* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download by Cristina Job Schmitt Dissertation submitted to the Faculty of the
Lithuanian grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Zulu grammar wikipedia , lookup
Udmurt grammar wikipedia , lookup
Old Irish grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
Old Norse morphology wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
English clause syntax wikipedia , lookup
Chinese grammar wikipedia , lookup
Georgian grammar wikipedia , lookup
Swedish grammar wikipedia , lookup
Lexical semantics wikipedia , lookup
Arabic grammar wikipedia , lookup
Old English grammar wikipedia , lookup
Modern Greek grammar wikipedia , lookup
Russian grammar wikipedia , lookup
French grammar wikipedia , lookup
Spanish grammar wikipedia , lookup
Turkish grammar wikipedia , lookup
Icelandic grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Portuguese grammar wikipedia , lookup
Romanian nouns wikipedia , lookup
Esperanto grammar wikipedia , lookup
Polish grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Determiner phrase wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Pipil grammar wikipedia , lookup
ABSTRACT Title of Dissertation: Cristina Job Schmitt, Doctor of Philosophy, 1995 ASPECT AND THE SYNTAX OF NOUN PHRASES Dissertation directed by: by Cristina Job Schmitt ASPECT AND THE SYNTAX OF NOUN PHRASES Professor Juan Uriagereka Department of Linguistics This dissertation shows that there are syntactic constraints involved in the aspectual interpretation of the VP that involve the internal structure of the verbal complements. Although a simple compositional semantics would predict certain interpretations, the syntactic computational system has its own mechanisms, and the relevant configurations do not always obtain. Consequently, certain readings end up being blocked. Chapter 1 Dissertation submitted to the Faculty of the Graduate School of the University of Maryland at College Park in partial fulfillment of the requirements for the degree of Doctor of Philosophy 1996 defines the relevant output conditions for aspectual interpretation based on the semantic work of Verkuyl (1995) and Krifka (1989), and proposes a syntactic configuration within the Minimalist Program for Linguistic Theory (Chomsky 1993) at which aspect is calculated. Chapter 2 shows that two classic cases of apparent grammaticalization of aspect, Finnish and Slavic can be reduced to independently needed elements in the Advisory Committee: grammar: A-quantifiers and movement to AgrO for Case checking. Chapter 3 examines Professor Juan Uriagereka, Chair Professor Norbert Hornstein Professor David Lightfoot Professor Ian Roberts Professor Bonnie Dorr Accusative Clitic Doubling in Spanish and provides a new analysis which accounts for the aspectual interpretations of the construction. Chapter 4 proposes a theory of Determiner Transparency in which definite determiners in relative clauses, certain kinds of adjectives and demonstratives do not select for their head NP and as a result, durative reading of the VP predicate are obtained if the head is non-quantized. Chapter 5 extends the analysis to "type" expressions, measure phrases and possessives (genitives). Chapter 6 provides an aspectual analysis of the copula verbs 'ser' and 'estar' in Brazilian Portuguese, and argues against their being the grammaticalization of the stage-level and individual-level distinction. © Copyright by Cristina Job Schmitt 1996 David Lightfoot was the best chair I have ever seen, a great teacher and a great friend. Acknowledgments I only regret not to have sat in more of his courses. I always like to read acknowledgments of other people's thesis because we can play the Norbert Hornstein is as much my advisor as my formal advisor. I learned an voyeur on people's lives and departments. It is my turn and I will do my share of enormous amount from him. He gave me a lot of his time and ideas, read with me a lot showing off how wonderfully surrounded I have been through my years in Maryland. Of about aspect and tense and discussed a lot of my stuff (which never sat for more than 24 course it will be impossible to thank everybody and I apologize for that. But let me first hours in his office) and always had questions for which not always I had answers. He thank the Department of Linguistics in Maryland as a whole because it is an impressive also was always very very supportive and always found some way to improve what I whole not only for its homogenous very high quality but also because it is composed by was doing, even when he didn't believe a word of it. first rate human beings. I need also to thank the department for the most amazing bit of Juan Uriagereka has been generous with ideas (which he has in profusion), criticism cultural shock I suffered in the US. I have experienced something in the department that and enthusiasm. He has read half English and half baked ideas one too many times. I was as unsettling as learning that the moon was not in the same place in Maryland as it would not be able to count how many times and in how many colors he has written in my was in my home town. I have learned that work on my stuff and work for the department papers: hectic!! I learned after a while and after some struggle that the reason he gave me were basically the same thing and that the faculty and the students were equally important so much shit was not because he hated me but because he thought I could fix the shit. I for things to work well. And everybody should be doing what they wanted so that the hope I did fix some of it. I am proud and thankful for it. He has been generous with ideas department could be healthy and productive. Everybody seemed to think that way and this (which he has in profusion) comments etc. He taught me to trust my intuitions when I is very amazing, because from where I come, if you are under, the public and the private couldn't put them into a coherent text. Thanks. are in opposite sides, and, if you are above, the public is dangerously made private. The Dave Lebeaux is also a very important figure. He has shaped a lot of the way I think idea that I could work, have fun working and learn from all these amazing people all at about linguistics. It was always great to have him around to discuss stuff and we spent a the same time was a novelty. Now I am spoiled and I think every place should run that lot of time wondering about aspect. He has the ability to work always with a big project way. I could talk about the colloquiums, the courses that were tailored to students under his hat and he also lets you go with your ideas as long as you start from the interests, the help with paper presentations, abstracts, general life and health insurance simplest case. You can't pretend or skip steps with him but always in a gentle and kind but I need to get now to the people. manner. His course on Language Acquisition was one of the greatest courses I have I will start from the beginning and thank David Lightfoot and Ian Roberts. I would never have got to Maryland if it were not for them. I am really really grateful to the unique opportunity they gave me and I hope not to have disappointed them too much. I took their ever taken. Amy Weinberg has always cheerful and encouraging. The only course I took with her was crucial to put the house in order. courses in Maryland and they were both in my committee. ii iii Sharon Inkelas left when I arrived and I suffered from that and Steve Crain and Roz Thornton were arriving when I was leaving, and I probably will suffer from that. But they provided the first big defense combo party in Maryland. Franci, Euzi Moraes and Lucia Lobato, who even called from Brasilia to congratulate me for having been accepted at Maryland. To my dorm friends Jane, Susan and Sergio thanks for the fun in the first weeks. Also here special very special thanks goes to my friend Ian Roberts who came from Also many thanks to my housemate and friend Jutta. She taught me all the practical Bangor for my defense. His courses in Maryland were great and I learned a lot from him. necessary things and was highly supportive in all senses, flowers included. Ellen-Petra My classmates and colleagues in the department of linguistics, made things nice Kester was my housemate for a month during her visit to Maryland and I thank her for (order of entrance): Joe Garman, Jeff Martin, Catherine, Stefan Martin, Peggy the company and friendship (and conversations about Spanish). Anthonisse, Li-Ling Chuan, Jairo Nunes, Marcela Depiante, Joel Hoffman, Ania Pelc- Many friends in Brazil kept my life there intact. It was always good to know I could Waters, Keiko Muromatsu, Susan Powers, of course Tor Ashen, Ricardo Echepare and come back if things screw up and they made sure I knew that. I need a couple more lives Juan Carlos. In this list I need to include in capital letters my dear colleage and friend to do to Henrique what he did for me. Thanks my dear friend, legal representative, handy Mark Arnold. Life was way better having him and Maria around. They helped me many, man, political advisor. I also need a couple of lives to do to my Ana what she did for me, many times and put me up for the defense (a scary thing to do). keys included. Also many thanks to Victor, my dearest friend, for his letters and My gratitude to my highly informed informants and their patience. For Finnish, Tor niceness. Also thanks to Pedrinho, T-shirts, etc. Thanks also to Fernando, e-mail and Ashen and Anne Vainikka. For Polish Ewa Jaworska and Ania Pelc-Waters. For Spanish irony, Rogério, posters, carinho, Seben, and first book, Arno. It is nice to know you are Ricardo Echepare, Claudia Albaladero, Marcela Depiante, Juan Uriagereka and many around. others. Thanks. It is getting hard. Now I want to thank my mother and my aunt. Their unconstrained Some other people are indirectly responsible for this work and I need to thank them. First, Leda Bisol in Brazil was my first encounter with real linguistics. I owe Leda almost moral, financial and practical support is amazing. This thesis is for them. Thanks to my sister for taking care of these two wonderful creatures. all I know about Phonology. Her and Celso Pedro Luft played a major role in making me And now we got to the most important piece in this work. Alan Munn. Without him a linguist. I owe him everything I know about the history of Portuguese. Thanks also to this thesis would have probably more pages but it would be unreadable. He heard it, read my colleagues in Brazil Maria Tasca and Rosa Hessel Silveira for their support. Some it, stepped on it (when it was on the floor and he was passing around with some idea or other Brazilian linguists have to be thanked either because the courses I took with them questions) too many times. He fixed it, helped with the computers. He was an advisor as kept me thinking about issues and or because they created events that helped me to do well on all grounds and he provided the life-support I needed to get it done. Many ideas linguistics in Brazil: Mary Kato, who organized a great winter institute in Sao Paulo, in this thesis I owe to him. He pushed me and supported me more than I deserved, in Miriam Lemle who organized a great Abralin Meeting in Rio where I met Ian Roberts, detriment of his own work I hope I can be as useful to him as he was to me. and of course David Lightfoot. Many thanks also go to Milton do Nascimento, Carlos Finally, I would like to thank Alan Beretta for kindly lending me an office and laser printer in the final stages of preparing the manuscript. iv v 2.3.2 A reflexive pronoun in Spanish..............................................108 Table of Contents Acknowledgments..................................................................................ii Table of Contents.................................................................................. vi Chapter 1 The syntax of the VP aspect..........................................................1 2.4 Chapter 3 Lack of iteration: Accusative Clitic Doubling, participial absolutes and have+agreeing participles ...........................114 Introduction ......................................................................................114 3.1 Basic Problems Posed by Accusative Clitic Doubling Constructions ............115 3.1.1 Main Issues in Previous Analyses and a Problem for Agreement in Specificity ......................................................................121 3.2 Accusative Clitic Doubling as a type of Small Clause .............................133 3.2.1. Properties of Identificational Small Clauses................................133 3.2.2 On the Notion of Strong Determiner ........................................143 3.2.2.1 Torrego's observation ...........................................143 3.2.4. The Structure for Accusative Clitic Doubling ..............................146 3.2.4 Summary .......................................................................153 3.3. The Aspectual Properties of Accusative Clitic Doubling...........................155 3.3.1 Terminativity and lack of iteration in accusative clitic doubling..........155 3.3.1.1 Case and Terminativity ..........................................155 3.3.1.2 Distinguishing Iteration from Quantification over Events . . .160 3.3.2. The Proposal ...................................................................163 3.3.3. A few observations about the data...........................................166 3.3.3.1 Habitual readings.................................................166 3.3.3.2 Plural subjects ....................................................167 3.3.4. Distinguishing clitic left dislocation and accusative clitic doubling ......168 3.4 Participial Absolutes and have+agreeing participles................................169 3.4.1. Participial Absolutes and Determiner Restrictions .........................171 3.4.1.1 Basic Facts........................................................171 3.4.1.2 Determiner Restrictions..........................................173 3.4.2. Have+agreeing Participles and Determiner Restrictions ..................177 3.4.2.1 Lack of Iteration in Absolutes and have+agreeing participles .......................................179 3.4.3. Implementing the identificational small clause for agreeing participles..181 3.5. Conclusion ..............................................................................185 Introduction .........................................................................................1 1.1 The Semantic Intuitions ..................................................................... 11 1.1.1 Basic cases of verb-complement interaction................................. 11 1.1.2 An intuitive account ............................................................ 16 1.2 Two ways of formalizing the VP aspect .............................................. 18 1.2.1 Verkuyl's feature proposal .................................................... 18 1.2.2 Krifka's analysis................................................................ 24 1.2.3 Parts and Wholes and Aspect ................................................. 27 1.3 Towards a minimalist syntactic theory of aspect..................................... 31 1.3.1 The syntactic questions ........................................................ 31 1.3.2 Basic advantages for assuming the Minimalist program ................... 36 1.3.3 Basic Assumptions on the Feature Checking Mechanism ................. 37 1.3.4 A few clear examples........................................................... 39 1.4 Previous Syntactic Accounts ........................................................... 44 1.4.1 Overview ........................................................................ 44 1.4.2 Borer (1993) .................................................................... 47 1.5 Summary ................................................................................. 51 Chapter 2 Markings on the verb and markings on the noun ................................ 52 Introduction ....................................................................................... 52 2.1 Finnish: Case Distinctions and Aspect ................................................ 57 2.1.1 Partitive vs. Accusative ........................................................ 57 2.1.1.1 The partitive/accusative distinction.............................. 59 2.1.1.2 Predictions ......................................................... 67 2.1.2 A note on the partitive of negation............................................ 72 2.1.3 Some Theoretical Implications for the Feature Checking Mechanism. . . . 80 2.1.4 Summary ........................................................................ 86 2.2 Markings on the Verb: Slavic Aspect.................................................. 86 2.2.1 Czech............................................................................. 89 2.2.2 Polish ............................................................................ 95 2.2.2.1 Perfective vs. Imperfective markings........................... 95 2.2.2.2 Accumulative na-.................................................. 99 2.3 More A-quantifiers .....................................................................104 2.3.1 Half- analyzed..................................................................104 vi Conclusions .............................................................................112 Chapter 4 Determiner transparency: relatives, adjectives and demonstratives ..........186 Introduction ......................................................................................186 4.1 Relative Clauses ........................................................................189 4.1.1 Basic Structure for Relative Clauses and its Effect on the Aspectual Interpretation: durative readings .............................................190 4.1.2 More evidence for determiner transparency ................................197 4.1.2.1 Evidence for indefinite operators in Relative Clauses........199 4.1.2.2 Evidence for definite operators in relative clauses............202 4.1.3 Terminative Readings with Plurals ..........................................204 4.1.3.1 The bounding nature of the relative clause ....................206 4.1.3.2 Summary..........................................................209 vii 4.2 The Wrong and the Difficult Adjectives..............................................210 4.2.1 Adapting Higginbotham (1985)..............................................211 4.2.2 The Difference Between Wrong and Difficult..............................214 4.2.3 Adjectives on the right: the case of Brazilian Portuguese .................219 4.2.4 Accounting for the aspectual role of wrong ................................223 4.2.5 What is so special about definite determiners? .............................225 4.3 Demonstratives..........................................................................228 4.3.1 Demonstratives as Disguised Definite Descriptions .......................229 4.3.2 A transparent determiner structure for demonstratives ....................232 4.4 Bare Nouns in Brazilian Portuguese and Aspect....................................233 4.4.1 Heads of definite relative clauses are NumPs and not DPs or NPs .......237 4.4.2 Evidence for Lack of Number in Bare Count Nouns......................239 4.4.2.1 Conjoined Bare Count Nouns: Evidence for DPs with no NumP ..............................240 4.4.2.2 Anaphora..........................................................246 4.4.3 Similarities between Bare Nouns and Bare Plurals........................251 4.4.3.1 Opacity and Scope ...............................................251 4.4.3.2 Lack of quantificational readings...............................254 4.4.4 Differences between the Bare Count Nouns, Bare Mass Nouns and Bare Plurals. ..............................................................259 4.4.4.1 Bare Noun and Mass Noun Restrictions ......................259 4.4.5 Summary .........................................................................267 4.5 Antecedent Contained Deletion as an argument for Determiner Transparency . .267 4.5.1 On the Necessity of Copies...................................................268 4.5.2 ACD and identity of AgrOs...................................................273 4.5.3 Empirical Support .............................................................277 4.5.3.1 Pied-piping........................................................277 4.5.3.2 Coordination......................................................278 4.5.3.3 Quantifier restrictions............................................280 4.5.3.4 Idioms .............................................................285 4.5.3.5 Focal Stress.......................................................287 4.5.4 Summary .......................................................................288 5.2.3 Measures and Definites .......................................................320 5.3 Chapter 6 Subjects and aspect: the case of ser and estar...................................352 Introduction ......................................................................................352 6.1 Subjects and the Stage-Level/Individual-Level distinction.........................353 6.1.1 The role of the subject.........................................................353 6.1.2 The stage level/individual level distinction..................................357 6.2 Against the VP/IP split hypothesis ...................................................360 6.2.1 The VP/IP split hypothesis ...................................................361 6.2.2 Against two different D-structures...........................................362 6.2.3 Against two different Infls....................................................364 6.2.4 Ser and estar and the stage-level contexts...................................370 6.3 An aspectual analysis...................................................................373 6.3.1 Reanalyzing the tests for Davidsonian arguments .........................373 6.3.2 Ser /estar: an aspectual distinction ...........................................380 6.3.3 Accounting for the lack of estar in stage-level contexts ...................386 6.4 Aspectual composition: the act be.....................................................386 6.4.1 Explaining the ability of ser to appear in SL contexts .....................387 6.4.2 Aspectual composition ........................................................388 6.4.3 The Perfective/Imperfective distinction in Brazilian Portuguese .........389 6.4.3.1 Pronoun interpretation and the perfective/imperfective distinction .........................................................390 6.4.3.2 Bare nouns and the perfective and imperfective distinction . 391 6.4.3.3 Stative/eventive verbs and the perfective/imperfective distinction .........................................................392 6.4.3.4 Existential Constructions in the Perfective ....................393 6.4.4 Deriving the ACT BE .........................................................397 6.5 Subject interpretations..................................................................400 6.5.1 Evidence for [+N] and [+V]..................................................401 6.5.2 The Ordered Argument Method and the Neo-Davidsonian Method . . . . .405 6.5.3 More evidence for subject differences.......................................408 6.5.3.1 Purpose clauses ..................................................408 6.5.3.2 Of John and John alternations ..................................409 6.5.4 Two types of small clauses: some speculations............................410 6.6 Final Chapter 5 A note on types, measures and possessives.....................................289 Introduction ......................................................................................289 5.1 5.2 Overt Type Expressions ...............................................................293 5.1.1 The Basic Semantic Intuitions................................................294 5.1.2 Structure ........................................................................297 5.1.2.1 Basic Syntactic Properties.......................................297 5.1.2.2 Adjectival Restrictions on type expressions...................301 5.1.2.3 The structure......................................................305 5.1.2.4 Analyzing two books of the type Mary likes..................306 5.1.2.5 Type Expressions in Brazilian Portuguese....................308 5.1.3 The Aspectual Properties of type expressions..............................311 5.1.4 Summary .......................................................................313 Measure Phrases........................................................................314 5.2.1 Main Differences between Type and Kilo ..................................314 5.2.2 Tentative Structure and Some Predictions ..................................319 viii Possessives and Partitives.............................................................323 5.3.1 John's books and the books of the library have a similar structure......324 5.3.1.1 -s Possessives ....................................................326 5.3.1.2 of Possessives....................................................328 5.3.1.3 Brazilian Portuguese Possessive de............................331 5.3.2 Two problems of John's is not John's two problems.....................332 5.3.3 Brazilian Portuguese possessive pronouns.................................339 5.3.4 Summary .......................................................................350 Summary..........................................................................414 Conclusion .......................................................................................416 ix References........................................................................................421 Chapter 1 The Syntax of the VP aspect Introduction The Minimalist Program (Chomsky 1993, 1994, 1995) represents a significant departure from previous versions of generative syntax in striving to radically reduce the amount of theoretical 'machinery' required by the grammatical theory. Chomsky poses the question in terms of what would constitute the minimal theory of the classical characterization of language as the relation that holds between sound/meaning pairs. Since neither sound (or for signed languages, the manual implementation of signs) nor meaning can be dispensed with in any grammatical theory, the minimal (and therefore, for Chomsky, the perfect) theory would satisfy simply the conditions imposed by the two cognitive interfaces that represent them: the Perceptual-Articulatory interface (i.e. sound) and the ConceptualIntentional interface (i.e. meaning). Chomsky calls these the "bare output conditions", and argues that the best grammatical theory should satisfy these conditions and no others. The problems are empirical "... and we can hope to resolve them by learning more about the language faculty and the systems with which it interacts. We proceed in the only possible way: by making tentative assumptions about the external systems and proceeding from there." (Chomsky 1995 p.6). This thesis is an exercise in this direction. I will make assumptions about the external system's minimal needs for aspectual interpretation and I will proceed from there. The body of the thesis was supposed to have two parts, and it does. However, as is often the way with dissertations, the first part expanded to the point that it encompasses five of the six chapters. Rather than have two rather lopsided parts, I have opted for five chapters on VP aspect and one chapter on a particular instantiation of sentential aspect. The matching four chapters on sentential aspect, and an envisioned third part will have to wait. x 1 Aspectual properties of verbal predicates have been investigated from various In syntax, it is the first line of investigation, based on the lexical properties of verbs, perspectives. In the semantics literature we can distinguish two main cuts in the kinds of that has had the most impact in the syntactic studies of aspect and not the line that data that constitute what has been called aspect. The first cut focuses on verbal properties emphasizes the compositionality. In fact, in the Principles and Parameters framework, the and can be taken as an attempt to understand how we store information about the world. little work done on aspect sharply maintained the idea that aspect is a purely verbal Various classifications of verbs have been proposed according to various different types property. Moreover, the work has maintained the traditional division between lexical of tests. The most influential proposal is the so-called Vendler-Dowty classification, aspect (aktionsart) and inflectional aspect, although the cut is not as clear as it may seem, which divides verbs into four classes: states (know, sit); processes (run, walk), since the same core distinction – bounded/unbounded – is present at the VP level and at accomplishments (build, draw) and achievements (die, for example). the IP level (the perfective/imperfective morphology in Portuguese and Spanish, for A problem with such classifications is that verbs seem to move from one class to example). another. Thus, while walk is a process, and consequently, although it progresses in time, In any case, more or less informally, syntactic treatments of aspect have focused on it has no culmination point, walk home or walk a mile belongs to the group of verbs that the aspectual morphology and have accommodated it into the grammar as a functional have a culmination point, just like build a house does. At the same time, build a house is head, on a par with Tense and Agreement, following Pollock's (1989) work (see for different from build houses in the sense that the first describes a temporally bounded or example, Belletti 1990). Various features have been proposed, and in all these analyses terminative event and the second, a temporally unbounded or durative event. the aspectual projections are taken to be part of the extended projection of verbs. In other words, verbs like walk and build have to be made ambiguous between a A more prominent role has, however, been given to aspect as a mediator between process (i.e., an event with no culmination point) and an accomplishment, (roughly an lexical semantics and syntax. In Grimshaw's (1992) definition of argument structure, for event with a culmination point). However, the work of distinguishing the ambiguity is example, aspect plays together with the thematic hierarchy, an important role on defining clearly on the nominal complement and not on the verb. These facts suggest that at least prominence relations among arguments. part of the classification is spurious and that what we are seeing here is a clear example of These studies acknowledge, but leave formally unaccounted for, the role that the compositionality. Aspect is compositional, i.e., it is a combination of verbal and nominal determiners in the argument position play in the composition of the VP aspect. The properties and not purely a property of verbs. reasons for this seem, at first glance, reasonable: first, why bother with such interaction The interaction between nominal and verbal elements (which has been observed in the in the syntax, if it seems that a very simple compositional semantics of verb+complement semantic literature by Verkuyl 1972; Krifka 1989 and others) constitutes the second type can derive the aspectual properties of the VP that relate to the internal arguments. In other of cut on what is taken to be the core aspect data. Here, it is more important to examine words, whatever way we interpret thematic roles and the compositionality between the the aspectual interpretations that result from the interaction between nominal and verbal internal arguments and the verb can also account for the aspectual properties of the VP. elements in a compositional way than to examine fine-grained lexical differences between The second reason is that the role of the complement is taken as secondary, since the verbs. 2 3 aspectual properties are taken to be verbal in nature, in that they characterize the internal (1) a. Tuula rakensi taloa. temporal configuration of an event. (Finnish) Tuula built house-PART This dissertation shows that there are syntactic constraints involved in the aspectual 'Tuula was building a/the house.' interpretation of the VP that involve the internal structure of the verbal complements. Although a simple compositional semantics would predict certain interpretations, the b. Tuula rakensi talon. syntactic computational system has its own mechanisms, and the relevant configurations Tuula built house-ACC do not always obtain. Consequently, certain readings end up being blocked. 