicon to account for the lexical
contrast between lowering and canonical epenthetic stems (also with the same final consonant):
The exceptional lowering construction is supposed to be the only lexical difference between the
licensing structure of the two entries above. Epenthesis should be taken as a result of the fact that
exceptional domain final licensing at the word level is available to a very small set of consonants
(i.e.j, v, h).
(48) Lowering as contrasting epenthetic stems
a. bokor 'bush' - bokrok 'bushes'
non-lowering epenthetic
b. sátor 'tent' - sátrak 'tents'
non-lowering epenthetic
(49) Lexical-level parsing of lowering epenthetic stems
264
Constructional CV phonology
But then the positive implication asserting that exceptional-cluster final stems are lowering falls
out from the interaction of two constraints. On the one hand, the exceptional consonant causes the
stem to be parsed, but if, on the other hand, it is preceded by a trochaic construction, it can only fall
back to exceptional licensing.75
(50) Lexical representation of bogus final stems
a. monmorphemic lowering stem kedv ‘mood’
b. analyic inflectional suffix76 dobj = dob +j(IMP-2ND-SING) ‘throw’
In conclusion, we conjecture that lowering stems are not extrametrical as their last syllable is
lexically marked (parsed) by the exceptional licensing construction. This enables a unified
representation of lowering stems:
(51) Lexical representation of a lowering stems hal 'fish'
Entailments about obligatorily lowering in the case of some stem classes were explained by the
compulsory parsing of the normally extrametrical stem final syllable and the unavailability of any
other construction than the exceptional construction.
This exceptional construction, however, is only allowed in the case of exceptional consonants
when on the interpreted level (actually word-finally). If it involves other than exceptional consonants,
it either resolves to a trochaic construction, or triggers (low-vowel) epenthesis:
(52) Parsed word level representation of lowering stem
a. free (base) form hal ‘fish’
75
76
Note that there is no /j/-final lowering epenthetic stem.
Suffixes themselves can have an effect on the licensing patterns they appear in. Inflectional and derivational
suffixes are equally found among synthetic as well as analytic suffixes. Interestingly, however, all inflectional
suffixes are lowering with respect to the epenthetic vowel appearing after them at least. Interestingly, the data in
(22) show that the analytic suffixes consisting of one consonant are all inflectional. As inflectional suffixes, they
are exceptionally licensed, thus they can appear in a single 'exceptional' CV domain word finally.
Constructional CV phonology
265
b. in synthetic suffixation halak ‘fishes’
Negative implications of lowering
In the following, we turn to some negative implications related to lowering:
(53) Properties implying non-lowering
a. A stem containing a nonshortening unstressed long vowel is non-lowering. No hypothetical
singular-plural pair of the form samár-samárak exists.
b. A real-cluster final stem is non-lowering.
No hypothetical singular-plural pair of the form samp-sampak exists.
Shortening and lowering First we have to emphasize that shortening (and thereby lower-\ in
bisyllabic stems with a long second vowel is in fact a lexically contrastive property —
even with the same vowels and consonants:
(54) Shortening/lowering as lexical contrast for bisyllabic stems
a. garázs 'garage' - garázst (id. ACC)
non-lowering + no shortening
b. parázs '(glowing) ember' - parazsat (id. ACC)
lowering + shortening
As evidenced by monosyllabic stems (cf. section (42)), it would be wrong to assume that a
preceding long vowel contradicts lowering in general:
(55) Shortening contrasting monosyllabic lowering stems
a. nyár 'summer' - nyarat (ACC)
b. vár 'castle' - várat (ACC)
Alternatively, one could claim that no lexically long vowels can appear in unstressed positions
(it would be a natural constraint to say that the inventory of available phonemes is larger in stressed
positions). This, however, cannot be right, since there are examples of monomorphemic stems in
which the unstressed long vowel is followed by a consonant cluster, therefore has to be lexically long
(cf. (4)):
(56) Unstressed lexically long vowels
a. parázna 'fornicator'
bogus cluster
b. kokárda 'cockade'
3rd type cluster
c. palánta 'seedling (plant)'
real cluster
Rendering an unstressed long vowel incompatible with lowering at the word level would also
be a wrong track to follow:
Constructional CV phonology
266
(57) Lowering after an unstressed long vowel alma - almákat 'apple-PL-ACC'
With virtually every phonologically encoded solution having been discarded, it is imperative to
turn to lexical constraints. It simply has to be stipulated that long vowels in unstressed positions of
lowering stems are not available.
(58) Constraint on vowels in lowering stems (I)
Long vowels cannot occur in the lexical representations of lowering stems.
This reflects that the exceptionality of lowering stems manifests itself in their more restricted
phonotactics relative to the productive pattern. The general ban on unstressed long vowels in the
lexicon cannot be extended to stressed vowels:
(59) Lexically long vowels in lowering stems
a. vár- várat 'castle'
b. tárgy- tárgyat 'object'
c. sátor - sátrat 'tent'
There is another lexical gap, however, namely that of shortening bisyllabic stems with a long
first vowel. Even when the vowel (e.g., a) in the first syllable could be represented as long lexically,
its occurrence is banned in shortening stems.
(60) Hypothetical stem ruled out by a constraint on across-the-board shortness
a. *mádály - mádalyat b. *mádaly - mádalyak
This contrast could be explained if we posited that the occurrence of lexically long vowels is
restricted even in stressed positions. They are allowed under the condition that they appear as
licensors in (at least some) forms of the paradigm (which is the case in (59)a-c):
(61) Constraint on vowels in lowering stems (II)
Lexically long (stressed) vowels of lowering stems are involved in trochaic licensing in some
forms of the paradigm.
This is illustrated below:
(62) Lowering and lexically long vowels sátrak sátor + k(PLUR) 'tents' (cf. (49))
As already discussed in the previous subsection, we also have to give an account of why the
vowels in shortening stems remain short as opposed to the general tendency of lengthening. This can
only be achieved by lexical stipulation. One could claim that shortness is forced across-the-board in
the whole stem in the synthetic paradigm of lowering stems.77
77
The straightforward tendency of across-the board shortening as well as the compulsory shortening of the second syllable is
exemplified by the dialectal lowering (and shortening) alternative of gúnár 'gander' (see (a) below), which has a short vowel
in the first syllable too, as well as the ironic non-standard (lowering) plural of tanár 'teacher' (see (b) below), where the
second long vowel shortens:
a.%gunarak 'ganders'
b.'.-()tanarak 'teachers'
Constructional CV phonology
267
Shortening, i.e., for a lexically short vowel to remain short, then, would be thought of as the
canonical, rather than the exceptional case. The long vowel of the base form madar and the
superessive case form madaron vowel should be enabled independently. If one does not wish to
explicitly restrict the ban on short vowels to an arbitrary class of synthetically suffixed forms (i.e., the
class excluding exactly the nominative and superessive), one should make it an option available only
as a 'last resort' at the word level.
(63) Situations triggering last-resort lengthening
a. madár ('bird'NOM)
end of domain
b. madáron ('bird'SUPERESSIVE)
followed by floating U
It seems obvious on the analogy of epenthetic stems as well as from the following data that the
restriction on last-resort lengthening refers only to the stem domain:
(64) Long vowels in suffixed forms nyara 'his/her summer' - nyarát (id. ACC) non-analytic
boundary nyara 'his/her summer' - nyarának (id. DAT)
analytic boundary
(65) Parsed form of shortening stems nyarak 'summers'
Real clusters in lowering stems If the long-vowel construction is thought to be a subtype of
(more specific than) the trochaic (cf. section 1.5), then it is relevant to ask whether the other subtypes
of the same construction are also made unavailable within the paradigm of lowering stems. By saying
that, similarly to the long-vowel construction, the real-cluster construction is not allowed in the lexical
representation and on the word level they parse the form only as a last resort, we obtain an immediate
explanation for the absence of real C clusters in lowering stems. This ban on real clusters in the
lexicon is also supported by across-the-board facts similar to that of vowels, i.e., real clusters do not
occur inside lowering stems.
Whatever way one would choose to explain that real clusters are not lowering (cf. (53)),22 we
have to note that actually real clusters can be followed by low-vowel epenthesis (i.e., can be lowering)
in suffixed forms:
(66) Real clusters with lowering in suffixed forms
a. karunkat (kar + POSS-1ST-SING + ACC)
‘our arm’
partial geminate in nominal inflection
b. csöngeni (csöng + INF) ‘to ring (intr.)’
partial geminate and verbal lowering suffix
c. hallottam (hall + PAST +1ST-SING)
‘I heard’ true geminate as verbal inflection
This is again a very strong argument for distinguishing between lexical (monomorphemic) and
word level constraints as well as against the phonological encoding of the original implication in (53).
268
Constructional CV phonology
Now we might turn back to the problem of the ambiguity of liquid + C clusters. Here, the
parsing of lowering stems ending in liquid + C clusters should be represented as containing no real
cluster in the lexicon (cf. the contrast between hold 'moon' (lowering) and zsold 'mercenary's pay'
(non-lowering)). In order to block epenthesis, the trochaic construction that licenses the empty V
between the Cs should be lexically present, in agreement with our earlier decision. Note that without
real-cluster licensing, the domain final C would remain unlicensed, therefore the word level parsed
form is achieved by the last-resort application of the real-cluster construction.
(67) Liquid + C stems
a. lowering hold ‘moon’
a. lexical
b. word-level
b. non-lowering zsold ‘mercenery’s pay’
a. lexical
b. word-level
In addition to our account of the static generalization on lowering stems, we might also hope to
explain why exceptional licensing triggers mandatory epenthesis in the accusative, a challenge
otherwise solvable by stipulation only. Real-cluster formation involving the segments of the stem is
demoted to a last-resort option, therefore, throughout the entire paradigm, suffixation is accompanied
by a mandatory (low-vowel) epenthesis.78
As we have mentioned already, this is relevant in the accusative only, because in the other
synthetic forms vowel epenthesis occurs irrespective of the stem class (cf. (42)).
78
This account gains further support if we consider a peculiar exception among lowering stems: oldal - oldalt - oldalak 'page
NOM - ACC - PLUR'
This lexeme is exceptional as it shows a paradigmatic diversity with respect to its accusative (normal) and other
non-analytic forms (lowering). Note that oldal is also the only exception within lowering stems in terms of
containing a stem-internal consonant cluster {Id).
269
Constructional CV phonology
(68) Accusative of a lowering stem
a. ill-formed accusative with real-cluster *halt (ACC) ‘fish’ (intended)
b. well-formed accusative with epenthesis halat (ACC) ‘fish’
2.5 Synthetic suffixation
In addition to the overall picture outlined in section 2.1, there is a lot more to say about suffixation. Throughout the earlier sections, it was assumed that synthetic suffixes in the nominal paradigm
(but not the accusative) trigger mandatory epenthesis after consonant final stems. The default vowel
appearing at epenthetic positions is mid as a rule, and only the (exceptional) lowering construction
can override this by requiring a low vowel. This obligatory epenthesis can be explained if the real
trochaic construction is allowed as a last resort only79
(69) Constraint on synthetic suffixation
Synthetically suffixed forms are parsed by the trochaic construction only as a last resort.
(70) Representation of a synthetically suffixed form of canonical VC final stem
a. ill-formed suffixed form with non-last-resort trochee *laptok lap + tok(POSS-2ND-PLUR) ‘your
sheet’
b. ill-formed suffixed form with epenthesis lapotok lap + tok(POSS-2ND-PLUR) ‘your sheet’
Verbal stems and synthetic suffixation
The phonological characterization of monomorphemic verbs is probably easier than that of
nouns, since there is no phenomenon similar to exceptional licensing, and the account can be directly
applied to derived verbs as well. Since verbal stems with internal epenthesis as well as ones ending in
long vowel + C exist, everything about nouns can be carried over to verbs (cf. sections 2.2 and 2.3) .
This is to say that we have to assume extrametricality at the lexical level. Though phonotactic
79
We consider this only a tentative solution, since its interaction with other constraints in the case of epenthetic
stems and vowel final stems is not entirely clear to us.
Constructional CV phonology
270
restrictions on the end of verbs are much stricter than those on the end of nouns, they can only be
accounted for in the form of word level constraints because of extrametricality. The empirical
generalization is that verbal stems in their synthetically suffixed forms end in a consonant. This is
apparent in stem final alternations, e.g., the one called v-epenthesis:80
(71) Evidence for consonant final verbal stems: v-epenthesis
a. lő 'he shoots' - lövök 'I shoot'
b. fő 'it cooks (intr.)' - fővök 'I cook (intr.)'
shortening
non-shortening
Actually, all synthetic suffixes in the verbal paradigm which trigger v-epenthesis also trigger
vowel epenthesis after the stem. This is true in the nominal domain also, but the fact that there are
nouns ending in a long vowel which do undergo v-epenthesis in synthetic suffixation is a contrast.26
This generalization could be explained if we said that the synthetic suffixation of verbs is such that
complex licensing domains involving three CV units (or, alternatively, overlapping constructions) are
demoted to a last-resort status.
(72) Constraint on verbal synthetic suffixation
The licensing domains of a synthetically suffixed verb consist of at most two CV-units.
This constraint could explain why there is epenthesis even after stems where the direct attachment of a consonant would not violate general phonotactics. These include stems ending in a
sonorant coda, in which case a real cluster would be possible, on the one hand, and shortening vepenthetic stems, on the other.
(73) Ill-formed verbal suffixation involving large licensing domains
a. *vonk = von +k(1ST-SING) ‘I drag’ ( OK vonok)
b. *l o +k(1ST-SING) ‘I shoot’ ( OK lövök)
But why are there no verbs ending in a canonical CV cluster, i.e., a short vowel? We have to
stipulate, additionally, that the base form of verbs (the lexical stem parsed at the word level as a single
closed domain) end in a generalized trochaic construction, which is either a long vowel or a consonant
(cf. 1.5).
(74) Constraint on verbal base forms
80
Other interesting but rather complex stem alternations, which would also be relevant here, are omitted for the