'Tuula built a/the house.' A parallel can be made here to quantifier scope properties (May 1985). Although scope is essentially a semantic property, it is dependent on the syntactic configuration. Is partitive case an aspectual marking, i.e., a feature [durative], or is partitive forcing For example, the set that a quantifier "lives on" (a semantic property) is determined by the a certain configuration that cannot be interpreted as having a culminating point? Or is NP the quantifier governs (a syntactic property). Similarly, wide and narrow scope partitive the lack of a [completed] feature as proposed by Vainikka (1993)? How do the readings of quantifiers (semantic properties) are determined via Quantifier Raising and c- facts in (1) relate to the English cases? Is there a parametric aspectual distinction between command (syntactic properties). In a very similar way, aspectual interpretations are Finnish and English or is there a unified way to account for the durative/terminative essentially semantic, but certain interpretations will be available or not depending on what alternations in interpretation without appealing to extra features? If we add the Slavic languages into the discussion, the same things that English does is forced or allowed by the syntax. Such an approach encounters immediately two challenges: first, the amount of cross- with determiners and Finnish does with case, Czech and Polish seem to do with their so- linguistic variation in the ways aspect appears to be encoded seems quite large. Second, if called aspectual morphology. How can these different "grammaticalizations" of aspect be we assume that certain configurations will determine certain readings, then we need to related? define what is relevant for aspectual interpretation and what are the relevant Could we talk about a [completed] feature in D for English, in the Case morphology configurations. A few examples at this point may be helpful to understand what I mean for Finnish and in prefixes for Polish? In other words, are we to assume that languages by a challenge. Consider first the case of Finnish. Why is it that the partitive case in encode terminativity on case, determiners or verbal markings, and are these yes/no type Finnish always forces durative readings, as exemplified by the contrast between (1a) and of parametric choices? The variation, however, does not stop there. In Spanish, a clitic can block iteration of (1b)? the VP. Consider the example below: 4 5 (2) a. Toqué la sonata hasta las 12, de hecho la toqué 10 veces. (4) John wrote the long papers #for years. (Cordoba Spanish) (I) played the sonata until 12, in fact (I) played it 10 times. In the example above, the only possible reading is what I will call a stretched reading. In such a reading John just took many years writing a defined set of long papers. In (3b), b. Lai toqué [a la sonata]i #hasta las 12, de hecho la toqué 10 veces. on the other hand, we do not need to have a pre-established number of papers that he (I) iti played [a the sonata]i until 12, in fact (I) played it 10 times wrote that happen to be wrong. We can have a reading in which John wrote papers and 'I played the sonata until 12, in fact I played it 10 times.' they were all the wrong ones. At this point we can do one of two things: we can either decide that aspect is an We can we force a repetitive reading of the object in (19a) but not in (19b), a case of absolutely empty label that includes determiners, case markings, clitics, adjectives and accusative clitic doubling in Spanish. Is there something special about this particular clitic adverbials (not to mention the actual morphological aspectual markers such as those that triggers some special feature in Spanish. Or is the LF configuration created by found in the Slavic languages). In this case, we might as well quickly change subjects. accusative clitic doubling, which is driven by its formal features and not by semantic Alternatively, we can try to understand what are the relevant properties for aspectual features, blocking the configuration that would allow iterative readings. Can we predict interpretations, what are the configurations that will produce certain readings and what are when iteration will be disallowed or is this a matter of pragmatics and context? the internal structure of the complements in the examples above that will produce certain The same question can be put forth for the contrast in (3) when compared to build readings or not. And from there we can decide whether there is something to the label examples. While the contrast between build houses and build the house seems to be aspect that has anything to do with the syntax or whether aspect is purely a semantic- related to the alternation between a definite determiner and a bare plural, in (3) the pragmatic property. alternation seems to be related to the presence of an adjective. Drawing on the work by Verkuyl (1993) and Krifka (1990) on aspect and Moltmann's (1994) conceptual view of the importance of part-whole relations in the (3) a. John wrote the papers #for years. semantics of natural language, I will make some basic assumptions about the relevant b. John wrote the wrong papers #for years. conditions required at LF for certain aspectual interpretations and I will propose that AgrO What is the adjective wrong doing to the complement that now durative readings are possible. Why should this be so? Is wrong an aspectual marker as well? Or is wrong forcing the internal structure of the complement to be such that the relevant features will not be in the appropriate position by the time aspect is calculated? It certainly cannot be is the locus for the aspectual calculation of the VP. Using minimalist assumptions, I examine the structural conditions and syntactic constraints that result in different AgrO configurations and consequently in different aspectual interpretations. My emphasis will be on the role that the syntax of noun phrase complements plays, and its consequences for the VP aspectual interpretations, rather than on different verb classifications. In fact I the case the all modifiers play the same role: will have very little to say about latter, since I am interested in the syntactic mechanisms 6 7 that will force one reading or another and not on lexical distinctions. In fact in the first movement to such a position. I also show that both Spanish and English have similar five chapters I will keep the verbs "constant" and I will vary only the internal arguments. constructions to the Czech, Polish and Finnish cases. These cases support the thesis that In chapter 6 I will vary verbs and subjects. aspect cannot be treated as parameterizable, in the sense that type A languages will encode My basic interest is to reduce various types of crosslinguistic variation of what has aspect in x way and type B languages will encode aspect in way y. been called aspectual markings to independently needed variation in how languages In Chapter 3, I deal with accusative clitic doubling in Argentinian Spanish. In terms encode quantificational elements and in how phrase structural properties affect case of aspect, accusative clitic doubling is, as the example above shows, an interesting case, marking. I will also make an argument for the use aspectual interpretations as a tool to since this construction blocks iteration of the verbal predicate. I propose that accusative investigate the syntax of noun phrases. clitic doubling constructions are identificational small clauses headed by the preposition a. A more general goal is to deny the idea of a parameter for aspect that the child has to The clitic is a pleonastic element that does not play a role in terms of interpretation but set. I will argue that it does not make sense to talk about a parameter of a semantic raises to the specifier of AgrO in order to get case. The structure proposed for accusative property that is absolutely compositional. In other words, we don't learn different clitic doubling allows us not only to account for the standard problems proposed by this aspectual features that a language chooses to activate or not. We learn words and constructions but also to explain the fact that accusative clitic doubling disallows iteration morphology: we learn whether features are strong or weak and we learn how the pieces of verbal predicates. The same analysis proposed for the accusative clitic doubling can of words are put together. But that is it. We can only talk about learning the aspectual account for two other apparently unrelated constructions: participial absolutes and have + features of a language to the extent that we can talk about learning what quantifiers and participle agreement constructions in Portuguese and Spanish. Moreover it provides us verbs mean. Because that is what aspect is: a combination of verbal and nominal features. with a window to understand certain differences among clitic doubling and clitic left This thesis is divided as follows. In this chapter I present the basic proposal and dislocation structures. discuss basic cases. Chapters 2 to 5 are case studies on different types of complements In Chapter 4, I propose a theory of determiner transparency in order to account for a and their role on the aspectual composition of the VP. Chapter 6 discusses the interaction set of apparent exceptions to the descriptive generalization that definite direct objects of between the verbal elements and the subject. eventive verbs always give rise to terminative readings. I show that there is a principled In Chapter 2 I discuss the case of Finnish aspect and case correlations in detail, and class of NP modifiers which, although appear to modify a definite NP, allow durative also discuss another well-known example of aspectual marking: the Slavic languages readings just in case the NP modified is a plural or mass. The structure I propose (similar (specifically Polish and Czech). These cases will allow us to start to test the proposal to Vergnaud 1985) is such that the head of the relative clause raises alone to the specifier made in this chapter and to address the problem of cross-linguistic variation. My goal is of AgrO in order to check case. This will force the VP to be interpreted as durative only to show that sometimes the cardinality of the internal argument is not visible for aspectual when the head noun is a "bare" plural. This analysis will allows also to investigate the calculations, because the right configuration does not obtain. I will do that, however, properties of bare count nouns in Brazilian Portuguese. without adding extra functional projections and without appealing to semantic reasons for 8 9 I argue that determiner transparency in relative clauses is independently required to data relating the interaction of perfective markers and subjects. Chapter 6 is a first account for Antecedent Contained Deletion and I show that, in addition to relative clauses, approximation on the role of subjects in the aspectual composition. demonstratives, certain types of adjectives (wrong, for example, but not blue) all show 1.1 The Semantic Intuitions determiner transparency effects. Chapter 5 is an extension of chapter 4. There I discuss other types of complements that force certain aspectual readings of the VP predicate: overt type expressions, which always force durative readings when in the object position of the VP predicate, possessives in English and Brazilian Portuguese which will vary according to the possessed noun characteristics and measure phrases (which force terminative readings). The analysis proposed supports the idea that depending on the internal structure of the DP complements, certain interpretations may be allowed or blocked. In Chapter 6, I make an attempt to extend the proposal I made for AgrO to AgrS . Most of the chapter is dedicated to a case study of two copula verbs, ser and estar, in Brazilian Portuguese. I argue against the idea that ser and estar are the lexicalizations of the distinction between individual-level and stage-level predicates. Instead, I argue, following the traditional view, that there is an aspectual distinction between ser and estar and that ser, being aspectualless, incorporates some other aspectual element in the sentence (by aspectual I mean an element that has to be interpreted in its relation with tense.) The emptiness of ser provides us with a clean place from which to start an investigation of the different verbal modifiers and their role on the composition of the aspect and will allow us to provide an analysis of the so-called ACT BE interpretations of The purpose here is not to give a fair and/or extensive review of the literature on VP aspect, but just (i) to describe the semantic intuitions related to the VP aspect; (ii) to present two ways of formalizing the verb-complement interactions in the semantics; (iii) to present the conceptual arguments for AgrO as the locus for calculating aspect; (iv) to discuss one syntactic treatments of aspect in order to contrast them with respect to my proposal and with respect to minimalist assumptions. Within VP, as I illustrated in the introduction, at least three apparently distinct syntactic objects go under the label of aspect: lexical properties of verbs, morphological affixes on verbs, and verb-complement interactions. The first two have received a more or less formal treatment within Government and Binding theory, while the third has received very little attention with the exception of Borer's (1993) work, as we will see. In the semantic literature, however, not only the morphological and lexical properties of verbs have been the goal of a fair amount of work, but also the interactions between verb and complements have been systematically part of the studies on aspect. In this section I present the basic facts from English that illustrate the verb and complement interactions, and the basic semantic intuitions. At the end of the section, I present two ways of formalizing the semantics of the VP aspect. ser (which are akin to the one found in John is being cruel. in English). This last chapter 1 . 1 . 1 Basic cases of verb-complement interaction is the most programmatic one and less coherent and should probably be split into two or three much more careful chapters. I opted for leaving it in the final version of this thesis The role of the complements in inducing different aspectual interpretations are the core of because I believe it makes a definitive point against the grammaticalization attempts of the study of this thesis. There are two ways in which the DP complement can affect the stage-level and individual-level predicates and because it brings forth an interesting set of 10 11 internal structure of the event: it can provide an end point and it can provide a way to partition the event into subparts.1 In (6a) we see that an eventive intransitive verb gives rise to a durative aspect as does a bare plural measure phrase (6b)3, since both sentences can be followed by the adverbial I will now run briefly through the classical examples of VP aspect that illustrate the role of verbs and objects in providing an end point to the event. These are given in (5), for an hour. However, (6c) shows that it is possible to derive a terminative reading if a complement is added and we know how much of it there is (in this case, one lap.) (6), (7) and (8) below. In these examples the verbs are kept constant, and the complement type varies producing different readings. Consider (5) first: In (7a) and (7b) we have, on the other hand, a stative verb. No matter what kind of complement these verbs take (either bare plurals, mass nouns or definite noun phrases ((7a) and (7b) respectively), the result will still produce durative readings. (5) a. #John built a house for a month. b. #John built the house(s) for a month. (7) a. John knew languages/French. c. John built houses for a month. b. John knew the man. Build a house, as we have already briefly discussed, is different from build houses in Finally, consider (8). In (8a) we can see that push yields a durative reading even the sense that the first describes a temporally bounded event (which is incompatible with when the object is a definite singular. However, in (8b) and (8c), we can see that the the adverbial for x time) and the second, a temporally unbounded event (which is complex verb push away will provide durative or terminative readings depending on what compatible with the for x time adverbial)2 . Notice that the verb is the same. The is in the object position. distinction between the two aspectual interpretations (bounded versus unbounded) is due (8) a. John pushed the cart for 3 hours/ #in an hour. to the fact that in (5a) we know how much of the object (houses) we have and in (5b) we b. John pushed the cart away #for 3 hours/ in five minutes. don't. Now we can consider (6). c. John pushed carts away for 3 hours/ #in five minutes. (6) a. John ran for an hour. b. John ran laps for an hour. The data from (5) to (8) illustrate the basic facts about the compositionality of aspect. c. #John ran one lap for an hour. Aspect is a combination of verbal and nominal properties and not purely a property of verbs. Certain verbs will always produce durative readings. Other verbs will be dependent on what is in the object position. 1 Throughout this thesis I will call events the V+internal argument structure. When the subject is included and other higher aspectual markers are included we have what I will call a situation, following Verkuyl 1993. Nothing particularly deep hinges on the distinction between events and situations here. It is a matter of the level in the phrase structure we are dealing with. durative/unbounded reading, but apparently there is only one way to derive a terminative 2 Following Verkuyl 1993, I will use the # to mark sentences that are unacceptable under the intended interpretation. Usually the intended interpretation is durative with for x time adverbials, and terminative with in x time adverbials. Examples will rarely be completely ungrammatical, as terminative predicates can take for x time adverbial if interpreted as iterated or with an artificially stretched meaning. 3 12 The data above also shows another important point: there are various ways to derive a Whether laps is interpreted as a measure or a direct object is not at issue here. 13 reading, and this involves the presence of certain nominal elements besides the presence of a certain verb. Consequently, verbs per se are not terminative.4 (10) a. John drank water for an hour. (#repetitive/ stretched reading) b. John ate apples for an hour. In English it seems that terminative readings, i.e., bounded readings, are partially the c. John hit the ball for an hour/ in an instant. responsibility of the determiner system, as the examples above illustrated. In languages d. John arranged sand for an hour. that do not have overt determiners, other mechanisms seem to be used. Thus, in Finnish e. #The coal miners hit rock for an hour. (which does not have overt determiners), as the example in (1) repeated here as (9) f. John arranged the sand for an hour. (#repetitive/ stretched reading) shows, case distinctions, at first sight, will force durative or terminative readings (but see Just as water has no individual parts but apples has as many individual parts as there Chapter 2). are apples, in a predicate like drink water, it is not possible to identify little parts of water (9) a. Tuula rakensi taloa. (Finnish) Tuula built house-PART (unless some external measure is added) (10a). We cannot consequently identify discrete subevents of drinking water. 'Tuula was building a/the house.' Now consider eat apples. In (10b) we can map subevents into individual apples. Apples, being countable, allows us to partition the event into discrete subevents. Now b. Tuula rakensi talon. consider a predicate like hit the ball in (10c). Here we can tell when the event is over, Tuula built house-ACC which would suggest that this predicate is compatible with in an hour but not with for an 'Tuula built a/the house.' hour. However, when we add the adverb for an hour, the result is acceptable with a repetitive reading, as illustrated in (10c). Because ball is a count noun, we can divide the The second property of the object that can affect the type of event we get is the event into subevents with discrete subparts of hitting the ball. In other words, while we object's ability or inability to partition the event into discrete subparts. cannot arrange sand for an hour (10d), or hit rock for an hour (10e), with repetitive readings, such a reading is possible if the complement is count and repetition is pragmatically acceptable. From what I have said above, it should be clear that the notion of part is basic in the discussion of aspect (viewed as the properties related to the temporal constitution of an event.) Although we think of time and verbs as having no intrinsic discrete partstructures, there are various ways to impose a part structure on events. The participant 4 Verbs like cough pose interesting questions with respect to their lexical aspect. One might be tempted to say that they are inherently terminative, since John coughed for five minutes yields an iterative reading. However, the relative unacceptability of ??John coughed in a minute leads me to suspect that they are inherently iterative, and therefore behave like most other intransitive verbs in being durative due to their lacking a complement. 14 related (i.e. argument related) part-structure is the part-structure that interest us. More 15 precisely, my interest here is how the internal argument can act so that it can impose a part-structure onto the event. But events can have parts that are associated with the properties of the internal arguments, and it is the ability to figure out discrete subparts of an event that allows us to perceive progress in time not as a dense property but as made of discrete subparts. 1 . 1 . 2 An intuitive account Intuitively we can think that a count noun in the object position of an eventive verb can One way to account for the facts above is based on two assumptions: first, verbs can't provide a measure just as kilo can provide a measure to flour in a kilo of flour (see count. In this sense, verbs are to be grouped with mass nouns in not having individual chapter 4). To produce a "count" VP, a property of the internal argument has to act as a discrete parts; second, only nouns have the property of individuation. Thus any aspectual natural measure on the verbal element that selects it as a measure. Moreover, count nouns property that depends on the individuation of participant related parts of the event, or in in object position can also specify when the event culminates by specifying the quantity of counting how many partitions of the participants there are in order to define the end point the object. Thus in eat three apples, once the third apple is eaten, the event is over. But in of an event, is to be associated with nouns. eat meat or eat hamburgers we cannot establish an end point for the event. Evidence that verbs are, in a sense, like mass terms comes from quantifiers such as In sum, the basic intuition is that eventive verbs can use the internal argument in such too much in English. These quantifiers, as observed by Moltmann (1994), can appear a way as to impose a part-structure onto the event. Depending on what is in the object with mass nouns but not with count expressions: we can say too much milk but not too position, the VP can be interpreted as having discrete parts or not. Moreover, depending much apples. We also can say run too much but not run a mile too much.5 The on whether we have information about how much of the object we have, we can decide quantifier too much can actually take scope over mass nouns and verbs, but not over whether the event has an end point or not. The most important distinction between eat verbs that combine with internal arguments and produce a count VP. These adverbs select three steaks and eat meat is the fact that only the latter is unconstrained because we do not for homogeneous predicates. A homogeneous predicate is a predicate that is both know how much meat we have. cumulative and divisive.6 By cumulative we mean that the sum of two events of eating Among the analyses proposed to deal with the semantic composition between the verb meat is still an event of eating meat. The sum of two events of eating three steaks is on the and the complement are Verkuyl (1972; 1993) and Krifka (1990). Verkuyl (1972) other hand an event of eating six steaks. Thus eat three steaks cannot be taken as homogeneous; it has a discrete part structure. proposes an analysis in terms of feature composition that is translated and improved in terms of model-theoretic semantics in Verkuyl (1993). Krifka, on the other hand, analyses the aspectual composition in terms of lattices.7 Although they differ quite radically on their ontologies—Verkuyl takes events as construed and Krifka departs from 5 The reading of running a mile too much that is equivalent to running a mile too far is irrelevant to the point I am making here. 6 Cumulativity and divisivity are defined as follows (from Moltmann forthcoming p.28) (i) a one-place predicate N is cumulative iff ∀xy (x ∈ [N] ∧ y ∈ [N] → sup< ({x,y}) ∈ [N]) Roughly, if x is water and y is water then the maximal quantity of x and y is also water. (ii) a one-place predicate N is divisive iff ∀xy (x ∈ [N] ∧ y<x → y ∈ [N]) Roughly, if x is water and y is part of x then y is water. 16 7 A lattice is a partial order A = <A,∧>. The pair <A,∧> is a relational structure where A is a non empty set and ∧ is the relation part-of (reflexive, transitive and antisymmetric). Intuitively, what a partial order does with a set is "to give a qualitative notion of comparison for it, an ordering in terms of structure, rather than in terms of numerical values" Landman 1991 p.85). 17 the assumption that events are primitives—their analyses share three properties, listed in will correspond to the role of adverbials and other elements. They can modify the "inner (11). aspect", but not undo it. The metaphor that Verkuyl (1993; 1995) uses is to say that the verb is like an (11) Aspectual Properties of the VP odometer in a car. Although we experience the progress in space in terms of the Real (i) aspect is compositional: properties of the Verb and properties of the Noun interact to derive the VP aspect; numbers (since space is infinitely divisible), we keep track of the progress we are making (ii) verbs have an additive property, i.e., they can compose with the internal in terms of the Naturals numbers. the natural numbers on the odometer, or objects along arguments in order to establish discrete subparts on the temporal axis; (iii) the internal argument provides crucial information for the calculation of the aspect of the VP. The boundedness of an event is defined in terms of the the road (the actual measuring units are irrelevant). The internal argument and the verb create a PATH, i.e., a set of indices and positions, which allows us a way to count the event. information provided by the internal argument. The external argument has to "go through" this PATH. I will concentrate on the role of the internal argument and its relation with the verb, since this is the focus of most of this 1.2 Two ways of formalizing the VP aspect thesis. In the following I summarize both Verkuyl's and Krifka's proposals, in order to show In other words, a function amalgamates the internal argument and the verb associating two different ways to formalize the semantics of aspect. I also add a discussion of part- indices with partitions of the set of the elements of the internal argument. Then another whole relations from Moltmann, which will bring about another dimension to the function takes the subject as the domain for the VP. discussion of part-structure. Note, however, that the particular formalization of the According to Verkuyl, the VP aspect can be reduced to a system of three features: (i) a semantics of the VP aspect is not particularly important for what I have to say, and, in parameter on the verb, which decides whether the verb can combine with the object in most cases, I will not choose one semantics over the other. Such is not the goal of this order to use the object as a counting device; (ii) a nominal parameter, which decides dissertation. What I am interested is in the syntactic interface conditions for the semantic whether the noun has its cardinality specified or not; and (iii) a thematic relation, which interpretation. decides whether the object has to be totally affected or not.8 The temporal parameter (verbal) is encoded by the feature [ADD TO] which, in a very rough manner, distinguishes states [–ADD TO] from events [+ADD TO]9. Events include 1 . 2 . 1 Verkuyl's feature proposal what have been called accomplishments, achievements and processes in the VendlerVerkuyl departs from the hypothesis that aspect is compositional. The "inner aspect" Dowty classification. corresponds roughly to the composition between the verb and its internal arguments. It is determined by verbal and nominal features. A major role is played by the internal argument, but a not irrelevant role is played by the external argument. The outer aspect 18 8 All three features will reappear in Krifka's system, as we will see below. 9 Both states and events will be taken as denoting eventualities. 19 The feature [+ADD TO] captures the notion of progress in time and space and can be b. An NP of the form Det N denotes a Unspecified Quantity of A in E relative to B if: (a) A ∩ B = ∅; or (b) | A ∩ B | cannot be determined. described by a successor function operating on intervals. Progress in time must be expressed by a sort of numerical index contributed by the verb which is to be "tuned to" c. A set S is bounded if there is an m ∈ Z+ (= Ν \ {0}) such that for all xi its internal argument. In other words, the feature [+ADD TO] specifies the verb not only as ∈ S, i ≤ m (i a number assigned to members of S). a non-state but also as able to combine with the nominal internal argument and create a (Verkuyl 1993:92) PATH . Then the denotation of the subject is mapped onto the VP, giving us the inner aspect. The intuition here is that if there is information about the quantity of the NP, be it The function that creates a PATH adds up measuring units to what is already given. It two, three or some, many , at least then the NP is to be interpreted as [+SQA]. is a successor function s: I → Ι, where I={(0,n): n ∈ N} and is defined as follows Otherwise, it is [-SQA]. If the intersection of A and B is null or cannot be determined, (Verkuyl 1993:222): then we do not know how much of the object we have and the set is taken to be unbounded. (12) ∀i ∈ Ι, if i = (0,n), s(i) = (0, n+1) (where I=interval, and i=subinterval) In other words, if there is no internal argument, as in John walked, the progress is indefinite and unconstrained, because there is no way to restrict the indices brought about by the [+ADD TO]. Remember, the function that creates a PATH has the cardinality of the Intuitively what (12) says is that at when we get to index n+1, the information of n is object as its range. If there is no object, all the members of the domain of the function l preserved. In the same way at n+2 the information of n and n+1 is also preserved. are mapped into an empty set and the result is durative. The atemporal parameter (nominal) roughly distinguishes whether the cardinality of If we have, on the other hand, Mary lifted four tables, here we can constrain the the nominal arguments can or cannot be specified in some way. It is encoded by the progress. A function l relates the [+ADD TO] to its internal argument. Its domain is the set feature [SQA ], which stands for "Specified Quantity of". This is a feature of the of indices provided by the [+ADD TO] verb. The range is the collection of cells of a Determiner system. In Verkuyl (1993) [SQA ] is defined in the Standard Theory of Generalized Quantification in the following way:10 partition of the set of four tables. A partition of a set W of four tables has at most as many cells as there are members of W. One of the possible partitions is exemplified below: (13) a. An NP of the form Det N denotes a Specified Quantity of A in E relative to (14) B iff A ∩ B is bounded. l = {< 1, {t1}>, < 2, {t2, t3}>, < 3, {t4}> Schematically we have the following: 10 These definitions are modified in Verkuyl 1993 to accommodate the assumption that NPs should be treated as <<e,t>,t>,t> and not as <<e,t>,t>. For our discussion here this is not crucial. 20 21 (15) not. Only if the properties of the verb and the properties of the internal argument NP meet NP {{t1}} {{t2, t3}} what Verkuyl calls the Plus Principle ([+ADD TO ], [+S Q A ]), is the predicate {{t4}} terminative.11 Function l Verb 0 For a situation to be bounded the verb has to be [+ADD TO] and the arguments have to •___________•___________• i i+1 i+2 be [+SQA]. In other words the Plus Principle says that not only does the verb have to have a positive value, but we must know the cardinality of its arguments. Here the PATH l is defined in connection with the function that makes the external argument x the domain of the function l. The function π relates the external argument to A schematic tree for the inner aspect in terms of the functions described above is given below: the VP. The pair <i,p> contains information about the positions at which the external (17) argument is at an index i. (16) PATH of S [+TerminativeS] NP[+SQA] VP [+TerminativeVP] π → l The PATH of l of a member x of the external argument denotation D is assigned to x by the function π: D→ (N → DL) defined by π(x )= l where: V[+ADD TO] NP[+SQA] l → The only other element we need to define the VP aspect is the thematic relation that l = {<i,p > : [[(p) (x)]]M,i =1} (Verkuyl 1995:8) holds between the [+ADD TO] verb and the object. The thematic relation can acquire two forms: = or ⊆. The first indicates that the object has been totally affected and the second Now, terminative aspect is the marked form. Durative is the "elsewhere condition" of aspect. In other words, although there is only one way to derive a terminative aspect, there are various ways to derive a durative aspect. According to Verkuyl, to derive a terminative aspect, the verb has to have [+ADD TO] feature, and the arguments have to have their cardinality specified, i.e., they have to be [+SQA]. For the predicate to be durative, it is enough for the verb to be [–ADD TO]. If the verb is [+ADD TO], then the predicate can still be durative if a bare plural, for example, appears in object position. Being [–SQA], the bare plural will create an unbounded PATH and the one that it has been partially affected by the verb. Either the verb is used up completely in the event or not. Some verbs that impose a thematic relation ⊆ with respect to the noun complement are the following: push, mow, iron, paint, help etc. For these verbs it doesn't matter what the cardinality of the object is. A durative reading will be always available. Among the verbs that impose = are eat, build, drink, write etc. For those verbs, the cardinality of the object is crucial do define whether the VP is durative or terminative. In most of this thesis, I will be dealing with the latter type of verbs, as they are the ones whose direct objects induce terminative readings. interpretation will be durative. For example, build houses is interpreted as durative but build the house or build every house will be interpreted as terminative. In other words, the arguments of certain types of verbs will determine whether the aspect is durative or 11 It should be noted here that Verkuyl does not present distinctions between types of durative predicates. His theory is a theory of terminative predicates. 22 23 1 . 2 . 2 Krifka's analysis The semantics of verbal predicates is parallel to that given to nominal reference and to temporal constitution. Events are primitives and thematic roles are primitives as well. The Krifka (1989) proposes a lattice-theoretic approach to aspectual composition in order to thematic relations such as AGENT and PATIENT relate the syntactic arguments to the event. account for how the aspectual values of NPs and verbs interact to yield an aspectual An example of the logical form Krifka proposes is given below: characterization for the larger constituent which contains them. The basic idea is to extend the lattice-theoretic framework for mass nouns and plurals (Link 1983) to the temporal (18) domain. To do this, Krifka postulates a set O of Objects and a set E of Events as drink λe [drink' (e) ∧ AG (e,z) ∧ PAT (e, z')] water water' primitives, following Davidson (1967). The extension of O and the extension of E have the structure of a complete join-semi-lattice12. The basic idea is that atelic expressions are drink water COMB (z') (λe [drink' (e) ∧ AG (e,z) ∧ PAT (e, z')]) (water') similar to mass nouns and bare plurals and telic expressions are similar to measure = λe∃z' [drink' (e) ∧ AG (e,z) ∧ PAT (e, z') ∧ water' (z')] constructions and count noun constructions.13 Telic predicates are quantized, and atelic predicates are cumulative just like nouns are quantized or cumulative. Nouns can either COMB (combine) takes a verbal predicate and a nominal predicate yielding a complex refer cumulatively or not, while verbs can only refer cumulatively. A mass term like wine verbal predicate. The complex verbal predicate will be cumulative if both the DP has the property of referring cumulatively, e.g., if there are two entities to which wine complement is cumulative and the thematic role has the property of being summative.14 applies, this predicate applies to their collection as well. A count DP such as five apples, An example is drink water. on the other hand, does not have this property; it is quantized. A quantized predicate can The influence of the reference types of nominal predicates on the temporal constitution be derived from a mass noun via the operation of a measure phrase. Measure phrases of the verbal predicates can be captured formally once we assume certain transfer serve "to 'cut out' entities of a certain size from the continuum of entities which fall under properties of the thematic relations. The patient relations are differentiated by the the head noun." (Krifka 1989) (see also Higginbotham 1994). following properties: summativity, graduality and uniqueness. If there are two events In order to derive a quantized predicate from a mass noun, two conditions must be John saw three horses and Mary saw four horses, and if the horses do not overlap, then met: the head noun predicate must refer cumulatively and the entity to which the whole the fact that John and Mary saw seven horses can be derived if one assumes summativity construction is applied must have parts which fall under the head noun but which are for the experiencer relation and the stimulus relation. measured as being less than the measure phrase. Expressions such as five ounces of gold Graduality holds for events like read a book: the book is subjected to the event of are quantized. reading in a gradual manner. Graduality does not hold for see a horse because seeing is instantaneous. 12 An i-join semi-lattice is a complete Boolean algebra with the 0 element cut out. 13 Krifka uses the terms atelic and telic instead of durative and terminative. I will keep to these terms in discussion of his idea, but the terms are to be thought of as interchangeable. 24 14 Summativity is defined as follows: ∀ R [SUM (R) ↔ ∀e∀e'∀x∀x' [R(e,x) ∧ R (e',x') → R (e ∪Ε e', x ∪Οx') ]] 25 Uniqueness of objects characterize the patient relations of effected and consumed (20) ∀e, x, R [ ITER (e, x R) ↔ Ρ(e, x) ∧ ∃ e', e", x' [e' ⊆Ε e ∧ e" ⊆Ε e ∧ ¬e' = e" ∧ x'⊆ Ο x ∧ R(e',x') ∧ R(e",x')]] objects.15 This of course only holds for object tokens and not for object types. I will discuss types further in chapter 5. Examples of the different patient relations is given below from Krifka: In sum, Krifka proposes that quantized noun phrases can provide natural measures that will combine with the verb through certain thematic relations such as the Patient (19) Patient relations relation. Just as in Verkuyl's system, defining the VP aspect necessarily requires defining the SUMMATIVITY GRADUALITY UNIQUENESS LABEL relation between the verb and its internal argument and the properties of the DP complement. write a letter X X X gradual effected patient eat an apple X X X gradual consumed patient read a letter X X _ gradual patient Krifka assumes that the notion of part has the following properties (where < is the part-of pat a cat X _ _ affected patient relation): see a horse X _ _ stimulus In both the domain of events and objects the notion of part is crucial. Following Link, (21) Properties of parts: (i) x < y ∧ y < z → x < z (transitivity) Verkuyl). The other properties are dependent on the lexical semantics of the verb and I (ii) x < x (reflexivity) will not discuss them any further. (iii) x < z ∧ z < x → x = z The patient relation has always the property of summativity (the ADD TO feature of The other case in which a complex verbal predicate is taken as cumulative is when the thematic role is summative and the noun phrase has singular reference. Here an iterative (antisymmetry) (iv) for any set A, ∃x x = sum < (A) (closure under sum formation) (v) (Az (z < x ↔ z < y)) → x = z (extensionality) reading obtains if the object is not effected or consumed as it is in write the letter or drink the wine. This is the case for example of read the letter which can be analyzed as an Such a notion of part relies on the assumption that an entity has exactly one part activity if the letter is read in an iterative way. Iterativity, stated formally in (20) is a structure and that there is only one dimension for parts of a given entity: space (for relation between an event e, an object x and a thematic relation R saying that at least one objects) and space-time (for events). part of x is subjected to at least two different parts of e. 1 . 2 . 3 Parts and Wholes and Aspect Moltmann (1994) presents a slightly different view of how we partition events. Her basic claim is that natural language relies less on the idea that there is a unique part15 An effected object is one which comes into existence during the course of the event. 26 27 structure for any given entity, but rather that events and objects can be partitioned according to a context. The important notion is of an integrated whole. "This is the notion (23) A part structure of an entity x in a situation s is a triple <A, Ps, Ws> satisfying (i-iv): of an R-integrated whole, which applies to an entity x just in case all the parts of x are (i) A is the set of entities y such that y < x and y is in s; connected under an appropriate relation R and no part is connected to an entity that is not (ii) Ps (parts) ⊆ A × A; part of x." (1994:6). (iii) For any W (context) ∈ Ws, there is a y , y < x such that W(y) is in s; An example is John's parents. John's parents form an entity whose parts are (iv) P satisfies the following: connected by the relation of having John as a son and are not connected to any other (a) ∀x y z ∃W (x Psy ∧ W(y) in s ∧ yPs z → xPsz entity under this relation. Integrity is a matter of degree and may be dependent on a (b) Restriction on sum formation (Moltmann 1994) particular dimension. The notion of integrated whole is defined more precisely below: If for an entity x, a situation s and a set X of parts of x, there is a whole property W for any x' in X such that W(x') in s, then for any non empty X' ⊆ X, sum < (X') is the (22) Integrated Whole sum of x with respect to Ps (sum Ps(x)) only if there is a whole property W such that W An integrated whole will be defined as the entities among whose parts such relations hold. x is an integrated whole if there is a division of x (sum< (X')) in s. Moltmann's notion of integrated whole allows her to dispense with transitivity and such that every member of that division stands in a certain relation to every other member and no member bears this relation to anything extensionality, two properties of parts that have been the target of some criticism, since other than member of the division. they do not seem to hold. With respect to transitivity, for example, although we can say (Moltmann 1994:6) John is part of the army and John's fingers are part of John, it does not follow that Whole properties play a role in the semantically relevant part relations. In other words, part relations must be relativized to situations. John's fingers are part of the army. With respect to extensionality, consider an expression like John swept the sand into a heap. Although the heap and the sand from which the heap is made arguably have the same parts, they are intuitively different entities. This particular notion of part-of structures is important in the sense that it brings us awareness that contextual and pragmatic effects can impose other part-of structures to the event and that what has no discrete individual parts at one level, may be constituted of individual discrete parts at another level. Thus although the fingers of John are part John and John belongs to the army, the relations are not the same. In the first case we have a body part relation and the second case we have a membership relation. Although we can work a semantics that will make these relations identical, we do not perceive them as the 28 29 same. The notion of part-of has then to be relativized to a context and not to ontological modify but not erase the part-of structure imposed by the internal argument. Moreover it considerations. In this thesis I will assume that we always evaluate the part-of in relation is very clear that we can find adverbials that can modify the VP or the IP level. to a context and the trick throughout this thesis is to maintain the context constant so that This notion of layers of aspect together with the idea that contexts have to be kept the effects that I will discuss do not get obliterated or shifted. To put it in another way, I constant will be particularly important when we deal with the accusative clitic doubling want to avoid putting into play pragmatic and contextual factors. This is not only facts in chapter 3 and when we deal with plural definites and the perfective and methodologically necessary but mandatory since context effects are post-syntax and imperfective markers in chapter 6. speaker dependent. I am only interested in the part-of of events that is dependent on 1.3 Towards a minimalist syntactic theory of aspect partitions related to the cardinality of the arguments. 1 . 3 . 1 The syntactic questions What is important though is that it is not the case that the aspectual composition is flat and has the property of associativity. To the contrary, if I combine A+B and then add C, In the Minimalist Program for Syntactic Theory (Chomsky 1993, 1994, 1995) not every this is not the same as combining A+(B+C). For syntacticians this is a truism, but there element that enters the computational system is born equal and not every element plays the are various analysis of aspect (De Praetre, ms, for example) that completely ignore the same role. In fact, if there exist purely syntactic (or formal) features, then they must be fact that in a sentence there is no hope but be in a certain hierarchical position. eliminated in the most economical way by the time a syntactic derivation reaches the In fact, even keeping context facts constant, within a sentence, it is necessary to interface, on the hypothesis that they would be uninterpretable. In addition, purely realize that there is more than one level of aspectual interpretation. The first level is the semantic features should play no role in the syntactic computation. If they have no formal one that involves the verb and its internal arguments. The second level will include the properties they are irrelevant for the computational system. subject, and although the subject can affect the aspect it cannot undo the properties that Given what we have seen above with respect to the interpretation of the VP aspect, derive from the verb and object interactions. Moreover, aspectual adverbs such as there are a number of very general questions we can ask based on general properties of repeatedly will allow a terminative VP to be perceived as durative. Also John ran is the Minimalist Program. different from John ran a mile in that the first event is unbounded and the other is not. However, if we add for an hour to the first sentence the result is a bounded event of onehour running. We provide an extra measure that is extrinsic to the event itself but is able to add a boundary. In the same way John ran a mile repeatedly can add a dimension of (24) (i) What are the minimal relevant conditions for aspectual interpretation? (ii) Are there aspectual features that could be considered syntactic or semantic or both? unboundedness since there is no indication of how many times the mile has been iterated. (iii) What syntactic configurations are necessary to satisfy the minimal conditions? However the iteration is still of a mile and not of two miles. With another adverbial, e.g. John ran a mile repeatedly until he got tired, we get another boundary established by the (iv) Can the diverse "grammaticalizations" of aspect found in various languages be given a unified explanation? adverbial element. What should be noted is that the addition of adverbial material can 30 31 To begin with the first question, whether we assume Verkuyl's model theoretic Suppose it is true that all the features relevant for aspectual interpretation are semantic. approach or the lattice approach of Krifka (as modified or not by Moltmann), we can What then remains for the syntax to do? In addition to the fact that φ-features are refine the more general conditions given in (11) above to the more detailed conditions implicated in [SQA], there is one important syntactic property implicit in (25): terminative given in (25). I take these to define the relevant conditions for aspectual interpretation. aspect requires both a nominal (semantic) feature [+SQA] and a verbal (semantic) feature [+ADD TO]. To do this, the verb and the noun have to be able to "see" each other, since (25) Relevant Conditions for Aspectual Interpretation (i) information about the verb (whether it is [+ADD TO] (Verkuyl) or they combine to form a PATH . What we have to understand is how the system allows the V and the N to "see each other". summative (Krifka)) In Chomsky 1993, functional features of lexical items have to be checked in (ii) (iii) information about the noun complement: whether it is count or not (Krifka/Moltmann) or whether its quantity is specified or not functional projections, since they cannot be formally licensed in a position occupied in the (Verkuyl/Krifka/Moltmann); initial representation. Thus if there are abstract features in an XP or in a X0 , these what relation holds between the verb and the argument (= or ⊆ (Verkuyl) elements will have to move to some external functional position. Unchecked features are or patient relation (Krifka)). uninterpretable and as such are excluded by the Principle of Full Interpretation which requires that every element of an output representation provide a meaningful input to the With the discussion above we can now begin to make some hypotheses with respect other relevant parts of the cognitive system. Strong uninterpretable features force overt to answering the other questions posed in (24). If we take (25) to characterize the movement and weak uninterpretable features force covert movement. Covert movement is relevant conditions, then we can ask to what extent the features implicit in (25) are always preferred. syntactic or not. It is quite clear that the properties of additivity and summativity and the The relation of a DP argument to the verb is mediated by an AgrP. Case and type of thematic relation that hold between the verb and its internal argument are purely agreement features are checked by moving the verb to adjoin to the head of the AgrP and semantic properties. We should expect them to have no direct syntactic effect. This by moving the DP to its specifier position. Agreement is determined by the φ-features in leaves the property of specified quantity and the count/mass distinction. Specified the head of a functional projection (AgrP) and case is assigned by the complex [V+ Agr] quantity per se is unlikely to be a syntactic feature either, as it puts in the same class bare or [T + Agr], for object and subject respectively. Thus nominal and verbal features are plurals and mass singulars (which are both [–SQA ]) for example. The count/mass checked in a Spec-head relation in a functional projection. distinction is perhaps syntactic, but it is also a semantic feature, since it is very relevant Given this, it follows that at AgrO the φ-features of the head of the internal argument for aspectual interpretation. It is important to note, however, that the feature [SQA] is are visible to the verb, since it is at this position that the formal nominal feature of the intrinsically bound up in the nominal agreements features (φ-features) commonly thought verb can be eliminated. This indicates that, at that position, the nominal features of the DP to consist of (at least) person, number and gender, and that the feature [ADD TO] is complement are visible to the V and that it is at this point that the relation of the verb and verbal. the nominal element can be interpreted as a PATH. And consequently it is at this point that 32 33 the PATH is defined as bounded or unbounded. More generally we can then treat Agr There are a number of conceptual advantages to proposing AgrO as the locus for projections as being necessary to satisfy the bare output conditions for calculating aspect, calculating aspect. First, AgrO is exactly the configuration where the features of the since this is the position were the φ-features of nominal and verbal projections can see internal argument (and therefore its head) can be checked against the verbal feature of the each other, as in (26). predicate. Second, the terminativity of the VP is independent of the properties of the subject. Thus we need a position in which the verb and the internal argument are in a (26) Agr projections are the locus for the interpretation of the relation between the φ-features of α, then made visible, and certain features in the target close relation that excludes the subject. Assuming a VP internal subject hypothesis head γ. α and γ do not belong to the same extended projection. Part-of (Kitagawa 1986, Koopman and Sportiche 1991) and a Larsonian analysis of internal relations are read off of Agr projections. arguments as specifiers of VP (Larson 1988), there is no other way to capture the relation between the verb and its internal argument in a systematic way, without including the Recall that there are four logical possibilities for specifications of [ADD TO] and [SQA]: subject in one way or other. Independently motivated morphological features—φ-features and Case—force both (i) the verb is eventive and the object has its cardinality specified (build the house); the verb and the internal argument respectively to move to AgrO. There, under the (ii) the verb is eventive and the object has its cardinality unspecified (build houses); mediation of Agr, features are matched and the aspect is calculated at LF. If for some (iii) the verb is non-eventive and the object has its cardinality specified (know the reason the DP complement does not move to AgrO, a durative reading will obtain. answer); It should be noted that I am not assuming that the DP moves to a certain position in (iv) the verb is non-eventive and the object has its cardinality unspecified (know French). order to get a certain interpretation, as this would be against minimalist assumptions. To the contrary, I am saying that if, for morphological reasons, the DP ends up in AgrO then Only in (i) is the VP interpreted as terminative. In fact if the verb is not eventive, for aspect calculus, it really doesn't matter whether the complement of the V moves to AgrO or not. certain features become visible for aspectual interpretation. If for some syntactic reason the DP does not move then a durative reading will be derived. In addition to the conceptual advantages, in the chapters that follow I will try to show So all we need is to establish when terminative readings obtain, given that that there are significant empirical advantages of the proposal which will allow us to treat terminativity is the marked case. With respect to VP aspect we can then impose the a wide range of seemingly different aspectual phenomena with a unified analysis. Before following condition on deriving a terminative interpretation. proceeding however, I would like to discuss a few of the conceptual arguments for adopting the minimalist program, and make clear some of the assumptions that I will be (27) Interpret VP as terminative (bounded) iff using to derive syntactic representations. AgrO contains a [+ADD TO] verb and a [+SQA] nominal element. 34 35 1 . 3 . 2 Basic advantages for assuming the Minimalist program conceptual system, which seems to be sensitive to the aspectual properties of sentences, LF seems to be the right place to calculate aspect anyway. One of the minimalist program's most important departures from the Government and The general advantage of adopting the minimalist framework is methodological: since Binding framework is its reduction of the number of levels of representation. Specifically in the minimalist program all movement has to be triggered by the feature checking there are no levels apart from the two interface levels PF and LF. There are no levels Dmechanism, i.e., nothing can move just to get a certain interpretation, we cannot resort to structure or S-structure. A derivation is constructed by a computational component that maps an array of lexical choices to the pair (π, λ). (A PF representation and an LF movement for interpretational reasons. Viewing syntactic operations as being driven simply by feature checking requirements allows us to evaluate to what extent aspectual representation). The array is a numeration N (a set of pairs of lexical items and indices, properties are a syntactic property, and to what extent aspectual properties are just understood to be the number of times that lexical item is selected). A derivation consists interpretational properties derived from independently motivated syntactic structures. of operations on phrase markers that are built up derivationally by the operations Select (which selects an item from the array and reduces its index by 1) and Merge (which takes 1 . 3 . 3 Basic Assumptions on the Feature Checking Mechanism two elements of the derivation and combines them.) A third operation Move, moves an element from within a phrase marker to some target within the phrase marker, extending the projection of the target. Movement operations, as we will see below, are triggered by the necessity of eliminating morphological features that are uninterpretable at an interface. Movement to satisfy a particular interpretation is thus ruled out in principle, unless that interpretation is exactly correlated with a morphological feature. There are two immediate general advantages in taking the minimalist program as a way to study the syntax of aspect: first, we are forced to treat aspect at LF as opposed to both at D-structure and at LF. Once there is no D-structure, aspect has be dealt with at LF More recently (Chomsky 1995) the feature checking mechanism has been rendered more precise. Functional features that are associated with the verb have two possible sources: they may be chosen arbitrarily as the verb enters the numeration or they might be the result of operations that form complex word associations with other elements. Features are classified as being [+interpretable] or [-interpretable]. The [-interpretable] features include strength; [+/- affixal]; Case; and φ-features of verbs and adjectives, i.e., φ−features of targets. Case and φ-feature checking is intuitively understood as an asymmetric relation: the verb assigns Case to the object and the φ-features are determined only. This is advantageous for the following reasons: first the system becomes less by those of the DP in the specifier of Agr. Only [-interpretable] features need to enter a redundant. With three syntactic levels of representation (as in Chomsky 1981 and work checking relation. based on it), part of the aspect was a D-structure property in that D-structure was taken in In the case of a DP moving to some Agr+V position, Case is the trigger for the some work (e.g. Grimshaw 1990 and Tenny 1987) to be the result of mapping aspectual movement. Being [-interpretable], Case is a feature that has to be checked either overtly properties into phrase structure. However, since, as we have seen above, aspect must or covertly. also refer to quantificational properties of DPs and possibly to inflectional elements, it must also be an LF property. Eliminating D-structure forces aspect to be calculated at LF and removes the redundancy. Second, since LF is the level that feeds the intentional36 37 Features enter into a checking relation if the moved element has unchecked features and can check some unchecked feature (not necessarily the same feature) on the target.16 point is that V and object have to match in a checking configuration and this is compatible with Chomsky's view of features. In addition, features must match. An example of this would be that of a non-nominative In addition, I will not assume that features move by themselves. DP which has raised to Spec of T. Thus the Case feature of the DP is in a checking So, with respect to the feature checking mechanism, the approach I will take is closer configuration but not in a Case relation with it. This causes a mismatch and the derivation to Chomsky's (1993) proposal with the following improvements/clarifications brought by terminates. If the DP is nominative, then it can enter a checking relation with T. The DP the 1995 proposal: raises to the Spec of T attracted by the need of T to check its D feature (which according to Chomsky is what explains the Extended Projection Principle (see fn. 16 above.) In the most recent version of the theory (Chomsky 1995), Agr is devoid of features because the features in the target that enter checking relations are [-interpretable] by (28) Feature checking (i) α moves only if morphological properties of α itself have to be checked; (ii) α can move to γ if α can check some feature of γ even if α does not check any feature of its own; definition. In fact, in Chomsky (1995), Agr is eliminated as superfluous, since when weak it has no interface properties. Instead, Chomsky proposes that, covertly, only features move, not XPs, since this is the minimal hypothesis. Only elements that need to be checked move covertly. Overtly however, an XP may move for convergence. (iii) φ-features such as number and person are[+interpretable] on the noun. φfeatures on the verb are [-interpretable] and need to be checked either covertly or overtly in Agr projections; (iv) strong features force overt movement of the head or the XP. Chomsky gets rid of Agr by claiming that there is no empirical motivation for it. I take the opposite route and will argue that Agr projections "exist" so that the bare output 1 . 3 . 4 A few clear examples conditions for aspect calculus can be met. Chomsky (1995) is, at first sight, incompatible Let's see how the proposal made above can be made to work. Let's examine first a simple with the thesis being pursued here that aspect is calculated at Agr projections. The regular transitive VP: incompatibility is, however, just at the level of technical execution, because the important (29) a. John ate the bagel. 16 The reason that the feature on the target need not be the one on the moved element comes from the need to account for multiply-embedded raising structures such as (i), or ECM constructions. (i) John seems t2 to be likely t1 to win Assume that John needs only to check Case. If Move were truly greedy, i.e. if John could only raise to a position in which it could get Case, movement from t1 to t2 would not be possible, since t2 is not a Case position. Lasnik 1993 argues that we can allow John to move if we assume that the Extended Projection Principle (EPP) requires each clause to have a subject, thus movement to t2 will satisfy the EPP in the higher clause. John, being needy, has nothing to lose, so to speak, by moving to t2. It is able to satisfy the EPP feature of the target. This version of Last Resort will correctly rule out (ii): b. John ate bagels. Here the verb is eventive and raises to Agr to check its features. The object raises to the Spec of AgrO to check its case. In (29a) the result is a terminative predicate and in (ii) John seems t2 likely that t1 will win In this example, John has no features to check, since it has checked its Case in its base position (t1), and as such cannot move simply to "help" out the higher clause satisfy the EPP. 38 39 (29b) a durative reading will be available since the object does not have its cardinality specified:17 (32) a. Táncoltunk egy-et dance-PAST-1PL one-ACC (30) 'We danced.' AgrO DP Agr' the bagel /bagels ate+Agr b. Sétáltunk egy-et VP subj V' (ate) (the bagel/bagels)T Now consider (31). walk-PAST-1PL one-ACC 'We took a walk.' A different case is that of Basque, where intransitives are formed with an auxiliary and a bare noun. Laka (1993) argues that we cannot treat the V+noun as a result of (31) John ran for three hours/ #in an hour incorporation, given the fact that they can be easily separated in the syntax as the examples in (33) illustrate. If the verb does not have an overt internal argument that needs case at AgrO, a durative reading will obtain. This is the case of (31). For Chomsky (following Hale and Keyser (1993)), intransitive verbs (as opposed to unaccusatives) contain an empty incorporated argument. If this is correct, it must not be [+SQA], or else a terminative reading would obtain.18 It is interesting to note that in Hungarian, overt expression of this argument appears as the numeral one in accusative case and forces a terminative reading (Kiefer 1994).19 17 I will assume that A-movement leaves copies behind, marked in ($30) in parentheses. Unless the copies are necessary for the point being made, I will omit them and use traces instead. 18 The null complement is also necessary so that a PATH can be built, according to Verkuyl. 19 This fact is also indicative that Enç's (1991) proposal is not tenable. According to her, specific complements appear in the accusative case. However we cannot say that in ($32) one is a specific object. Nonetheless it appears in the accusative. 40 41 (33) is calculated. This is possible because in (a) and (b) the preposition incorporates at LF a. nork egin behar du lan? into the verb and the DP complement checks its case by moving to the specifier of AgrO. who-E done must have work A schematic tree is given in (35): 'Who must work?' (35) AgrO b. nork egin du lan? DP Agr' the store/stores to+ran+Agr who-E done have work 'Who has worked?' VP subj V' (to+ran)T c. Oso ondo egin duzu lan. very well done have-you work 'You have worked very well.' PP (to)T store/stores)T (the Finally, consider (36) and (37): (36) a. John ran near parks for three hours/ #in an hour d. Ez dut lan-ik egin. b. John ran near the park for three hours/ #in an hour not have work-any done 'I haven't worked.' (37) a. John ran towards danger for three hours/ #in an hour She proposes instead that the bare noun receives case in situ and does not move to b. John ran towards the bridge for three hours/ #in an hour AgrO. In this case the result is a durative predicate, since the cardinality of the In the above cases a durative reading is always produced no matter what is in the complement (a bare noun) is not specified (Ricardo Echepare, p.c.). object position of the preposition. Unlike to, near is a preposition that cannot be stranded Now I will consider very briefly a couple of cases of intransitive verbs with and thus nothing will be at AgrO by the time aspect is calculated. PPs headed by near are prepositional objects. A full analysis of prepositional objects is outside the scope of this probably in a configuration that disallows overt or covert incorporation (in the cases of thesis. languages that do not allow overt preposition stranding like Brazilian Portuguese and (34) a. John ran to the store in five minutes/#for hours b. John ran to stores #in five hours/ for 5 hours French.) The result is a durative predicate. Towards, on the other hand, is a preposition that can be stranded (see Hornstein and Weinberg 1981). The relevant facts for this claim are given below: These examples behave as predicted if the object can be used as a measure to the event. To obtain such an effect it is necessary that the object be in AgrO by the time aspect 42 43 overt morphological markings, tend to treat morphological aspectual markers as part of (38) a. *What park did you run near b. What park did you run towards? the functional projection of verbs on a par with tense, following Pollock's (1989) work. c. What park did you run to? Among the authors that postulate aspectual projections are Heycock (1995), Borer (1993) and Hendrick (1991), among others. I will not deal with these proposals here, since they How can we maintain the explanation we gave for the cases of run to and at the same deal mainly with higher level aspect than the VP aspect that is my main concern. time explain (37)? I will assume here that towards is a preposition that specifies the ⊆ To my knowledge, however, there is no systematic discussion on how many of these relation on its internal argument. Even though it incorporates, it turns the verb into a aspectual functional projections a sentence can have (since they clearly can have more push type of verb, i.e. a verb that is always allows durative readings because the thematic than one), where they are in the tree or what role they play cross-linguistically in relation relation between the verb and the internal argument is not = but ⊆. Push (a ⊆ verb) has to other syntactic processes. In other words, aspectual morphology has played a to be distinguished from push away (an = verb) which behaves similarly to walk to, also somewhat marginal position among inflectional features in the Government and Binding an = verb. framework. According to the hypothesis being pursued here, aspectual features per For the case of near we can just say that near does not incorporate into the verb and se should not exist, i.e. there is no feature "durative" or "terminative" with respect to the the result is that nothing moves to the Spec of AgrO. Case is checked by the preposition VP aspect. In addition, the approach proposed here (following Verkuyl) explicitly denies alone (either because it has its own checking domain) or because the argument the possibility of grammaticalizing in the syntax the (in)famous Vendler-Dowty taxonomy incorporates into it. A durative reading will thus be obtained. of aspectual classes. Thus, in general the mechanism works. A terminative reading will obtain if the verb The reason for this should be quite clear by now: the aspectual classes (States, or the verb complex is of the right type and if the object has its cardinality specified and Activities, Achievements and Accomplishments) merely provide convenient labels for they are at AgrO at LF. event or situation descriptions (which was in fact what Vendler intended) rather than 1.4 characterizations of the categoricity of particular verbs. Thus, it makes little sense to say Previous Syntactic Accounts that eat is an accomplishment verb, since eat apples is an activity while eat the apples is an 1 . 4 . 1 Overview accomplishment. But if eat apples is an activity, what do we do with eat apples for an So far I have made a proposal for the interpretation of VP aspect focusing on the hour which is now an accomplishment. The mere fact that the Vendler-Dowty aspectual interaction between the verb and the internal argument. This is a somewhat novel classes can apply at various levels of the VP depending on the direct object and further approach to aspect in the syntax literature, since most of the work done on aspect in the adverbials makes the notion of lexical classes superfluous. Government and Binding framework has viewed aspect as a lexical property of verbs that Illustrative of the study of aspectual properties of lexical items as driving the syntactic mediates the mapping from the lexicon to the syntax or as a functional head that belongs mapping is the work of Grimshaw (1992) and Tenny (1987). Grimshaw, for example, to the extended projection of the verb. The authors who deal with languages that have proposes a level of argument structure where prominence relations among arguments are 44 45 represented. "The prominence relations are jointly determined by the thematic properties projections: not only does aspect determine the position that arguments will end up in the of the predicate (via the thematic hierarchy) and by aspectual properties of the predicate" syntax but it also shapes the VP internal structure.20 (1992: 4). The discussion on the aspect remains however somewhat cursory. For Tenny, 1 . 4 . 2 Borer (1993) "the mapping between cognitive structures and syntactic argument structure is governed by aspectual properties. Only the aspectual part of cognitive structure is visible in the Borer (1993) argues that lexical projections do not exist at all and a Vmax headed by syntax." (1987: 247) For these authors, aspect is, in a sense, one of the driving forces derive, for example, will be projected containing unprojected and hence unordered for the mapping of arguments in the syntax and there is a very tight relation between unhierarchically arranged arguments. thematic roles and aspect. Much of their analysis focuses on various constructions such as passives, middles and dative shift alternations that appear to be sensitive to aspectual (39) VMax derive NP, NP properties of particular classes of verbs. To the extent that Grimshaw and Tenny argue that "measures" of events must be direct objects, they are compatible with the approach I am proposing, since the AgrO Spec/head relation is in minimalist terms the direct object relation. However, it is not enough to know that there is a measure, but one wants to know how the measure is allowed to interact with the verb. My main focus here is on the role of internal arguments in the aspectual composition of the VP and in the morphological aspectual properties that impose constraints on the internal argument properties. Given this, I will abstract away from the most part on the idiosyncratic properties of individual lexical items. I will also assume that thematic roles have to be dissociated to a large extent from the aspectual properties of the VP predicate. The order of the arguments is re-constituted as a result of a set of aspectual constraints on the realization of specifiers associated with particular interpretations, traditionally associated with specific grammatical functions. The hierarchical ordering of arguments will be achieved by movement of such arguments to some specifier of some functional projection. Above VP there is an aspectual projection associated with event measurement. If AspP is left unspecified, the result is a stative verb. Otherwise we have an eventive verb. To account for the distinction between unaccusatives, transitives and unergatives, Borer proposes two more distinctions associated with the aspectual projection. The Aspectual projection can have a specifier or not. If the AspP has a specifier, it can be The only verbs I will be interested in are verbs that are [+ADD TO] and impose a thematic relation of the = type. In the same line of reasoning, namely that aspect of particular classes of verbs is relevant to the mapping from the lexicon to the syntax, there is one other proposal—Borer (1993)—that I will consider here as it goes a step further in relating aspect and syntactic 20 Travis 1992 also makes a proposal for relating the internal structure of VP to aspectual properties by assuming lexical decomposition into states and processes (following the work of Dowty 1972 and Pustejovsky 1989) where a state corresponds to one part of a Larsonian VP shell (the innermost VP) and a process involves the addition of another VP shell encoding Dowty's CAUSE operator. To the extent that this proposal encodes the Dowty-Vendler aspectual verb classes in the lexical decomposition, it is susceptible to the same criticisms given above with respect to compositionality and the role of determiners in calculating aspect. Also, there is no reason, a priori to assume that States are in any way more primitive than Events. Dowty (1972) adopts this idea for the following reason: in standard logic time is encoded as a fixed sequence of frames (states). Change is an illusion that comes from distinct states succeeding each other. Kamp (1979) however, proposes that it is possible to have dynamic primitives in the theory, namely it is possible to establish changes as the primitives constitutive of time. Our notion of time is then nothing else than changes temporally ordered, and changes are events. To the extent that Travis' proposal could be incorporated into Hale and Keyser's 1993 view of lexical structure, it may be compatible with certain properties of particular lexical items. Here I am concerned with the syntactic interactions of aspect rather than with specific lexical properties. 