sake of simplicity. We refer the reader to Rebrus (in print).
Constructional CV phonology
271
a. stem-final long vowel l o ‘shoots’
b. stem-final consonant lop ‘steals’
4
c. stem-final cluster leng ‘swings’
Pseudo-analytic suffixes
Interestingly, there is a class of verbal suffixes in Hungarian showing a dual behaviour in terms
of analyticity. These include verbal person inflections and the infinitive marker.81
Clearly, they can not be treated as normal analytic suffixes, for two reasons:
1. They can attach to the usual synthetic stem form of epenthetic verbs:
(75)
a. ugorva = ugor + va
analytic
'jumping'
b. ugranak = ugor + nak(3RD-PLUR-INDEF) 'they jump'
pseudo-analytic
2. They trigger vowel epenthesis after consonant clusters:82
(76)
a. mondva = mond + va
'saying
analytic
b. mondanak = mond + anak(3RD-PLUR-INDEF) 'they say
pseudo-analytic
Actually, the above suggest a dual behavior inasmuch as these suffixes behave differently from
analytic suffixes, hence their name: pseudo-analytic. If we were to treat them as analytic suffixes, we
could never hope to explain why they trigger vowel epenthesis in some cases. If, on the other hand,
81
Probably the verbal past-tense inflection also belongs here, but for reasons of its very complicated and unique
behavior, we ignore it in the present paper.
82
There is a tendency to violate this generalization by dropping the epenthetic vowel when the cluster is also a
good stem, e.g., bánt(a)lak 'I hurt you' (cf. pántlika 'ribbon', footnote 10).
272
Constructional CV phonology
we treated them as synthetic, we would have to find some stipulative account for the lack of
epenthesis in some suffixed stems:
(77) Contrast of synthetic and pseudo-analytic suffixes
a. l 'he shoots' - lövök 'I shoot' - l nek 'they shoot'
b. lop 'he steals' - lopok 'I steal' - lopnak 'they steal'
This suggests that pseudo-analytic suffixes constitute a special class within the class of suffixes.
There are two further aspects any analysis of this type of suffixation has to explain: on the one hand, it
is interesting that the epenthetic vowel in the case of pseudo-analytic suffixes is low, as opposed to
the canonical mid vowel (cf. (78)).83
On the other hand, pseudo-analytic suffixes show a unique effect of paradigmatic integrity.
This is to say that, out of the two phonologically well-formed alternatives of an epenthetic stem, the
one that matches the third-person singular base form is chosen. This is relevant in the case of a class
of verbs the base form of which is not the same as the third-person singular form, the latter derived by
exceptional non-zero suffixation by -ik (hence they are traditionally called '-ik- verbs'):
(78) Paradigmatic integrity effect in pseudo-analytic suffixation
a. ugrik ‘he jumps’ epenthetic -ik-verb
ugrani/
ugorni ‘to jump’
ugranak/
ugornak ‘they jump’
*ugralak/OK ugorlak ‘I jump (above) you’
ugra[ssz]/
ugorsz ‘you jump’
ugrotok/
ugortok ‘you (plur) jump’
ugrott/*ugort ‘he jumped’
ugrottam/
ugortam ‘I jumped’
b. söpör ‘he sweeps’ canonical epenthetic verb
*söpreni/OK söpörni ‘to sweep’
*söprenek/OK söpörnek ‘they sweep’
*söpörlek/OK söpörlek ‘I sweep (above) you’
*söpre<ssz>/OK söpörsz ‘you sweep’
83
Actually, it is mid in the case of -tOk (e.g., mondotok 'you(plur) say') but, in accordance with earlier assumptions, this can be attributed to the floating U element in the suffix, which really does not occur in the others.
273
Constructional CV phonology
*söprötök/OK söpörtök ‘you (plur) sweep’
*söprött/OK söpört ‘he sweeped’
*söpröttem/OK söpörtem ‘I sweeped’
It is also worth mentioning that, since verbs can end in virtually any consonant, and the pseudoanalytic suffix attaches to them without epenthesis, virtually any (bogus) consonant cluster could
emerge. This in principle could result in an OCP violation, like in the case of a whole lot of analytic
suffixes (cf. (30)). But as pseudo-analytic suffixation takes place within a synthetic domain, it is not
surprising that each of these suffixes start in alveolar consonants (/t,n,sz,l/) which are exempt from the
OCP.
Is there any hope to derive the special behaviour of these suffixes within other synthetic ones?
In the light of the fact that all pseudo-analytic suffixes contain a lexically associated vowel (i.e., they
constitute syllables themselves),84 our account of verbal synthetic suffixation is ready with an
explanation of the lack of epenthesis at the suffixation boundary in some cases. Since there is no need
to license the consonant in the suffix, licensing domains are not required to form across the
stem/suffix boundary, therefore there is no need for epenthesis in the simple cases (cf. (77)).
Epenthetic stems are peculiar, however, since in principle they allow two equally well-parsed
variants. Paradigmatic integrity determining the choice, then, needs to be stipulated additionally,
anyway. 85
(79) Pseudo-analytic suffixation
This, apparently, is not the whole story, however, and there remain a great deal of phenomena
to explain in relation to pseudo-analyticness. The exhaustive typology of suffixes and suffixation
types is still to be explored, and it is as yet a challenge for our future research. We believe, though,
that the ideas advocated throughout this paper can serve as a promising framework for the future
analysis of a wider range of Hungarian morphophonological phenomena.
References
Harris, John (1990) Segmental complexity and phonological government. Phonology 7, 255-300.
Harris, John (1994) English Sound Structure. Blackwell, Oxford.
Hulst, Harry G. van der & Grazyna J. Rowicka (1997) On some parallels between (un)stressed vowels and
(un)realized empty nuclei. In: Booij & van de Weijer (eds.) Phonology in progress - progress in phonology.
Holland Academic Graphics, the Hague. 125-49.
Itô, Junko and Armin R. Mester (1995) Japanese phonology. In John A. Goldsmith (ed.). The handbook of
phonological theory. Cambridge, Mass. & Oxford: Blackwell. 817--838.
Kaye, Jonathan D. (1995) Derivations and Interfaces. In: Durand, J. -- F. Katamba (szerk.) Frontiers of
phonology: atoms, structures, derivations. Longman. London. 289-332.
Kaye et al. (1985): Kaye, Jonathan D., Jean Lowenstamm, Jean-Roger Vergnaud: The internal structure of
phonological elements: a theory of charm and government. Phonology Yearbook 2, 305-28.
84
The exceptions are geminate /s/ and /t/, showing exceptional behavior, anyway. It is also true that all canonical
synthetic suffixes consist of a single consonant.
85
The uncertain grammatical status of these examples corroborate our analysis based on the idea that a basic
phonological optionality is resolved by an independent principle.
274
Constructional CV phonology
Kaye et al. (1990): Kaye, Jonathan D., Jean Lowenstamm, Jean-Roger Vergnaud: Constituent structure and
government in phonology. Phonology 7, 193-231.
Lakoff, George (1995) Cognitive phonology. In Goldsmith (ed.) The last phonological rule. The University of
Chicago Press. Chicago and London.
Lowenstamm, Jean (1996) CV as the only syllable type. In Durand, J. & B. Laks (szerk.) Current trends in
phonology: models and methods. Vol. 2. 419-41. Salford, Manchester.
Polgárdi Krisztina (1999) Hungarian is strict CV. Manuscript.
Rádai, Gábor (2000) Implementing Construction Grammars in GIN. To appear in Huba Bartos (ed.). Papers on
the mental lexicon.*
Rebrus, Péter and Miklós Törkenczy (1998) Phonotactics and the morphophonology of the Hungarian word.
Talk given at the International Conference on the Structure of Hungarian 4, Pécs.*
Rebrus Péter and Törkenczy Miklós (1999) Defektivitás. Talk given to the Budapest Phonology Circle, April
1999.*
Rebrus Péter (in print) Morfofonológiai jelenségek a magyarban [Morphophonological phenomena in
Hungarian]. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 3. kötet. Alaktan [Hungarian structuralist
grammar. Vol. 3. Morphology]. Akadémiai Kiadó, Budapest.
Rebrus Péter (to appear) Kormányzás-fonológia kormányzás nélkül [Government Phonology without
government]. Proceedings of the 3rd Hungarian Linguistics Phd Conference. Szeged. JATE.
Rebrus Péter (1999) A hely-elemek egyeztetése a CV-fonológiában [Agreement of place elements in CVphonology]. Proceedings of the recent methods in research on the Hungarian language conference.
Rowicka, Grazyna (1999) On Ghost Vowels. A Strict CV Approach. Doctoral dissertation, Univ. of Leyden.
LOT Dissertations; 16, Holland Academic Graphics, The Hague.
Ritter, Nancy N. (1995) The Role of Universal Grammar in Phonology: a Government Phonology Approach to
Hungarian. Doctoral dissertation. New York University. Manuscript.
Scheer, Tobias (1998) A unified model of phonological government. The Linguistic Review
Szigetvári Péter (2000) Why CVCV. The Even Yearbook. ELTE SEAS. Budapest. 117-52.
Szigetvári, Péter (2000) Deconstructing syllable structure. Ms., Eötvös Loránd University. To appear in Huba
Bartos (ed.). Papers on the mental lexicon.*
Törkenczy Miklós (1992) Vowel-zero alternations in Hungarian: a government approach. In: Kenesei István -Pléh Csaba (szerk.) Approaches to Hungarian 4: The structure of Hungarian. JATE. Szeged. 157-76.
Törkenczy Miklós (1994) A szótag. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 2. kötet. Fonológia.
Akadémiai Kiadó, Budapest.273-392.
Törkenczy, Miklós (2000) Phonotactic grammaticality and the lexicon. To appear in Huba Bartos (ed.). Papers
on the mental lexicon.*
Kormányzás-fonológia kormányzás nélkül
275
Kormányzás-fonológia kormányzás nélkül
Rebrus Péter
Az alábbi cikk célja az, hogy a hagyományosan szótagszerkezetinek nevezett jelenségek
elemzésére egy újfajta fonológiai reprezentációt adjon. Kiindulópontunk a kormányzás-fonológia nev
fonológiai elmélet (l. KLV (1985), (1990), Harris (1990), (1994)), melynek két központi fogalma a
kormányzás (government) és az engedélyezés (licensing). Ezek fonológiai pozíciók között definiált
relációk, melyek a prozódiai reprezentációra és a szegmentális tartalom interpretációjára írnak el
feltételeket. A kormányzás-fonológia továbbfejlesztése az ún. szigorú CV-fonológia (l. Lowenstamm
(1996)), mely a prozódiai szerkezetnek a korábbinál egységesebb és egyszer bb reprezentációt ad.
Ennek érdekében az elmélet elveti az elágazó összetev ket; így a rím, a szótagmag és a szótagkezdet -mint szótagszerkezeti összetev k -- eredeti jelentése értelmét veszti. A prozódiai szerkezet ennélfogva
kizárólag kezdet és mag (azaz C V) szekvenciákból áll.
A cikk két részb l áll: az 1. részben a CV-fonológián belül vizsgáljuk meg a mássalhangzókapcsolatok viselkedését, és megmutatjuk, hogyan lehet a kormányzás-fonológia egyes fogalmait a
CV-fonológiában értelmezni. Az elemzések a szerz nek a magyar morfofonológiáról írt dolgozatán
(Rebrus (megj. alatt)) alapulnak. A 2. rész a mássalhangzó-kapcsolatok további vizsgálatából
kiindulva elemzi a szegmentális tartalommal nem rendelkez magánhangzó-pozíciók interpretációs
feltételeit. Az eredmény a CV-fonológiának egy olyan felfogása, amelyben az engedélyezés típusai a
lehet legegyszer bben vannak megadva; és ez a kormányzásnak mint speciális engedélyezési
relációnak a megszüntetése révén egyszer bben implementálható fonológiai reprezentációt
eredményez
A mássalhangzók engedélyezése
Valódi mássalhangzó-kapcsolatok
A kormányzás-fonológia bizonyos fogalmainak kifejezhet eknek kell lenniük a CV-fonológia
keretein belül is: ilyenek azok, amelyek a nyelvekben megfigyelhet szótagszerkezeti (prozódiai)
összefüggéset magyarázzák. Egy ide tartozó jelenségkör a mássalhangzó-kapcsolatoknak a nem
egységes viselkedése. A mássalhangzó-kapcsolatok szempontjából a nyelvek tipizálhatók: egyes
nyelvekben nincsenek mássalhangzó-kapcsolatok (ezek az ún. CV-nyelvek); más nyelvekben csak
bizonyos mássalhangzó-kapcsolatok lehetségesek (ilyen nyelvek pl. az ún. Prince-nyelvek, melyekben
csak gemináta vagy nazális + homorgán obstruens lehetséges). Vannak nyelvek, melyekben
intervokalikusan szinte minden mássalhangzó-kapcsolat el fordul86, szóvégen (vagy t végen) azonban
a lehetséges kapcsolatok korlátozottak (ilyen nyelv a magyar). Azokat a mássalhangzó-kapcsolatokat,
melyek szó végén és intervokalikusan is el fordulnak, valódi kapcsolatoknak, azokat, amelyek csak
intervokalikusan fordulnak el , álkapcsolatoknak87 hívjuk. A magyarban valódi kapcsolatnak
számítanak a gemináták, a részleges gemináták (homorgán nazális + C), és a likvida + C
kapcsolatok88.
A kormányzás-fonológia a fenti különbséget úgy ábrázolja, hogy a valódi kapcsolat els eleme
az el z rím-összetev módosítója (a rím feje a mássalhangzó-pozíció el tti szótagmag), A kapcsolat
második eleme pedig kezdet (l. (6a)). A rím módosítóját azonban az utána álló kezdetnek
engedélyeznie kell, melynek következménye egyfel l az, hogy ilyen pozíció csak mássalhangzó el tt
lehetséges (szó végén vagy magánhangzó el tt nem); másfel l az engedélyezésre szegmentális
feltételek vonatkoznak, amelyek korlátozzák az engedélyezési viszonyba lép magánhangzók
86
Valójában a mássalhangzók egymásutánjára a fentiekt l független korlátozások vonatkozhatnak, mint a
kötelez hasonulások vagy az azonos jegyek egymás melletti el fordulását tiltó kötelez kontúrelv (OCP) stb.
87
88
Pontosabban, ezek közül azokat, amelyek szó(tag) elején nem fordulhatnak el .
A három típuson kívül további szóvégi kapcsolatok vannak, részletekre itt nincs mód kitérni, l. Rebrus
(el készületben)
Kormányzás-fonológia kormányzás nélkül
276
min ségét. Az álkapcsolatban mindkét mássalhangzó kezdet-pozícióban áll, közöttük nincsen
engedélyezés. Van viszont egy nukleáris pozíció, melyhez nincs rendelve szegmentális tartalom (ez az
ún. üres mag). Az üres magok interpretációját az engedélyezési viszonyok határozzák meg: az
engedélyezett üres mag nem kap interpretációt, míg a nem engedélyezett üres magnak kötelez en meg
kell jelennie. Az engedélyezés ebben az esetben az ún. alapos kormányzás révén valósulhat meg.
Alapos kormányzás csak a megjelen mag és az el tte álló üres mag közott lehetséges jobbról balra (l.
err l (6b), részletesen KLV (1990). Ennek az ábrázolásnak a következménye az, hogy álkapcsolat csak
intervokalikusan állhat el , hiszen egyébként alapos kormányzás híján az üres magnak interpretálódnia
kell.
(6) Mássalhangzó-kapcsolatok ábrázolása a kormányzás-fonológiában
a. valódi kapcsolat
b. álkapcsolat
mag: kormányoz (→)
kezdet: engedélyez (←)
(üres magot a mögötte álló
alaposan kormányozza)
R
|
N
←
N
N
O
N
N
O
N
O
N
|
|
|
|
|
|
|
|
x
x
x
x
x
x
x
→
x
←
x
|
|
|
|
|
|
|
a
l
t
a
t
l
a
A valódi kapcsolatok fenti elemzése azonban következetlen két szempontból is. A kormányzásfonológiában a kormányzás olyan speciális engedélyezési reláció, mely általában (i) összetev n belül a
fej és a módosító között van adva, (ii) szegmentális hatása van. Az (i) feltétel biztosítja, hogy ne
legyenek fej nélküli elágazó konstituensek, a (ii) feltétel pedig konceptuális különbséget tesz a
kormányzás és a többi engedélyezési reláció között. A (6a)-beli valódi kapcsolatok esetében a két
feltétel egymásnak ellentmond: a rím mint összetev feje (a mag) és módosítója (a mássalhangzókapcsolat els eleme) között általában nincs szegmentális fonotaktikai kapcsolat. Vannak viszont
fonotaktikai korlátozások a mássalhangzó-kapcsolat két eleme között, ezek viszont nem alkotnak egy
összetev t. Ennek következtében a mássalhangzó-kapcsolatok esetén az (i) és (ii) feltételt egyszerre
nem lehet kielégíteni. A klasszikus kormányzás-fonológiai elemzés az (i) feltételt részesíti el nyben
(ii)-vel szemben, melynek következtében (6b)-ben a mag intrakonstituens kormányzási relációban van