46 47 [+CASE] or [–CASE]. Three structures can thus be derived as exemplified by the trees in (42) TP (40) to (42), the fourth (+CASE, –SPEC) being ruled out. T' SPEC T (40) AspP TP Asp SPEC VP T' V, NP T AspP Asp' SPEC1 [–Case] Asp VP V, NP Delimiters, i.e., directional PPs in unergative derivations, will be projected as adjuncts to AspP. Borer acknowledges that the direct objects play an important role in defining the aspect and proposes an account for it that relies on Diesing's 1992 mapping hypothesis In (40) the argument moves to SPEC1 where a measure interpretation obtains. But this given in (43): is not a Case position, thus the NP has to move to Spec of TP. The result is an unaccusative verb. Now consider (41): (41) (43) Mapping Hypothesis Material in the IP area of a clause (external to the VP) maps onto the *TP restrictive clause and material in the VP maps onto the nuclear scope. T' SPEC T SPEC1 [+Case] AspP She proposes that NPs do not in fact move to the specifier of Asp to get Case. Asp' Asp Instead, they move to obey the Mapping Hypothesis. In AspP position the NPs can be VP interpreted as measure/delimiters. This movement is then what activates the event V, NP measurement properties of the aspectual node. Non-specific NPs have no reason to move In (41), the argument gets Case in SPEC1 since this is a [+CASE] aspect. Given that there is no other argument, the result will be unacceptable given that nominative must be obligatorily assigned. If there were another argument then the result would be a regular to AspP. They can move to a position below AspP where partitive is assigned or incorporate into the verb. This latter option is what happens to bare plurals and mass nouns. Non quantificational NPs and non-specific NPs remain in the nuclear scope: they transitive construction. In (42), on the other hand, the specifier is not projected and so an NP in need of case must raise to Spec TP to receive nominative. The event in this case has never been measured out. This is the structure associated with unergative (intransitive) verbs. 48 either incorporate or move to some intermediate projection, where they are still subject to existential closure. This intermediate projection, which Borer labels as an F projection is where partitive is assigned in Finnish according to Borer on the (as we shall see in 49 chapter 2) misguided assumption that objects in the partitive are non-specific and subject (45) a. Kaksi miestä söi joka kakun. to existential closure. two-nom men-PART ate every cake-ACC (44) 'Two men ate every cake.' TP spec T' T b. Kaksi miestä söi joka kakkua. AspP Asp two men ate every cake-PART FP spec 'Two men were eating every cake.' F' F VP (46) V, NP, NP Juha Marrjaa näki. Juha-NOM Marrjaa-PART saw 'Juha saw Marrjaa.' As Borer herself points out, her proposal is incompatible with minimalist assumptions. First, for her it is crucial that DPs move to get certain interpretations. Such Although Borer's proposal has some problems beside being incompatible with movement is barred in principle in the minimalist framework. Second, projecting a VP minimalist assumptions, there is an intuition that sometimes what distinguishes a durative that has no internal structure is not an option that is easily reconciled with minimalism. versus a terminative predicate is that the cardinality of the argument is not visible/visible Although this proposal obviates the need for equidistance, as pointed out to me by Ian depending on the syntactic position it appears, which is part of the motivation to use Roberts, it is definitely incompatible with the bare syntax of Chomsky 1995. AgrO as the locus for terminative predicate. Borer's proposal also seems to have too many degrees of freedom that would be difficult to constrain. For example, it should be possible in Finnish to allow one object to 1.5 Summary receive partitive and one object accusative, something that is impossible. Finally it has some empirical problems: for example, it is unable to account for the fact that even a non- In this chapter I have outlined the relevant conditions for calculating aspect, and made a specific apple, which would only move to the Spec of F can measure the event of eating proposal that the correct syntactic configuration for aspect calculation is AgrO. In the rest an apple, an interpretation only possible in the Spec of AspP This fact is associated to the of the thesis I will provide a number of case studies that support this view and lend assumption that partitives in Finnish are always non-specific, a hard position to argue for, further confirmation to the idea that the syntax of the verbal complement is crucial for given that strong quantifies and proper names can appear in the partitive as exemplified determining what aspectual interpretations will obtain. below and fully discussed in chapter 2. 50 51 Given that these languages do not have an overt definite/indefinite determiner Chapter 2 distinction, they pose an interesting challenge to the proposal made in chapter 1 that Markings on the verb and markings on the noun* aspect is a combination of nominal and verbal features. In Finnish it seems as if partitive Introduction and accusative are the overt markings of durative and terminative aspect respectively. In It is impossible to deal with VP aspect and not touch at least briefly on two apparent cases Slavic it seems that the perfective versus imperfective markings (or the lack of the latter) of its grammaticalization: the case of Finnish and the case of the Slavic languages, will produce terminative versus durative readings. Thus it seems that in one language exemplified below: durativity/ terminativity is marked on the noun, and in the second case, on the verb. However, as we have seen, terminative aspect is compositional: i.e., verbal properties (1) a. Tuula rakensi taloa viisi vuotta. (Finnish) interact with nominal properties. Moreover we have seen that aspect is a semantic Tuula built house-PART five years-ACC property that like scope depends on specific syntactic configurations. Thus we cannot talk 'Tuula was building a/the house for five years.' about aspect as a categorial element that is present or absent in a language, nor can we talk about a division among languages in terms of aspectual marking on the verb or on the b. #Tuula rakensi talon viisi vuotta. noun, given the compositional nature of aspect. Tuula built house-ACC five years-ACC In this chapter, I will argue that the markings in these languages are not encodings of 'Tuula built a/the house for five years.' a durative/ terminative feature per se. Instead, the morphological markings on the noun or on the verb will allow or block the right elements to be at AgrO by the time aspect is (2) a. Ota pil vino. (Czech) Ota drank wine. calculated. Languages' different abilities to check case may produce different LF configurations and thus different interpretations. Also, languages' use of adverbial quantifiers will force or inhibit [SQA] information at AgrO, creating effects that at first b. Ota vypil vino. sight look like aspectual markings. Ota PERF-vypil vino The discussion here is not to be taken as an in depth analysis of Finnish and Slavic 'Ota drank the wine.' complementation structure but rather as an exercise on how facts like (1) and (2) are expected if we assume the proposal made in chapter 1 that says that (i) the VP aspect is compositional and calculated at AgrO and (ii) that in order to obtain a terminative reading it is necessary that both the right nominal and the verbal features are at AgrO by the time In this chapter I need to thank Anne Vainikka and Tor Ashen for promptly clarifying me the Finnish facts and Ewa Jaworska for the kindness with which she answered my questions about Polish. I also thank Chris Piñon for sending me his papers. I also need to thank Alan Munn for pointing out the facts from English and hearing about Finnish and Polish and Czech ad nauseam. * 52 aspect is calculated. 53 The first case I will discuss is the case of Finnish. A common description of Finnish prefixes half- and over- in English and on the case of reflexive clitic markers in Spanish. case in the syntactic literature (dating back at least to Ross 1967) is that partitive case is These elements impose restrictions on the VP predicates that are similar to the Czech and used for durative predicates while accusative case signals a terminative predicate. I will Polish prefixes and I will argue that they should probably be given a similar analysis. In show that both claims are only partially correct, since it is not the case that an accusative other words, we can find in English and Spanish cases of A-quantifiers and cases where case in the sentence will force a terminative reading. Accusative case is compatible with a durative reading will obtain because the case of the complement inhibits the [+SQA] durative readings. At the same time, every partitive case DP will produce durative element to reach AgrO.21 readings since I will argue that partitive case prevents the movement of the DP object to AgrO. To put it more bluntly, in this chapter I want to argue that what looks like parametric distinctions on aspect, with Finnish using case to mark aspectual distinctions and Slavic The second example I will discuss concerns the verbal markings in the Slavic using A-quantifiers, are just reflexes of independent properties of languages. In other languages. In both Czech and Polish there are no definite or indefinite determiners. words there is nothing to be learned by the child that is specifically and purely aspectual. Instead, there is a set of morphological "perfective" markings on the verb that affect the What has to be learned by the child is how information about quantities are to be encoded interpretation of the object and appear to restrict what can appear in the object position. As in a language, what are the A-quantifiers in a language and what case is checked at AgrO a result, these perfectivizing markers will force terminative readings and in some cases and what case is not checked at AgrO. This information has to be learned independently force a specified quantity interpretation on the object. The imperfective (in the non-modal of aspect in the same way as the morphology of the language has to be learned reading) will force durative readings by imposing that objects do not have a specified independently. quantity. Given that the system presented in chapter 1 denies the existence of a A second goal of this chapter is to discuss the feature checking mechanism of terminative/durative feature, the Slavic examples are an important test of the theory. There Chomsky's minimalist program. I will argue that Finnish case markings support the idea are two points I want to make here. I will argue that the perfective markers are essentially that features are asymmetrical with respect to their needing to be checked, i.e. intrinsic the grammaticalization of SQA features realized on the verb. Rather than simply encoding features of verbs do not need to be checked off of the verb. Only features not intrinsic to a terminativity, they specify other more lexical properties similar to the properties that category need to be eliminated. Specifically, I will argue that accusative case features do quantifiers or determiners exhibit. This makes them amenable to being analyzed as A- not need to be checked on the verb, in the same way as number features on NPs do not quantifiers (in the sense of Partee 1993). Such an analysis is perfectly compatible with the need to be checked off of the NP. In other words I will argue that accusative case is an hypothesis made in chapter 1. intrinsic feature on verbs, but not on nouns. Thus it has to be eliminated from the nouns We might think that, if a language does not have overt determiners, it will make use of A-quantifiers to signal the terminativity of the VP or will use case, as the Finnish and but not from the verbs. This will allow us to account for the ability of Finnish verbs to check two accusative cases. the Slavic cases seems to suggest, but if a language has overt determiners such alternatives are not available. In the third section, I want to briefly comment on the verbal 54 21 The latter cases from English and Portuguese/Spanish have been discussed in Chapter 1 and I will not come back to them in this chapter. 55 Before we move to the case studies, I should mention that in the semantic literature, the Finnish and the Slavic cases have been treated as semantically similar, although perfective aspect presupposes a quantized verbal predicate and has scope over the complex verbal predicate. Similar patterns arise both in Polish and Russian. particular analyses have been quite different from each other. For example, both Verkuyl It is important to notice that the semantic analysis given by Krifka and Verkuyl are (1993) and Krifka (1989) have suggested that Finnish partitive/accusative case quite different for these cases. For Krifka, the partitive acts over the event and for distinctions in eventive verbs and Slavic imperfective/perfective morphological Verkuyl, it only affects the relation between the verb and the object. In this chapter I will distinctions should be treated in a parallel way, given their semantic similarities. For not try to discuss which one of the analyses is the correct one. It is important to note that Verkuyl, the case markings in Finnish mark the argument as affected (accusative) or both Verkuyl and Krifka try to collapse the Slavic and the Finnish cases, in that they have partially affected (partitive). The perfective markings on the verb in Polish will mark the been treated as having similar semantics. Given their syntactic differences, as we will see object as totally affected and imperfective as partially affected. The notion of partial below, I suspect that collapsing the semantics is not the right approach, and that Krifka's affectedness and total affectedness is to be encoded as a lexical property of the verbs. analysis best suits the Slavic case while Verkuyl's analysis is more suited to the Finnish Krifka associates the partitive and the imperfective with progressivity in English. He also suggests that both Finnish and Czech should be treated similarly. For him, languages facts. 2.1 Finnish: Case Distinctions and Aspect can mark progressivity via periphrastic constructions as in English or by some special case in the object as in Finnish. Progressivity involves a part-of relation and can be 2 . 1 . 1 Partitive vs. Accusative associated with the operator in (3), which basically says that a predicate like be drinking a As noted above, Finnish does not have a definite/indefinite determiner distinction. In glass of wine applies to events which are parts of events to which a glass of wine applies. other words, we cannot decide at first whether the cardinality of the complement is Partitivity on the noun is associated with a similar operator, given in (4). He then goes on specified or not. Thus a bare noun will be interpreted as a definite, indefinite, or bare to prove that the partitive operator can have similar effects in the verbal predicate, depending on stress, scrambling and the context. Examples (5) and (6)22 illustrate the producing durative readings. In Finnish then, partitive case can signal progressivity. apparent correlation between case and aspect. These examples show that the aspect of the (3) (4) predicate seems to depend not on the type of the determiner (there isn't one) but on the λPλe'∃e [P(e) & e'⊆E e] (progressive operator) case of the internal argument: accusative23 will force a terminative event but partitive will λPλx'∃x [P(x) & x'⊆ O x] (partitive operator) create a durative event (Heinämäki 1984; Vainikka 1989; Tenny 1987; De Hoop 1992; and others). In the case of Czech, according to Krifka, as exemplified in (2), the opposite happens. The verbal predicate operator affects the meaning of the nominal predicate. If the perfective marking is present in the verb, the object is interpreted as quantized. Otherwise it is interpreted as non-quantized or cumulative. According to him, the 22 The examples from Finnish are from Heinämäki (1984), unless I specify otherwise. 23 Accusative is a label for various morphological possibilities. See Vainikka (1989) for details. 56 57 (5) a. Maria kantoi kirjaa. (7) a. Kaksi miestä söi joka kakun. Maria carried book-PART two-nom men-PART ate every cake-ACC 'Maria was carrying a/the book.' 'Two men ate every cake.' b. Maria kantoi kirjan. b. Kaksi miestä söi joka kakkua. Maria carried book-ACC two men ate every cake-PART 'Maria carried a/the book.' 'Two men were eating every cake.' (6) a. Maria kantoi kirjaa tunnin. (8) Juha Marrjaa näki. Maria carried book-PART hour-ACC Juha-NOM Marrjaa-PART saw 'Maria carried a/the book for an hour.' 'Juha saw Marrjaa.' b. *Maria kantoi kirjan tunnin. In what follows I will argue that partitive is assigned by an independent case marker Maria carried book-ACC hour-ACC and thus blocks movement of the partitive object to AgrO, while accusative is evidence of 'Maria carried a/the book (to some place) for an hour.' movement to AgrO. This makes the predictions that not all instances of accusative case force terminative readings and that all instances of partitive case on the DP complement In (5a) we have a durative event. In (5b) we have a terminative event. Therefore for will force durative readings. I will show that in fact these predictions are correct, and will an hour adverbials can be added in to (5a) giving (6a). (6b), on the other hand, is be used to argue that partitive and durative are not simple encodings of durativity and incompatible with for an hour. Note here that the adverbial in (6a) is in the accusative, terminativity, but rather independent case properties of Finnish that have aspectual although the complement is in the partitive. effects, given that aspect is calculated at AgrO. Finally I discuss two other cases of Assuming in both cases there is a DP or a NumP, since in both cases we interpret obligatory partitive case which support the hypothesis that partitive is not the encoding of book as the internal argument of the verb carry, the question is then what is the partitive a feature [+durative]. doing so that it prevents the cardinality of the internal argument being 'visible'. That partitive blocks visibility of the cardinality rather than just specifying "no cardinality" is 2 . 1 . 1 . 1 The partitive/accusative distinction shown by the examples below, which in spite of the cardinality of the noun complement There are at least three hypotheses that can be entertained to account for the basic being overt (in (7b) because of the quantifier joka 'every', and in (8) because of the distinction above between partitive and accusative case. One hypothesis is that case proper name) a durative reading is still available if the object is in the partitive. assignment is a lexical property of verbs, and some verbs assign accusative, some 58 59 partitive and some both. This means that for every verb that can assign both cases, there (10) a. Manne kehui hevosta. are really two verbs, one which assigns accusative and another which assigns partitive.24 M praised horse-PART A second hypothesis is that partitive is lack of case and/or partitive objects are 'Manne was praising the/a horse.' incorporated objects. A third hypothesis is that accusative is checked at AgrO by the verb and partitive is checked in situ by an independent head. In what follows, I will argue against the first two hypotheses and will implement the third one. The idea that case assignment is related to particular verbs seems initially plausible b. Minä pelkään sotaa. I fear war-PART 'I am afraid of war.' given that there are many verbs that seem to take objects only in the accusative or only in In (10) accusative objects would be unacceptable. However, when a resultative is the partitive. This restriction is only apparent, however. First, negation productively triggers partitive case in the object (see §2.1.3 below.) added, as the examples in (11) show, the event becomes terminative, and the accusative form is possible.25 (9) En lukenut kirjaa (Kangasmaa-Minn 1984: 86) I-NEG read-PAST.PTC.ACT book-PART (11) a. Manne kehui hevosen maasta taivaaseen 'I was not reading the book' M. praised horse-ACC earth-from heaven-to 'I did not read the book to the end' 'M praised horse the horse from earth to heaven.' Thus, even if a verb is of the type that regularly takes accusative objects, these will appear in the partitive under negation. b. Pelkäsin itseni puolikuoliaaksi I-feared myself-ACC half-dead-to Second, non-eventive verbs usually take partitive objects. However, if a resultative is 'I scared myself half to death.' added, then a non-eventive verb can assign accusative. Examples are given below: It is not the case that the resultative itself is responsible for turning the verb into one which can only assign accusative, since resultatives can appear with partitive objects as well, independent of the negation facts discussed above, as shown in (12). 24 I will use the term 'assign' in my description of the case properties in a (somewhat) theory-neutral way. Where issues of implementation arise, however, I will use 'check' in accordance with the minimalist view of case being checked rather than being assigned. 60 2 5 See Vainikka (1989) for a list of verbs that usually only check partitive case, unless a resultative is added. According to her some verbs check only accusative. However, Heinämäki points out they can take a partitive as well. Thus it seems that most verbs have in fact the ability to check both cases. 61 (12) a. Juha löi ihmisiä kuoliaaksi. (Tor Ashen p.c.) We cannot say that accusative in the durational adverbial is an inherent case, because when nominative is not assigned the adverbial will appear in the nominative, as other Juha hit people-PART dead 'Juha was hitting people dead.' accusative objects do when nominative is not assigned: (15) Muistele matkaa vuosi! b. Juha ei liönyt ihmisiä kuoliaaksi Reminisce trip-PART year-NOM Juha no-3SG hit-PCT people-PART dead 'Reminisce about the trip for a year!' 'Juha wasn't hitting people dead.' I will therefore reject the hypothesis that there are two different verbs that can check The data (negation and resultatives) above show that partitive and accusative cases are either partitive or accusative given that both accusatives and partitives can co-occur in the relatively independent of lexical items. In order to implement the hypothesis that both sentence. were assigned by the verb we would have to make the assumption that there were two I will move then to the second hypothesis, namely the hypothesis that partitive is the lexical entries for every verb. But then we would have to find an independent explanation lack of case. It is possible to suppose that DPs that seem to have partitive case in fact for the following facts, noted by Maling (1993): don't have case at all. Partitive would then "mark" lack of case. This hypothesis raises a (13) theoretical and an empirical problem. First, such a hypothesis would violate the Case Liisa muisti matkan vuoden. Liisa-NOM remembered trip-ACC year-ACC 'Liisa remembered the trip for a year.' filter or some alternative version of it. Second, nominative case is the case that seems to be the representative of default case in Finnish (see Vainikka (1989) and Milsark (1985)). To have two different morphological forms for default case is quite unreasonable. An The verb muistaa 'to remember' does not allow partitive in the affirmative. When alternative would be to say that partitive objects always incorporate into the verb. combined with a duration adverb, accusative will appear also in the adverb. A related However the ability of the objects to scramble freely with partitive or with accusative in verb, muistella 'to reminisce', takes only partitive and not accusative on its object. Finnish suggests that this is not a valid proposal:26 However, the adverbial will appear in the accusative case, as illustrated in (14): (14) Liisa muisteli matkaa vuoden. Liisa-NOM reminisced trip-PART year-ACC 'Liisa reminisced about the trip for a year.' 26 I omit translations of the subtle meaning differences among that different orders, as they are not relevant for the present point. 62 63 (16) a. Juha sö kakkua (Tor Ashen, p.c.) Juha eat cake-PART (17) a. Katsoin häntä. I-looked him-PART b. Juha kakkua sö ' I was looking at him' c. kakkua sö Juha d. kakkua Juha sö b. Katsoin Pekkaa. e. sö Juha kakkua I-looked Pekkaa-PART f. sö kakkua Juha ' I was looking at Pekkaa.' (Kangasmaa-Minn 1984:88) An alternative to this proposal is to assume with De Hoop (1992) that partitive is The fact that universal quantifiers such as every can also appear with the partitive weak case and accusative is strong case. Weak case will produce non-specific readings object, as in (7b) above, repeated here as (18) for convenience, also makes this and strong case specific readings. Both would be assigned by the verb, one in situ and hypothesis untenable. the other at the specifier of VP, which we can assume to correspond to the specifier of AgrO. However, such a hypothesis is untenable under minimalist assumptions. The (18) a. Kaksi miestä söi joka kakun. complement cannot have its case checked in situ since checking does not happen in the two-NOM men-PART ate every cake-ACC internal domain of the verb. To treat partitive as inherent case is also unlikely, because as 'Two men ate every cake.' we have seen it is not associated with a particular semantic role. Moreover, according to De Hoop, weak case would be assigned to non-specific b. Kaksi miestä söi joka kakkua. elements. However, given that proper names (e.g. (8) above and (17b)) and pronouns two men ate every cake-PART ((17a) below) can appear in the partitive, the idea that only non-specific elements check 'Two men were eating every cake.' partitive is highly unlikely. In both (18a) and (18b), both scope orderings are possible, thus any connection between case and the specificity or the strong/weak distinction is completely spurious. As Chesterman (1991:140) states: "the sense in which partitives are indefinite is purely a quantitative one. The partitive does not have anything directly to do with the distinction between known and unknown referents." Let's then in turn to the third possibility, namely that accusative is checked by V+Agr in AgrO and partitive is checked by some other head, without mediation of AgrO. Historically, this is quite plausible, as partitive case derives from a Uralic suffix with an 64 65 ablative meaning, signifying movement away from, in which case it was probably an the complement of P raises to the Spec of P (or a higher AgrP above PP) to check its case inherent case. or just the D or the noun incorporates into P in order to check its case, since P is affixal. I do not think that partitive is an inherent case since partitive is also the case of nouns This is unimportant for the point I want to make here. in many partitive NPs, as the examples below show. The quantifier head will appear in (20) a. AgrO b. VP whatever case it is required to by the position in the sentence and the head noun will DPi appear in the partitive. Agr' Vj+Agr (19) a. pala juustoa V' VP V V' (Chesterman) Vj piece-NOM cheese-PART P DP DPi Accusative case 'a piece of cheese' PP Partitive case The main consequence for checking partitive by the P is that it removes from the DP b. kilo voitaa (Seppänen) the need to move to the specifier of AgrO for case reasons. Thus, unless the complement kilo butter-PART in the partitive is marked by topic or focus, it will have no reason to move. If partitive 'a kilo of butter' case is not checked at AgrO, durative readings will be naturally derived, since in this case no nominal form will provide a measure to the VP nor an end point to the predicate. Thus c. liian paljon voita the association of partitive with durativity is rather a consequence of the fact that nominal too much butter-PART features are not in AgrO at LF where aspect is calculated. Partitive is not encoding a 'too much butter' feature [durative]. Partitive case is a case that is not checked at AgrO. The consequence is no movement to AgrO, which will force the VP to be interpreted as durative. Given that nouns are not taken to be case checkers in general, it follows that there must be some independent case checker for the noun complement.27 Accusative, on the other hand, is the sign of movement to AgrO, but again, accusative is not the encoding of terminativity. At AgrO we have the necessary conditions for I propose therefore that partitive is checked by a P element (much like of in English, terminative predicate readings to arise. Note that I said conditions. Remember, according except much more productive) and that accusative is checked at AgrO. The basic to my hypothesis we need the right verb and the right complement to be at AgrO by the structural configurations for case checking are given below. (20a) represents accusative time aspect is calculated in order to obtain a terminative reading. case checking and (20b) partitive case checking. I will leave open the question of whether 2 . 1 . 1 . 2 Predictions 27 If a specified quantity of cheese in the example ($19a) is inferred then the object will appear in the ellative case. I will not deal with these cases here. For some discussion of partitive constructions see chapter 5. 66 67 The proposal I have just made makes two predictions. First, since accusative is the is important is that it is the adverbial all summer that is providing a boundary here and sign of movement of the object to AgrO, the prediction is that a terminative readings can not the object cactus, in spite of cactus and the adverbial appearing in the accusative. In arise with accusative. Given that plurals tend to always show up in the partitive if the fact, the object can also appear in the partitive in this example. quantity is not specified, terminative readings will be the norm if the verb is of the right The examples above have illustrated the fact that not all accusatives induce terminative type. However, if the verb is not of the right type, it should still be possible that readings. In Vainikka (1993) Vainikka proposes that accusative is checked by a accusative will not induce terminative readings as a consequence of using the object to [COMPLETED] feature. However, not all cases we have accusative we have a terminative measure the event. In fact this can be shown to be true from the examples below: predicate.28 If we were to assume that accusative case is the realization of the terminative aspect, we would have no way to explain the fact that with a verb like weigh or keep the (21) Kaali painoi kilon (Heinämäki p. 174) cabbage weighed kg-ACC object can appear in the accusative. Moreover we would have to be able to distinguish partitive objects with an accusative measure and accusative objects with an accusative 'The cabbage head weighed one kilogram.' measure. However, if we proceed compositionally, we can predict that a durative predicate (whose object is in the partitive) can be made bounded by the addition of an Here the verb is weigh and the measure phrase is in the accusative. Nonetheless we external temporal boundary. What is interesting about Finnish is that it is possible to have do not have a terminative reading, since weigh is a stative verb. more than one DP checking for accusative case and have also one DP checking for The fact that time adverbials can act as measurers of an event also provide evidence accusative and one for partitive. that accusative case on the object is not necessarily a mark of terminativity. Consider the It should be noted that with respect to partitives, it is also not true that every noun that case of (22). shows up superficially in the partitive will force a durative reading. In fact there is a case (22) a. Pidin kaktusta parvekkeella. (Heinämäki) I kept cactus-PART balcony-on in which an apparent partitive complement can induce terminative readings. Consider (23). According to Chesterman (1991) , a sentence such as (23), allows three readings: 'I kept the cactus on the balcony.' (23) Söimme kakkua (Chesterman) ate-1PL cake-part b. Pidin kaktuksen parvekkeella koko kesän a. 'We were eating a/the cake.' (durative) I kept cactus-ACC balcony-on all summer-ACC b. 'We ate some cake.' (terminative) 'I kept the cactus on the balcony all summer.' c. 'We were eating some cake.' Here the verb pitää 'keep' implies duration. An accusative object is possible if it implies that there is a specific amount of time that the cactus stayed in the balcony. What 68 28 Stative verbs like remember, as we have seen, check accusative and nonetheless do not force a completed action reading. 