a rím módosítójával szegmentális következmény nélkül, a kezdet pedig engedélyezési viszonyba lép
ugyanezzel (ez az ún. ”kóda”-engedélyezés, l. Kaye (1990)), mely viszonynak azonban a
szegmentális tartalomra nézve következményei vannak. A kormányzás-fonológiában a szótag mint
összetev ellen szóló egyik érv éppen ehhez hasonló volt: mivel a kezdet és a mag között nincs
szisztematikus fonotaktikai kapcsolat, ezért nincs szükség a szótag szerepeltetésére az elmélet
primitívumai között. Ezt a gondolatmenetet a rímre is alkalmazhatjuk: ha a valódi mássalhangzókapcsolatokban nem tételezünk fel rím-összetev t, akkor az (i) feltétel nem akadályozza meg, hogy a
V→CC szekvenciában a jelölt relációt engedélyezésnek (és ne kormányzásnak) tartsuk.
A két mássalhangzó közötti kapcsolat ezzel szemben alapvet en szegmentális. A valódi
kapcsolatban részt vev két mássalhangzó között szonoritási és hasonulási kapcsolatokat találunk. Ezt
a folyamatot a legcélszer bb az autoszegmentális tengelyeken kifejezni, azaz a prozódiai ábrázolásban
eddig feltételezett “kóda”-engedélyezést felválthatjuk egy újfajta kormányzási viszonnyal, mely a
Kormányzás-fonológia kormányzás nélkül
277
szegmentális tartalmak között van adva. Ez az ábrázolás azt is megengedi, hogy a két C-pozíció
között az álkapcsolatokhoz hasonlóan a valódi kapcsolatokban is üres V-t tételezzünk fel. Hiszen a két
C-pozíció szegmentális tartalma a köztük lev üres V-pozíció ellenére szomszédos lesz. Az alábbi (7)
ábrázolás ezt a szigorú CV-elméleten belüli elemzést mutatja. (Az engedélyezési viszonyt →, a
kormányzási viszonyt ←, a kormányzási tartományt [ ] jelöli. A kezdet, illetve mag-pozíciók
jelölésére a C, illetve a V bet t használjuk.)
(7) Mássalhangzó-kapcsolatok ábrázolása CV-fonológiában
a. valódi kapcsolat
b. álkapcsolat
(nincs rím, a kezdetek szegmentális
(ugyanaz, mint (6b))
tartalma között kormányzás van)
V ← V
V
→
C
|
|
a
[l
V
←
C
V
V
→
C
|
|
|
t]
a
t
V
C
|
l
V
|
a
Ekkor a (7a)-beli valódi kapcsolat és a (7b)-beli álkapcsolat reprezentációja közötti különbség
nem az összetev s szerkezetben lesz, hanem abban, hogy van-e a C-pozíciók szegmentális tartalma
között kormányzás.89
A kezdetek közötti kormányzási reláció (interonset government) feltételezése nem új,
megtalálható többek között a lengyel mássalhangzó-kapcsolatok elemzésében (l. Gussman & Kaye
(1993), de magyar adatokkal is alátámasztható. A magyarban van két olyan toldalék, amelynek
felszíni alakja a t végi mássalhangzótól függ: ha a t végi mássalhangzóval a toldalék képes valódi
kapcsolatot alkotni, akkor ez a kapcsolat jelenik meg szó végén. A tárgyrag, például, ha a t végi
mássalhangzó koronális szonoráns vagy folyamatos hang (r, l, j, n, ny, s, zs, sz, z) köt hangzó nélkül,
ha a t vég más mássalhangzó, akkor köt hangzóval jelenik meg.90 A hagyományos kormányzásfonológiai elemzés szerint a t alak végén álló C nem lehet a rím része (8a), hiszen utána nem áll C,
ami “kóda”-engedélyezni tudná (l. (6a)). Így a t alakvégi C kezdet-pozícióban áll, l. (8c); ezzel
szemben a köt hang nélküli toldalékolt alakban a t végi C a rím része (l. (8b)), hiszen ha kezdet lenne,
akkor alapos kormányzás híján az utána álló V-nek meg kellene jelennie (l. (8d)). Ennélfogva a t alak
a (8c), a tárgyragos alak a (8b) ábrázolást kapja, ez viszont a projekciós elvet sérti, mely szerint a
prozódiai szerkezet nem változtatható meg. Így -- ha a projekciós elvet változatlan tartani akarjuk -- a
klasszikus kormányzási fonológia is arra kényszerül, hogy kezdet-pozíciók közötti viszonyt tételezzen
fel. A CV-elemzés a fenti problémát megoldja, hiszen a toldalékolt alak nem tartalmaz rímet, ezért a
projekciós elv nem akadályozza meg, hogy a t alakot és a toldalékolt alakot egységes prozódiai
szerkezettel ábrázoljuk.91
89
Ezt a viszonyt jogosan hívhatjuk kormányzásnak, hiszen a kormányzásra vonatkozó el bbi (ii) feltétel teljesül,
az (i) feltétel viszont a CV-fonológiában -- nem lévén elágazó szerkezet -- értelmét veszti.
90
Nyitótöveknél -- néhány kivétellel -- mindig van köt hangzó, err l a jelenségr l még szólunk.
91
A múlt id jelöl je ehhez hasonlóan viselkedik, bizonyos eltérésekkel, melyre itt nem térünk ki.
Kormányzás-fonológia kormányzás nélkül
278
(8) A t végi msh. min ségére érzékeny toldalékolás
a. t alak (rímmel)
b. toldalékolt alak (rímmel)
R
R
|
|
O
N
|
|
x
x
|
|
d
a
*
→
N
O
N
N
|
|
|
|
x
x
x
|
|
|
|
l
d
a
l
x
c. t alak (rím nélkül)
→
x
←
x
|
t
d. toldalékolt alak rímmel
N ←*N
O
N
O
N
|
|
|
|
|
|
|
|
|
|
x
x
x
x
x
x
x
x
x
x
|
|
|
|
|
|
|
d
a
l
d
a
l
t
→
O
N
O
N
O
N
*dalot
A kóda engedélyezése
A CV-fonológiában a valódi és álkapcsolatok prozódiai reprezentációja azonos: mindkett
kizárólag CV-szekvenciákat tartalmaz. Az egységes ábrázolás alkalmat ad arra, hogy egységes
magyarázatot adjunk azokra a folyamatokra, amelyek mindkét típusú mássalhangzó-kapcsolatokat
egyaránt érintik. Ilyen folyamat a magyarban a V + n + folyamatos C szekvenciákban lejátszódó
fakultatív nazalizáció92. Ez félig nazalizált hosszú magánhangzót hoz létre a nazális mássalhangzó
törlése mellett, pl. vonhat ~ v[o:~]hat, vonsz ~ v[o:~]sz, vonlak ~ v[o:~]lak, Kunság ~ K[u:~]ság, kanra
~ k[a:~]ra, régens ~ rég[ε:~]ns. A nazalizáció független attól, hogy az n valódi vagy álkapcsolatot
alkot a rákövetkez mássalhangzóval (vö.: ns, nsz, ill. nh, nr). Ennek CV-fonológiai elemzése a
következ : a nazalitás N eleme vagy C-pozícióban jelenik meg alveoláris nazálisként (9a), vagy az
üres V-pozícióba kerül, amelyben aztán az el z V-pozíció szegmentális tartalmával együtt hosszú
nazalizált magánhangzóként realizálódik (9b).
92
A folyamat lezajlása a környezett l és a nyelvi regisztert l függ, részletesebb leírást ad Siptár XXX.
Kormányzás-fonológia kormányzás nélkül
279
(9) Az n interpretációi folyamatos mássalhangzó el tt
a. mássalhangzóként
b. nazalizált magánhangzóként
V C V C ~
V C V C ~
| |
|
|
| |
| N
|
|
N |
α
β
[α
]β
interpretációja:
[α n β]
[α:~ β]
A fentihez hasonló a pótlónyúlás jelensége. Ez a folyamat V + l/r szekvenciákban játszódhat le:
a mássalhangzó elt nik, a magánhangzó megnyúlik, függetlenül attól, hogy a likvida milyen
mássalhangzó-kapcsolatot alkot: pl. arra ~ [a:]ra, elmegy ~ [ε:]megy 93. Az interpretációs lehet ségek
a (9)-hez hasonlók: a likvida szegmentális tartalma vagy a C-pozícióhoz kapcsolódik (10a), vagy nem
kapcsolódik sehova, ekkor az üresen maradt CV-szekvencia V-pozícióját a magánhangzó
szegmentális tartalma foglalja el (10b).
(10) Likvida interpretációi magánhangzó után
a. mássalhangzóként
V
C
|
|
V
b. hosszú magánhangzóként
~
V
C
V
~
|
| l/r
α
|
(l/r)
[α
]
interpretációja:
[α l] vagy [α r]
[α:]
Abban az esetben, ha a nazális vagy likvida el tt álló magánhangzó eleve hosszú, kizárólag
nazalizáció, illetve törl dés zajlik le, pl. színház ~ sz[i:~]ház, szánra ~ sz[a:~]ra ill. állat ~ [a:]lat. Ezt
úgy magyarázhatjuk, hogy a reprezentációból egy prozódiai egység, egy egész CV-szekvencia
interpretálatlan marad (a (11)-beli ábrákban zárójelbe téve), ezt a folyamatot prozódiai
egyszer södésnek nevezzük.
93
l. a 7 lábjegyzetet.
Kormányzás-fonológia kormányzás nélkül
280
(11) Prozódiai egyszer södés hosszú magánhangzó után
a. nazalizáció
b. likvida-törlés
V C V(C V)C ~
V C V(C V)~
|
|
|
|
|
N
|
|
[α
]
β
(l/r)
[α
]
interpretációja:
[α:~ β]
[α:]
A prozódiai egyszer södés, a degemniáció elemzésében is szerepet kap. A degemináció olyan
hármas mássalhangzó-kapcsolatban zajlik le, amelyben valamely két szomszédos mássalhangzó
azonos, pl. kapott -- kapta, sa[k]tól, gom[b]an. A folyamat elemzése azt az elképzelést használja fel,
mely szerint a gemináták (melyek valódi kapcsolatok) szótári ábrázolásában az els C-pozíció nem
rendelkezik szegmentális tartalommal, azaz üres C (l. Harris (1994)). Abban az esetben, ha a gemináta
el tt magánhangzó áll, ezt az üres C-t az utána álló teli C kormányozza, a valódi kapcsolatoknál
megszokott módon (l. (7a)). A kormányzás következménye az, hogy az üres C interpretációja a teli Cvel megegyez lesz, azaz együtt hosszú mássalhangzóként jelennek meg, l. (12a). Abban az esetben,
ha a gemináta el tt mássalhangzó (azaz a CV-szerkezetben üres V) áll, akkor a gemináta els üres Cpozíciója és az utána álló üres V nem interpretálódik (l. (12b))94. Az egyszer södés kiváltó okát az
engedélyezési viszonyokban kereshetjük: (12a)-ban a teli V-pozíció a gemináta els C-pozícióját
képes engedélyezni, az eredmény hosszú mássalhangzó (kapott). (12a)-ban -- ahogyan (7a)-ban is -ezt jobbra mutató nyíllal (V→C) jelöltük. Abban az esetben azonban, ha a gemináta el tt üres Vpozíció áll, az engedélyezés nem lehetséges. A degeminációt elszenved alakban pontosan ez a
helyzet: az egész alak csak akkor lehet jólformált, ha minden pozíció engedélyezve van, ennek egyik
lehetséges megvalósulása (12b): a gemináta els elemét tartalmazó CV-szekvencia nem
interpretálódik: ekkor a múlt id jele utáni magánhangzó képes a t utáni V-t alaposan kormányozni,
melynek következtében az nem jelenik meg.95
94
A degeminációnak az az esete, melyben nem a gemináta el tt, hanem utána áll üres V (pl. sakktól), hasonlóan
elemezhet .
95
Valójában a köt hangzós, geminátát tartalmazó alakban (*kapotta) is minden pozíció engedélyezve van, ez az
alak azonban nem grammatikus. Ennek oka talán az, hogy a köt hangzó megjelenése “drágább” mint a prozódiai
egyszer södés (N.B. CC-vég igéknél ebben a környezetben sincs mindig egyszer södés, pl. ajzotta, ugrotta,
csuklottam, hiszen itt az egyszer södés révén olyan CCC állna el , amelyben nem lenne minden pozíció
engedélyezve). A részletes vizsgálattól itt eltekintünk.
Kormányzás-fonológia kormányzás nélkül
281
(12) Gemináta interpretációja
a. teli V után
b. üres V után
(hosszú msh.)
(degemináció)
V ← V
C V C V→C V C V
C V→C V(C V)C V
| | |
| | |
k a p o[
kapott
|
←
t]
| |
k a p
t a
kapta
A fenti elemzésben feltételeztük, hogy a mássalhangzókapcsolatok els elemét az el tte álló
megjelen V-pozíciónak kell engedélyeznie.96. Így ez a viszony megfelel annak a viszonynak, melyet a
klasszikus kormányzás-fonológiában rímen belüli (intrakonstituens) kormányzásnak hívnak (l. (6a)).
Ennek a viszonynak -- mint említettük -- nincs fonotaktikai következménye, így joggal tekinthet a
prozódiai reprezentáció részének. Ezért -- és mivel a CV-fonológiában nincs rím -- a szóban forgó
relációt egyszer engedélyezésnek tekinthetjük. Másfel l, a nazalizáció és a pótlónyúlás szempontjából
a valódi és álkapcsolatok hasonlóan viselkednek, és a fonotaktikai szabályok is a hasonlóságot
mutatják. Valóban, a szonoránst és szibilánst nem tartalmazó álkapcsolatok sem állhatnak szó elején,
még akkor sem, ha intervokalikusan léteznek, pl. atka, labda, bakfis, akció, kapca, nátha de *#tk,
*#bd, *#kf, *#kc, *#pc, *#th. Néhány kivétellel ugyanez a tiltás érvényesül a mássalhangzó utáni
el fordulásra. (A szonoránst tartalmazó nem valódi kapcsolatok egy része szó elején lehetséges, pl. Cr,
Cl, Cj, Cv, Csz; egy nagyobb osztályuk el fordul mássalhangzó után, pl. partner, pántlika, arktikus.
Ezzel a jelenséggel itt nem foglalkozunk, részleteket az olvasó Rebrus (megj. alatt)-ben talál). Így
nem követünk el hibát, ha az álkapcsolatokra is érvényesnek tekintjük azt a feltételt, mely szerint
minden mássalhangzó-kapcsolat els elemét az el tte álló interpretált magánhangzónak kell
engedélyeznie. Ezt az engedélyezést úgy tekinthetjük, mint az engedélyezési elv last resort (“végs
esetben alkalmazott”) megvalósulását. Az engedélyezési elv szerint minden pozíciónak engedélyezve
kell lennie; a C-pozíciókat alapesetben az utánuk álló interpretált V-k engedélyezik, CC esetén
azonban ez nem lehetséges, ekkor az el tte álló interpretált V engedélyez. A kormányzási
fonológiában minden (akár üres) V is engedélyezheti az el tte álló C-t, így a nem valódi kapcsolatok
els elemének balról való engedélyezése jelent s változtatás. A balról jobbra való engedélyezést mivel
a hagyományosan kódának tekintett C-pozíciót engedélyezi, kóda-engedélyezésnek97 hívhatjuk:
(13) KÓDA-ENGEDÉLYEZÉS: megjelen V engedélyezi az utána álló, jobbról nem engedélyezett C-t
A kóda-engedélyezésre további érvet szolgáltat a mássalhangzó-kapcsolatra (is) végz d
igetöveknek az ún. kvázianalitikus toldalékok el tti viselkedése. A jelenség a következ 98: egyes igei
toldalékolt alakok köt hangzója ún. szintetikus toldalékok el tt nem jelenik meg, az ún. analitikus
toldalékok el tt megjelenik. Egyes, ún. hangkivet tövekben a köt hangzó megjelenése maga után
vonja a t belseji V elt nését. Az igei toldalékok kvázianalitikus csoportja (-nAk, -lAk, -sz, -na, -ni, 96
A magyar fonotaktikai szabályok azt mutatják, hogy semmilyen valódi kapcsolat nem állhat sem
mássalhangzó után, sem szó elején. Itt eltekintünk a szibilánsoktól (pl. e[kszt]ra, kunszt, o[b]struál, instant) és a
két szonoránsra végz d tövek tárgyragos alakjaitól (fájlt, görlt, konszernt, Kölnt stb.)
97
Ne keverjük össze a klasszikus GP-beli “kóda”-engedélyezéssel, amelyben a rím módosítóját az utána álló
kezdet engedélyezi jobbról balra, l. (6a). Ez utóbbi reláció a mi elképzelésünk szerint a CV-fonológiában
kormányzás, l. (7a).
98
A részletes vizsgálatot l. Rebrus -- Törkenczy (1998), (1999)-ben és Rebrus (megj. alatt)-ben.
Kormányzás-fonológia kormányzás nélkül
282
tOk) el tt a köt hangzó a t osztálytól függ en vagy a szintetikus, vagy az analitikus csoporttal
megegyez en viselkedik. l. (14).
(14) Igei hangkivetés kvázianalitikus toldalékok el tt
E/3.(alapalak) szintetikus
a. hangkivet , nem ikes:söpör
kvázianalitikus analitikus
söpröm söpörni söpörhet
b1. hangkivet , ikes
(ugor)
ugrom
b2.
ugrik
ugrom ugrani
--
siklani
--
c. defektív (ál-CC vég )siklik
d. valódi CC-vég
siklom
tart
ugorni
tartom
ugorhat
tartani
tarthat
A releváns általánosítás a következ : ha a t az alapalakban (ez általában az E/3. alak) CC-re
végz dik (14b2, c. és d. t osztályok), akkor minden alakban CC-re végz dik. A CV-fonológiában ez egy
paradigmatikus szinten érvényesül szabályban fogalmazható meg, l. (15).
(15) KÓDA-INTEGRITÁS: az alapalak kóda-engedélyezése az összes alakban megmarad
A kvázianalitikus alakokban a (14a,b1) t osztályok esetén a toldalék magánhangzója alaposan
kormányozza a t és a toldalék közötti üres V-t, így ez nem jelenik meg, és nem képes alaposan
kormányozni a t belseji V-t, így a t belseji V megjelenik ((16a)). Ezzel szemben a (14b2,c,d)
t osztályba tartozó alakokban a kóda-integritási feltétel megakadályozza a kvázianalitikus alakok
el z vel párhuzamos képzését: a grammatikus alakban a t és a toldalék közötti V megjelenik, így
alaposan kormányozza a t belseji V-t, ami így nem jelenik meg, tehát nem képes engedélyezni az
el tte álló C-t. A kóda-engedélyezés így életbe lép (16b).
(16) Kvázianalitikus alakok
a. nincs t belseji kóda-engedélyezés
V
V←/V
←V
b. van t belseji kóda-engedélyezés
V
V←V←/V
C V C←V C V C V
C V→ C V C V C V
| | |
|
| |
| | |
|
| |
s ö p
r
n i
s i k
l
n i
söpörni vö.: söpör
(l.még ugorni, de *sikolni)
siklani
vö.: siklik
(l.még tartani, ugrani,de*söpreni)
A fenti elemzés magyarázatot ad arra, hogy három különböz igei t osztály kvázianalitikus
alakjaiban miért kötelez a köt hangzó megjelenése: a (15) kóda-integritás csak akkor elégíthet ki, ha
a V-projekción.az alapos kormányzás (16b)-beli sorrendben, és nem a (16a)-beli sorrendben megy
végbe. A defektív (14c) t osztályban a t vég mindig álkapcsolat (pl. siklik, hámlik, sejlik, kétli), a
Kormányzás-fonológia kormányzás nélkül
283
(14d) osztályba tartozó tövekben a t vég valódi kapcsolat (tart, ajz, mond, küld stb.), a hangkivet
(14b2) osztályba tartozó tövekben a t utolsó két mássalhangzója alkothat valódi vagy álkapcsolatot is
(pl. ugrik, omlik, fürdik). A három osztály közül a valódi kapcsolatra végz d töv ek köt hangzójának