69 The first reading is unbounded and can be modified by for an hour but the second one English the fact that the selected head is a bare noun does not imply a durative reading, as isn't and can be modified by in an hour. This fact seems puzzling since I assumed above in (25). Instead, it is the measure that affects the aspect and that has its cardinality the partitive case is not checked at AgrO. However, there is an important fact about the specified (see chapter 5). partitive that allows terminative readings, which will make it different from regular partitive objects. Only divisible NPs have the potential of expressing partial quantity via the partitive case. Divisible NPs are homogeneous predicates, either mass nouns like vesi (25) a. Mary ate two pounds of rice in a week b. #Mary ate two pounds of rice for a week 'water' or plural count nouns pojat 'boys'. In these cases, the partitive indicates a partial One could ask whether this null measure has a specified cardinality or not and quantity of the object. To account for these cases, I propose that there is null some that is receiving accusative case while the partitive only checks the case of the head noun. The partitive is thus internal to the DP which as a whole receives accusative case. The tree below illustrates the case in which the partitive can be interpreted as terminative. (24) DPi P whether it needs to check case I will assume that it does, given that other measures in Finnish do show accusative case: V+Agr DP milk-PART (Heinämäki p.174) Kristiina waited hour-ACC Agr' PP amount of something, which allows us to infer a specified quantity. With respect to (26) a. Kristiina odotti tunnin. AgrOP some-ACC whether it needs case or not. I will assume here that some is to be understood as a certain 'Kristiina waited an hour.' VP V DPi b. Kaali painoi kilon. Although the hypothesis of a null element seems at first ad hoc, remember that partitive is the case checked by the complements of piece/kilo as illustrated in (19) above. cabbage weighed kg-ACC 'The cabbage weighed one kilogram.' Moreover in Polish, as we will see below, there is evidence for the same structure, even In sum, partitive case on the DP complement of a verb uniformly provides us with a though Polish does not have a productive partitive case (actually genitive) like Finnish. In durative reading, since the complement does not raise to AgrO. The partitive (and lack of Polish, an object of a verb that usually assigns accusative can appear in the genitive, but movement to AgrO) has nothing to do with definiteness or specificity, as the examples when it does, it means some of the X, and is consequently compatible only with the with strong quantifiers, pronouns and proper names show. perfective forms. This will be discussed in detail in §2.2.2 below. As noted briefly in chapter 1 in the discussion of Borer (1993), the Partitive within the noun phrase complement, thus, does not affect the durative or partitive/accusative case distinction cannot be linked to the Diesing's Mapping terminative reading of the complements. This is not surprising. In the same way, in Hypothesis, since both partitive and accusative objects may or may not be specific. If it 70 71 is correct that partitive case does not force the movement of the DP complement to some The obligatoriness of partitive with negation in Finnish makes it is crucial to position outside the VP, as I have proposed, then an analysis of scope facts that depends distinguish this phenomenon from the apparently similar case of Russian. In Russian on A-movement to AgrO cannot be correct, since, as we have seen, the object can be in with sentential negation a nominal expression can appear in the genitive but need not as in the partitive and allow specific and non-specific readings. The mapping hypothesis per se (i) and (ii) (data from Pesetsky 1982).30 may or may not be correct, i.e. there may be some position to which specific DPs raise, but the data from Finnish clearly show that the position cannot be AgrO and the syntactic (28) a. Ja ne poluc̆al pisem. I NEG received letters-GEN-PL motivation cannot be case.29 2 . 1 . 2 A note on the partitive of negation b. Ja ne poluc̆al pis'ma I NEG received letters-ACC-PL Although it is way outside the scope of this dissertation, I would like to mention one other place where we find partitive objects in Finnish: the partitive of negation. This brief Krifka 1989 suggests that one should treat negation as triggering the part-of relation with discussion is intended to add some support for the idea that we should not tie partitive vs. respect to the whole event. Thus a negated event is one among the events that may have accusative case directly to durative vs. terminative interpretations, and will also provide happened. Recall that the part-of operator can be interpreted as relative to the set of events more evidence for my analysis that partitive is checked without movement to AgrO. E or the set of objects O. Krifka argues that negation marks the scope of the partitive Descriptively, negation will always trigger partitive in the object of the sentence. operator, thus although it appears on the noun, when negation is present, it takes scope Partitive is obligatory if the sentence is negated: over the entire event. Although this maybe a semantic way to treat these cases, it seems (27) a. Juha ei liönyt ihmisiä kuoliaaksi. Juha NEG-3SG hit people-PART dead 'Juha didn't hit people dead.' that there is a simpler analysis for the Finnish facts and its nature is syntactic. In negative sentences in Finnish, the negative element has the stem form e-. It functions like an auxiliary verb in that the agreement morphology (person and number of the subject) shows up on it, as illustrated in (29), while the main verb remains in a neutral "participial-like" form that appears to inflect for tense and number. Examples of this are b. Juha Marjaa ei nähnyt. given in (30). Juha Marjaa NEG-3SG see 'Juha didn't see Marjaa.' 29 See also the discussion on Antecedent Contained Deletion in chapter 4 for independent evidence of this claim, and the discussion in chapter 6 for independent arguments against the Mapping Hypothesis as it relates to the stage level/individual level distinction. 72 30 The same is true for the partitive of negation in Basque (Ricardo Echepare, personal communication) and for the partitive in French. 73 (29) 1SG e-n (31) a. Ole-n asu-nut täällä viikon. 2SG e-t have-1SG lived-SG here for a week 3SG e-i 'I have lived here a week.' 1PL e-mme 2PL e-tte 3PL e-ivät b. Ole-mme asu-neet täällä viikon have-1PL lived-PL here for a week 'We have lived here a week.' (30) a. e-n puhu To account for the fact that negation and person features of the subject appear above no-1SG speak the verb, Mitchell (1991) proposes the following structure32 which corresponds to 'I don't speak.' example in (30a): b. e-n puhu-nut (32) no-1SG speak-PAST.PTC.SG CP AgrP 'I didn't speak.' NP Agr Agr NegP (1SG) Neg T/MoodP c. e-mme puhu-neet no-1PL speak-PAST.PTC.PL +Neg 'We didn't speak.' T VP [+pres] V NP puhu-SG The same past participial form used with negation is a verbal adjective that also occurs in complex tenses. It agrees in number with its subject and is declined in all cases.31 Examples without negation are shown in (31). The negative verb does not have tense features, only person features. It is the present or a past participial form that will indicate whether the sentence is in the present or past. If a modal is added, the negative element with the inflection will still show up before the modal element, as the examples below show: 32 Basically the same structure is also proposed by Holmberg et al. (1993). 31 I will not deal with the case distinctions here. 74 75 (33) a. Minä puhu-isi-n (from Mitchell p.376) verb to AgrS. The verb then must remain low yet still must get rid of its number agreement (which it must do by agreeing with the subject), and its tense/mood features. I-NOM speak-COND-1SG Since the verb does not have person features, it has no need to raise any further. The 'I would speak.' number features, I assume, will be checked by the participial form against the subject in a specifier head relation on an Agr node. The subject then moves to the specifier of AgrS, b. Minä e-n puhu-isi to check the person features on the negation head. Tense/mood can be checked by raising I-NOM no-1SG speak-COND to T. Since AgrS is being used to check the agreement features on the Negation, there 'I would not speak.' must be another Agr projection to check the agreement on the lower verb. My proposal is In the complex tenses (present perfect and past perfect) the auxiliary olla in the that AgrO is the Agr projection that serves this purpose. A schematic tree is given in (35). affirmative shows the agreement with the subject and the main verb appears in the I will assume here, following Bobaljik and Jonas (1993) that specifier of TP is not participial. In a negated clause both the main verb and the auxiliary appear in the available in Finnish since Finnish is a language that does not have transitive expletive participial form and the person features will appear in the negative verb. constructions (Schot-Saikku 1993). (35) (34) a. Minä ole-n tul-lut I-NOM be-1SG come-PAST.PTC/SG CP AgrSP SUBJ 'I have come.' Agr' NEG+AGR 1SG b. Minä ol-i-n tul-lut I-NOM be-PAST-1SG come-PAST.PTC/SG tNEG NegP T/MoodP T+V+AGR +PRES AgrOP tSUBJ 'I had come.' Agr' V+Agr VP tSUBJ V' c. Minä e-n ol-lut tul-lut tV I-NOM no-1SG be-PAST.PTC/SG come-PAST.PTC/SG 'I hadn't come.' DP The verb moves up to T/MP to check its tense features but does not move any further. It has no person features to check at AgrS and it is unable to check nominative case on the Given Mitchell's analysis of the basic clausal facts, we can ask why partitive case is subject. (In Finnish, if the verb does not have person features it will be unable to check obligatory with negation. Given that person features cannot ever show up on the verb nominative case on the subject, and the subject will appear in the genitive.) The specifier when there is sentential negation, I will assume that negation blocks the movement of the 76 77 of AgrO is equidistant from both the subject and the object, once the verb has moved to object position, the derivation will crash because the DP will not be able to check its case AgrO. Given that, we can either move the subject to the specifier of AgrO or the object. features.33 The intuition here is that the subject will move through the specifier of AgrO in its way to In structures with auxiliaries without negation, we do not force partitive case on the AgrS. We need now to prevent the object to move to AgrO and force the subject to move object. This can be explained by the difference between Negation and the Auxiliary. If through it. By moving the subject to the specifier of AgrO we trap the object downstairs we assume that Aux is generated in T, we can allow the verb to move to T, since it has T and the only way the object can check case is if there is a partitive case checking head. features to check, and then to AgrS. The presence of negation will block the auxiliary Why should the object not move to AgrO? The reason lies on the fact that if the verb does from moving to AgrS because the verb has no features to check in Neg. This means that not check number features in this Agr position, it has no other Agr position to check the subject does not need to move through AgrO, and accusative case can be checked by number features against the subject. Thus if we move the object to the Specifier of AgrO the object. A schematic tree is given in (37). At LF, the verb will raise to T+Aux and the derivation will crash because there are number features on the verb that have been left probably from there to AgrS to check number features.34 unchecked. (37) (36) CP AgrSP CP AgrSP SUBJ SUBJ Agr' NEG+AGR 1SG Agr' T/MoodP AGR NegP tNEG Aux+T T/MoodP T+V+AGR +pres +pres AgrOP AgrOP Agr' V+Agr VP OBJECT Agr' tSUBJ V+NUM VP V' tV DP VP SUBJ We can summarize this section by re-examining the ordering of partitive case vs. tV tOBJ accusative case. Partitive case appears obligatorily in the object if the sentence is negated, since in those cases that is the only way to check the case of the object. If the sentence is Given this analysis, it will be impossible to move the direct object to AgrO. The net result is that the object gets trapped into its base position. Luckily, Finnish has a way out: it can assign partitive case freely. If an accusative case-marked DP shows up in 33 There are certain prepositions such as ilman/vailla 'without' that also force partitive case. We can either say that these prepositions lack case or that they are decomposable into with+not and not blocks case checking by the preposition. 34 Note here that I am allowing the V to move at LF to a position which is apparently a copy, since presumably the verb moves overtly to AgrS. As it will become clear in the next chapter this movement contrary to Chomsky's assumption, has to be allowed for a variety of cases, one of which will be discussed in the next chapter. 78 79 affirmative, then depending on the verb, the object can appear in the partitive and a both verbs and nouns. I will argue that it is preferable to see (accusative) case as an durative reading will be derived because the object does not move to AgrO. Finally, if the intrinsic feature of verbs that assign external theta roles, and therefore a feature that only verb is [+ADD TO] and the object is mass or plural, we can get a partitive interpretation of needs to be checked on nouns. the object with a null some. In this case, the result is terminative. Accusative case, on the Suppose, for sake of argument, that case must always be checked off of both nouns other hand, allows terminative readings, but doesn't force them; it will depend on the and verbs. Since the data from Finnish show that accusative is always available, if it verb. were a feature that needed to be checked on the verb, then it follows that every verb The partitive/accusative case distinction is thus only indirectly tied to aspect, as the discussion in this section also shows.35 would need to be listed twice, once with the case feature, and once without. If this were not the case, partitive objects would cause the derivation to crash; they do not raise to AgrO and the verb would reach LF with unchecked features, unless an adverbial were 2 . 1 . 3 Some Theoretical Implications for the Feature Checking Mechanism adjoined to AgrO and checked the case feature on the verb. Assuming such a proposal, In the previous sections, I proposed that partitive case in Finnish was checked by an i.e., that accusative case has to be checked off from the verb, every Finnish verb would independent head and accusative was checked by V+Agr at AgrO. The evidence I used need two forms. This is more or less the same argument I gave against allowing verbs to for proposing this can be summarized below: assign both partitive and accusative case. The problem, however, is not restricted to Finnish. The same point can be made with English resultative constructions, as in (38) (i) negation always triggers partitive case; (ii) most verbs can check both partitive and accusative case even if resultatives are added or if temporal measure adverbials are added; (iii) complements of partitive nominals also appear in the partitive. (38) a. John laughed himself silly b. *John laughed himself c. John laughed This proposal raises an interesting theoretical question with respect to how case The unacceptability of (38b) does not come from the fact that himself has no case but works within the minimalist feature checking mechanism since it explicitly departs from from the fact that himself does not receive a role from laugh. (38a) is acceptable because the assumption that verbs could have objects in either partitive or accusative, with only himself is getting a role from silly. We could try to say that laugh silly is a complex verb the latter being checked at AgrO. This assumption brings into question Chomsky's and only the complex verb laugh silly can check for accusative.36,37 However, the same suggestion that case is a [–interpretable] feature and must always be checked off from explanation would be harder in cases such as (39): 35 The partitive also appears in predicative constructions. In these constructions, the subject always appears in the nominative case and the predicate can appear in the partitive or in the nominative. Partitive will be assigned optionally if the subject is plural. Partitive will be assigned obligatorily if the subject is a mass noun. This fact is clearly unrelated to aspect since we do not get an aspectual interpretation relating to case. Although I do not have an account for this fact, it should be clear that it again shows that partitive/accusative distinction is independent of aspect. 36 There is a large literature on resultatives which I will not go into here (cf. Simpson 1983; Carrier and Randall 1992, Levin and Rappaport Hovav 1995, Neeleman and Weerman 1993, Hoekstra 1988). The exact analysis of resultatives is not really at issue. What is important is that seemingly intransitive verbs can assign accusative case under certain circumstances. 80 37 We have already rejected this for the Finnish partitive/accusative distinctions based on the fact that resultatives in Finnish can also have partitive objects. 81 need to be checked off on the verb. This solves the problem of an unchecked accusative (39) a. John ran case on the verb.38 b. John ran a mile Assuming that accusative is not checked on a verb, just as number is not checked off of a noun phrase, entails that the verb can license a specifier for accusative checking Notice that the Finnish counterparts of the sentences above also show accusative in without getting anything checked off on the verb, just like a noun can check number the object of the transitive sentence and in the measure : features in more than one adjective. This makes the prediction that more than one accusative can be checked by the same Verb+Agr. This should be possible if the element (40) a. Pirjo juoksi kengät hajalle Pirjo ran shoes-ACC pieces-to that raises to check case gets a role some other way or does not need a thematic role. This 'Pirjo ran the shoes to pieces' seems to be the case for temporal measures in Finnish and perhaps English (e.g. John slept an hour). Of course, not all languages allow this possibility and more investigation is necessary to verify under which conditions two accusatives are a possibility. Note that b. *Pirjo juoksi kengät in the example (43) two accusatives are only possible if a directional adverb is added. Pirjo ran shoes-ACC 'Pirjo ran the shoes' (41) Pirjo juoksi kilometrin Pirjo ran kilometer-ACC 'Pirjo ran a kilometer' To explain these cases, we would need two of each verb again. We would need run1 and run2 and laugh1 and laugh2 depending on whether a measure was available or not, or a resultative was available or not. Alternatively we would need to say that case features are optionally present on the verb and are added before they get to the syntax. This is plausible but it makes Burzio's generalization, according to which verbs will be able to assign accusative if they have external arguments, a mystery. Suppose now that Burzio's generalization is in fact a generalization about transitive verbs, i.e., an intrinsic property of transitive verbs. We could interpret this observation as meaning that accusative is an intrinsic feature of transitive and unergative verbs. i.e., it is basically a category label. If it is a category label, i.e. if it means transitive verb, then it does not 82 3 8 This forces us to look for another explanation for the following contrast: (i) (ii) (iii) *John believes to be intelligent. John believes himself to be intelligent. I believe him to be intelligent. Following what I have proposed it must be the case that what rules (i) is the lack of null case to license PRO in the embedded clause. 83 (42) Pidin salaisuuden viisi vuota one is to say that the verb has no number features for object, if the object is in the I kept secret-ACC 5 years-ACC partitive. This would entail that the verb can enter the syntax with or without number features. If the verb enters the syntax with number features, and the object has partitive 'I kept the secret for 5 years' case, the derivation will crash because there will be unchecked number features on the (43) a. *Opas vei minut viisi kilometriä guide took I-ACC 5 kilometers-ACC 'The guide took me five kilometers' verb. This would be plausible, since, while the ability to have an external argument and therefore to check accusative is an intrinsic property of a certain class of verbs, number features that agree with the object are not an intrinsic property of verbs. This way we make a distinction between intrinsic features, which are lexically predicted and non- b. Opas vei minut viisi kilometriä etelään guide took I-ACC 5 km-ACC south-to 'The guide took me five kilometers south' lexically predicted features. Note that this is not the same as saying that there are two different verbs in terms of argument structure or case checking abilities. Rather it says that Agr features and case features have to be dissociated and that accusative case may be an inherent property of verbs that check nominative.39 It is interesting to note that regular double object constructions in Finnish, however, An alternative would be simply to say that Agr checks the number features of the verb cannot have both objects in the accusative, nor can they have one object in the accusative and in fact that is why the verb raises to Agr. However Agr is a checking position, not a and one in the partitive. This is due to the fact that only one object gets a role from the checker. We could then say then that if the verb does not show object agreement overtly, verb while the other argument is always marked by a post-position that inherently assigns then it does not have Agr features that need to be erased. The verb will still move through case to its complement. Examples are given below: Agr because Agr needs V features in order to be able to check tense at TP. If Agr does not have verbal features, it will not move to T and the derivation will crash because the T (44) a. Saat paketin minulta features will not be checked.40 get-2SG package I-ABL All that is important here is that I assume that accusative is always available as an 'You'll get a package from me' intrinsic feature of verbs and therefore need not to be wiped out and that partitive is not checked at AgrO. The discussion above, if in the right track, forces us to establish an b. Ole' hyvä ja anna' tuo laukku minulle'-ALL asymmetry between two types of features: features that need to be checked and features 'Please give that bag to me' 3 9 Urdu/Hindi shows overtly need to distinguish agreement and case. Objects can appear in the Once we assume that accusative case does not need to be checked off of the verb, the question arises as to what happens to the number features of the verb when the object carries partitive and does not raise to AgrO? There are two possibilities at least: the first 84 nominative or in the accusative case. If the object appears in the nominative object agreement is possible (if the subject is not in the nominative) and definite and non-specific indefinite interpretations are possible. If, on the other hand the object is marked accusative, then object agreement is impossible and the only interpretation is specific. See Butt 1993. 40 I will not consider here the possibility that Agr can be skipped. 85 that do not need to be checked because they are interpretable or because they are category markings. Moreover, it predicts that more than one DP can appear in the accusative case. To an imperfective verb the perfective morphology can be added. To a perfective marked verb, further imperfective morphology can be added, about which I will have nothing to say here. What never arises is a perfective verb with another perfective 2 . 1 . 4 Summary morpheme added, nor can a perfective marker be added on top of the second type of With respect to aspect, the Finnish case shows that what is really important is what is at imperfective morphology. AgrO by the time aspect is calculated. Sometimes a measure can be at AgrO and that is all that is necessary for the VP aspect to be externally bounded. Because Finnish seems to have two ways to check for the case of the internal argument, the result is that some objects will not raise to Spec of AgrO and a durative reading will be available. There are two main questions to be answered with respect to the Slavic aspectual system. The first is what do the perfective markers encode? Descriptively, they encode terminativity, but given the theory of aspect proposed in chapter 1, terminativity per se cannot be grammaticalized; it must be derivative of the semantic properties of either the verb or the object or both. Similarly, the same question can be asked for imperfective 2.2 Markings on the Verb: Slavic Aspect verbs. If they are not encoding durativity directly, how do they force durative readings? In this section I want to examine two cases where the durative/terminative readings seem The basic answer that I will give to these questions is that the perfective/imperfective to be encoded on the verb with consequences for the types of DPs that can appear in the distinction is encoding [SQA] features on the verb, rather than on the noun. In effect, object position. The two cases I will examine are the cases of Czech and Polish, two they are the equivalent of the determiner/quantifier system that a language like English languages that, like Finnish, do not have overt determiners. These languages have a has, but instead of quantizing the NP, they quantize the VP. Pushing this further, we can varied range of morphological ways to encode perfectivity and imperfectivity. The consider the imperfective verbs as "bare" verbs, i.e. non-quantized, just as mass nouns common trend on the analysis of these morphological processes is to not to treat them as and bare plurals are "bare" nouns in English. In both cases, durative readings will be independent functional heads in the syntax but rather as derivational affixes. As noted by derived. Spencer (1991), it is hard to argue that they are independent inflectional heads in the To do this I will adopt the analysis suggested by Partee (1993) and implemented by syntax, since not always does the addition of a prefix to a verb produce a transparently Filip (1992) for Czech and by Piñon (1993) for Polish and show how it fits well with the compositional result. It is also not the case that every verb can combine with every suffix. idea that aspect is calculated at AgrO. It very much depends on what verb the prefix is added to get a certain meaning or other. The basic structure for the three languages' derivational processes is given below: Partee (1993) discusses two types of quantificational elements: D-quantifiers like every and A-quantifiers like adverbials such as always in English. The A-quantifiers seem to be divided in two groups: (i) true A-quantification "with unselective quantifiers and a (45) V imp perf syntactic basis for determining, insofar as it is determinate, what is being quantified over" V and (ii) lexical quantification where "an operator with some quantificational force (and impV perhaps further content as well) is applied directly to a verb or other predicate at a lexical 86 87 level, with (potentially) morphological, syntactic and semantic effects on the argument how the proposal that terminativity is calculated at AgrO can be advantageous in that it structure of the verb. " (Partee 1993: 21) will allow us to unify English, Finnish and Slavic languages into a single proposal. The perfectivizing and imperfectivizing morphology in Czech and Polish, as Filip 2 . 2 . 1 Czech (1992) notes, is like the narrower scope A-quantification. Unlike an adverb like always which can unselectively bind a number of elements in its scope, these elements tend to The examples below illustrate two properties of the Slavic aspectual morphology: (i) interfere only with the interpretation of the direct argument of verbs with incremental pefectivizing morphemes have different semantic content; (ii) perfectivizing morphemes themes. 41 To explain the narrow scope of these verbal modifiers, Filip suggests, affect the interpretation of the object in different ways. Consider the examples below: following Partee (1990), that "since the verb and its arguments are in the relation of predication and since the predication is necessarily local, V0 modifiers have 'local semantic effects', and in some cases ' local quantificational effects'" (Filip 1992: 153). It is not clear what Filip means by predication in this context. (46) a. PilI kávu43 drank-3SG-MASC coffee-ACC 'He was drinking coffee.' I will follow the assumption that these elements should be treated as V0-modifiers that have the ability to take scope over certain types of complements. The selective scope of the adverbials will follow from the fact that the VP aspect is calculated at AgrO; thus only the verb and the argument that reaches AgrO can be directly affected by the head adverbial b. Vy-pilP kávu PREF-drank-3SG-MASC coffee-ACC 'He drank up (all) the coffee.' that is adjoined to the verb. This section is divided as follows: first I present the basic Czech cases in order to (47) a. PletlaI svetry show that the prefixes and suffixes that create perfective forms have different meanings, knitted-3SG-FEM pullovers-PL-ACC all of them quantificational, and affect the DP object in one way or another. Second I will 'She was knitting pullovers.' show that if we assume that these elements are interpreted at AgrO, no further stipulation needs to be made with respect to their scope.42 Then, in §3.2., I discuss two cases of perfective markers in Polish concentrating this time on the case of the complements and its interaction with different verbal morphology. b. U-pletlaP svetry PREF-knitted-3SG-FEM pullovers-PL-ACC 'She knitted (all) the pullovers.' [i.e. she finished knitting all the pullovers.] Again this section is not to be considered an in depth analysis of the so-called aspectual system of the Slavic languages. Instead it should be taken as a demonstration of 41 An incremental theme (Dowty 1991) is the basically the same as the = (rather than ⊆) relation on the thematic role for Verkuyl. 4 2 I will ignore so-called perfectivizing morphemes that change the argument structure of the verb. 88 4 3 Following the tradition in the Slavic literature, the superscript I stands for imperfective and the superscript P for perfective. 89 If the DP object has a specified measure, it is unacceptable with the imperfective verb (48) a. U-pilP kávu (ze šálku) PREF-drank-3SG-MASC coffee-ACC (from cup) 'He drank a little coffee (from the cup).' drink.44 (49) a. *PilI všechnu kávu drank-3SG-MASC all-ACC coffee-ACC b. Na-pilP se kávy PREF-drank-3SG-MASC REFL coffee-GEN b. *PilI dvě kávy 'He drank some coffee.' drank-3SG-MASC two coffee-GEN Consider first (46) and (47). Each pair (a) and (b) of these sentences differs only in c. *PilI hodně kávy the main verb. The (a) examples have the imperfective form and the (b) examples have the drank-3SG-MASC a-lot-of coffee-GEN perfective forms. What is important is that the only formal clue as to how the mass and the plural DPs in object position are to be interpreted comes from the verb. In the (b) The above examples are far from all the possibilities, but should suffice to show that examples, the DP complement has to be interpreted as bounded (i.e. as quantized) due to the so called perfective morphology in fact is a collection of perfectivizing morphology the perfectivizing morphology on the verb. In the (a) examples, on the other hand, the with different semantic import. In the cases above, the perfective marker induces a direct objects have to be interpreted as non-quantized, given the imperfective morphology terminative reading of the predicate. However, the terminativity is really coming from the (actually the absence of perfective morphology). fact that the perfective marker has quantificational import over both the verb and the Now, when we compare (46b) with (48a) and (48b) we can see that different object. perfective prefixes have different effects on how the object is to be interpreted, although What is important here is that this possibility holds only if the object can measure the they all give rise to terminative readings. Thus (46b) behaves as if there were a universal event, i.e. if the = (rather than the ⊆) relation holds between the internal argument and the quantifier in object position. In (48a) the prefix u-, when added to the verb pil 'drink' verb. If no homomorphism can be established between the verb and the DP object, then causes the DP to be interpreted as denoting a small portion (this is not to say that with the quantificational V0 modifier does not force a quantized reading on the direct object and other verbs it will have exactly the same meaning.) In (48b), the prefix na- gives rise, on the other hand, to the reflexive perfective napit seP. In this case, as the glosses show, it a durative reading will be derived. Note that this happens with exactly the class of verbs that in English do not care about the quantity of the object (push being one of them), has the quantificational force of the unstressed some in English. 4 4 If some temporal adverbial is added such as 'at the same time', then a quantified object is allowed with the reading that the object is produced or consumed gradually, if pragmatically plausible. Thus in the example below, there is a possibility of distributing over the 10 sweaters so that they all come gradually into existence: (i) PletlaI deset svetrů najednou knitted-3SG-FEM ten pullovers-PL-ACC at-the-same-time 'She was knitting ten pullovers at the same time' 90 91 namely the verbs that specify the ⊆ relation rather than the = relation on their thematic if, instead of having the indirect object as the lowest argument, we had it adjoined to VP, role. Thus in (50b), although there is a perfective marker, the noun can be interpreted as it would be impossible to have the perfective A-quantifiers modifying only the direct quantized or bare, depending on context. This is not the case for verbs which specify the object. = relation on their objects. (51) AgrO Agr (50) a. MíchalaI jsem polévku [perf+V]+Agr stirred-1SG-FEM AM-AUX-1SG soup-ACC VP subject 'I was stirring (the) soup' V VP object b. ZamíchalaP jsem polévku V stirred-1SG-FEM AM-AUX-1SG soup-ACC 'I stirred (the) soup' V' PP The assumption I am making here is that the verb+perfective, being a morphological head that is part of the verb can only take scope over the checking domain of the verb. We can account for these facts45 straightforwardly simply by calculating aspect at AgrO. Given that the VP aspect is calculated at AgrO, it follows that the affixal element will be able to take the V+object in its scope. In fact almost all the objects in both the That will include in the tree above only the specifier of AgrO. Evidence for differences in scope between a derivational head and a syntactic head can be seen in English as the contrast below shows: perfective and imperfective verbs are marked for accusative (I will deal with the genetivus partitivus in the next section). The fact that only the direct object can be affected (52) a. John isn't lucky at a single sport he plays. by the perfective markers, and not other internal arguments, is further evidence that the b. *John is unlucky at a single sport he plays. Spec/head relation at AgrO, rather than the VP, is what is important. If the aspect interpretation were simply VP internal, it is hard to see how an indirect object could not (53) John is unaware of a single problem. also be unselectively bound by these modifiers. If we took the indirect object to be generated in a VP shell and all that the A-quantifier needed was to take scope over the VP, In (52), the sentential negation can license the negative polarity item a single, but the it would be very hard to see how the indirect argument would have been omitted as the prefix un- cannot license it in the same context. However, when the polarity item is in the schematic tree in (51) for a double object verb (based on Larson's analysis) shows. Even complement of an un-adjective, then it can be licensed. This follows from the fact that in (52a,b) a single sport is not a complement of unlucky but is a complement of at, and thus 45 These examples are from Filip. As Ian Roberts pointed out, those cases have long-head movement which suggest that the participle is in C. It remains to be seen whether there are aspectual differences in cases with and without head raising and what is the appropriate analysis for these constructions. 