megjelenését a klasszikus kormányzás-fonológiában az ún. kormányzás-engedélyezés magyarázza: a
valódi CC-kben az ezek után álló mag a második C-t (kezdet) engedélyezi (l. (6a)), az engedélyezést
az els C-nek (a rím módosítójának) “kóda”-engedélyezés formájában tovább tudja adni. Ilyen
kormányzás-engedélyezésra azonban csak megjelen vagy tartományvégi üres mag képes (l. Charette
(1990)-t a francia svá megjelenésér l). Ez azonban nem ad magyarázatot arra, hogy miért kötelez a
köt hangzó a nem valódi kapcsolatra végz d tövek fenti alakjaiban is. A CV-fonológiai reprezentáció
és a kóda-engedélyezés lehet vé teszi, hogy a három osztály azonos viselkedését az alapalakok azonos
CV-szerkezetével és engedélyezési viszonyaival magyarázza, függetlenül attól, hogy a t végi
mássalhangzó-kapcsolat milyen típusú.
Az alábbi (17) táblázat összefoglalja a CV-fonológiában eddig feltételezett engedélyezési
relációkat. Alapvet elv, hogy az engedélyezés lokális, adott irányú reláció. Minden engedélyezés egy
megadott szerkezeten belül van definiálva: a magprojekció a V-ket tartalmazza, a prozódiai szerkezet
a CV szekvenciákat, az autoszegmentális reprezentáció a szegmentális tartalmat (az újításokat
aláhúzással jelöltük; int. V jelentése interpretált V).
(17) C és üres V engedélyezése
hol?
mi mit merre?
elnevezés
hatás
• üres V ← ]
tartományvégi eng.
üres V nem interpretálódik, C-vég alak
• üres V ← int. V
alapos kormányzás
üres V nem interpret.,álkapcsolat, l. (7b)
• C ← int. V
C-engedélyezés
C megjelenik, l. (12b), (16b)
• int. V → C
kóda- engedélyezés
CC megjelenik, l. (16a)
• szonoráns C ← C
kormányzás
V nem interpret., valódi CC, l. (7a)
• üres C ← C
kormányzás
V nem interpret., gemináta, l. (12a)
a. magprojekción
b. prozódiai szerkezeten
c. autoszegmentális szerkezeten
Az üres mag engedélyezése
Kivételes engedélyezés
A klasszikus kormányzás-fonológiában minden V-pozíció képes az el tte álló C-t engedélyezni.
Az el z fejezetben vázolt CV-fonológiai keret ett l eltér: a prozódiai szerkezeten csak az interpretált
V-k tudnak engedélyezni, l. (17b). A fonológiai tartomány végén álló üres V azonban a magyarban
további problémákat vet fel. Az eredeti elképzelés szerint a tartományvégi üres V is képes az el tte
álló C engedélyezésére; abban az esetben viszont ha a tartomány végér l elkerül, és nincs alaposan
kormányozva, kötelez en megjelenik. Ez történik a nem-analitikusan toldalékolt alakokban: ha a tövet
nem-analitikus toldalék követi, akkor a t végi üres V köt hangzóként realizálódhat. A magyarban
viszont a köt hangzók kétféle formában interpretálódhatnak: középs (o, e, ö) vagy alsó (a, e)
Kormányzás-fonológia kormányzás nélkül
284
magánhangzóként; azokat a töveket, amelyek után a köt hangzó a produktív mintától eltér en,
kötelez en alsó, nyitótöveknek hívjuk.99 A nyitótövek t végi V-je azonban nem csak min ségében tér el
a többi t végi V-t l, hanem engedélyezési képességük is nagyobb. Az alábbi (18a-d.)-ben olyan t vagy toldalékváltakozásokat soroltunk fel, melyek valamilyen szempontból a produktív mintától
eltérnek (b. és c. kivételes f névi t osztályokat mutat, ezekbe az osztályokba tartozó tövek mind
nyitótövek, d. inflektált igetöveket ad meg, ezek szintén mind nyitótövek). A (18a.,c.,d.) csoportokban
az a közös, hogy a nem nyitótövekt l eltér en a toldalékolt alakokban a köt hangzó mindig megjelenik.
(18b.,c.)-ben azt látjuk, hogy itt a t végi V er sebb engedélyez : alaposan kormányoz, ezzel t belseji
mgh.-rövidülést (nyarak) vagy mgh.-kiesést (tetvek) vált ki, illetve engedélyezi a t végi üres C-t,
amellyel v-betoldódást okoz (füvek, tetvek). Az er s engedélyezési képességnek fonotaktikai hatása
van: (18e.,f) olyan t -, illetve inflektált alakokat mutat, melyek végén álló CC csak nyitótövekben
fordul el .
(18) Nyitót végén a V-pozíció er sebb engedélyez
nyitót .
a. tárgyrag
b. rövidülés
nem nyitót
halat (vö.: halak)
nyár ~ nyarak
elemzés
C V C←V C V]
dalt (vö.: dalok)
| | |
|
h a l
t
V← V
várak, károk
C V C V C←V C V]
c. v-tövek
f ~ füvek
tet ~ tetvek
| |
|
|
nya
r
k
C V C←V C V]
tk
| |
|
f ü
k
V← V
bet k
C V C V C←V C V]
| | |
t e t
d. inflektált t
e. t végi Cv
Cj
99
rójanak, l jenek stb.
|
ü
k
C V C V C←V C V C V]
vájnak, fújnak
rójalak, l jelek vájlak, fújlak
| |
|
| | |
rójatok, l jetek vájtok, fújtok
r[o
]j
n a k
kedv,-szenv,könyv,elv,ölyv,örv stb.
--
C V C V C←V]
szomj,alj,fürj,ujj,gally stb.
--
| | |
|
k e d
v
Nem célunk a köt hangzók és a nyitótövek kimerít jellemzése, a teljes leírásnak összetett morfofonológiai
feltételeket kell feltárnia, melyt l hely hiányában eltekintünk.
Kormányzás-fonológia kormányzás nélkül
285
f. C+C t vég
fogj, dobj, óvj stb.
--
C V C V]C←V]
hagyd, nyomd, óvd
--
| | |
|
f o g
j
A jelenség magyarázatot nyer, ha feltételezzük, hogy a nyitótövek végén álló üres V a többi t t l
eltér en, kivételesen, képes prozódiai engedélyezésre, l. (19).
(19) Kivételes engedélyezés: nyitót végén az üres V engedélyezi az el tte álló C-t
A t alakban ez azt jelenti, hogy a tartományvégi engedélyezés folytán ez a t végi üres V nem
jelenik meg, de engedélyez. Toldalékolt alakokban azonban ez az engedélyezési viszony megmarad:
tartományvégr l elkerülve a t végi CV szekvencia minden esetben interpretálódik.
A nem nyitótövek végén álló üres V problémája azonban megmarad. A klasszikus kormányzásfonológiában a tartományvégi üres V -- mint minden üres V -- engedélyezi az el tte álló C-t. Azonban
a tartományvégi üres V-nek ezen túl kormányzás-engedélyez nek is kell lennie, hiszen a t végi valódi
kapcsolatok csak így elégíthetik ki az engedélyezési elvet. Egyes nyelvekben (pl. francia) ezért a
t végi és a t belseji üres V engedélyezési képessége eltér, hiszen itt a t belseji üres V nem kormányzásengedélyez (azaz valódi CC után nem állhat C), és ugyanezt láttuk a magyarban a kvázianalitikus
köt hangzóknál is 100. A tartománybelseji és -végi üres V-k ilyen megkülönböztetése nem t nik
vonzónak. Van azonban egy másik probléma a t végi üres pozíció engedélyezésével. Ahogyan (18)ban láttuk a nyitótövek különleges morfofonológiai és fonotaktikai tulajdonságai magyarázhatóak a
(19)-beli kivételes engedélyezéssel. A szabályos tövek t végi C-jének engedélyezése így nem
származhat a t végi üres V-t l, hiszen ekkor nem lenne különbség a kétféle t típus között. Ennélfogva a
nem nyitótövek t végi C-je a (13)-beli kóda-engedélyezéssel történik. Ez az elemzés az egy
mássalhangzóra végz d tövek esetén kielégít eredményt ad, l. (20a). A valódi kapcsolatra végz d
tövekben nyitót esetén az els C-re a kóda-engedélyezés, a második C-re a kivételes engedélyezés
vonatkozik (20b.i.). A nem nyitótövekben azonban -- kivételes engedélyezés híján -- a második Cpozíció nem lehet engedélyezve. Ezen az engedélyezés örökl désének felhasználásával sem lehet
segíteni, hiszen a valódi kapcsolatokban feltételezett kormányzás jobbról balra halad, azaz a második
C kormányozza (engedélyezi) az els t, l. (7a). A probléma a (7a)-beli kormányzási viszony
átfogalmazásával oldható meg: ha a valódi kapcsolatok szegmentális kormányzását felváltjuk egy
inverz irányú engedélyezéssel, akkor a CC els tagján keresztül a kóda-kormányzás örökl dhet az új
inverz irányú szegmentális engedélyezés révén; (20b.ii.)-ben ezt az örökl dési útvonalat vastagon
jelöltük.
100
l. a (16) alatti magyarázatot. A monomorfemikus alakokban viszont lehet valódi CC+C szekvencia, a
harmadik viszont szinte mindig szonoráns, l. err l Rebrus (megj. alatt).
Kormányzás-fonológia kormányzás nélkül
286
(20) A kóda engedélyezése
a. ~C#
nyitót
nem nyitót
i. kivételes engedélyezés
ii. kóda-engedélyezés
C V C← V ]
C V→
→C V ]
| | |
| | |
h a l
b. ~CC#
d a l
i. kóda-eng. + kivételes eng.
ii. kóda-engedélyezés örökl dik
C V→
→C V C← V ]
C V→
→C V C V ]
| | |
|
| | |
h o l
d
f o[l → t]
|
A szóvégi kapcsolatok fenti elemzése felveti azt a kérdést, hogy mik a feltételei az inverz
szegmentális engedélyezésnek. Ahogyan korábban említettük, a valódi kapcsolatoknak három f
típusa van: a gemináta, a részleges gemináta (homorgán nazális + C) és a likvida + C kapcsolatok. Az
els két típus szótári ábrázolása az elfogadott elképzelés szerint üres szegmentális tartalommal történik
(l. Harris (1994)): (21a)-ban a gemináta, (21b)-ben a nazális kezdet valódi kapcsolatra mutatunk
példát (ez utóbbiban lexikálisan szerepel egy N nazalitás elem is). A kóda-kormányzás örökl dése
ezen az üres pozíción keresztül történhet.101
(21) A (részleges) gemináták interpretációja
a. obstruens/nemfoly. gemináta b. nazális + obstruens/nemfolyamatos
(kóda: lexikálisan üres)
(kóda: lex. üres helyelemmel)
C V→
→C V C V ]
C V→
→C V C V ]
| | |
| | |
|
cse[ → p]
k o[N → p]
csepp
komp
|
A valódi kapcsolatok interpretációja egy általános elv figyelembevételével magyarázható. Ez az
elv azt mondja ki, hogy ami engedélyez, annak interpretálódnia kell (l. Rowicka (1999)).
(22) AZ ENGEDÉLYEZ LÁTHATÓSÁGA: az engedélyez nek meg kell jelennie.
101
Bizonyos nyelvekben (pl. japán) csak ez a két típus lehetséges. A likvida-kezdet (l,r,j) valódi kapcsolatok a
magyarban a likvidák szegmentális szerkezetével vannak összefüggésben, ami azonban nem tisztázott.
Figyelemre méltó az az univerzális tendencia, hogy ebben a pozícióban a likvidák vokalizálódásra hajlamosak,
ami az ilyen kapcsolatok jelölt voltát mutatja.
287
Kormányzás-fonológia kormányzás nélkül
A (21)-beli szerkezetekben ez azt jelenti, hogy az (inverz) szegmentális engedélyezéssel együtt
jár a valódi kapcsolat els elemének megjelenése: az üres C szegmentális tartalmát (pl. a
helyelemeket) a szomszédos C-t l kapja. Ez a magyarázata annak, hogy a (21)-ben lev kapcsolatok
homorganikusak.
A mássalhangzó-kapcsolatra végz d alakok elemzésének (20b)-beli megkülönböztetését a (18)ban felsoroltakon kívül további empirikus adatok támasztják alá102. Ha megvizsgáljuk a nyitótövek
végén álló CC-ket, akkor azt tapasztaljuk, hogy ahogyan n a t végi CC-k jelöltsége (azaz
univerzálisan kevésbé gyakoriak), ugyanúgy n a nyitótövek aránya az összes t höz viszonyítva.
Valóban, a valódi kapcsolatok prototípusa a (21)-beli (részleges) gemináta. Annak ellenére, hogy a
f névi CC-tövek túlnyomó többsége ilyen, egyetlenegy nyitótövet sem találunk köztük.103 A
likvidakezdet ek közül a jC-tövek nem nyitótövek (kivétel, ha a második j vagy v, pl. ujj, gally, ölyv),
a rC-tövek nagy része nem nyitót , a kivételek nagy részében hosszú magánhangzó van és (vagy) a
második C nem alveoláris vagy szonoráns104. Az lC-tövek közül a zöngétlen mássalhangzósok nem
nyitótövek (1 kivétel: talp), a zöngés obstruenssel rendelkez csoport nagy része nyitót 105, az ll-tövek
lehetnek nyitótövek, és nem nyitótövek is. A C + j,v,h kapcsolatra végz d t csak nyitót lehet106, l.
(18e). A jelenség a nyitótövek (20b.i.)-beli ábrázolásával megmagyarázhatóvá válik. Nyitót ben nincs
szoros fonotaktikai kapcsolat a t vég utolsó két mássalhangzója között, éppen azért, mert közöttük
nem áll fenn engedélyezési viszony (ebb l a szempontból álkapcsolatok). Természetesen ez nem
jelenti azt, hogy a második C bármilyen lehet: obstruens után csak j, v (és marginálisan h); ezt a
kivételes engedélyezésnek a szokásos engedélyezésnél gyengébb volta magyarázza (az üres V-nek a
telinél kisebb az engedélyezési képessége). Nyitót nem lehet (részleges) gemináta, mert a gemináta
(l. (21)) els eleme a szegmentális tartalmát a második C-t l kapja a szegmentális kormányzás
következményeként, a (22) elv szerint. A közbens típusokban (a likvida-kezdet ekben) mindkét
típusú engedélyezés lehetséges: minél jelöltebb a kapcsolat els eleme (sorrendben: j < r < l), és minél
jelöltebb a második elem (sorrendben: zöngétlen obstruens < zöngés obstruens < nazális < likvida, ill.
alveoláris < labiális < palatális) annál valószín bb, hogy a második C nem a hosszabb engedélyezési
útvonalon van engedélyezve, hanem közvetlenül a kivételes engedélyezés révén, azaz nyitót . A
jelöltséget még befolyásolja a CC el tti magánhangzó hosszúsága is: mivel a hosszú magánhangzók
szótári ábrázolásában a második V-pozíció üres, ezért ennek kóda-engedélyez képessége kisebb
(annak ellenére, hogy megjelenik).
Trocheikus engedélyezés
A szóvégi mássalhangzó-kapcsolatok fenti elemzésével kapcsolatban felmerül egy további
probléma: mi engedélyezi a szóvégi CC belsejében lev üres V-t? A probléma lényege az, hogy a
t végi üres V nem képes jobbról balra alaposan kormányozni az el tte lev üres V-t, l. (20b). Egy
lehetséges megoldás a (20b.i.) kivételes engedélyezés esetén a t végi V-t ruházná fel alaposan
kormányzónak, a (20b.ii.) valódi kapcsolatokra pedig a korábbi elemzést alkalmaznánk: a két C
közötti üres V a C-k engedélyezési tartományának belsejében van, ezért nem kell megjelennie (err l l.
pl. Scheer (1998)). Egy másik lehetséges megoldás az alapos kormányzás inverzét használja fel: ha a
magprojekción definiált alapos kormányzás nem jobbról balra, hanem balról jobbra megy, akkor a
(20b)-beli alakokban a CC közötti üres V engedélyezve van (azaz nem interpretálódik). Ennek az ún.
trocheikus engedélyezésnek a feltétele a CC el tti interpretált V, ez pedig minden szóban forgó
102
Az adatok forrása Papp (1975) és (1994)
103
Nem számítjuk ide az inflektált alakokat (pl. kalapunk, l tt, unt) és a mellékneveket, melléknévi igeneveket
(deviáns, megl tt, megunt stb.). Ezek független okokból nyitó tulajdonságúak, l. Rebrus-Polgárdi (1997) .
104
Pl. térd, tárgy, társ, hárs, nyárs, árny, szárny, férj, érv, sérv, szarv, szerv, terv, örv, sarj, fürj, szörny, törzs.
105
Pl. tölgy, hölgy, völgy, hold.
106
Az üdv és a bolyh ellenpélda, ha egyáltalán szabad alakok.
Kormányzás-fonológia kormányzás nélkül
288
mássalhangzó-kapcsolat el tt megvan. A trocheikus alapos kormányzás mellett szóló érveket l. van der
Hulst--Rowicka (1997) és Rowicka (1999)107.
A balról jobbra ható kóda-engedélyezés és a CV-tengelyen definiált jobbról balra való
engedélyezés egyaránt a CV-szinten van definiálva. A CVCC szekvenciákban a megjelen V mind
balra, mind jobbra engedélyez; míg ez a többi szinten az engedélyezéssel nem fordul el . A kódaengedélyezés last resort tulajdonsága is különleges: a kóda-engedélyezés csak akkor lép érvénybe, ha
a C-t az utána álló V nem képes engedélyezni (mert nem interpretálódik). Felt n , hogy a trocheikus
engedélyezés és a kóda-engedélyezés forrása és feltétele ugyanaz: az interpretált V-pozíció. Ha
feltételezzük, hogy a nem megjelen üres V képes az el tte álló C-t jobbról balra engedélyezni
(ahogyan azt a standard elmélet teszi), akkor a kóda-engedélyezés levezethet a trocheikus
engedélyezésb l: a nem megjelen V trocheikusan engedélyezve van, amely engedélyezés örökl dik az
el tte álló C-re. Így a kóda-engedélyezés feleslegessé válik: a mássalhangzó-kapcsolatok els eleme és
a tartományvégi mássalhangzók mind az örökl d trocheikus engedélyezés révén vannak
engedélyezve, l. (24).
.
(24) Kóda engedélyezése örökl d trocheikus engedélyezéssel
a. valódi kapcsolat
V
→
V
V
b. álkapcsolat
V
→
V
c. szóvégi C
V
V
→
V
C←V C← V C←V
C←V C← V C←V
C←V C← V
| | |
| |
| | |
|
| | |
b a
s z o k
nya
l a p
g o N
gomba
→
szoknya
|
lap
Az örökl d engedélyezés és a C-nek teli V általi jobbról való engedélyezése között kizáró
viszony van: az el bbi akkor lehetséges, ha a C utáni V trocheikusan engedélyezve van, az utóbbi
akkor, ha nincs trocheikusan engedélyezve. Ennélfogva a kóda engedélyezéséhez nincs szükség last
resort mechanizmusra. Az engedélyezési képesség függ az engedélyezési útvonaltól: ez az út minél
hosszabb, az engedélyezés annál gyengébb. Ez a magyarázata annak, hogy a kóda-helyzet gyenge
pozíció: a nazalizáció és a pótlónyúlás/törlés (l. (9), (10), (11)) is ezt példázza. A kétféle engedélyezés
különbsége azonban egy másik jelenségben is megjelenik: a /v/ és a /h/ kezdet-pozícióban
szonoránsként, kóda-pozícióban obstruensként jelenik meg108.
107
A magyarban ilyen jelenség lehet a hangkivet f nevek tárgyesetének kötelez köt hangzója (bokrot, foglyot,
hasznot, poklot, taknyot stb.), annak ellenére, hogy ezek nagy része nem nyitót , vö (8).
108
A jelenségról b vebben Siptár (1994), Szigetvári (1998), Zsigri (1996) ír, CV-fonológiai elemzést Ritter
(1999)-ben találunk.
Kormányzás-fonológia kormányzás nélkül
289
(25) A /v/, /h/ változatai109
a.
b.
c.
kzt.
megjelenés
#_
V_
C_
engedélyezés
~_V
szonoráns
vár
avat
hatvan
teli V által
ház
pohár
nátha
--
sáv
érv
--
doh
enyh
(vlach)
sávtól,bovden érvt l
(Hradzsin)
dohtól,jacht
~_#
obstruens
~_C
obstruens
örökl d
örökl d
enyht l
A trocheikus engedélyezés az alapos kormányzás inverze, a magprojekción van értelmezve két
szomszédos V között. A trocheikusan engedélyezett üres V nem jelenik meg, az engedélyez re pedig a
(22) feltétel vonatkozik: az vagy lexikálisan szegmentális tartalommal bír, vagy lexikálisan üres, de
interpretálódó V. A trocheikus engedélyezés a trocheikus hangsúlymintázat lexikalizált változatának
tekinthet (Hulst -- Rowicka (1997)): két egymás utáni V-pozíció közül az els a fej, a második a
módosító. Két V-pozíció azonban nem csak a mag-projekción lehet szomszédos, hanem szegmentális
szinten is, ebben az esetben közöttük üres C-pozíció van. Ha e két V szegmentális tartalma között
kapcsolat van, akkor annak interpretációja hosszú magánhangzó vagy diftongus (a klasszikus
kormányzás-fonológiában ez az elágazó mag). A kapcsolatot a szegmentális szinten bevezetett balról
jobbra irányított engedélyezési reláció fejezi ki, mely a C-k közötti hasonló reláció analogonja110. A
hosszú magánhangzók lexikális ábrázolásában -- a geminátákhoz hasonlóan -- az egyik V-pozíció
üres. Ennek az interpretációja az engedélyez pozícióval megegyezik, l. (26a). Abban az esetben, ha a
hosszú magánhangzó után kóda van, a trocheikus engedélyezés ez utóbbi V-pozícióból indul, l.
(26b,c).
(26) Hosszú magánhangzós alakok
a. nyílt szótag
V
V
b. szóvég
V
V
c. zárt szótag
→
V
V
V
V
→
V
C←V C V C←V
C←V C V C←V
C←V C V C←V C←V
| |
| |
| |
|
| |
]p a
p[o
]k
p[a
p[o
→
pópa
→
pók
|
→
][r
| |
→
t]a
párta
A hosszú magánhangzók fenti ábrázolása magyarázatot adhat arra a magyar fonotaktikai
jelenségre, hogy a tövekben a VVCC szekvencia korlátozott: néhány kivétellel a magánhangzó á
109
A kóda gyengülése (vokalizáció) és az itt lejátszódó er södés (obstruentizálódás) közötti különbség abban
kereshet , hogy az el bbi olyan szegmentumokat érint, melyek állhatnak valódi CC-k els elemeként (n,l,r), míg
az utóbbiak (v, h) nem. Ezért ez utóbbiak csak álkapcsolatban állhatnak, így szonoráns változatuk nincs
engedélyezve. A részletes szegmentális elemzést itt elhagyjuk.
110
Ezekkel és az CC-kkel való kapcsolatával részletesen foglalkozik Rebrus (megj. alatt) és (el készületben).
Kormányzás-fonológia kormányzás nélkül
290
vagy é.111 Mivel az ilyen szerkezetekben a CC második tagjának az engedélyezése a hosszú
magánhangzó második V-jét l indul (l. (26c), így a (26)-beli VV-szerkezetek közül ennek a
leghosszabb az engedélyezési útvonala (így engedélyez képessége a legkisebb). A VCCszekvenciával összehasonlítva (amelynek magánhangzója nem limitált) a VVCC annyiban
különbözik, hogy az el bbi engedélyezési út eleje lexikálisan üres V, amely a telihez képest szintén
gyengébb engedélyez . Ezt az üres V-t viszont az el tte álló teli V engedélyezi, amely akkor a
leger sebb engedélyez , ha a szegmentális tartalma a legszonoránsabb á vagy é.112
Engedélyezési tartományok
Az alábbi (27) táblázat tartalmazza az eddig tárgyalt engedélyezési típusokat.113 Mivel az
elméletben nincsenek összetev k, és a fonotaktikai kapcsolatot a szegmentális engedélyezés fejezi ki,
ezért nincs értelme a kormányzás fogalmát ett l megkülönböztetni. Így olyan elméletet kapunk,
amelyben a kormányzás-fonológia által magyarázható általánosításokat meg tudjuk fogalmazni a
szigorú CV-fonológia absztraktabb reprezentációiban. Ezen kívül az engedélyezési típusok
egyszer bbé válnak.
(27)
hol?
Engedélyezés (új változat, vö. (17))
mi mit merre?
elnevezés
hatás
trocheikus
V nem interpretálódik
prozódiai
C megjelenik, üres V: CC
a. magprojekción
• interpretált V → üres V
b. CV-szerkezeten
• C ← V (feltétel nélkül)
teli V: CV
c. autoszegmentális vonalon
• szonoráns/üres C → C szegmentális
valódi CC megjelenik
• megjelen V → üres V szegmentális
hosszú VV megjelenik
Az alapvet engedélyezési elvek korábbiaknál egységesebben fogalmazhatóak meg. Az
engedélyezés három elvet követ: (i) lokális (azaz az adott szinten szomszédos pozíciók léphetnek
viszonyba), (ii) örökl dik (azaz az engedélyez az engedélyezést tovább adhatja egy másik
engedélyezési viszonyban; minél hosszabb az útvonal, annál kisebb az engedélyezési képesség), (iii)
tartomány által meghatározott (azaz a viszonyba lép elemek meghatározzák az engedélyezés típusát,
irányát). Ez utóbbi el nyös tulajdonság felhasználható egy deklaratív fonológiai formalizmus
felépítésére. Ez az új formalizmus az engedélyezési viszonyok helyett kizárólag azok tartományait
jelöli. Az alakok interpretációját az üres és teli C és V pozíciókon definiált tartományok határozzák
meg; az egyes alakokban olyan tartományok vannak adva, melyek kielégítik a tartományok
létrejöttének feltételeit (l. (27a,c). Több ilyen alak általánosítható: hasonló CV-vázú alakok gyakran
111
Err l l. Törkenczy (1994), CV-fonológiai elemzését l. Polgárdi (1999). Figyelemre méltó, hogy az angolban a
hasonló szekvenciákban a mássalhangzók koronálisak.
112
113
A szintén nem fels ó, ábrázolása ezekt l eltér, részleteket l. Rebrus (megj. alatt).
Nem tárgyaltuk a szó(tag)kezd CC-ket, melyekben egy új típusú szegmentális engedélyezés m ködik (l., pl.
Scheer (1997)). és a teli V-k közötti engedélyezési viszonyokat, melyek hangsúllyal és a magánhangzóharmóniával vannak összefüggésben (l. pl. Burzio (1994), Bárkányi (199X), Dienes (1997).
Kormányzás-fonológia kormányzás nélkül
291
ugyanazokkal a tartományokkal járnak, de bizonyos esetekben különböznek114. A cél a legáltalánosabb
alakok megadása, melyeknek az egyes alakok az instanciái. Fordított irányban: egy alak akkor
jólformált, ha valamilyen általánosítás speciális esete. Ebben a formalizmusban a prozódiai
reprezentáció már maga kétdimenziós: a C és V egységeket egymás alatt jelöli, a CV-szekvenciákat
egymás mellett (ezt a CV-n belüli engedélyezés többitól eltér volta indokolja). A CV-n kívüli
kormányzási tartományokat szögletes zárójellel jelöljük (az egységesség miatt a nem kormányzó, de
megjelen V-k is zárójelbe kerültek). A lexikálisan üres pozíció jele •, a telié C vagy V (az egyszer bb
olvashatóság kedvéért a megjelen -- azaz trocheikusan kormányzó -- üres V-t ∅-vel jelöljük).115
(28) Engedélyezési tartományok reprezentációja (példák)
instancia (alakok)
általánosítás (CV-váz)
konstrukció
m
[i]
p
[a][u]
C
[V]
• C
[V][V]
b r
[o •]
m t r
[o][o •]
C C
[V •]
C C C
[V][V •]
b k r
[o][∅ •]
b k r t
[o •][∅ •]
C C C
[V][∅ •]
C C C C
[V •][∅ •]
r [• k]
[o •][a]
f [l t]
C [• C]
[V •][V]
C [C C]
gemináta:
egy szóvégi
[• C]
[V •]
[l C]
[o
•] •
[V
•] •
valódi CC:
[V
h
[a
l
•]
C
[V
C
•]
kivételes eng.:
(szabály)
a. t végi teli V
mi
apu
[V]
b. tartományvégi üres
bor
motor
kóda:
[V
C
•]
c. t belseji üres
bokor
bokrot
d. valódi CC
rokka
folt
•]]
e. nyitót
hal
[•]]
halat
kedv
kedvet
h l t
[a][∅ •]
k d v
[e •][•]
k d v
C C C
[V][∅ •]
C C C
[V •][•]
C C C
kivételesen eng megjelenik
álkapcsolat szóvégen
114
A különbség adódhat paradigmatikus elvekb l, mint amilyen a (15) kóda-integritás.
115
Így a nyílt szótag [V], a zárt szótag [V
•] vagy [∅
•]
Kormányzás-fonológia kormányzás nélkül
292
[e
•][∅
•]
[V
•][∅
•]
f. hosszú mgh.
pópa
• C
]
V [V]
• C
]
[V •]
• [C C]
]
[V •] •
hosszú mgh.:
C C C C
[V][∅ •][V]
C C C C
[V •][∅][V]
C [C C] C
[V •][∅][V]
hangkivetõ, nem ikes
C
[
mák
[
párt
[
V
C
V
C
V
•
]
V V ,
(két konstrukció a VV-n)
[
g. kvázianalitikus alakok
söpörni
ugrani
tartani
hangkivetõ, ikes (kóda-integritás)
CC-vég (kóda-integritás)
Köszönetnyilvánítás
Köszönöm Polgárdi Krisztinának és Trón Viktornak az el adással és a kézirattal kapcsolatos
megjegyzéseiket.
A cikk megírásában a szerz t a Research Support Scheme (1687/320/1998) támogatta.
Irodalom
Bárkányi Zsuzsa
Burzio, Luigi (1994) Principles of English Stress. Cambridge. Cambridge University Press.
Dienes Péter (1997) Hungarian neutral vowels. The Odd Yearbook
Charette, Monik (1990) License to govern. Phonology 7, 233-53
Gussmann, Edmund & Jonathan D. Kaye (1993) Polish notes from a Dubrovnik café. I. The yers. SOAS
Working Papers in Linguistics and Phonetics 3, 427-62.
Harris, John (1990) Segmental complexity and phonological government. Phonology 7, 255-300.
Harris, John (1994) English Sound Structure. Blackwell, Oxford.
Hulst, Harry G. van der & Grazyna J. Rowicka (1997) On some parallels between (un)stressed vowels and
(un)realized empty nuclei. In: Booij & van de Weijer (eds.) Phonology in progress - progress in phonology.
Holland Academic Graphics, the Hague. 125-49.
Kaye, Jonathan D. (1990) “Coda” Licensing. Phonology 7, 301-30.
KLV (1985): Kaye, Jonathan D., Jean Löwenstamm, Jean-Roger Vergnaud: The internal structure of
phonological elements: a theory of charm and government. Phonology Yearbook 2, 305-28.
KLV (1985): Kaye, Jonathan D., Jean Löwenstamm, Jean-Roger Vergnaud: Constituent structure and
government in phonology. Phonology 7, 193-231.
Lowenstamm, Jean (1996) CV as the only syllable type. In Durand & Laks (eds.). Vol. 2. 419-43.
Papp Ferenc (1975) A magyar f név paradigmatikus rendszere. Akadémiai Kiadó, Budapest
Papp Ferenc (1994) (szerk.) A magyar nyelv szóvégmutató szótára. Akadémiai Kiadó, Budapest
Polgárdi Krisztina (1999)
293
Kormányzás-fonológia kormányzás nélkül
Rebrus Péter -- Polgárdi Krisztina (1997) Two default vowels in Hungarian? In: Booij & van de Weijer (eds.)
Phonology in progress - progress in phonology. Holland Academic Graphics, the Hague. 257-75.
Rebrus Péter (megj. alatt) Morfofonológiai jelenségek a magyarban. In: Kiefer Ferenc (szerk.) Strukturális
magyar nyelvtan. 3. kötet. Alaktan. Akadémiai Kiadó, Budapest.
Rebrus (el készületben) Egyeztetés a CV-fonológiában? kézirat (az Mai magyar nyelv leírásának újabb
módszerei IV., Szeged konferencián el adott anyag)
Rebrus Péter -- Törkenczy Miklós (1998)
Rebrus Péter -- Törkenczy Miklós (1999)
Ritter, Nancy N. (1999)
Rowicka, Grazyna (1999) On Ghost Vowels. A Strict CV Approach. Doktori disszertáció, Leideni Egyetem.
LOT Dissertations; 16, Holland Academic Graphics, The Hague.
Scheer, Tobias (1998) A unified model of phonological government. The linguistic Review
Siptár Péter (1994) A mássalhangzók. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 2. kötet.
Fonológia. Akadémiai Kiadó, Budapest. 183-272.
Szigetvári Péter (1998) Voice assimilation in Hungarian: the hitches. The Even Yearbook 3, 223-36.
Törkenczy Miklós (1994) A szótag. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 2. kötet. Fonológia.
Akadémiai Kiadó, Budapest.273-392.
Zsigri Gyula (1996) A zöngésségi hasonulás és a t nékeny mássalhangzók. Néprajz és nyelvtudomány XXXVII.
A helyelemek egyeztetése a CV-fonológiában
294
A helyelemek egyeztetése a CV-fonológiában
Rebrus Péter
Bevezetés
Ennek a dolgozat a célja megvizsgálni, hogy egy fonológiai elméletben az egyeztetés fogalma
milyen mértékig értelmezhet . Elméleti keretként a kormányzás-fonológia egy ága a CV-fonológia
szolgál. Az egyeztetés, mint általános terminus a nyelvészetben azt jelenti, hogy valamely szinten
szomszédos elemek valamely jegyüket megosztják. Ennek a fonológiai értelmezéséhez jó terepet
adnak az asszimilációs jelenségek. Ebben az írásban a mássalhangzók közötti helyhasonulást és a
magánhangzó-harmóniát vizsgáljuk, és megpróbálunk általános típusokat felállítani. Nem célunk a
magyar asszimilációs jelenségek részletes számbavétele, els sorban a szerkezeti és értelmezésbeli
analógiákat kerssük.
A 2. részben a CV-fonológiának a prozódiai (szótagszerkezeti) fogalmait vesszük sorra. Az itt
leírt elemzések els sorban Rebrus (megj. alatt) morfofonológia dolgozatán alapulnak. 2.1.-ben a
mássalhangzó-kapcsolatok típusait és az ezek közötti jelöltségi viszonyokat vesszük sorra. 2.2.-ben
ugyanezt tesszük a magánhangzó-kapcsolatokkal, bemutatva, hogy a CV-fonológia alkalmas eszköz a
kett szerkezeti párhuzamainak kifejezésére. A 3. részben a helyhasonulásokat vizsgáljuk
mássalhangzók (3.1.) és magánhangzók (3.2.) között.
Az alábbiakban bemutatjuk a cikkben szerepl fonológiai jegyeket, melyek a kormányzásfonológia és a CV-fonológia m vel i áltál többé-kevésbé elfogadottak.
(1) Elemek
a.
?
(zár-elem) nem folyamatos hangok [-continuant]:
zárhangok
affrikáták
p,b,t,d,ty,gy,g,k c,dz,cs,dzs
b.
h
nazálisok
laterális
m,n,ny,n,η
l
(zörej-elem) obstruensek [-sonorant]:
zárhangok
affrikáták
réshangok
f,(v),sz,z,s,zs(x/h)
c.
N
nazálisok
d.
U
labiális, nem-alsó
p,b,m,f,v,l
u,o,ü,ö
e.
I
palatális, nem-hátsó
ty,gy,ny,cs,dzs,s,zs,r
i,é,e,ü,ö
f.
A
alsó,szonoráns
l,r
á,a,e,é,o,ö
Kormányzási tartományok
Valódi mássalhangzó-kapcsolatok
A kormányzás-fonológia a mássalhangzó-kapcsolatok két f típusát különbözteti meg: a valódi
és az álkapcsolatokat.116 A valódi kapcsolatok univerzálisan jelöletlenebbek: egyes nyelvekben csak
116
Err l l. többek között KLV (1985), (1990), Harris (XXX), (XXX)
A helyelemek egyeztetése a CV-fonológiában
295
valódi kapcsolatok lehetségesek, más nyelvekben intervokalikusan álkapcsolatok is lehetségesek, de
szó végén tipikusan valódi kapcsolatok állnak. A magyar ez utóbbi nyelvek közé tartozik.
A magyarban (és számos más nyelvben) a szó végén álló két mássalhangzó közül a második
nem lehet szonoránsabb mint az els 117. A kivételek három csoportba sorolhatók: (i) nem morfémán
belüli kapcsolatok (pl. a felszólító módú igealakokban a C+j és C+d kapcsolatok), (ii) a
morfofonológiai szempontból is kivételes f névi nyitótövek osztálya, (iii) C + zöngétlen koronális
réshang (sz, s) kapcsolatok. A fenti kivételekt l eltekintve a kapcsolatok jelöltség szerint tipizálhatók:
a legjelöletlenebb típusokban a lehet legtöbb lehetséges képzési hely és zöngésség mássalhangzó
részt vesz szóvégi kapcsolatokban. Ez alapján a jelöltségi hierarchia a következ : (a) homorgán nazális
+ nemfolyamatos/réshang, (b) nemfolyamatos/réshang gemináta, (c) approximáns + nemfolyamatos,
(d) réshang + zárhang, (e) affrikáta/zárhang + zárhang. Az els két csoportból csak zöngés réshangok
hiányoznak (*nzs, *nv, *mv, *zz, *zzs, *vv). A (c) típuson belül is hierarchia van: minél szonoránsabb
az els hang, annál jelöletlenebb a kapcsolat, így az ellenpéldák többsége lC alakú: zöngés réshanggal
(*lzs, *lz) vagy labiális/palatális hanggal (*lb, *lty). A (d) és (e) típusban a második szegmentum
szinte mindig alveoláris (kivétel az szk és a ck kapcsolatokban), és két zöngés obstruens szóvégen
ritka vagy nem létezik.
A fenti kivételes (i), (ii), (iii) és nem következetesen viselked (d) és (e) osztályokat a további
vizsgálatokból itt kizárjuk. A fenti (a), (b) és (c) csoport CV-fonógiai ábrázolása (6-8)-ban látható: a
két mássalhangzó közötti kapcsolatot az autoszegmentális vonalakon megadott kormányzási
tartományok teremtik meg ([ ]-vel jelölve). Az (a) és (b) csoportba tartozó kapcsolatokban a lexikális
ábrázolás az els C szegmentális tartalmát nem adja meg (l. err l Harris XXX).
(6) Alveoláris valódi kapcsolatok (hely-elem nincs)
a. gemináta
b. részleges gemináta
c. approximáns+C
(homorgán)
(homorgán nazális)
(nem homorgán)
CVCVCV]
CVCVCV]
CVCVCV]