92 cannot raise to some position where it enters a spec/head relation with unlucky. In the case of (53), on the other hand, a single problem is the complement of unaware of and 93 has to raise to some Agr position to check case against the complex head unaware of. In quantized if the PATH is quantized. In this case, the DP complement must have its this case the negative polarity item, which needs to be in the scope of a negative element, cardinality specified. If the verb and the object do not build a PATH then the can be licensed. The relevant structures at LF for (52b) and (53) are given in (90) and perfectivizing morpheme and the imperfectivizing morpheme will affect the verb only. (55) respectively. On the assumption that contentful prepositions such as at check their The proposal that aspect is calculated at AgrO is entirely consistent with the facts cases via an independent Agr projection, the negative polarity item will not be in the about Czech aspectual morphology. Unlike Finnish, objects of both terminative and checking domain of unlucky and therefore not in the scope of un- in (90). On the other durative predicates are marked for accusative case. The perfective morphology will hand, if of only checks case via incorporation into the adjective and therefore via the Agr impose a terminative reading on the VP only if it attaches to the right kind of verb. If a projection of the AP itself, then the negative polarity item will be in the checking domain bare NP object checks case against a perfective marked verb whose argument structure of the unaware and therefore in the scope of un-. specifies the = relation, the object will be interpreted as quantized relative to the lexical content of the perfective marker (some, all, a little, etc.) AgrO immediately turns into the (54) AP right position for this to happen, given that these elements only take scope over the A AgrP object. unlucky DPi Agr a single sport Agr P atj (55) PP Agr P tj 2 . 2 . 2 Polish ti 2 . 2 . 2 . 1 Perfective vs. Imperfective markings The brief discussion of Czech has given us a feeling for the range of the different AgrP DPi perfectivizing morphemes. In this section I will try to give a clearer idea of the role of the Agr' a single problem Agr Aj Agr imperfective morphology based on Wierzbicka's (1967) and Piñon's (1992, 1993) work AP P+Aj P Aj ofk unaware on Polish. Then I briefly make a note on the so-called genetivus partitivus that is only PP P tk ti possible with the object of some perfectivizing morphemes. Basically the imperfective verb denotes a sequence of actions. Such sequence is treated as "an innumerable" set and the "successive acts of this sequence cannot be A perfectivizing morpheme is then an A-quantifier that can only take scope over the counted" (Wierzbicka 1967: 2237). In a sentence with a perfective verb, on the other checking domain of the head. It looks for a quantized predicate; an imperfectivizing hand, "some element distinguishing the object as a separate element (set of elements) morpheme looks for a homogeneous predicate. The predicate will be homogeneous if the must be explicitly or implicitly present)". (1967: 2238) PATH built by the verb and the object is homogeneous, which entails in other words that The imperfective has more than one interpretation. It can have an imperfective nonthe DP complement is homogeneous (or can be interpreted as such). The predicate will be contemporaneous meaning or a progressive meaning. The first meaning is illustrated 94 95 below in (56a) which contrasts with the perfective version in (56b). The English glosses (58) a. On zjad P misk kaszy show that in English it is the article that solves the ambiguity between an imperfective and 'He ate a bowl of porridge.' a perfective reading. This is one of the readings of the imperfective. In Polish, the only element that distinguishes these two readings is the marking on the verb. b. On zjad P kasz = On zjad P ca a kasz 'He ate all the porridge.' (56) a. On jad oliwki 'He has eaten olives (all his life).' This implicit measure, the so-called genetivus partitivus is obligatory with the perfective forms of some verbs (59a), and impossible with the imperfective form of the b. On zjad oliwki verb (59):46 'He has eaten the olives.' (59) a. Zjad emP sobie truskawek The other reading of the imperfective, which will concern us throughout the rest of 'I ate (have eaten) some strawberries.' this chapter, is one that is similar to the progressive in English. With such a reading, the imperfective morphology is incompatible with measures such as some, a bowl , a b. *Jad emI truskawek spoonful etc. I ate (have eaten) strawberries-GEN (57) a. *On jad I w tym momencie troch kaszy c. Jad emI truskawki 'He was then eating some porridge.' 'I ate (have eaten) strawberries-ACC' b. *On jad I wtedy misk (lizk ) kaszy 'He was then eating a bowl (a spoonful) of porridge.' With the perfective form of [+ADD TO] verbs, not only are measures acceptable, but they are required, although the measure is not always overt. 46 With certain verbs (mainly I believe [–ADD TO] verbs) either accusative or genitive is possible, but here it is not a partitive: (i) a. Po ycza pieni dze na prawo i na lewo s/he-borrows-IMP money-ACC on right and on left s/he borrows money left right and center b. Po ycza pieni dzy na prawo i na lewo s/he-borrows-IMP money-GEN on right and on left s/he borrows money left right and center 96 97 If a numeral is added with the object, and the verb is in the imperfective, the resulting reading is one in which the verb cannot take the internal argument as a measure. Thus, in with a count singular interpretation) all that is required of the VP is that it be homogeneous; the object will thus be interpreted accordingly.48 (60a) the two pears have to be eaten simultaneously. (60a) is paraphrasable as (60b).47 (61) (60) a. On jad wtedy 2 gruszki (Wierzbicka) Bo ena pisa aI list, kiedy Jan wróci P Bo ena write.PAST.F letter.ACC when Jan return.PAST.M 'He was then eating two pears.' 'Bo ena was writing a/the letter when John returned.' Following Piñon, we can treat the genetivus partitivus in Polish as referring to proper b. On jad wtedy 2 gruszki 'He was then eating two pears at the same time.' parts of some object, which is incompatible with reference to the whole. This explains why the accusative is acceptable but the genitive is not with imperfective verbs. These basic facts show that Polish is quite similar to Czech in that perfective markers require quantized VPs, and force a specified quantity to be present in the direct object. In the next section I take a closer look to the genetivus partitivus in order to rule out one alternative analysis existing in the literature. With respect to the imperfective, we can treat these as forming a "bare" VP, i.e. one 2 . 2 . 2 . 2 Accumulative nawhich is interpreted as homogeneous. In this way, we can explain why quantized objects are unacceptable unless the quantity is taken to be a whole "lump", so to speak. This The discussion in this section is going to be based on the semantic analysis of Piñon allows us to account for singular count nouns as objects of imperfective eventive verbs as (1993) for the na- prefix. Like Filip, he analyses na- in Polish as an A-quantifier that has in (61). Rather than imposing homogeneity on the object (which would be incompatible the property of applying to a class of imperfective verbs adding a sense of accumulation to the interpretation of the direct object and a sense of perfectivization to the whole VP. 4 7 One could think that in these cases the head noun would incorporate into the verb leaving behind the quantity marker. However, there is no apparent differences between the perfective and the imperfective paradigms in terms of scrambling as the acceptability of the examples below show: (i) a. Maria zjad a dwa jab ka. Maria ate-PERF two apples (62) b. Dua Maria zjad a jab ka. Two Maria ate-PERF apples Examples are given below: a. drukowa I > na.drukowa P print > print a lot of b. obiera I > na.obiera P peel > peel a lot of c. rwa I > na.rwa P pick > pick a lot of c. Jab ka Maria zjad a dwa. Apples Maria ate-PERF two (ii) a. Maria jad a dwa jab ka. Maria ate-IMP two apples 48 The imperfective seems to yield odd readings similar to those noted in English for the progressive. Thus, (i) is somewhat odd compared to (ii). In (ii), we prefer a reading that is sort of simultaneous, i.e. it doesn't need to be the case that one is reading the papers literally at the same time, but simply that they are being read as a group. For example we would judge (ii) to be false if the speaker had only one paper in his possession at the time. (Alan Munn, personal communication.) b. Dua Maria jad a jab ka Two Maria ate-IMP apples c. Jab ka Maria jad a dwa Apples Maria ate-IMP two (i) I was eating two apples when Mary arrived (ii) I was reading three papers when Mary arrived 98 99 Intuitively, the idea is that entities accumulate in the course of an event if their quantity increases in the course of the event. Thus pick, print and peel produce a result that accumulates over time. Eat, read, love on the other hand do not produce more food, more (64) *Marek na-gada- g upstw-a Marek NA-said nonsense-ACC 'Marek said a lot of nonsense.' books or more love. These verbs will be incompatible with the accumulative na-. 'read enough The examples below, however, show that the genitive marker in the object has of' with the obligatory reflexive pronoun are perfectly acceptable. "Such verbs make nothing to do with the na- prefix itself. Numerals or measure phrases are perfectly reference to the experience of the agent, who by eating something comes to feel full acceptable with na- provided they denote large amounts of the object. When an overt enough etc." (Piñon 1993:15 fn11)49 measure appears, it cannot be in the genitive; instead, it appears in the accusative. The However, the verbs 'eat enough of' and Na- is interesting in that at first sight it seems that it blocks accusative case, as the genitive is therefore an internal marker of the partitive, rather than a case assigned by the example below from Hale and Bittner (1994:12) shows. This would pose a potential verb. The DP complements in fact are checking accusative case and the result is a problem for the analysis I am proposing that links terminative readings to movement to terminative reading. AgrO. Hale and Bittner propose that the lack of accusative case follows from treating naas an incorporated measure meaning a lot, which leaves behind a genitive marked complement (see also Baker 1988 who first proposed this analysis of na-.). (63) a. Marek gada- g upstw-a Marek talked nonsense-ACC 'Marek said nonsense.' b. Marek na-gada- g upstw Marek NAi-talked [ti nonsense-GEN] 'Marek said a lot of nonsense.' This predicts that accusative objects with na- should be unacceptable as (64) shows. 4 9 This case is very similar to the reflexive "aspectual" marker in Spanish I will discuss in section 4. 100 101 (65) a. *Irenka na. uska aP stu orzechów (66) a. *Nadrukowa em ka dej strony Irenka NA-shelled hundred-GEN nuts-GEN I NA-printed each page-GEN-SG b. Irenka na. uska aPsto orzechów b. *Nadrukowa em wszystkich stron I NA-printed all pages-GEN-SG Irenka NA-shelled hundred.ACC nuts.GEN 'Irenka shelled a hundred nuts.' This is not to say that all perfectivizing morphemes are incompatible with all and each. c. *Irenka na. uska aP koszyka orzechów If the prefix is do- (add more) is the perfectivizing morpheme, then the result is acceptable Irenka NA-shelled basket.GEN nuts.GEN with perfective forms. Imperfective forms are unacceptable with each and marginal with Irenka na. uska aP koszyk orzechów all. It is necessary to have the accusative form, as illustrated in (68). . Irenka na-shelled basket.ACC nuts.GEN (67) a. *Dodrukowa em ka dej strony 'Irenka shelled a basket of nuts.' I-more-printed-PERF each page-GEN-SG d. *Irenka na. uska aP trzech koskyków orzechów b. Dodrukowa em wszystkich stron Irenka NA-shelled three.GEN baskets.GEN nuts-GEN I printed all pages-GEN-SG Irenka na. uska aP trzy koskyki orzechów Irenka na-shelled three.ACC baskets.ACC nuts.GEN (68) a. Dodrukowa em ka da strone 'Irenka shelled three baskets of nuts.' I-more-printed-PERF each page-ACC-SG These examples show that na- is not substituting for a measure. Rather it behaves just like other perfective morphemes in the sense that it seems to be part of the verb, rather than a quantifier incorporated from the object. Given the ability to co-occur with other b. Dodrukowa em wszystkie strony I printed all pages-ACC-SG measures, it seems to be better to treat na- as another instance of an A-quantifier that I believe that the reason na- is incompatible with each and all has to do with meaning modifies a quantized VP. Essentially it selects for quantized objects that accumulate to a of the prefix itself. In the case of all, it seems that the meaning of accumulation above degree that is above expectation. The prefix na- is, however, incompatible with each and expectation that na requires cannot be satisfied. In the case of each distributivity may be all. causing the incompatibility. 102 103 The genetivus partitivus is then a complex complement with a null some as its head. This null head checks for accusative case at AgrO just as the overt measured DPs do. 2.3 More A-quantifiers unacceptable, while a quantized mass noun is perfect. Similar contrasts can be seen in (70-72).50 (69) a. John half-ate an apple b. ?John half-ate apples To speakers of Germanic and Romance languages, the perfective morphemes in Polish c. *John half-ate bread seem quite odd, given that determiners do most of the work in determining terminativity d. John half-ate the bread of the VP in those languages. In this section I would like to briefly point out two examples, one from English and one from Spanish which are likely to be analyzed as Aquantifiers much in the same way as in Slavic. The first case is half- in English. The (70) a. *John half-read junkmail b. ?John half-read novels second case is the case of a reflexive clitic in Spanish. I will not give a detailed analysis c. John half-read the novel of either case, but will simply show that the kinds of restrictions they place on the interpretation of the object make them amenable to an analysis involving A-quantification. (71) a. ?John half-built houses b. John half-built a house 2 . 3 . 1 Half- analyzed At least two English prefixes impose restrictions on their objects: re- and half-. The (72) a. ??John half-discussed the problem relationship between re- and telicity has been discussed by Wechsler 1991. I know of no b. *John half-said something work on half- (although Tenny 1987 discusses halfway). Half- can be prefixed to a variety of verbs with varying degrees of acceptability. For Although the ability of a verb to be modified by half is highly idiosyncratic, as the many speakers it simply means 'sort-of', and acts as what Quirk et al. 1985 call a examples above show51 (eat is better than drink and read and build are better than 'downtoner'. This use of half seems not to impose much restriction on the object of the verb. However, for some speakers, half- is much more quantificational, and imposes restrictions on the direct object. Consider the examples below. In (69a/b) there is a contrast between a singular and a bare plural in the object position. Half- seems to be requiring a [+SQA] object. This becomes more apparent with the contrast in (69c/d) where a mass noun bread is totally 50 Re- also shows the same restriction with respect to mass nouns, a fact that I have not seen mentioned in the literature: (i) (ii) (iii) (iv) *John reread mail. John reread the mail. *John re-kneaded dough. John re-kneaded the dough. The same effects hold in Brazilian Portuguese: (v) (vi) *João renegociou terra. 'João renegociated land.' João renegociou a terra. 'João renegociated the land.' 5 1 It seems that many verbs allow only the downtoning reading. 104 105 discuss), with consumption verbs, it is necessary that the quantity of the object be specified in order to obtain a reading in which the event is halfway completed. There is a clear contrast between mass noun and bare plurals. Mass nouns are clearly worse as (half ¬) b. I didn't half eat the bread. complements of half-V. There are two possibilities to analyze half. We can think of half as a partitive on the object that incorporates into the verb, schematically illustrated below: (73) (75) a. I didn't eat half of the bread. AgrOP (¬ half) An even more clear example comes from (76). (76) a. I only half-saw John and Mary Agr' b. I only saw half of John and Mary [half +V] VP To the extent that (76b) is acceptable, it either means that you saw one of John or V' Mary, but not both, or it means that you literally saw only half of them, for example from half+V DP half the waist up. (76a), on the other hand, does not give either of these readings. Instead, it D' D NP means that you saw John and Mary, but it was more like a "half a seeing of them" than a "seeing of half of them". This would be similar to Hale and Bittner's analysis of Polish na-. In fact, at first These examples argue for treating half as an A-quantifier that modifies a quantized sight, this might seem plausible, as half often does not require an overt case marker when event which will force a specified quantity on the object, and derive terminative readings. modifying an NP; thus we can say half the bread. Given the fact that the process is highly dependent on the types of verbs and the result However, it is not clear that half- has the same quantificational meaning as it does meaning of the expression varies from partially doing something to doing something inside the partitive DP. Consider the difference between (74a/b). While (74a) entails that sloppily (the downtoner reading) the parallel to the Polish perfective markers is striking. both apples get partially eaten, (74b) is consistent with one apple being completely Half is not the only A-element that forces a quantized object52. The examples below uneaten. show that out has a very similar behavior: (74) a. I half-ate two apples b. I ate half of two apples Negation will also show the differences between the prefix half and the half partitive, as pointed out to me by Ian Roberts. Half can take scope over negation only when it is quantifying over the object rather than as an A-quantifier half-. 106 52 I owe examples ($77) and ($78) to Ian Roberts. 107 (77) a. John out-ran a contestant. (80) Yo ME comí diez manzanas, y ahora me duele el estómago b. ?John out-ran contestants. I CL.1SG ate ten apples and now me hurts the stomach c. *John out-ran deer. 'I ate ten apples, and now my stomach hurts.' d. John out-ran the deer. (81) (78) a. Norbert out-argued a student. José y yo ya NOS sabemos toda la lección Jose and I already CL.1PL know all the lesson b. ?Norbert out-argued students. 'Jose and I already know the whole lesson.' c. *Norbert out-argued ideology. d. Norbert out-argued the ideology. If in example (79) instead of una copa de vino we just have vino 'wine', the result is unacceptable. In the same way, if instead of diez manzanas or toda la lección we have just The case of out is in a sense better than half to make the point that these elements are not originated in the DP complement, since out-a student is clearly not a well-formed manzanas 'apples' or el Espanõl 'Spanish', the sentences are unacceptable as exemplified below: complement. (82) 2 . 3 . 2 A reflexive pronoun in Spanish *Juan SE tomó vino anoche antes de acostarse Juan CL.REFL drank wine last night before going to bed 'Juan drank wine last night before going to bed.' Spanish also has a likely candidate for some sort of A-quantifier similar to the Polish and English cases. The basic fact is that a reflexive pronoun can be added to achievements and accomplishments provided the object has its quantity specified. The net effect of (83) *Yo ME comí manzanas, y ahora me duele el estómago. adding the reflexive is that the object has to be entirely affected in the event. Consider the I CL.1SG ate apples and now me hurts the stomach following examples from Nishida (1994): 'I ate apples, and now my stomach hurts.' (79) Juan SE tomó una copa de vino anoche antes de acostarse (84) *José y yo ya NOS sabemos el Español. Juan CL.REFL drank a glass of wine last night before going to bed Jose and I already CL.1PL know the Spanish 'Juan drank a glass of wine last night before going to bed.' 'Jose and I already know Spanish.' Of course without the clitic, no problem arises with these objects. A clear contrast is shown in (85) and (86). Read is a verb that allows a partial or a completed affectedness 108 109 on the internal argument. Thus in (85) Juan doesn't need to have finished the book. In (86), however, this is not possible. The object must have been totally affected. Thus if the object has been generated in object position (for the relative clause case, see chapter 4) and its quantity is specified and the object is totally affected the clitic is acceptable. (85) Juan leyó el libro un poco más anoche, y ahora le faltan dos capítulos 'Juan read the book a bit more last night, and now he has two chapters left.' According to Nishida we cannot argue that the reflexive pronoun is argumental, but it should, instead, be seen as an aspectual marker. He presents two reasons for rejecting the argument analysis: first, the sentences above are perfectly acceptable without the reflexive (86) *Juan SE leyó el libro un poquito más anoche, y ahora le faltan dos capítulos pronoun, and second, the reflexive pronouns cannot be substituted for their strong 'Juan read the book a bit more last night, and now he has two chapters left.' counterparts. Nishida argues that these reflexive elements are base generated with the verb and have the property of forcing the object to be quantized in the sense of Krifka For some speakers, if todo, 'all' is added then the reflexive is obligatory: (1989, 1991). Assuming such a hypothesis, we can account for the fact that what is (87) *?Mi abuelito fumó todo el puro Mi abuelito SE fumó todo el puro 'My grandfather smoked the whole cigar.' relevant is that it takes scope over the VP only, as the other A-quantifiers did. (90) AgrO Object Agr [V+SE]+Agr (88) VP *?Los invitados tomaron todo el vino Los invitados SE tomaron todo el vino 'The guests drank up all the wine.' The difference between the SE and the cases of half- or na- is that SE is not a derivational affix but a clitic. For morpho-phonological reasons the clitic has to raise to some higher position to check certain features. This implies that the clitic is excorporated If the object receives a type reading, which always allows durative readings (see in the syntax given that clitic climbing is possible. Alternatively, the verb is generated chapter 5 for fuller discussion), the clitic is unacceptable. with a null A-quantifier that is licensed by being bound to a non-thematic clitic pronoun, (89) *En esos días Juan no quería comerSE las uvas cultivadas en esa región on those days Juan NEG wanted to-eat-CL.REFL the grapes grown in that which shares person features with the subject. The latter solution avoids excorporation but potentially creates a problem for scope. Why can't the A-quantifier have as much scope as its binder? One possibility is that the binding has no LF purposes and the non- region 'At that time, Juan didn't want to eat the grapes grown in that region.' thematic binder is deleted at LF. I don't see at this point that the latter alternative is any better than excorporation and I will assume with Roberts 1991 that the clitic can excorporate. Such excorporation is driven by PF requirements and thus its final position is irrelevant for interpretation. 110 111 What is important for present purposes is that the reflexive clitic behaves much like features in Finnish and the verb to check them in Polish and Czech. At the same time we the A-quantifiers in that it modifies particular kinds of quantized predicates, being highly would still need a position for the DP complement to check its case. If we collapse the lexically dependent. two we seem to get the best of both worlds. In addition, by assuming that only terminative readings have a specified configuration, durativity can be thus be obtained in 2.4 various ways: in Polish by having an object with unspecified quantity and in Finnish by Conclusions having the partitive object not move at AgrO. In this chapter I have put forward an analysis for Finnish partitive/accusative case and Finally, the Spanish and English cases of A-quantifiers are interesting in that they Slavic perfective/imperfective that gave empirical support to the proposal that terminative show that it is not a prerogative of Slavic languages to have A-quantifiers selecting for readings are obtained only if a certain syntactic configuration holds at LF, where aspect is quantized or non-quantized VPs. A-quantifiers that affect the object only are possible if interpreted. We showed that just by assuming that the partitive object in Finnish does not the language allows head-A quantifiers, otherwise it does not. It has nothing to do with a raise to AgrO, immediately entails that the predicate will be interpreted as durative. parameter of aspect. The same is true for the case distinctions in Finnish. As we have Accusative, on the other hand, allows terminative readings because the object will be a seen in the previous chapters, both in English and in Brazilian Portuguese we can make a specified quantity and will also be at AgrO by the time aspect is calculated. Both the verb predicate durative as long as we have other ways to check the case of a quantized object. and the object move to AgrO not to be interpreted but to satisfy morphological requirements. Not only is there no "aspect" parameter, but "aspect" is nothing special that has to be learned by the child. What the child has to learn is the particular case assigning properties In Polish and Czech, the object always receives accusative case and terminativity is, therefore, dependent on what appears at AgrO by the time aspect is calculated. Cases of its language and the specific lexical items such as A-quantifiers. What the child does not have to learn is how the language "encodes" aspect per se. where the DP complement appears with no overt specified quantity can be disambiguated In the next chapters we will move one step further and look at some more complicated by perfectivizing morphemes. Different perfectivizing morphemes in Polish and Czech verbal complements and show that the internal structure of the complement can affect the will select for quantized readings of the objects and the result is a terminative readings. In VP aspectual interpretation. In the next chapter I discuss the aspectual properties of other words, if a perfectivizing morpheme appears on a verb whose complement does not accusative clitic doubling in Spanish, and in the following chapters I will show how appear to be quantized, we know that we are to interpret the VP as quantized. relative clauses, demonstratives, certain kinds of adjectives and other constructions give Imperfective morphemes can A-quantify over homogeneous predicates. Thus, the object rise to surprising aspectual effects. cannot be quantized or has to be interpreted as non-quantized in order for the result to be acceptable. The result will be a durative reading for the VP. Notice that, if we had an aspectual phrase with a feature durative/terminative, we would have to make this feature loose enough to allow the object to check the aspectual 112 113 Chapter 3 Lack of iteration: Accusative Clitic Doubling, participial absolutes and have+agreeing participles* 3 . 1 Basic Problems Posed by Accusative Clitic Doubling Constructions Consider the following sentences: Introduction (1) a. Toqué la sonata. The goal of this chapter is to present an analysis of accusative clitic doubling that can (I) played the sonata account for some of its traditional problems and also for a new problem related to its 'I played the sonata.' (Cordoba Spanish) aspectual properties. The analysis I will be proposing will provide further evidence for b. La toqué. the claim that the syntax of direct arguments crucially affects the aspect of a predicate, (I) it played since it crucially affects what is at AgrO by the time the terminativity of the VP is 'I played it.' calculated. This chapter is divided into four parts: in section 1 I lay out some of the traditional problems associated with accusative clitic doubling and I add a new problem c. Lai toqué [a la sonata]i related to its aspectual properties. I also show that there are two other apparently unrelated (I) iti played [a the sonata]i constructions that display the same aspectual properties: participial absolutes and 'I played the sonata.' have+agreeing participles. In section 2, I show that accusative clitic doubling has similar properties to certain "identificational" small clauses. By assuming that accusative clitic doubling is structurally an identificational small clause, it is possible to take care of the traditional problems. In section 3, I deal with the aspectual properties of the construction. There I show that the aspectual properties of the accusative clitic doubling construction follow directly from its internal structure. In section 4 I extend the analysis to participial absolutes and have+agreeing participles in Brazilian Portuguese and Spanish. (1a) illustrates a regular transitive construction with a DP object; (1b) illustrates the clitic version of the simple transitive construction; and (1c) illustrates what has been called accusative clitic doubling. Besides the DP argument, we also have a clitic that is interpreted as coreferential with the object. Accusative clitic doubling constructions in Spanish have posed at least three problems for the theory of grammar: first we face the problem of having two DPs apparently competing for the same thematic role. I will call this the thematic role problem. Why don't * This chapter was born by reading Marcela Depiante's paper on clitic doubling in Argentinian Spanish. I thank her for data and very helpful discussions. I also thank Claudia de Albaladejo Galus, Ricardo Echepare, and Gabriela Alvarez, respectively, for discussion and data on conjoined accusative clitic doubling, quantifier scope interactions and identificational small clauses. For data on conjoined accusative clitic doubling I am also thankful for the many Cordoba speakers around the world that replied very kindly to my questions. A version of this chapter has been presented at Going Romance 1993 in Utrecht, and at the University of Connecticut, and a published version is scheduled to appear in Probus. I thank the audience of the Conference, particularly Denis Delfitto and Elena Agnoustopoulu. I also thank Norbert Hornstein, Dave Lebeaux, and Ian Roberts for discussion and comments on an earlier version of the paper. Finally I would like to thank Alan Munn and Juan Uriagereka for discussion, editorial comments, help with English and a very deep patience. I also thank Margarita Suñer and the anonymous reviewers for Probus for very helpful and careful comments on the previous versions of the forthcoming paper. 114 we interpret (1c) as I played itj and [the sonata]i? Given that this is not the interpretation we get, the picture is similar to constructions with pleonastic elements, i.e., constructions in which one of the DPs does not have a thematic role and is just present for predication reasons. If the intuition that one of the elements in accusative clitic doubling is pleonastic 115 is in the right track, then the questions are the following: which of the elements is the predicate will always be terminative. Unlike regular terminative predicates, the accusative dummy element, and how can the dummy appear in object position? clitic doubling constructions, however, disallow iteration of the VP predicate. The second problem is what is the role of the prepositional element a in the accusative Consider the predicate play the sonata. As we have seen, adverbials such as for x time clitic doubling. Note that the a does not show up in the regular transitive constructions or until x time can do two things to a terminative VP predicate: either they stretch the (1a). Let's call this the "a" problem. event so that duration of the event can cover the period described by the adverbial (as The third problem that these constructions have traditionally posed is the fact that not illustrated schematically in (4a)) or they force the mapping of subevents of the event all types of NPs can appear as the doubled object (Torrego ms.; Jaeggli (1986); Suñer described onto the stretch of time covered by the adverbial (4b). In the first case, we play (1988); Sportiche (1993); Uriagereka (1995)). I will call this the NP restrictions problem. a sonata in slow motion and in the second case we repeatedly play the sonata so that it For example, while a definite noun phrase, a demonstrative and a possessive DP can be fills the duration of the adverbial (let's say three times, as in the drawing): doubled (2), a bare plural, many and few, for example, cannot, as illustrated in (3): (4) (2) a. Loi vi [[al/a este/a su] hombre]i (Cordoba Spanish) (I) iti saw [a [the/this/ his] man]i play play the sonata [_______] for one hour the sonata [__/__/__/] for one hour 'I saw the/this/his man.' b. Loi vi [[al/ a esto/ a su ] libro]i (I) iti saw [a [the/this/his] book]i If such an adverbial is added to an accusative clitic doubling construction, however, the only possible reading is the one in which a single event has been artificially stretched to the point of covering the duration of the modifying adverbial. 'I saw the/this/his book.' Thus we cannot get subevent readings of the event described by the VP if the direct (3) a. *Losi vi [a hombres/ a libros]i object is doubled, as the contrast in (5) demonstrates. Without the clitic it is possible to (I) themi saw [a men/a books]i play the sonata ten times; with the clitic, this reading is impossible. The only possible 'I saw men/books.' reading of (5b) is the stretched reading. b. *Losi vi [a muchos hombres/pocos hombres] (5) a. Toqué la sonata hasta las 12, de hecho la toqué 10 veces. (I) them saw [a many men/few men] 'I saw many/few men.' (Cordoba Spanish) (I) played the sonata until 12, in fact (I) played it 10 times. There is a fourth problem that accusative clitic doubling constructions pose, which I will call the aspect problem. Because the doubled DP is (roughly) definite, a doubled 116 117 b. #Lai toqué [a la sonata]i hasta las 12, de hecho la toqué 10 veces. b. *Maria tinha feitas camas. (I) iti played [a the sonata]i until 12, in fact (I) played it 10 times Maria had made-FEM.PL. beds-FEM.PL 'I played the sonata until 12, in fact I played it 10 times.' 'Maria had made beds.' The same restrictions on the types of DP complements and the same restrictions on (8) a. Convocada la reunion, María se levantó. the aspectual interpretation that hold for accusative clitic doubling constructions also hold called-FEM.SG the-FEM.SG meeting-FEM.SG in two other constructions in both Argentinian Spanish and Brazilian Portuguese.53 'With the meeting called, Maria stood up.' The constraints on the types of determiners for participial absolutes constructions are illustrated in (6) and (8) for Brazilian Portuguese and Spanish respectively; the constraints on the DP arguments in the have+agreeing participle constructions are illustrated in (7) (Spanish) b. *Convocadas reuniones, María se levantó. called-FEM.PL the-FEM.PL meeting-FEM.PL 'With meetings called, Maria stood up.' and (9) for Brazilian Portuguese and Spanish, respectively. Finally the lack of iteration in both constructions are illustrated in (10) for Brazilian Portuguese and (11) for Spanish . (9) a. María tenía convocada la reunion. (Spanish) Maria had called-FEM.SG the-FEM.SG meeting-FEM.SG (6) a. Resolvidos os problemas, podemos sair. (Brazilian Portuguese) 'Maria had called the meeting.' solved-MASC.PL the-MASC.PL problems-MASC.PL, (we) can leave b. *María tenía convocada reuniones. 'With the problems solved, we can leave.' Maria had called-FEM.PL meetings-FEM.PL b. *Resolvidos problemas, podemos sair. 'Maria had called meetings.' solved-MASC.PL problems-MASC.PL, (we) can leave (10) a. *Tocada a sonata por uma hora, Maria se levantou. 'With problems solved, we can leave.' (7) a. Maria tinha feitas as camas. (Br.Port. iterative reading) (Brazilian Portuguese) Maria had made-FEM.PL. the-FEM.PL. beds-FEM.PL 'Maria had made the beds.' Played-FEM.SG the-FEM.SG sonata-FEM.SG for an hour Maria stood up b. *Maria tinha feitas as camas por dois dias. (Br. Port. iterative reading) Maria had made-FEM.PL the-FEM.PL beds-FEM.PL for two days (11) a. *Tocada la sonata hasta las doce, María se levantó. (Sp. iterative reading) Played-FEM.SG the-FEM.SG sonata-FEM.SG until 12, Maria got up 53 Brazilian Portuguese lacks third person accusative clitics in spoken language. Consequently it cannot have accusative clitic doubling. European Portuguese clitics behave differently altogether, and I will have nothing to say about them in this thesis. 