ma
h
haN h
szö
h
[
?]
[ ?]
[
]
[
matt
117

hant
]
?
[I ]
sajt (l. még folt, kert)
A szonoritási hierarchia: likvida>nazális>rés>affrikáta>zár, ezen belül zöngés>zöngétlen.
A helyelemek egyeztetése a CV-fonológiában
296
(7) Labiális valódi kapcsolatok (hely-elem: U)
a. gemináta
b. részleges gemináta
c. approximáns + C
(homorgán)
(homorgán nazális)
(nem homorgán)
CVCVCV]
CVCVCV]
CVCVCV]





cse
h
koN h
go
h

[ ?]
[ ?]
?
[ U]
[ U]
[U U]
A
csepp
komp
golf (l. szörp, skalp)
(8) Palatális valódi kapcsolatok (hely-elem: I)
a. gemináta
b. részleges gemináta
c. approximáns + C
(homorgán)
(homorgán nazális)
(nem homorgán)
CVCVCV]
CVCVCV]
CVCVCV]



 
pö
h
poN h

ko
h
[ ?]
[ ?]
?
[ I]
[ I]
[I I]
A
pötty
ponty
korty
A fenti ábrázolásokat (9)-ben összegezzük.
(9) Szóvégén lehetséges kapcsolatok ábrázolása
a. gemináta
b. részl.gemináta
c. approximáns + C
(bármi+bármi)
(nazális+nemfoly.)
(appr.+nemfoly./rés)
CVCVCV]
CVCVCV]
CVCVCV]






α (h)
α N (h)
α (h)
zár-tengely:
[ (?)]
[ ?]
(?) (?)
hely-tengely:
[ (U/I)]
[ (U/I)]
[U/I (U/I)]
A helyelemek egyeztetése a CV-fonológiában
297
A kóda
• interpretációja:
zár-tengely:
ua.
ua. (van)
bármi
hely-tengely:
ua.
ua.
ua.
• szótári ábrázolása:
üres
N
r/j/l
A fenti esetekben a kormányzási tartomány els C-pozíciója a második C-pozícióéval
megegyez en interpretálódik. Likvidák esetén az els pozíció hely-eleme minden esetben ki van töltve,
így az interpretációs szabály csak az üres pozíciókra vonatkozik.
(10)
Kóda-tartomány interpretációja:
Ha kóda-tartományban az els pozíció helyeleme lexikálisan nincs megadva, akkor
interpretációja a második pozíciójéval azonos.
Szerkezeti párhuzamok
A CV-fonológiai ábrázolás lehet séget ad arra, hogy kifejezzünk a mássalhangzó- és a
magánhangzó-kapcsolatok között fennálló több hasonlóságot. Az el bbi esetben a két pozíció között
üres V, az utóbbi esetben üres C áll. A párhuzam a típusok között is fennáll: valódi CC-nek a
hagyományosan elágazó magnak nevezett típus felel meg, míg az ál-mássalhangzó-kapcsolatoknak a
hiátus. Hiszen az el bbi esetben a két szegmentum között szigorú fonotaktikai kapcsolat van, addig az
utóbbi esetben ilyen nincs. Az el z részben láttuk, hogy a valódi mássalhangzó-kapcsolatokon belül a
fonotaktikai kapcsolatnak három típusa van: (a) melyekben minden elem közös, (b) melyekben a
hely-elemek közösek és (c) melyben a kormányzott pozíciónak más hely-eleme is lehet (l. (9)).
Magánhangzó-kapcsolatoknál az (a) típusnak a hosszú magánhangzó felel meg, ábrázolásukban a
kormányzott (második) V-pozíció üres: ezt látjuk (11b)-ben.
(11) kormányzott üres pozíció
a. gemináta
b. hosszú mgh.
CVC
VCV


h
[A

zár-tengely
[
hely-tengelyek:
[ (U)]
[(U) ]
[ (I)]
[(I) ]
interpretáció:
?]
]
[t:], [p:], [ty:]
[a:], [e:], [o:], [ö:]
A (b) típusnak a közös hely-elem (záródó) diftongus felel meg, ezek egyes magyar
nyelvjárásokban a hosszú magánhangzók realizációi. Ebben a típusban csak az I és/vagy az U elemek
vonalán van a kormányzási tartomány megadva: ezek az elemek mindkét V-pozícióban
interpretálódnak, míg az A elem csak az els ben (l. (12b)).
A helyelemek egyeztetése a CV-fonológiában
298
(12) helyelem nélküli kormányzott pozíció
a. részleges gemináta
b. diftongus közös helyelemmel
C V C
V C V



N
h
A

zár-tengely:
[
? ]
hely-tengelyek:
[
(U)]
[(U)
]
[
(I)]
[(I)
]
interpretáció: [nt], [mp], [nyty]
[ei], [ou], [öü]
A harmadik, legjelöltebb típusban a második V-pozíció lexikálisan rendelkezik szegmentális
tartalommal, így el fordul, hogy a diftongus két pozíciójának nincs közös hely-eleme. A jelöltséget
mutatja CC esetén az, hogy ezen típusnak több példánya t végén nem létezik (l. (13a)), VV esetén
pedig az, hogy a magyarban ilyen diftongust egyáltalán nem találunk (l. (13b)).
(13) helyelemmel rendelkez kormányzott pozíció
a. approximáns+C
b. diftongus közös helyelem nélkül
C V C
V C V




(A)
h
A

? ]


zár-tengely:
[
hely-tengelyek:
[(U) (U)]
[(U) (U)]
[(I) (I)]
[(I) (I)]
interpretáció
[lt],[jt],[rt],[ldy],[rp]
de nincs: [jty],[jdy],[jp],[jb],[lty]
[au], [ai], [eu], [oi] stb.
(magyarban nincsenek)
Az álkapcsolatoknak megfelel típus az, amelyben a két V-pozíció nem alkot kormányzási
tartományt: ebben az esetben két különböz szótag V-jér l beszélünk. Ezek ábrázolása (14a,b)-ben
található.
A helyelemek egyeztetése a CV-fonológiában
299
(14) pozíciók kormányzási tartomány nélkül
a. intervokalikus CC
b. hiátus
C V C
V C V




α
β
γ
δ
Az a tény, hogy az álkapcsolatokban nincs kormányzási tartomány nem jelenti azt, hogy
intervokalikusan minden olyan mássalhangzó-kapcsolat megjelenhet, amely nem állhat szóvégen.
Valójában az intervokalikus CC-kre több korlátozás is vonatkozik (ezekr l l. a Függelékben (26-27)et). Itt a korlátozások közül kett t említünk. Az els a kötelez kontúrelv, mely az azonos hely-jegy
álkapcsolatokat tiltja (l. (16)). Azaz tiltja a labiális+labiális és a palatális+palatális álkapcsolatokat118,
l. (17a,b). Nem tiltja viszont a hasonló alveoláris kapcsolatokat (tn, dn, tl, dl, nl), mert az alveolárisak
ábrázolása hely-jegy nélkül történik.
(15)
KÖTELEZ KONTÚRELV:
a szótári alakban két szomszédos, C-pozícióhoz kötött helyjegy nem lehet azonos.
(16) tiltott kapcsolatok
a. palatálisra
b. labiálisra
* C V C
* C V C




I
I
U
U
Figyeljük meg, hogy a homorganikus valódi kapcsolatok a fenti elvet nem sértik, hiszen ezek
ábrázolásában csak a második C-pozícióban szerepel hely-elem (11a) és (12a).119 A (15) elv a szótári
ábrázolásra tesz kikötést, így még arra az esetre is vonatkozik, ahol a C-pozíciók közötti lexikálisan
üres V a felszínen megjelenhet. Valóban, ezekben az ún. hangkivet tövekben (pl. szobor -- szobrok,
majom -- majmok, köröm -- körmök, fátyol -- fátylak, öböl --öblök) sem lehet a lexikálisan
szomszédos két C mindegyike labiális vagy palatális (részletes elemzésüket l. Törkenczy (XXX),
Rebrus (megj. alatt)).
A kötelez kontúrelv a szomszédos magánhangzó-pozíciókra is áll. A hosszú magánhangzók és
a közös hely-elem diftongusok ábrázolásában a kormányzott pozíció lexikálisan üres (l. (11b), (12b)).
A (14b)-ben bemutatott hiátusra a kontúrelv kizárja a palatális+palatális magánhangzószekvenciákat:
valóban, a palatális hiátustöltés az átvett szavakban (is) kizárja az ii, ié, éi, ei, ie szekvenciákat (l. err l
118
Azaz *pm,*bm,*fm,*vm,*pv,*bv,*fv,*fp,*vb,*vv, illetve *jty,*jdy,*jny (de: tyny, dyny létezik, ezekr l kés bb
szólunk).
119
Külön meggondolást igényelnek a (13c) típusú valódi kapcsolatok közül azok, amelyekben szerepelnek
azonos hely-elemek (rty,rgy,lp,lf,lm). Ezek fonetikailag nem homorganikusak, a probléma az r és az l
reprezentációjának átgondolását teszi szükségessé. Ezzel szemben a (15) elv vonatkozik a jty, jgy, jny valódi
kapcsolatokra, ezek viszont fonetikailag azonos helyen képzettek.
A helyelemek egyeztetése a CV-fonológiában
300
Siptár XXX)120. Mind álkapcsolat, mind hiátus azonban megjelenhet (analitikus) morfémahatáron. A
kontúrelvet sért CC-k morfémahatáron megmaradnak (pl. hegynyi, döfve, savban, ujjnyi), a hiátusbeli
azonos hely VV-ket viszont a (palatális) hiátustöltés megszünteti (eké[j]i, ki[j]é stb.). A különbséget
az magyarázza, hogy míg a hiátus a V-pozícióból a C-pozícióba való terjedéssel megszüntethet
(amely folyamat független érvek szólnak), addig a CC-kre ilyen eszköz nem áll rendelkezésre. A
részletes elemzés meghaladná a cikk kereteteit.
Asszimilációk
Az álkapcsolatokra vonatkozó másik korlátozás a kötelez hasonulásokkal kapcsolatos.
Hasonulásokat általában morfémahatáron tapasztalunk (dinamikus folyamatok), azonban ezeknek
lexikális (statikus) általánosításként is hatásuk van: tövön belül sem találunk olyan kapcsolatot,
melyre kötelez hasonulási szabály vonatkozik. Ebben a részben a helyhasonulások két típusát: a
palatális és a labiális hasonulást vizsgáljuk meg. Egyéb hasonulásokat -- mint a zöngésségi és
szibiláns hasonulás, affrikálódás, geminálódás -- itt nem vizsgálunk (a C-hasonulásokról részletesen l.
Siptár XXX).
Hely-hasonulás mássalhangzók között
A palatális hasonulás egy típusa az, amely alveoláris+palatális mássalhangzó-szekvenciákban
zajlik le: az alveoláris magánhangzó palatálissá válik. Ez a folyamat álkapcsolatokban
morfémahatáron át is lezajlik (pl. á[ty]nyúl, va[dy]nyúl, rá[dy]jár, szí[ny]játék)121. Mivel az
álkapcsolatok nem tartalmaznak kormányzási tartományt, ezért egy ett l független interpretációs
tartományt kell feltételeznünk: a palatális mássalhangzó I hely-eleme kötelez en interpretálódik a vele
balról szomszédos alveoláris mássalhangzón (melynek ábrázolása hely-elem nélküli), l. (17).
(17)
Az I elem kötelez terjedése
C V C
{


α

I}
A fenti konstrukciót a tövön belüli palatális+palatális álkapcsolatokra (pl. sa[ty]nya,
naro[dy]nyik) is alkalmazhatjuk: ezekben a lexikális ábrázolás nem két palatális, hanem
alveoláris+palatális mássalhangzó-szekvenciát tartalmaz. Így a szótári alak nem sérti a (15) kötelez
kontúrelvet. Hasonló problémát találunk azokban a mássalhangzó-kapcsolatokban, melyek f névi
tövekben nyitótövek végén állnak. Rebrus (2000) szerint a nyitótövek végén álló mássalhangzókapcsolatok álkapcsolatok. A palatális szonoráns gemináták ilyenek: jj és nny csak nyitótövek végén
állhat (pl. ujj, gally, könny). Ekkor viszont ezeknek a geminátáknak álgeminátaként való ábrázolása
sérti a (15) kontúrelvet. Azonban ezek az alakokról is feltételezhet , hogy az els C-pozícióban az I
elem nem része a szótári ábrázolásnak, viszont a (17)-beli folyamat révén interpretálódnia kell.
120
121
A labiális+labiális V-szekvenciák tövön belül ritkán fordulnak el , ezek problémáját itt nem érintjük.
Nem foglalkozunk az igék felszólító módú alakjaiban lezajló, ett l eltér palatalizációval. Ezenkívül a
toldalékolt alakokban lezajló palatalizáció különbözik a szóhatáron tapasztaltnál: az el bbi esetben a C+j
szekvenciák végeredménnye mindig palatális gemináta. A j-nek a geminálódásával itt nem foglalkozunk.
301
A helyelemek egyeztetése a CV-fonológiában
A (17) palatalizáció aktív jellegének következménye, hogy az alkalmazódik alveoláris nazális +
palatális C (n+ty, n+gy, n+ny, n+j)122 szekvenciákon is. A palatalizáció vagy a szóvégi n-et (pl.
vé[ny]tyúk, ké[ny]gyár), vagy a szóvégi kapcsolatban lev n-et (pl. pi[nyty]nyi (= pintnyi), bolo[nydy]
nyár) érinti. Az utóbbi két eset azt mutatja, hogy a (17) konstrukció a szóvégi valódi CC-t alkotó
kormányzási tartományon belül is kifejti hatását.
A palatális hasonulással ellentétben, mely bármilyen alveoláris mássalhangzón lezajlik, a
labiális hasonulás csak nazálisokra vonatkozik. Ez a korlátozás vonatkozik morfémán belüli (pl.
ködmön, ritmus, de: kampó, ember), és morfémahatáron álló (pl. vadbarom, átporol, szétmegy, de
ké[m]por, ka[m]buli, nyelvta[m]ban) szekvenciákra is. Így a (17) palatális terjedéshez hasonló
általános labializációs szabályt nem adhatunk. Az egyik lehet ség az lenne, ha a labiális hasonulási
szabályt nazálisokra specifikálnánk. Ekkor két labiális hasonulást kell feltételeznünk: tövön belül a
valódi kapcsolatok interpretációs szabálya, a morfémahatáron pedig a labiális hasonulási szabály
eredményeképpen kapnánk m-et.
Egy másik lehet ség a morfémahatáron lejátszódó labiális hasonulást a valódi kapcsolatokban
feltételezett kormányzási tartomány interpretációs szabályának aleseteként fogná fel. Ekkor a
morfémahatárral elválasztott n + labiális kapcsolatokat a szóvégi valódi kapcsolatokhoz hasonlóan
kormányzási tartománynak kell tekintenünk; azaz fel kell tételeznünk, hogy CC-kormányzási
tartomány morfémahatáron is létrejöhetnek, s t az említett szegmentumok esetén ez kötelez 123.
Esetünkben ez azt jelenti, hogy a t végi n lexikális ábrázolása a valódi kapcsolatbelivel azonos módon,
lebeg N elemmel történik. Ellenkez esetben a nazális elem C-pozíciója nem lenne üres, ami a nazális
C valódi kapcsolat el feltétele (l. 12a)124. Ez utóbbi megoldás el nye, hogy nem kell két külön labiális
hasonulási szabályt feltételeznünk, elegend a kormányzási tartomány interpretációs szabálya.
Magánhangzó-harmónia
A magánhangzó-harmónia a magyarban a toldalékbeli magánhangzónak a t belihez való
illeszkedését jelenti. A mássalhangzó-hasonuláshoz hasonlóan ez is asszimilációs folyamat, azzal a
különbséggel, hogy a folyamat iránya fordított (balról jobbra). A folyamat a szomszédos
magánhangzókat érinti, függetlenül attól, hogy közöttük hány mássalhangzó áll. Ebb l a szempontból
különbözik mind a mássalhangzó-hasonulástól, mint a hosszú magánhangzó vagy diftongus
kormányzási tartományon belüli interpretációjától (l. (11b), (12b)). A palatális harmónia egy V-hez
asszociált I elem terjedése a t le jobbra álló V-pozícióba; feltétele, hogy a második V-pozíció üres
vagy a toldalékban van.125 Az I elem interpretációs tartománya nem alkot kormányzási tartományt, l.
(18).
122
Utóbbi esetén geminálódás lezajlik, a végeredmény palatális gemináta (nny).
123
Hasonló folyamatot a morfémahatárral elválasztott gemináták esetén is fel tételezhetünk, hiszen a
degemináció jelensége így egységesen magyarázható, l. Rebrus (megj. alatt)
124
Erre mutató független érv, hogy a morfémahatárral elválasztott n + folyamatos msh. szekvenciákban (pl.
vonsz, vonlak, vonhat, tanszék) a nazális a V-pozícióban interpretálódik, melynek eredménye nazalizált
magánhangzó.
125
Feltétel még, hogy az I elemnek inter- vagy intraszegmentálisan engedélyezve kell lennie. A palatális
harmónia összetett jelenségér l l. pl. Nádasdy-Siptár (XXX), Dienes (XXX)
A helyelemek egyeztetése a CV-fonológiában
302
(18)
Az I elem terjedése (egyszer sítve)
V C V
 
 β
{I
}
A palatális terjedés fenti szabálya a C-közötti (17) hasonulás analogonja, hiszen nem
kormányzási tartományban zajlik le, így még (analitikus) morfémahatáron is keresztülmehet.
A hely-elemek egyeztetésének egy a palatális harmóniával összefügg példáját mutatják az igei
definitumjelöl alternánsai. Ez a toldalék a palatális harmóniának megfelel kétalakú -ja/-je vagy
egyalakú -i alak helyett -j(a)/-i váltakozást mutat. A morfémaváltozatok közötti választás a palatális
harmónia szabályait követi: ha a t utolsó V-pozíciójában nincs a harmónia szempontjából számba jöv
I elem, akkor a toldalék -ja, ha van, akkor a toldalékalternáns az -i (pl. lopja -- lepi, lopják -- lepik,
lopjátok -- lepitek). Rebrus (megj. alatt) szerint az E/3. tárgyas ragozású igealak toldaléka CVszekvenciából és egy lebeg (V-hez nem asszociált) I elemb l áll. Ha a t megfelel módon tartalmazza
az I elemet, akkor ez engedélyezi a lebeg elem V-pozícióban való megjelenését, és az eredmény az -i
alternáns (l. (19a)). Ellenkez esetben a lebeg elem az üres C-pozícióban jelenik meg mint j, és az
üres V-pozíció a-ként interpretálódik (l. (19b)).
(19) A definitumjelöl alternánsai (egyszer sített reprezentáció)
a. van engedélyezés
b. nincs engedélyezés
C V C V C V
C V C V C V
  