118 119 b. *María tenía discutida la sonata por 3 horas (Spanish iterative reading) Maria had discussed-FEM.SG the-FEM.SG sonata-FEM.SG for three hours Cordoba dialect does not allow it (as in have+agreeing participles). Specific indefinites are also acceptable, as pointed out by Suñer (1988) (see also Everett 1992 for a careful discussion of the dialectal variation in River Plate Spanish). The optionality of a and the In this chapter I argue that the restrictions on the types of noun phrases that can possibility of specific indefinites are both completely unacceptable in Cordoba Spanish in appear in accusative clitic doubling, participial absolutes and have+agreeing participles are accusative clitic doubling. What accounts for this variation is unclear, and only a much instances of the same phenomenon. I propose that the constraints on the types of noun more in-depth analysis of the subtle syntactic and semantic differences between the phrases are to be derived from the fact that the internal argument in all three constructions determiners and the preposition a across dialects may be able to tease them apart. is not a simple DP but is a small clause with a non-thematic subject. With respect to the For the purposes of this chapter, however, I will concentrate on the Cordoba dialect aspectual properties, I propose that lack of iteration is to be associated with the fact that at for accusative clitic doubling, since in this case there is no variation: (i) the a is obligatory; LF the DP complement is not in a specifier-head relation with the complex [V+Agr] at (ii) specific indefinites are not acceptable; and (iii) doubling is possible with any regular AgrO. This structural property will block subevent readings. direct object. Data from Brazilian Portuguese will become relevant when we extend the Before we move to section 1.1, an observation about the data I will be discussing is analysis to participial absolutes and examine the behavior of bare plurals in these necessary. There is a lot of variation among Spanish dialects. Many dialects of Spanish constructions, since Brazilian Portuguese, unlike other Romance languages, allows allow accusative clitic doubling of pronouns and animate objects only. This is true, for generic and specific readings of bare plurals quite freely, a fact which will become useful instance, for many dialects of Spanish spoken in Spain and in some dialects spoken in in arguing against existing analyses of participial absolutes. Uruguay. The dialect I am going to be discussing here is the Spanish spoken in the central part of Argentina (Cordoba more specifically) where accusative clitic doubling is 3 . 1 . 1 Main Issues in Previous Analyses and a Problem for Agreement in Specificity equally possible with both animate and inanimate objects. I will not have much to say here as to why an animacy restriction should be crucial to license accusative clitic In this section I will briefly discuss the similarities among some of the existing analyses doubling in other dialects, and know of no current treatment of these matters. of accusative clitic doubling, glossing over differences and technical details of It should be noted that the main distinction between dialects that allow doubling of animates only versus the dialects that allow doubling of inanimates and animates does not affect what I have to say about the aspectual properties of accusative clitic doubling. Lack of iteration holds equally well in the animate only dialects. implementation. My goal here is to outline what the issues are, and what the main intuitions behind the proposed solutions are. Analyses of accusative clitic doubling constructions have dealt with three of the problems mentioned above: (i) the thematic role problem: what guarantees the doubled The variation, however, does not stop there. In the dialects that double animates and object and the clitic not to be interpreted as having independent reference from each other; inanimates, as, for example, one of the dialects of River Plate Spanish, the status of a is (ii) the NP restrictions problem: why only some kinds of DPs can be doubled; and (iii) very unclear. It is apparently optional for some speakers, and can also appear where the 120 121 the a problem: what role the preposition a54 plays in doubling constructions (obligatory with the doubled phrase in the dialect we are examining). With respect to problem (iii), the presence of a, the a has been analyzed as a case marker (Jaeggli (1986)), or as a specificity marker (Sportiche (1993)). In the latter case, Existing analyses treat the clitic as (i) an agreement marker per se (Suñer (1988), for the a is taken to appear whenever there is agreement in specificity between the clitic and example); or (ii) treat the relation between the doubled phrase and the clitic as Spec-head the doubled phrase. The constraint on the types of DPs that are acceptable in accusative agreement (Sportiche (1993), for example). The evidence for some kind of agreement clitic doubling (problem (ii)) is understood yet again as the product of the agreement comes from the fact that the clitic agrees in gender, number and arguably in specificity relation between the doubled phrase and the "specific" clitic head (Suñer (1988), with the doubled phrase. Sportiche (1993) and Uriagereka (1995), for example). Crucially, for Sportiche (1993), For Suñer, the thematic role problem is elegantly solved, since for her the clitic is a specificity is the driving force for the movement of the DP to the specifier of the mere agreement marker, and the doubled phrase alone receives the thematic role. For Accusative phrase55 . Thus, "specifics" (definites, demonstratives and possessive Sportiche (1993) and Uriagereka (1995) the doubled phrase occupies a specifier position pronouns) can occupy the head of the doubled DP; non-specific noun phrases (bare and enters a specifier-head agreement relation with the clitic head as in the schematic tree plurals) cannot, as exemplified below. below: (13) a. Loi vi [[al/a este/a su] hombre]i (Cordoba Spanish) (I) iti saw [a [the/this/ his] man]i (12) XP a DP 'I saw the/this/his man.' X' b. Loi vi [[al/ a esto/ a su ] libro]i clitic (I) iti saw [a [the/this/his] books]i For Sportiche the clitic is base-generated as the head of an Accusative phrase to 'I saw the/this/his book.' which the DP complement moves. His analysis assumes that only the Doubled DP has a theta role, the clitic being part of the functional complex. In Torrego's analysis, quoted by Uriagereka, the XP illustrated in (12) is generated as the complement of the verb. In this case the thematic role problem arises again, since the specifier of DP is not a position where an argument gets a role via predication (in possessive constructions in English, for example John's friend does not mean John is a friend). Thus, another source for theta marking of the doubled phrase must be present. 54 It is far from clear how to gloss this element, so I will leave it unglossed. What is clear from the data shown above is that it cannot be taken as an animacy marker as a study of a dialect that only doubles animates might suggest. 122 c. *Losi vi [a hombres/ a libros]i (I) themi saw [a men/a books]i 'I saw men/books.' The hypotheses (i) that there is agreement in specificity between the doubled phrase and the clitic and (ii) that the a is a way to encode agreement in specificity brings up the question of what is in fact the nature of this agreement. In other words, is specificity a 55 Sportiche and Uriagereka are following Diesing's (1992) Mapping Hypothesis, according to which specifics are forced outside the VP. 123 sufficient condition for accusative clitic doubling to be acceptable or is specificity a While the quantifier todos 'all, every' followed by a definite determiner (14a) is necessary but not sufficient condition to license the accusative doubling? Is it the case that acceptable, todo followed by a "bare noun"59 in (14b) is unacceptable60 in clitic doubling all specific DPs are acceptable in accusative clitic doubling?56 constructions: The answers to the questions above are highly dependent on what we take specificity to be. In spite of the variety of proposals to account for specificity effects, two main lines (14) a. Losi vi [a todos los [hombres/ libros]]i (Cordoba Spanish) can be clearly distinguished: (i) specificity is a matter of LF or S-structure scope (Fodor (I) them-MASC.PL saw all-MASC.PL the-MASC.PL men/ books-MASC.PL and Sag, 1982), i.e., specificity is basically a label for wide scope; or (ii) specificity is a 'I saw all the men/books.' property of the semantics of NPs. This latter view is advocated by Enç (1991). For her, b. *Loi vi [a todo hombre/ libro]i specificity is not a matter of scope and is not to be confused with definiteness. Specificity (I) him/it-MASC.SG saw all-MASC.SG man/ book-MASC.SG is a way to link an argument with previously established referents, that of being a subset 'I saw every man/book.'61 of or standing in some recoverable relation to a familiar object. In other words, specificity is linked to partitivity, sometimes spelled out by case.57 It is possible to show that, whether we take scope or familiarity/partitivity as the Since todo and todo los behave differently, we should expect them to behave differently in contexts where specific DPs are allowed or disallowed. relevant characterization of specificity, specificity is too loose a term to characterize the Todo can appear with a bare noun, agreeing in gender with the noun, as exemplified constraints on the kinds of DPs that can appear as doubled objects. Evidence for this in (15a,b). It can also be followed by a definite noun phrase, as exemplified in (16a,b). comes from the distinction between todo and todos los, which share many (but not all) In this case it also agrees in gender and number with the article and the head noun. properties with every and all, respectively.58 59 I am using the bare noun in its naive descriptive meaning, as will become clear below (see chapter 4 for more discussion of bare nouns.) 60 Some speakers do not accept todo plus a bare noun in object position in general, unless it is modified or some adverbial element is added afterwards. However, even if we modify ($14b) the result is still unacceptable with clitic doubling: (i) 56 Also, if the a is a specificity marker, this marking has to be somewhat redundant to the properties of the DP, since the a cannot make a bare plural acceptable, although a "strong" reading (i.e. a generic reading) should be possible with bare plurals. 57 A third way to view specificity is to assume that presupposed material is always specific (as in Diesing 1992, for example). For problems with both the presuppositional view and the partitivity view of specificity, see Abbott (1995). 58 I will, in this section, add data from Brazilian Portuguese to show that todo and todos os behave exactly in the same way. The Brazilian Portuguese data will be relevant when we extend the analysis proposed to other constructions in section 4. 124 *Lo vi a todo hombre en la playa himi saw-1P.SG to every mani at the beach I himi saw every mani at the beach (ii) *Lo vi a todo niño que comía un helado himi saw-1P.SG to every childi that was eating an ice-cream I himi saw every childi that was eating an ice-cream 61 A similar distinction can be found in the behavior of mucho, although the facts are much less clear given the fact that mucho can have strong and weak readings. For some speakers the strong version (as in Milsark 1977) of mucho is acceptable in the doubled construction (see Suñer 1988). This is not true for the Cordoba dialect speakers I consulted. I will assume that mutatis mutandis the acceptable mucho can be given a similar analysis of todos los. However, speakers I consulted were not consistent on their judgments and a complete analysis of mucho and the variation around it will be left as an open question. 125 b. Toda alumna menos yo llegó temprano. (15) a. Siempre resolví todo problema. (Spanish) every-FEM.SG student-FEM.SG except I arrived early Always (I) solved every-MASC.SG problem-MASC.SG 'Every student, except me, arrived early.' 'Always I solved every problem.' c. Todas as alunas menos eu chegaram cedo. b. Eu sempre resolvi todo problema. (Braz. P.) (Brazilian Portuguese) all-FEM.PL the-FEM.PL students-FEM.PL except me arrived early I always solved every-MASC.SG problem-MASC.SG 'All the students, except me, arrived early.' 'I always solved every problem.' d. Toda aluna menos eu chegou cedo. (16) a. Siempre resolví todos los problemas. (Spanish) every-FEM.SG student-FEM.SG except I arrived early Always (I) solved all-MASC.PL the-MASC.PL problems-MASC.PL 'Every student except me arrived early.' 'Always I solved all the problems.' (18) a. Todos los alumnos presentaron un artículo diferente. b. Eu resolvi todos os problemas. (Spanish) (Brazilian Portuguese) all-MASC.PL the-MASC.PL students-MASC.PL presented a different paper I solved all-MASC.PL the-MASC.PL problems 'All the students presented a different article.' 'I solved all the problems.' b. Todo alumno presentó un artículo diferente. Semantically, all, every and todo exhibit various similarities. They are all universal quantifiers, monotone increasing in the second argument and decreasing in the first every-MASC.SG student-MASC.SG presented a different article 'Every student presented a different article.' argument, are proper filters, and allow terminative interpretations of a predicate when in object position (see section 3.2.2). Evidence for treating todo (with or without a c. Todos os alunos apresentaram um artigo diferente. (Braz. P.) determiner) as a universal quantifier comes also from the fact that it can license exceptives all-MASC.PL the-MASC.PL students-MASC.PL presented a different paper (17), same and different (18) and can be modified by almost (19): 'All the students presented a different article.' d. Todo aluno apresentou um artigo diferente. (17) a. Todas las alumnas menos yo llegaron temprano. (Spanish) every-MASC.SG student-MASC.SG presented a different article all-FEM.PL the-FEM.PL students-FEM.PL except I arrived early 'Every student presented a different article.' 'All the students, except me, arrived early.' 126 127 (19) a. Casi todos los alumnos leyeron el programa minimalista. (Spanish) almost all-MASC.PL the-MASC.PL students-MASC.PL read the minimalist program b. *Havia todas as pessoas na praia. (Brazilian Portuguese) was all-FEM.PL the-FEM.PL people-FEM.PL at the beach 'There was all the people at the beach.' 'Almost all the students read the minimalist program.' c. *Habia toda gente en la playa. b. Casi todo alumno leyó el programa minimalista. (Spanish) was every-FEM.PL people-FEM.PL at the beach almost every-MASC.SG student-MASC.SG read the minimalist program 'There was every people at the beach.' 'Almost every student read the minimalist program.' d. *Havia toda gente na praia. c. Quase todos os alunos leram o programa minimalista. (Br. Portuguese) almost all-MASC.PL the-MASC.PL students-MASC.PL read the minimalist (Brazilian Portuguese) was every-FEM.PL people-FEM.PL at the beach 'There was every people at the beach.' program (21) a. Todo hombre es inteligente. 'Almost all the students read the minimalist program.' (Spanish) Todo homem é inteligente. (Brazilian Portuguese) d. Quase todo aluno leu o programa minimalista. every-MASC.SG man-MASC.SG is intelligent almost every-MASC.SG student-MASC.SG read the minimalist program 'Every man is intelligent.' 'Almost every student read the minimalist program.' b. Todos los hombres son inteligentes. Moreover, as universal quantifiers, they are unacceptable in there constructions (20), a well-known test for specificity, since there can be no presupposed set nor can the elements be scoped out. In addition, both are perfectly acceptable as subjects of Todos os homens são inteligentes. (Spanish) (Brazilian Portuguese) All-MASC.PL the-MASC.PL men-MASC.PL are intelligent 'All the men are intelligent.' individual-level predicates, a context that only allows scoped-out-of-the-VP DPs. (See chapter 6.) If specificity were a sufficient condition for doubling, todo and todos los should be both acceptable in accusative clitic doubling constructions. (20) a. *Habia toda la gente en la playa. was all-FEM.SG the-FEM.SG people-FEM.SG at the beach 'There was all the people at the beach.' (Spanish) The contrast in (14) shows that specificity cannot be the sufficient condition for accusative clitic doubling, assuming scope/familiarity-partitivity definitions. Ideally we need to find another construction in which, while todos los is acceptable, todo is unacceptable. In the next section I discuss such a construction. 128 129 But before moving onto that, it should be noted that there is an important semantic distinction between todo and todos los that does not correlate to specificity but that is important to discuss so that not only the similarities but also the differences between todo and todos los become clear. The semantic distinction is akin to the distinction between all and every in English, with the exception of scope facts.62 (23) a. Todo hombre cree que pro es/*son inteligente. (Spanish) every-MASC.SG man-MASC.SG believes that (he) is/*are intelligent b. Todo homem acha que pro é/*são inteligente. (Brazilian Portuguese) every-MASC.SG man-MASC.SG believes that (he) is/*are intelligent All and every in English, Spanish and Portuguese can be distinguished by the fact that 'Every man thinks that pro is intelligent.' all allows group readings and every is distributive, as can be shown by (22c,f). Similarly can todo plus the definite article and todo plus the bare noun be distinguished in Spanish c. Todos los hombres creen que pro son/*es inteligentes. (Spanish) all-MASC.PL the-MASC.PL men-MASC.PL believe that (they) are/is and Brazilian Portuguese (22a,b) and (22d,e): intelligent-PL (22) a. Todos los soldados rodearon la ciudad. (Spanish) All-MASC.PL the-MASC.PL soldiers-MASC.PL surrounded the city 'All the men believe that (they) are/ *is inteligent.' d. Todos os homens acham que pro são/*é inteligentes. (Brazilian Portuguese) b. Todos os soldados cercaram a cidade. (Brazilian Portuguese) all-MASC.PL the-MASC.PL men-MASC.PL believe that (they) are/is All-MASC.PL the-MASC.PL soldiers-MASC.PL surrounded the city intelligent-PL 'All the men believe that (they) are/ *is inteligent.' c. All the soldiers surrounded the city. d. *Todo soldado rodeó la ciudad. (Spanish) every-MASC.SG soldier-MASC.SG surrounded the city The quantifier in (23a,b) binds a singular pronoun. The plural reading is unacceptable. The converse is true in (23c,d). The quantifier followed by a determiner phrase in the plural only binds plural pronouns. A similar fact is exemplified in (24): e. *Todo soldado cercou a cidade. (Brazilian Portuguese) every-MASC.SG soldier-MASC.SG surrounded the city f. *Every soldier surrounded the city. (24) a. Todo profesor salió bebido/*bebidos de la fiesta. Todo professor saiu bêbado/*bêbados da festa. (Spanish) (Brazilian Portuguese) every-MASC.SG professor-MASC.SG left drunk-SING/*drunk-PL from the Todo can also license singular pronouns. First, consider the bound readings of a null party 'Every professor left the party drunk.' bound pronoun: 62 Both todo and todos los behave as all for matters of scope, i.e., when in object position they do not take scope over the subject. See Schmitt 1993 for a discussion and an analysis. 130 131 b. Todos los profesores salieron bebidos/*bebido de la fiesta. (Spanish) b. Recibí de todos los profesores esta carta de recomendación. Todos os professores saíram bêbados/*bêbado da festa. (Br. Portuguese) (Spanish) Eu recebi de todos os professores esta carta de recomendação. every-MASC.PL the-MASC.PL professor-MASC.PL left drunk- (Brazilian Portuguese) MASC.PL/*drunk-MASC.SG from the party 'I received from every the professors this letter of recommendation.' 'All the professors left drunk from the party.' Consider (26) in the situation where a single letter has been received. In (26a) the Examples (22) and (24) are suggestive of the lack of group readings for the quantifier sentence is unacceptable given that a unique letter requires a collective sender. Given that, plus bare noun. Consider the behavior of todo with the verbs that take collective I think we can safely assume that todo does not allow collective readings. In (26b), on predicates: the other hand, the quantifier can receive a collective reading and therefore the sentence is acceptable. The same letter has been sent from all the professors as a group. (25) a. *Todo profesor se reunió en el parque. (Spanish) From the data above, it is clear that there is a distinction between todo and todos los b. *Todo homem se reuniu no parque. (Brazilian Portuguese) but this distinction can be basically summarized as follows: todo plus a bare noun allows *Every man met in the park. distributive readings and todo plus a DP allows both distributive and collective readings. c. El grupo se encontró en el parque. d. O grupo se encontrou no parque. (Spanish) (Brazilian Portuguese) 'The group met in the park.' This difference is certainly not a difference in specificity.63 If specificity is not what is at stake then what is the distinction that is relevant for accusative clitic doubling? 3.2 Here we cannot appeal to the lack of morphological plural to account for the 3.2.1. Accusative Clitic Doubling as a type of Small Clause Properties of Identificational Small Clauses unacceptability of (25a) since (25c) is fully acceptable. With respect to distributivity, The intuition I shall pursue here is a refinement of what has been proposed and then, todo behaves more like every than like all. summarized above, but will have the advantage of accounting for both the DP restrictions The following example clearly indicates distributive readings for the quantifier todo followed by the bare noun: (26) a. *Recibí de todo profesor esta carta de recomendación. *Eu recebi de todo professor esta carta de recomendação.(Br. Portuguese) 'I received from every professor this letter of recommendation.' 132 (Spanish) 63 Suñer (1988) proposes that there is in fact a difference between todo and todo los. In the first case we have quantification over kinds and in the second case we have quantification over sets. The first would be interpreted as non-specific and would be disallowed in there constructions for reasons other than specificity. And in fact Suñer (p.c.) believes that a list reading may be possible in there constructions if there is modification of some sort of todos los. The relevant contrast is as follows: (i) a. Había todos los niños del mundo en ese parque (ok with the so-called list reading) b. *Había todo niño en ese parque (no list reading) It is not clear to me however how a distinction between quantification over sets vs. kinds accounts for the inability of todo to appear in there constructions. If it doesn't explain this, then there is no reason to reject an analysis that todo and todos los are specific, as we have shown above) and something else accounts for the DP restrictions in clitic doubling. Since I will be making a positive claim about what this property is, it seems to me that it is preferable to leave the standard accounts of specificity untouched. 133 and the aspectual properties of accusative clitic doubling, without making use of a notion that definite determiners in certain languages can be pleonastic in the sense that they do such as agreement in specificity. not provide reference for a noun phrase (i.e. they do not bind the <R> position of The question is how to allow the so-called doubled phrase and the clitic to be nominals in Higginbotham's (1985) sense) and therefore have no semantic role.66 They interpreted as "the same" and, at the same time, reject the idea that the clitic is an have only a syntactic role, being licensed by the agreement with the NP. The NP gets its agreement marker per se. Treating the clitic as an agreement marker is undesirable <R> element by being bound by another element (see Vergnaud and Zubizarreta 1992, because such agreement must be stipulated to have both the NP restrictions and the and chapter 4). Thus, although the determiner shows agreement with the noun (unlike aspectual properties encoded in it, in stark contrast with other agreement markers in it/there which seem to be invariable), it is still treated as a pleonastic element by Vergnaud Spanish. Once we refuse the agreement hypothesis, we have to deal with the thematic and Zubizarreta. role problem, since now we have two DPs.64 There is yet another class of constructions in which the pronominal element in subject I would like to argue here that the clitic is the non-thematic subject of a small clause position could also be taken as pleonastic. It is this type of construction that I will (i.e. a pleonastic element), in the same way that certain pronominal elements can be concentrate on. This construction, exemplified in (28) for English and (29) and (30) for thematic or non-thematic, as the examples below (from Rothstein 1995) show: Spanish and Portuguese, respectively, has been grouped by Higgins (1985) under the heading of Identificational sentences and are typically used for teaching the names of (27) a. It is obvious that John will arrive late. people. b. It is obvious. (28) a. This is John. In the first case it is not an argument but in the second case it is definitely an b. It is John. argument, for while we can ask what is obvious? we cannot ask what is obvious that c. These are the Smiths. John will arrive late. Pleonastics are the canonical case of pronominal non-arguments in d. This is John and Mary. case positions, and it is standardly assumed that pleonastics can only appear in subject d. It is the Windsors. position.65 Besides the canonical cases of pleonastic elements there and it, Vergnaud and Zubizarreta (1992), in their discussion of inalienable possession in French, have argued 64 There is another problem with the agreement hypothesis to which I will come back when we take a look at conjoined DPs in clitic doubling constructions, which is not akin to regular subject verb specifierhead agreement. 65 The standard assumption that expletives always appear in subject position has been challenged by Pullum and Postal (1988). However, Rothstein (1995) argues convincingly that the cases discussed by Pullum and Postal can be divided in two groups and that the pleonastic elements are in fact subjects of small clauses. In the other cases, the it element is argumental. 134 66 For example, in inalienable constructions a singular definite article does not imply a semantic singular. In the following example the doctor didn't examine a single stomach. The reference of estomac is given by leur and not by the definite singular determiner. (i) Le docteur leurs a examiné l'estomac. the doctor them examined the stomach 'The doctor examined their stomachs.' 135 (29) a. Es Juan. (Spanish) pro is Juan (31) Who is this/that/it? a. This/that/it is John. b. This/that/it is the Mayor. b. Son los Vilas. c. *John is this/that/it. pro are the Vilas d. *The Mayor is this/that/it.67 c. Es Pedro y María. e. Es el Prefecto (Spanish) pro is Pedro and Maria f. É o prefeito (30) a. É o João. (Brazilian Portuguese) (Brazilian Portuguese) prois the Mayor pro is the João g. *El prefecto es (pro) (Spanish) h. *O prefeito é (pro) b. São os Silva. (Brazilian Portuguese) the Mayor is (pro) pro are the Silvas Moreover, we cannot naturally question the pronoun in the cases above: who is John c. É o Pedro e a Maria. pro is the Pedro and the Maria cannot be answered by it or that. Higgins notes yet another property of the pronominal element in these constructions. The pronominal element has what he calls "common In (31) we have a pronominal element in subject position followed by a proper name gender". To illustrate this, consider (32a) and (32b). or a definite description. As in regular expletive constructions, the pronoun cannot appear after the copula, without inducing a radically different interpretation. This is illustrated in (31c,d) for English and in (31e-h) for Spanish and Portuguese : 67 The only way to accept this in the complement of BE is by adding one in English. In this case, however, we have a different construction (an equative) because the demonstrative followed by one can appear as the complement of BE with the same meaning. A reviewer for Probus points out that one could argue that the demonstrative always appears higher in predicative constructions due to some sort of scale of referentiality, as proposed by Heggie (1988). However, the cases from Brazilian Portuguese and Spanish, illustrated above, rule out this hypothesis, because all accounts of referentiality that I am aware of would place null pronouns as being less referential than overt NPs (see e.g. Burzio 1989, 1991). 136 137 (32) a. That woman is the Mayor of Cambridge. b. That woman is Mayor of Cambridge. c. That is the Mayor of Cambridge. from the Mayor of Cambridge. It is there to satisfy the predication relation as in other constructions with pleonastic elements. The contrast between (32c) and (33a) again illustrates the distinction between the so- d. *That is Mayor of Cambridge. called common gender pronoun and the regular pronoun. (33a) can appear in the reversed e. It is the Mayor of Cambridge. order, as illustrated in (33b). The ability to be reversed means that the personal pronoun f. *It is Mayor of Cambridge. is not playing the role of an expletive, since expletives cannot appear in object position.69 In (32a) we can establish and equivalence between two DPs in which they either have (33) a. He is John. the same extension or we can establish an intensional equivalence by assigning a property b. John is him. to that woman. In (32b) we have only the latter possibility. In other words, in (32a) the c. Mayor Barry is the Mayor for life. definite description after BE68 can be interpreted either as an argument of BE, in which d. The Mayor for life is Mayor Barry. case it is not a predicate, or as a predicate that selects for an argument, in this case the In fact we should consider sentences in (33a,b) as comparable to sentences in (33c,d) subject that woman. In (32b) the indefinite can only be interpreted as a predicate. Thus where identity of reference is being established. (32c) differs from (33) in that in (32c) that woman functions as its subject. we are not establishing identity of reference; we are not asserting existence. Instead, we Higgins has noted that, if that woman is substituted for that or it in (32a), the result is are identifying a referent. The pleonastic pronominal element it has no reference, and that acceptable, as we can see in (32c). However, if we substitute that woman for that in has only the locative information that is in the demonstrative part of the DP (which can (32b), the sentence is unacceptable as shown in (32d), since Mayor of Cambridge is a vary).70 predicate that requires a [+human] subject. That does not satisfy this requirement, and In sum, the pronominal element in subject position of the sentence it's John bears consequently the sentence is unacceptable. case but no thematic role and is related to an element that it c-commands.71 The schematic Higgins' observation can be reinterpreted in the following way: that/it can appear in structure for such identificational small clauses is given in (34). (32c) because in this case it need not to be an argument of the Mayor of Cambridge. In fact, I would like to argue that the Mayor of Cambridge is an argument of BE and the pronominal element is an expletive in that it does not receive a theta role from the 69 Treating this as an expletive will also account for the inability of it to appear in other object positions as the following example due to Greg Carlson: (i) #I'd like to introduce you to this. This is John. predicate is the Mayor of Cambridge. Thus it will not have to obey selectional restrictions 70 What I have in mind here is an analysis of demonstratives constituted of two parts: a location and a nominal element. Thus the real expletive part of the demonstrative is the nominal element and not the locative part. This is John means basically here it is John. The pleonastic part is the it. Semantic support for an analysis of demonstratives comes from Bennett (1978) and for syntactic evidence see chapter 4. 68 Since much of the discussion that follows is valid for various languages including Brazilian Portuguese, French and Spanish, I will use the notation BE as a cover term for the copula in this construction. 71 A thorough discussion of the ways that have been proposed for interpreting pleonastics (e.g. Chomsky 1993) is beyond the scope of this chapter. The technical problems of implementation are irrelevant for the discussion at this point. Overtly, the DP with the demonstrative has checked its strong 138 139 The DPs that are banned in accusative clitic doubling are also banned in (34) a. This is the doctor/John. identificational small clauses. (36) shows that non D-determiners can only give rise to b. predicational interpretations: AgrP this Agr' (36) a. *These are friends. DP b. *This is every friend.73 the doctor/John c. *These are all friends.74 Now consider the DPs that can appear in the complement position of BE in So far three generalizations can be brought up from the discussion: first, the same identificational sentences. As we can see in (35), the same noun phrases that can appear determiner restrictions that hold for accusative clitic doubling hold also for the in identificational small clauses are exactly those that can appear in accusative clitic doubling: definites, proper names and pronouns.72 I will call the determiners that can complement of BE in identificational small clauses. Second, in both constructions a pronominal element is related to a full DP that is c-commanded by the pronoun. Third, in appear in the complement position of BE in identificational sentences D-determiners. both constructions the pronominal element (if agreement is possible) will agree in phifeatures but cannot be considered to obey semantic selection restrictions from the full DP (35) a. This is the mayor. b. These are all the prisoners. that it c-commands.75 In this sense, it is like the expletive determiners discussed by c. This is John. Vergnaud and Zubizarreta. Given these similarities between accusative clitic doubling DP d. This is him. restrictions and identificational small clause restrictions, my proposal is that accusative clitic doubling is an instance of an identificational small clause, and that the clitic, just like 73 Norbert Hornstein (p.c.) has pointed out that the following sentence is perfectly acceptable: (i) morphological phi-features by Spell-Out. I will come back to the issue of expletive replacement when I deal with the aspectual properties of accusative clitic doubling and case. 72 Margarita Suñer (p.c.) pointed out that examples such as (i) undermine the claim that indefinites cannot appear in identificational small clauses: (i) This is a cat. I would like to argue here that such cases are not identificational. This can be shown by the contrast in (ii). (ii) a. This is the Mayor of Cambridge. b. #This is a person. The oddity of (iib) as opposed to (i) shows that we have here a restriction on [+human], which shows that a person is imposing a selectional restriction on the subject. In other words, the demonstrative in (i) and (iib) is argumental. 140 This is every prisoner. In fact, I believe this is correct but only with a group reading, which we have seen to be impossible for todo: (i) *É todo prisioneiro (It) is every prisoner 74 Alan Munn (p.c.) notes that this sentence is grammatical with a predicative interpretation with all modifying the predicate similar to They all left. I am concerned here with the identificational reading in which all would be modifying just friend. This is perhaps more clearly shown with the example *This is all milk, in which the predicative reading is much harder to obtain. 