  
l A p
l A p
I
 →I
lepi
U
I
lopja
A (19a)-beli palatális engedélyezés a következ :
(20)
I-engedélyezés:
a t beli I engedélyezi a vele jobbról szomszédos lebeg I V-pozícióban való megjelenését
A palatális engedélyezés azt fejezi ki, hogy az I elem V-pozícióban egyedül instabil, azaz egy
t le balra álló megjelen I-nek kell t engedélyeznie. Ez az elképzelés átvihet a palatális elemnek a
t ben való megjelenésére is, ebben az értelemben a (20) engedélyezés a (18) palatális terjedés lexikális
analogonja. Másrészt az I-engedélyezés a (16b) kontúrelv duálisa is: az engedélyezés az I elem
szomszédos V-pozíciókban való többszöri megjelenését preferálja, a kontúrelv ugyanezt C-pozíciókra
tiltja.
Az analógiák sora a labiális harmónia egyérték jegyekkel való elemzésével folytatható.
Polgárdi-Rebrus (1998) szerint a háromalakú toldalékok (pl. -hoz/hez/höz, -tok/tek/tök) és a középs
A helyelemek egyeztetése a CV-fonológiában
303
köt hangzók által mutatott ún. labiális harmónia nem az U elemnek a t b l való terjedése, mert a
háromalakú toldalékok magánhangzója már lexikálisan tartalmazza az U-t126. Ebben az esetben a
harmóniát az el z példához hasonlóan a lebeg elem balról való engedélyezésének kényszere váltja ki.
(21)
U-engedélyezés :
a lebeg U-nak az I-vel együtt való megjelenését a balról szomszédos U engedélyezi
A (21) feltétel következménye az, hogy a t után a középs rövid magánhangzók palatálislabiálisként csak egy szomszédos t beli labiálissal együtt jelenhetnek meg (pl. sünhöz, tökhöz vs.
képhez, perhez, hithez). A kötelez palatális harmónia miatt a (21) szabály minden palatális t utáni
rövid középs magánhangzót127 tartalmazó toldalékban alkalmazódik. A labiális engedélyezés
bizonyos értelemben a (20) palatális engedélyezés párja. Duálisa a (16a) kontúrelv.
Összegzés
A helyhasonulások vizsgálata a magyarban azt mutatja, hogy ezek alapvet en két csoportba
sorolhatók: a prozódiával kapcsolatosak, és az attól függetlenek típusába. A CV-fonológia kertében a
prozódiai következménnyel járó viszonyokat a kormányzási tartomány fejezi ki, melynek prototipikus
formáiban a reprezentációs elvek folytán inherensen hasonulást feltételezünk. A prozódiával nem
összekapcsolható helyhasonulások elemzésére szolgálnak az interpretációs tartományok, melyek egy
fonológiai tartományon automatikusan megvalósuló konstrukciók. A palatális és a labiális
helyhasonulás -- mint a hasonulások általában -- az egyeztetés fonológiai formájának tekinthet k
annyiban, amennyiben az egyes szinteken (V-szint, autoszegmentális szint) szomszédos entitások
közötti jegy-egyezésre mondanak ki feltételeket. A CV-fonológiai reprezentációban a C-k és a V-k
közötti kapcsolatokat leíró strukturális hasonlóság a kormányzási tartományokkal fejezhet ki. A
prozódiai jelöltségi viszonyok egységes kifejezésén túl azonban ez a hasonlóság a
helyhasonulásokban is tükröz dik: a C-k közötti és V-k közötti interpretációs tartományok egymásnak
megfeleltethet ek: a palatális hely-elemre mindkett , a labiális hely-elemre egyik esetben sincs
konstrukció. Az interpretációs tartományok tekinthet k úgy mint a kormányzási tartományok
“gyenge” formái, ennyiben követik azok irányát: C-k esetén jobbról balra, V-k esetén balról jobbra. A
lexikális általánosítások egy speciális fonológiai korlátozáshoz, a kötelez kontúrelvhez vezet. Ez az
elv a szomszédos C- (ill. V-) pozíciók közötti lexikális jegyazonosságot tiltja. Ennek duálisa a V-k
közötti palatális és labiális-engedélyezés éppen az ellenkez jét fejezi ki: a különböz hely-elemekkel
rendelkez egymásutáni V-k nem preferáltak. Míg a kontúrelvnek a C-k esetén nincs dinamikus
hatása, V-k esetén a hiátustöltés és egyéb morfofonológiai folyamatok a kontúr elv sértése ellen
hatnak. A labiális és palatális engedélyezés közül az el bbi tövekben, az utóbbi toldalékolt alakokban
fejt ki nagyobb hatást.
126
Az elemzést kissé más megfogalmazásban említi Rebrus (megj. alatt)
127
A fels labiális magánhangzókban és a hosszú középs magánhangzókban az U elem nem lebeg
A helyelemek egyeztetése a CV-fonológiában
304
Függelék
(Törkenczy (1994) alapján)
(23) Lehetséges szóvégi mássalhangzó-kapcsolatok morfémán belül
a. obstruens-vég ek (dz, dzs, j, v és h nélkül)
p/b
f
t/d
c
sz/z
cs
s/zs
ty/gy
kolomp tromf
hant
konc
sansz
kilincs kliens ponty
gomb
rend
csepp muff
matt
jobb
kedd
k/g
i. homorgán nazális:
pénz
--
tank
rongy hang
ii. gemináta:
vicc
passz
meccs plüss
--
pötty
bükk
--
meggy segg
sors
korty
park
törzs
tárgy
dramaturg
iii.approximáns:
r
szörp
turf
szerb
j
?
selyp
skalp
?
dölyf
sajt
golf
folt
kommersz
korcs
borz
Svájc
majd
-v
perc
kard
-l
part
polc
?
hajsz nefelejcs (Majs) --
hüvelyk
rajz
félsz
kulcs
pajzs
--
cajg
fals
--
halk
küld
--
--
völgy
?
--
--
--
--
X
--
maszk
X
--
(rezg)
↑
--
--
--
--
--
(NOSZF)
koszt
rivalg
iv. koronális réshang128:
sz/z
-s/zs
--
--
--
must
--
↑
gerezd
--
↑
--
pünkösd
X
(borscs)
X
↑
--
--
snapsz --
taps
--
--
--
--
--
--
X
X
(Detk)
X
X
--
--
↑
(Batyk)
v. zárhang129:
p/b
t/d
↑
(copf) recept (Apc)
↑
(Tabd)
--
--
--
X
↑
-ty/gy
↑
--
--
X
X
X
--
--
--
128
nemzárhang+obstruens els pozíciójában ezen kívül: szaft,tomaha[f]k, (MAFC), jacht,barack,(Recsk)
129
ezek az el bbiekt l eltér en nem csökken szonoritású kapcsolatok
--
A helyelemek egyeztetése a CV-fonológiában
305
-k/g
--
--
--
--
akt
---
fax
↑
--
voks
--
--
--
↑
-(Szakcs)
smaragd
--
ny
v
l
j
r
stramm finn
könny
--
hall
ujj
orr
v
--
--
--
↑
--
--
--
l
film
(Köln) --
elv
↑
X
--
j
slejm
kombájn
ölyv
fájl
↑
--
r
sárm
konszern szörny
terv
görl
fürj
↑
m
↑
--
--
?
n
X
↑
X
-szenv (ajánl) X
--
ny
--
--
↑
könyv --
--
b. szonoráns-vég ek
m
n
i. gemináta:
ii. approximáns:
--
iii. nazális130:
hamv --
szomj --
X
Általánosítások
(24) Lehetséges szóvégi kapcsolatok a magyarban (a h-t nem számítva)
term. osztály:
kivétel131:
a.
bármilyen gemináta (de:
zöngés réshangok: vv, zz, zzs
b.
i. homorgán nazális + nemfolyamatos
mp,mb, nt,nd,nc,ncs,nl, nyty,nygy, nk,ng -ii.
+ réshang132
mf, nsz,nz,ns
c.
i. approximáns + nemfolyamatos
(zöngés réshang:) nzs
v + bármi
rp,rb,rm, rt,rd,rc,rcs,rn,rl, rty,rgy,rny, rk,rg,
jm, jt,jd,jc,jcs,jn,jl, jk,jg,
j + pal.(/lab.):
jty, jgy, jny, jp, jb
lp,lm, lt,ld,lc,lcs,ln, lgy, lk,lg
l + (pal./lab.):
lty, lny, lb
130
ezek mind emelked szonoritásúak (ezeken kívül még ilyen: kedv, enyh, ?bolyh)
131
a kötelez hasonulásokat nem számítva
132
a nazális mássalhangzó helyett nazális mgh. jelenhet meg
↑
A helyelemek egyeztetése a CV-fonológiában
306
ii. approximáns + réshang
l + zöngés réshang: lz, lzs
rf,rsz,rz,rs,rzs, jf,jsz,jz,js,jzs, lf,lsz,ls
d.
veláris réshang: rch, jch, lch
i. réshang + alveoláris zárhang
ft, szt,zd, st,zsd, cht
ii. alveoláris obstr. + veláris zárhang
szk,ck, csk
e.
--
zárhang + alveoláris zárhang/réshang
palatális+bármi: tyt,gyd,tysz,gyz,tys,gyzs
pt,bd,psz,ps, kt,ksz,ks,gd
f.
bármi + zöngés réshang: bz, bzs, gz, gzs
zöngés nem érdes msh. + v / j 133
rv,jv,lv,nv, mv,nyv,dv,rj,mj
pal./vel.obstruens:gyv,bv,gv, bj,gj
(25) Intervokalikus mássalhangzó-kapcsolatok monomorfemikus alakokban
a. obstruens-vég ek
p/b
f
t/d
c
sz/z
cs
s/zs
↑
(tamtam)
--
szomszéd csámcsog emse
ty/gy
k/g
--
tömkeleg
i. approximáns
ii. nem homorgán nazális:
m
↑
↑
ny
--
nyomda
--
--
--
vamzer
--
--
(kényszer)
--
kámzsa --
--
(manysi)↑
--
--
--
↑
--
ofszet --
--
--
cafka
--
--
--
afgán
--
keszty ↑
--
--
↑
↑
ostya
iskola
↑
(uzsgyi) vizsga
↑
--
iii. réshang134:
f/v
--
↑
-sz/z
--
bovden
aszpik aszfalt ↑
naszcens
↑
azbeszt
s/zs
↑
kaspó (násfa) ↑
kosbor
↑
--
↑
--
--
↑
--
↑
kapca ↑
--
iv. zárhang:
p/b
↑
↑
--
↑
lépcs
obzervál
133
ezek mind egy kivételes t osztályhoz tartoznak (nyitótövek)
134
intervokális affrikáta+obstruens csak szó végén megengedett van
lepke
habzsol --
--
A helyelemek egyeztetése a CV-fonológiában
307
t/d
ty/gy
--
(hétf ) ↑
X
X
X
X
X
patkó
--
↑
X
X
X
X
X
--
pitypang (fityfiritty)
--
--
--
--
--
↑
pletyka
akció
(jegyz )
-↑
bakcsó ↑
↑
--
-↑
--
--
↑
r
j
h
--
-bakfis ↑
--
↑
lagzi
bugyborék
k/g
b. szonoráns-vég ek
m
n
ny
v
l
l
↑
↑
--
tolvaj ↑
X
X
málha
r
↑
↑
↑
árva
↑
↑
varjú
marha
j
↑
↑
(ejnye)
golyva ↑
135
i. approximáns
majré ↑
kályha
ii. nem homorgán nazális:
m
lomha
↑
amnézia
n
X
↑
X
konvoj (jelenleg) --
X
(inhalál)
ny
--
--
↑
ponyva (tényleg) --
X
konyha
--
sufni
--
--
kifli
cifra
ifjú
--
--
--
luvnya
--
bóvli
sevró
szovjet
(nyámnyila)
(nyamvadt)homlok
kamra tömjén
iii. réshang:
f/v
sz/z
s/zs
eszme disznó tarisznya
(köszvény) maszlag
csizma pózna --
--
üzlet
ismer
(fösvény)
pislog --
--
--
vizsla --
--
masni rusnya
zsolozsma alamizsna
--
--
csoroszlya --
(ezred) -kushad
iv. zárhang/affrikáta:
p/b
t/d
c
135
--
flepni --
(képvisel) káplár
--
(abnormis) --
--
sablon abrak gereblye
ritmus etnikum X
ótvar
katlan matrac X
ködmön bodnár X
dudva nudli
nadrág X
kecmereg fecni
--
--
--
itt szó végén lehetséges, j-re és v-re végz d alakokat is megadtunk
spicli
apró
kopja
--
abház
nátha
--
A helyelemek egyeztetése a CV-fonológiában
308
cs
kocsma plecsni --
--
ty/gy
petyhüdt
fityma --
satnya
katyvasz
hagyma --
narodnyik
--
kagyló --
lakmuszakna
szoknya
ekvi-
nyakló bokréta csuklya ekhó
(dágvány)
nyegle egres
k/g
magma bognár --
(becsl ) --
--
(fátylas)
---
X
máglya
Általánosítások
(26) Nem létez intervokalikus kapcsolatok a magyarban
kivételek136:
tiltások:
a.
i. az 1. palatális nemfolyamatos
nyp,nyb,nyf,nyt,nyd,nyc,nysz,nycs,nys,nyzs,nyk,nyg,nyl,nyr
de: (nyv),nyh
tyf,tyt,gyd,tyc,tysz,gyz,tycs,tys,gyzs,gyg,tyl,tyr,gyr,tyh de:typ,gyb,tyk,tyny,(tyv),gyl
jny,
de: jv,jr,jh
ii. a 2. palatális nemfolyamatos
lny,mny,fny,zny,sny,pny,bny,cny,csny,gny
de:rny,vny,szny,zsny,tygy,gyny,kny
mty,mgy,fty,vgy,zgy,zsgy,cty,csty,pty,bgy,kty,g+gy
b.
de: szty,sty
i. érdes + érdes (mind. kor.)
szcs,sz+s,z+zs,sc,s+sz,zs+z,scs,c+cs,c+sz,c+s,cs+sz,cs+c,cs+s
de: szc
ii. koronális + r / j
nr,szr,zr,sr,zsr,cr,csr
de: tr,dr
zj,sj,zsj,cj,csj
de: (rj),szj,
iii. affrikáta + obstruens
cp,cf,ct,c+cs,c+sz,c+s,cty,csp,csf,cst,cs+sz,cs+c,cs+s,csty
(de: ck, csk)
és gzs
c.
i. labiális + labiális
fp,vb,pf,fm,vm,fv,vv,pm,bm,pv,bv
(de:mv)
ii. labiális + obstruens/nem foly.
136
mt,mc,mty,mk
de: md,msz,mz,mzs,mcs,ms
fn,vn,fny,fc,fcs,vsz,fs,vzs
de: vny,(ft),vd,fsz,fk,vg
pty,bgy,bg,bn,pny,bny
de: pk,pn,pc,pcs,bz,bzs
a szóvégen megengedetteken és a kötelez en hasunulókon kívül
X
A helyelemek egyeztetése a CV-fonológiában
309
d.
iii. zár + labiális obstruens
de: typ, gyb, kf
koronális zár + nemkor. zár
de: tk
Köszönetnyilvánítás
Köszönöm Siptár Péternek és Trón Viktornak a kézirattal kapcsolatos megjegyzéseiket.
A cikk megírásában a szerzõt a Research Support Scheme (1687/320/1998) támogatta.
Irodalom
Harris, John (1990) Segmental complexity and phonological government. Phonology 7, 255-300.
Harris, John (1994) English Sound Structure. Blackwell, Oxford.
Hulst, Harry G. van der & Grazyna J. Rowicka (1997) On some parallels between (un)stressed vowels and
(un)realized empty nuclei. In: Booij & van de Weijer (eds.) Phonology in progress - progress in phonology.
Holland Academic Graphics, the Hague. 125-49.
Itô, Junko and Armin R. Mester (1995) Japanese phonology. In John A. Goldsmith (ed.). The handbook of
phonological theory. Cambridge, Mass. & Oxford: Blackwell. 817--838.
Kaye, Jonathan D. (1995) Derivations and Interfaces. In: Durand, J. -- F. Katamba (szerk.) Frontiers of
phonology: atoms, structures, derivations. Longman. London. 289-332.
Kaye et al. (1985): Kaye, Jonathan D., Jean Lowenstamm, Jean-Roger Vergnaud: The internal structure of
phonological elements: a theory of charm and government. Phonology Yearbook 2, 305-28.
Kaye et al. (1990): Kaye, Jonathan D., Jean Lowenstamm, Jean-Roger Vergnaud: Constituent structure and
government in phonology. Phonology 7, 193-231.
Lakoff, George (1995) Cognitive phonology. In Goldsmith (ed.) The last phonological rule. The University of
Chicago Press. Chicago and London.
Lowenstamm, Jean (1996) CV as the only syllable type. In Durand, J. & B. Laks (szerk.) Current trends in
phonology: models and methods. Vol. 2. 419-41. Salford, Manchester.
Polgárdi Krisztina (1999) Hungarian is strict CV. Manuscript.
Rádai, Gábor (2000) Implementing Construction Grammars in GIN. To appear in Huba Bartos (ed.). Papers on
the mental lexicon.*
Rebrus, Péter and Miklós Törkenczy (1998) Phonotactics and the morphophonology of the Hungarian word.
Talk given at the International Conference on the Structure of Hungarian 4, Pécs.*
Rebrus Péter and Törkenczy Miklós (1999) Defektivitás. Talk given to the Budapest Phonology Circle, April
1999.*
Rebrus Péter (in print) Morfofonológiai jelenségek a magyarban [Morphophonological phenomena in
Hungarian]. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 3. kötet. Alaktan [Hungarian structuralist
grammar. Vol. 3. Morphology]. Akadémiai Kiadó, Budapest.
Rebrus Péter (to appear) Kormányzás-fonológia kormányzás nélkül [Government Phonology without
government]. Proceedings of the 3rd Hungarian Linguistics Phd Conference. Szeged. JATE.
Rebrus Péter (1999) A hely-elemek egyeztetése a CV-fonológiában [Agreement of place elements in CVphonology]. Proceedings of the recent methods in research on the Hungarian language conference.
Rowicka, Grazyna (1999) On Ghost Vowels. A Strict CV Approach. Doctoral dissertation, Univ. of Leyden.
LOT Dissertations; 16, Holland Academic Graphics, The Hague.
Ritter, Nancy N. (1995) The Role of Universal Grammar in Phonology: a Government Phonology Approach to
Hungarian. Doctoral dissertation. New York University. Manuscript.
Scheer, Tobias (1998) A unified model of phonological government. The Linguistic Review
Szigetvári Péter (2000) Why CVCV. The Even Yearbook. ELTE SEAS. Budapest. 117-52.
310
A helyelemek egyeztetése a CV-fonológiában
Szigetvári, Péter (2000) Deconstructing syllable structure. Ms., Eötvös Loránd University. To appear in Huba
Bartos (ed.). Papers on the mental lexicon.*
Törkenczy Miklós (1992) Vowel-zero alternations in Hungarian: a government approach. In: Kenesei István -Pléh Csaba (szerk.) Approaches to Hungarian 4: The structure of Hungarian. JATE. Szeged. 157-76.
Törkenczy Miklós (1994) A szótag. In: Kiefer Ferenc (szerk.) Strukturális magyar nyelvtan. 2. kötet. Fonológia.
Akadémiai Kiadó, Budapest.273-392.
Törkenczy, Miklós (2000) Phonotactic grammaticality and the lexicon. To appear in Huba Bartos (ed.). Papers
on the mental lexicon.*