75 The agreement facts at first sight suggest that to call the demonstrative or the clitic an expletive is not correct, since expletives tend not to show agreement (Rothstein 1995 and observation from Probus reviewers) At the end of this section I will show, however, that the agreement exhibited by the clitic is not semantic but syntactically determined and identical to the agreement found in there constructions in English. 141 the demonstrative, is not receiving a thematic role from the predicate a+DP. Consequently (38) a. Jean est le médicin. whatever explains the need for a D-determiner to be in the complement of BE in the 'Jean is the doctor.' (French) identificational sentences explains the need for a D-determiner in the doubled phrase.76 My assumption is that being a D-determiner is a semantic property that can only be visible if the D-determiner has checked case in an Agr Projection. The need for a Ddeterminer to have its case checked is expected if it is to be interpreted as an argument of BE. It also encounters empirical support from the following contrast, adapted from Higginbotham (1987): b. Jean est médicin. John is doctor 'John is a doctor' c. C'est Jean/le médicin. This is Jean/ the doctor d. *C'est médicin. (37) a. *I consider [that [the man] b. I consider [that [to be[ the man]] This is doctor 'This is a doctor' When a copula is not present, a definite description in a complement clause such as (37a) must have a predicate reading rather than an argument interpretation. That is then Being a D-determiner is a necessary condition to appear as an argument in BE the argument of the property the man, and we get an odd reading that roughly identificational constructions and in accusative clitic doubling constructions. Obviously, corresponds to that thing is the man. In (37b) the copula is present. In this case, the D- however, it cannot be the case that any language that has D-determiners will have determiner can check case in the specifier of AgrO of BE. Thus the man can be interpreted accusative clitic doubling. A solution to this problem is suggested by Torrego who ties as the argument of BE, and that can be interpreted as a pleonastic element. the ability to license accusative clitic doubling and the ability to license null complements. Given the English data above, it is clear that both English and Spanish have Ddeterminers, i.e. determiners that can appear in the identificational small clauses discussed above. Since English does not have clitics, the question of why English does not have 3 . 2 . 2 On the Notion of Strong Determiner 3 . 2 . 2 . 1 Torrego's observation accusative clitic doubling doesn't arise. However, there are languages such as French Torrego (in a manuscript cited by Uriagereka 1995) notes that the determiners that can with clitics that do not license accusative clitic doubling and yet have D-determiners, as appear in the doubled phrase in accusative clitic doubling constructions can license an illustrated below: empty pronominal element identified by the head D and bound in the discourse. The examples in (39) show that the ability of the determiners to appear in accusative clitic doubling correlates to its ability to license empty categories in the nominal complement. 76 I would like to suggest that one of the properties of D-determiners is that they are never interpreted as having distributive readings, only group readings, but a more thorough semantic analysis of these constructions is in order to actually define D-determiners. 142 143 (39) a. lo de Paris (Argentinian Spanish) doubling. What seems to be necessary is that the determiner is both a D-determiner and has the property of being able to license null anaphora, i.e., be strong in Torrego's sense. the-MASC.SG (one) of Paris It is very intuitive that English would have weak definite determiners and Spanish 'the one from Paris' would have strong determiners, if we take strength to relate to the presence of overt b. todos los de Paris feature distinctions. Spanish determiners inflect for gender and number (lo, la, los, las) all-MASC.PL the-MASC.PL (ones) of Paris and English the is invariable. The difference between French and Italian, in one hand, and 'all the ones from Paris ' Spanish, in the other hand, is, however, more subtle, since in both cases there seems to be number and gender markings, although only Spanish allows for null anaphora and c. *todo de Paris accusative clitic doubling. every-MASC.SG of Paris Null anaphora, as I see it, is akin to pro-drop, an overt syntax phenomenon. The *'every one from Paris' general principle that seems to rule pro-drop phenomena in various languages is the For Torrego, the ability to license empty categories and doubled phrases is tied to the following: if Agr has enough features, it can license null subjects; if it has not enough strength of the determiner. Strong determiners can license empty categories and features, it cannot. Suppose licensing of null complements in DPs mirrors the pro-drop specifiers. The definite determiners in French and English, for example, which do not phenomenon.78 Then it must be the case that D carries enough features to license a null license null pronominal elements, would be weak determiners.77 element in its specifier. What to "carry enough features" means is that the D can be proxy for the null complement identified in the discourse. (40) a. lo de Paris (Spanish) One way of thinking of the distinction between French and Spanish determiners is to b. *le de Paris (French) relate it to Delfitto and Schroten's (1991) idea that noun phrases are not created equally c. *the from Paris (English) with respect to morphological information (see also Szabolcsi 1994). They propose that 'the one from Paris' French, in particular, does not have morphological number on the noun, and, therefore, the noun never has to raise to NumP to check features and therefore cannot move to D, The ability to license null complements is a necessary but not a sufficient condition to which would mean that D never has all the NP features. In the same way, we can think license accusative clitic doubling, as there are determiners that license null complements that not all determiners raise to the highest position in the DP (I will assume it to be an (muchos 'many', una 'one', for example) and yet are unacceptable in accusative clitic AgrDP projection above the D, as an AgrCP projection has been proposed for the CP, as in Roberts 1993), since the determiner does not always carries all the necessary 77 This notion is related to strength of morphological features and should not be confused with strong and weak determiners in the semantic sense as used by e.g. Milsark (1977), De Hoop (1992) and references there. In Milsark's sense both the definite determiner in French and Spanish are strong determiners. 144 78 I am not saying that every language that has pro-drop of subjects in tensed clauses will have null complement anaphora. Null complement anaphora depends on the properties of the licensing head D+Agr or T+Agr and it is clear that Ds and Ts are not necessarily alike in a given language. 145 morphological information to force movement to that position. Although an analysis αP (42) a. along these lines would take us too far afield, my suggestion here is that Ds in Spanish DP* and Portuguese (unlike Ds in Italian or French) can reach an AgrD projection overtly, clitic b. α' a DP DP^ DP^ which I will assume is what allows licensing of a pro in the specifier of AgrDP or deletion D D' D* NP under recoverability. A schematic tree is given below: The DP marked DP^ in (42) corresponds to the doubled phrase and the DP marked (41) Null complement anaphora structure DP* corresponds to the clitic, which, being a pleonastic element has to be licensed as the AgrDP non-thematic subject of a predicate. The a element corresponds to the BE in the pro Agr identificational small clause and has DP^ as its complement. I will call the small clause Agr+D DP D projection αP for expository purposes. Being an argument of BE/a , DP^ needs to check ... case features. DP^ will check case by incorporating into a, as we will see below. The Whatever this property is, it allows the determiner to act as a pro-nominal and check case for the DP. In French and Italian this is only possible in the case of clitics. In cases in which the DP can act as proxy for the DP, movement of the D as a head is possible to check for case features for the whole DP. clitic will in turn agree in number and gender features with the D element that has incorporated into the a. The first advantage of the structure in (43a) is that once we assume that the doubled phrase is the complement of the a and not its specifier, we can tie the inability of todo+NP What determines whether D0 or DP raises for case checking? My proposal is that a strong determiner has two options: it can either raise as a DP to check case or it can raise as a head, since both movements can arguably have the same cost. Weak determiners, on to appear in accusative clitic doubling constructions with its inability to appear in the identificational small clause with BE. The structure for todo+NP and todo+DP is given below: the other hand, can only raise as XPs. In section 3 I will discuss the consequences of (43) a. AgrP b. AgrP incorporating the determiner, as opposed to raising of the DP to AgrO for the aspectual toda Agr' todas Agr' interpretation of accusative clitic doubling. Agr 3.2.4. The Structure for Accusative Clitic Doubling DP D all-FEM.SG Agr amiga friend-FEM.SG la amiga all-FEM.PL the-FEM.PL friends-FEM.PL The structure for accusative clitic doubling I am proposing is given in (42a) and Torrego's proposal is repeated in (b): 146 DP 147 Two points are important for our discussion. First, as Suñer (1988) argues (see also (44) a. Juan y María son médicos. (Spanish) Schmitt 1993), todo is a specifier, not a head.79 The second point that is important in this b. O João e a Maria são médicos. discussion is that the head of the DP can be filled with a D-determiner or not. The lack of c. John and Mary are doctors. (Brazilian Portuguese) a D-determiner in todo +NP is responsible for its absence in accusative clitic doubling constructions. Also, the lack of an overt definite determiner will block group readings. (45) a. Ellos son la médica y el médico que se graduaron el año pasado. (Spanish) The structure proposed in (42a) differs from the structure in (42b), in two ways: first b. Eles são a médica e o médico que se formaram no ano passado. we now have a full small clause headed by an a head. Thus the relation between the two (Br. Port.) DPs is one of a pleonastic subject and a complement of an a phrase and not between a They-MASC.PL are the doctor-FEM.SG and the doctor-MASC.SG, that head and its specifier.80 graduated last year Second, in this structure, the clitic is an XP and not the head of the projection (as in c. They are the female and the male doctor that graduated last year. Sportiche (1993)), since treating the clitic as the head of the projection and the doubled DP as its specifier is incompatible with my proposal that the clitic is a pleonastic element. Agreement patterns in conjoined noun phrases that have been clitic-doubled provide empirical evidence against treating the clitic and the double as being in a Spec/Head relation directly as Torrego's structure requires. A conjoined noun phrase in the canonical subject position triggers plural agreement on There is another pattern of agreement found cross-linguistically, which descriptively takes place under government rather than the Spec-head relation: this type of agreement has been widely discussed in the literature (see McCloskey (1986) for Irish; Bahlouhl and Harbert (1992) for Arabic; Munn (1994) for English and a minimalist account of these facts). One of the most distinguishing factors of agreement under government is that it the verb and masculine plural agreement on the adjective, as illustrated in (44) and (45) gives rise to first conjunct agreement with conjoined DPs. Thus, in [V [DP1 and DP2]] for Portuguese, Spanish and English. This is taken to be the canonical specifier-head order, agreement will be with DP1. In [[DP1 and DP2] V], agreement cannot be with the first conjunct. A clear case of agreement under government arises in English there agreement relation. constructions as in (46) from Munn 1993: (46) a. There is a man and a woman in the garden. b. *There are a man and a woman in the garden. 79 Todo unlike every in English, can appear with a determiner following it. Every is in complementary distribution with the definite determiner the (hence the unacceptability of *every the ). Also by maintaining the distinction between todo and every as a distinction between a modifier and a head we can may capture scope differences between the two languages. Todo, unlike every, in object position cannot take wide scope over the subject, nor does it allow pair list answers in questions of the type who did everyone see, despite having the same properties in terms of distributivity (see Schmitt 1993). 80 The a being the head of the identificational small clause will allow distinctions between dialects. As we mentioned before, there are dialects in which only animate DPs can be doubled. In such cases we might assume that a has animacy features present that need to be checked. 148 c. There are two men and a woman in the garden. d. *A man and a woman is in the garden. Here, the verb agrees with the post-verbal subject, and first conjunct agreement is obligatory. If the subject is pre-verbal, first conjunct agreement is impossible. 149 The same pattern arises in the identificational small clauses discussed in 2.1 above, in specifier head relation as in (49a) and the agreement with the first conjunct obtains by incorporation of the head of the first conjunct onto the verb, as in (49b):81 English, Spanish and Brazilian Portuguese, as exemplified below: (49) a. (47) a. This is John and Mary. AgrS el nene y la nena b. These are the Windsors and the Smiths. AgrS Agr' son c. *These are John and Mary. b. Agr' Di+ es XP muy lindos-masc.pl (48) a. Es Juan y María. DP (Spanish) É o João e a Maria. XP DP (Brazilian Portuguese) Di el '(pro) is J. and M.' NP y nene X' BP DP la nena c. El nene y la nena son muy lindos. b. *Son Juan y María. (Spanish) The boy and the girl are very pretty. *São o João e a Maria. (Brazilian Portuguese) d. Es el nene y la nena. '(pro) are J. and M.' (It) is the boy and the girl. c. Son los Clintons. (Cordoba Spanish) (Spanish) São os Clintons. (Brazilian Portuguese) '(pro) are the Clintons' Spec-head configurations only allow agreement with both conjuncts. Agreement under government on the other hand allows agreement with the first conjunct. In identificational small clauses what we find is agreement under government, in which If the identificational small clause has a conjoined noun phrase as its complement, agreement with the first conjunct is possible, although not always obligatory.82 If first conjunct agreement is obligatory, as (47) and (48) show. Note that both the demonstrative and the copula show agreement with the first conjunct and not agreement with both conjuncts. Munn (1992, 1993) proposes that conjoined structures are adjunction structures in which a Boolean phrase (BP) headed by a conjunction is adjoined to the first conjunct which is the head of the whole construction. According to Munn (1994), a way to deal first conjunct agreement in Minimalist terms, is to assume that in (44) and (45), the agreement with the verb obtains in a 150 81 As pointed out by one of the speakers I consulted, nenas and nenes is more common in Cordoba than niñas and niños. This is the reason I used them throughout this chapter. This lexical distinction does not alter the agreement facts discussed here. 82 What is important here is that agreement under government can trigger first conjunct agreement but Spec-head agreement cannot. Judgments on agreement under government are not always clear cut. Prescriptive rules seem to get in the way and somehow plural agreement (i.e. agreement with both conjuncts) in government configurations is also accepted by some speakers, even when the first conjunct is singular. This phenomenon is attested in various languages, Arabic being one of them (see Munn 1993). One of the speakers I consulted for the Spanish identificational small clauses and for the clitic doubling coordinated objects told me (of first conjunct agreement) "it makes no sense, but this is fine." The others accepted the facts in ($50) without question, although they do not accept first conjunct agreement when the coordinated DP is a subject of a clause in its canonical specifier of AgrS position. Variation exists with respect to whether the a has to be repeated in the second conjunct. Some speakers accept (i) but others only accept (ii). I will leave this matter open (for a discussion of how the second conjunct would get case in (i) see Munn (1993)). (i) a. Lo vi al muchacho y la chica 151 accusative clitic doubling has the same structure as identificational small clauses, as we c. Las vi a las nenas y a los nenes. proposed above, the prediction is that the clitic would allow agreement with the first (I) them-FEM.PL saw a the-FEM.PL girl-FEM.PL and a the-MASC.PL boys- conjunct. This follows from the fact that the head of the DP that is incorporated into the a, MASC .PL is the D of the first conjunct. The clitic in the specifier of Agr will have to agree with it. In d. Lo vi al profesor y las alumnas. fact this is exactly what we find, as illustrated in (50): (I) him-MASC.SG saw a the-MASC.SG teacher.MASC.SG and a theFEM.PL student-FEM.PL (50) a. Lo vi al muchacho y a la chica. (Cordoba Spanish) (I) him-MASC.SG saw a the-MASC.SG boy-MASC.SG and a the-FEM.SG The structure is given in (51): girl-FEM.SG αP (51) b. Los vi a los hombres y a las mujeres. α' DP* (I) them-MASC.PL saw a the-MASC.PL and a the-FEM.PL women-FEM.PL clitic lai+a DP^ DP^ lai b. La toqué a la sonata y el adagio (ii) a. Lo vi al muchacho y a la chica b. La toqué a la sonata y a el adagio BP NP αP B Note that the structure in which the doubled DP is in the specifier (as proposed by Torrego) would only allow agreement with both conjuncts. First conjunct agreement in Margarita Suñer (p.c.) does not accept ($50b). She also allows accusative clitic doubling with indefinite specifics and without the a, which my speakers from Cordoba did not accept. As pointed out by Everett 1992, there is more than one dialect that allows doubling. The ability to drop the a is probably related to the ability to get the plural clitic in accusative clitic doubling. To account for this dialectal variation, one possibility is that in Suñer's dialect los is adjoined to the DP al muchacho y a la chica in ($50), much as subjects are adjoined to VP in some analyses (see Koopman and Sportiche 1991). The structure would then be as follows: (iii) [DP los [DP al muchacho y a la chica]] Assuming this to be correct, we can get the clitic to agree with both conjuncts in the same way together agrees with both conjuncts in the following constructions, since it requires a plural, in spite of the fact that the verb agrees with the first conjunct in (iva). Similar facts hold for juntos 'together' in Brazilian Portuguese, as illustrated in (v). Example (vb) is acceptable with an anaphoric reading, i.e., the man is with somebody else previously specified in the discourse, in which case no agreement shows up (cf. vc). (iv) a. There is a man and a woman together in the room. b. *There is a man together in the room. c. There are children together in the room. (v) a. Estava um homem e uma mulher juntos na sala. was-SG a-MASC.SG man and a-FEM.SG mulher.FEM.SG together-MASC.PL in the room. b. *Estava uma homem junto na sala. was--SG a-MASC.SG man together-MASC.SG in the room. c. Estava uma mulher junto na sala was-SG a-FEM.SG woman together-MASC.SG in the room. 152 clitic doubling thus provides independent motivation for both the small clause analysis and the incorporation of the doubled DP's determiner into a. 3 . 2 . 4 Summary The proposal that accusative clitic doubling is an instance of an identificational structure allows us to account for three classical problems: (i) the thematic role problem: the clitic is an expletive kind of element of an identificational small clause, thus it does not have a thematic role assigned by the predicate. 153 (ii) the 'a' problem: BE in identificational small clauses is a case assigner. The a in accusative clitic doubling structures plays the same role as a case assigner to its 3 . 3 . The Aspectual Properties of Accusative Clitic Doubling 3 . 3 . 1 Terminativity and lack of iteration in accusative clitic doubling complement, licensing the D-determiner. (For details of implementation see As we have seen, terminative aspect is the result of a combination of certain verbal and 3.2.1.) certain nominal features mediated by a theta role in a checking configuration. Depending (iii) the NP-restrictions problem: identificational small clauses require D-determiners on the verbal feature and on the nominal feature, the interpretation will be that of a as their complements. Thus we can associate the determiner constraints to the durative or a terminative predicate. The proposal made in chapter 1 and supported in identificational properties of the small clause. chapter 2 by the data from Finnish and Slavic is that AgrO is the locus for interpretation of terminativity of the VP. Finally, the lack of accusative clitic doubling in French/Italian follows from the fact that the determiners in French/Italian are not strong in Torrego's sense and therefore cannot move as a head for case checking purposes.83 3 . 3 . 1 . 1 Case and Terminativity Accusative clitic doubling constructions with [+ADD TO] verbs are terminative. Since Moreover by treating the doubled phrase as the complement of the small clause and these constructions force the presence of D-determiners for independent reasons, a not as its specifier we have accounted for the agreement facts. The agreement facts are terminative reading may in principle be derived provided we can arrive at the perfectly compatible with the idea that the clitic is a pleonastic element that displays configuration proposed in (55) at LF. syntactic and not semantic agreement, just as Vergnaud and Zubizarreta's (1992) analysis of expletive determiners. In order to do this we first need to show how the DPs in the identificational small clause check case and agreement features according to Chomsky's (1993) feature The account proposed implies, though, that regular transitive verbs can take small clause complements, instead of simple regular DPs. However this seems to be more and checking mechanism. Let's see first how subjects and objects check case, so that the feature checking mechanism interacting with the economy considerations become clear. more a necessary move, given the complexity of possessive (Szabolcsi 1983; 1994; also chapter 5), partitive (Uriagereka 1993; also chapter 5) and relative clause constructions (see chapter 4 and Kayne 1994). Moreover, we will see that this is not only necessary but a positive result of this proposal, since it will allow us to account for the aspectual (57) AgrP spec1 Agr' Agr VP subj properties of these constructions. V' V obj The chain V in (57) has in its minimal domain subj and obj. If V adjoins to Agr, 83 Note that this is not to say that there is no phonological cliticization of the determiner into a preposition. I will assume that those cases are prosodic facts due to mere adjacency. 154 forming the chain (V, Agr), then the minimal domain of this complex head is {spec1, 155 subj, obj}. Both the subject and the object, in this case, are equidistant from spec1 given can move overtly to the specifier of AgrO to check case, as in instances of Exceptional the definition of equidistance given in (58). If the subject raises to the spec1, then the Case Marking. From there the clitic moves to whatever position it has to move to attain its object cannot check its case in a higher projection because it would have to cross two PF position.84 filled specifiers to do so, violating the economy consideration of shortest move. Thus DP^ cannot move anywhere as a phrase, since any movement of it—even if a moves only the object can raise to spec1 and the subject can raise higher, as the verb moves up, to V—will cross two specifiers: the clitic and the subject. But the D of DP^ is a strong extending the domain. In both cases the relation between the DP and the V is mediated by determiner in Torrego's sense and has the option of moving as a head to check for the Agr. Following Chomsky (1993), in nominative-accusative languages, nominative is the case of the whole projection. Incorporation of the D to a is then obligatory, because a can active case, i.e., is always available, and accusative will only be checked if nominative is check the case of the Di. checked. The proposal here is that a in accusative clitic doubling is a case assigner, following Jaeggli (1986), if it can at LF be in a checking position. Here I assume with Baker (1988) (58) If α and β are in the same minimal domain, they are equidistant from γ. In particular two targets of movement are equidistant if they are in the that there are prepositions that depend on the verb to be able to assign case and that there are others that can check the case of their complement independently. In minimalist terms same minimal domain. (Chomsky (1993)) we could think that prepositions that can check for their case independently would have Now we can consider the accusative clitic doubling case checking mechanism. their own Agr projection. The prepositions that are parasitic on the verb would lack their own Agr projection. In order for its complement to have its case checked at LF such a (59) AgrO preposition has to incorporate onto the verb and move with it to the closest Agr. The LF Agr movement of the preposition is triggered by the need to check features of its own. VP subj Assuming that the a in accusative clitic doubling is one of the parasitic prepositions V' V that have to adjoin to a verb in order to become a case checking element, the derivation at αP DP* LF proceeds as follows: D raises to a. This is possible because D is strong.85 The feature α' clitic a complex[D+a ]raises to the verb and from there the complex head [V+[D+a ]] moves to DP^ Di AgrO0 in order to check features of a, and case on the incorporated D. NP Structurally, the complement of the verb is the small clause αP headed by a. A potential problem arises here. Verb movement in Spanish is a pre-Spell-Out operation. However, what I am proposing is that the complex D+a will move to the verb, According to the case filter, there are two DPs in need of case: the doubled phrase (DP^) needs case as does the clitic DP*. The verb moves to AgrO before spell-out and from 84 The actual position that the clitic ends up after AgrO is irrelevant for the aspectual interpretation of the aspect of the VP. there to a higher position; the pleonastic clitic, which also needs to check its case features, 85 I am assuming that the PF order of the complex head is dependent on the morphological rules of the language and does not necessarily mirror the syntactic ordering. 156 157 but the verb is not there anymore. It is in a higher position because movement has been The adverb quase sempre, as illustrated in (60b), intervenes between considero and triggered by strong features. At the base position we have a copy of the moved verb and infeliz, the predicate that supposedly incorporates at LF. at AgrO we have an Agr+V copy. In Chomsky 1994, movement to a copy is stipulated as an illicit operation. Various other cases seem to pose the same problem and force manipulation of the copies. (60c) illustrates a case where the anaphor can be bound by John. Assuming that However, there are various cases that seem to suggest that such a movement should anaphors raise to some IP position (as in Chomsky 1993 following Lebeaux 1988), this be allowed. For example: in complex predicate constructions, as illustrated in (60a,b) and raising can only happens at LF from the intermediate trace in the specifier of the C that anaphor binding, illustrated in (60c) heads the embedded clause. This is a clear case in which part of a copy is moved at LF. Thus, whatever solution we may end up finding for these cases will account for the (60) a. Jean ne faisait pas manger à Pierre. Jean (not) made not eat to Pierre 'Jean didn't make Pierre eat.' b. Eu considero quase sempre a Maria infeliz. facts in the accusative clitic doubling constructions. One possibility is that all we do is to move the features directly to the checking position (as in Chomsky 1995, although there Agr projections are taken to be not necessary). It should be noted that he movement of the [a+D] to Agr is also necessary in order for I consider almost always the Maria unhappy the pleonastic element to be replaced, since pleonastic elements are uninterpretable at the 'I almost always consider Maria unhappy.' interface. The LF derivation is illustrated below in (61a) and (61b); (61b) shows the LF after c. Which picture of himselfi did Johni say that Bill liked? the elimination of the pleonastic element: The case of causatives is illustrated in (60a): if causative constructions are analyzed as verb incorporation at LF, we have to allow for adjunction of the lower verb to the copy of the causative verb which has raised overtly to a T or to the AgrS position in languages with strong verb movement. The example (60a) above, from French, illustrates that the incorporation cannot be a pre-Spell-Out phenomenon, since pas 'not' can intervene between the causative verb and the main verb. Also, in consider small clauses (illustrated in (60b)), the same problem arises: if we assume an analysis in which the predicate undergoes complex predicate formation and therefore incorporates onto consider at LF (see Baker 1988), we have to allow incorporation of the predicate to the copy of consider, since in the Brazilian Portuguese case, for example, consider has moved before Spell-Out to a higher checking position. 158 159 (61) a. AgrOP DP* (62) a. Toqué la sonata hasta las 12, de hecho la toqué 10 veces. (Cordoba Spanish) Agr' [[[Di+a]V]Agr] (I) played the sonata until 12, in fact (I) played it 10 times VP subj 'I played the sonata until 12, in fact I played it 10 times.' V' [[Di+a ] V] αP b. #Lai toqué [a la sonata]i hasta las 12, de hecho la toqué 10 veces. DP* α' clitic Di+a (I) iti played [a the sonata]i until 12, in fact (I) played it 10 times DP^ 'I played the sonata until 12, in fact I played it 10 times.' Di NP c. #Loi choqué [a lo coche]i repetidamente. b. (Cordoba Spanish) AgrOP (I) iti hit [a the car]i repeatedly Agr' 'I iti hit the cari repeatedly.' [[[Di+a]V]Agr] VP By LF, [[[D+a]+V]+Agr] are in a specifier head relation with a copy of the DP*. The Let me first make the distinction between terminativity, iteration and quantificational case of D is checked and D allows the elimination of DP*. I assume here that the of events clear. There is another piece of information that the DP complement gives to the specifier-head relation with its associate (the D of DP^) is enough for that. event: it determines whether we can partition the event into discrete subevents or not. Milk does not have subparts discretely identifiable; in the same way, drink milk does not 3.3.1.2 Distinguishing Iteration from Quantification over Events have individual subparts. On the other hand, in drink three cups of milk we can have three subevents discretely identifiable. The count/mass distinction decides whether we In chapter 1 I proposed that terminativity obtains if an eventive verb and a DP argument whose quantity is specified are at AgrO by the time the VP aspect is interpreted. If we know how much of the object there is, we know when the event culminates. Now, it is easy to see, with respect to the accusative clitic doubling facts, that a terminative reading have subevents based on the individual discrete parts of the DP complement or not. Moreover, only VPs that have discrete individual parts can be iterated, because iteration is the ability to repeat subevents, and that depends on the ability to identify discrete subparts of the object. can be obtained, since a D-determiner is in fact at AgrO by the time aspect is calculated. Thus the terminativity of the predicate is derived smoothly. The question now is how to derive the lack of iteration, repeated below: (63) a. John hit the ball for an hour. b. #The coal miner hit rock for an hour. 160 161 Thus while we can say (63a) with a repetitive reading, it is impossible to say (63b) with a repetitive reading. The same is true for intransitive verbs with no object. If a an hour. In the (c) example, every will force a multiple event reading. The same facts can be shown for todo in Brazilian Portuguese: measure is added, then an iterative reading is possible, as in (64a,b). (67) a. Eu toquei sonatas por uma hora. (64) a. #John ran repeatedly. I played sonatas for an hour b. John ran a mile repeatedly. b. Eu toquei a sonata por uma hora. (Iteration) I played the sonata for an hour It seems then that terminative readings will allow repetition of the VP in the general case. Durative readings, however, can allow or disallow repetitive readings depending on c. Eu toquei toda sonata por uma hora. the head noun. If a bare plural is present at AgrO by the time aspect is calculated, a (Multiple events) I played every sonata for an hour repetitive reading is possible, but a mass noun will block iteration unless the mass is preceded by a determiner, since mass nouns preceded by definite determiners are made 3.3.2. The Proposal count (see Higginbotham 1994 and references there). From this discussion we can say that the configuration in (68) will allow iteration, (65) a. *John played sonatas in five minutes. b. John played sonatas repeatedly. provided we have an eventive verb and an object whose quantity is specified. (68) AgrOP the sonata Agr' A question arises at this point: why should we say that in (63a) and (65b) we do not play+Agr VP have a multiple event reading as opposed to this partition of the event in subevents? The evidence for this will come from the contrast between the following sentences: Thus, in Spanish, iteration is possible with a regular object whose quantity is (66) a. John played sonatas for an hour. b. John played the sonata for an hour. specified, a clitic or a bare plural, as expected. (69) a. Toqué la sonata frecuentemente.86 (Cordoba Spanish) c. John played every sonata for an hour. (I) played the sonata frequently. In (66a) and (66b) the event must have lasted one hour, but in (66c) each sonata was played for an hour and the whole event will last as long as there are sonatas. In the (a) and (b) cases the event is subdivided into subevents and the sum of their durations will be 162 86 One should be careful here, since adverbials like frequently allow two readings: a VP reading and a sentential reading. I am interested here in the VP readings. The sentential reading is an habitual reading that is irrelevant for what we are discussing here, since it involves some sort of quantification over events. 163 b. La toqué frecuentemente. be mediated by the Agr features.87 External boundaries can force iteration (adverbs like (I) played it frequently for x time, for example), but they are not themselves part of the subevents. Terminativity, on the other hand, just cares about the end point of the event. It need not to see the c. Toqué sonatas frecuentemente. (Cordoba Spanish) (I) played sonatas frequently. The question is what is wrong with the accusative clitic doubling structure that disallows iteration, in spite of forcing terminative readings. subparts, just the maximum amount. Some evidence for the correlation between determiner incorporation and lack of iteration comes from Galician. Although Galician disallows clitic doubling, it allows overt determiner incorporation (see Uriagereka 1988). When the determiner incorporates to the verb, the result is the same as in accusative clitic doubling, as the pragmatic unacceptability of (72) exemplifies below: (70) AgrOP DP* Agr' (72) since you were eighteen years old, you crashed-the cars [[[Di+a]V]Agr] VP First it is important to notice that the DP* is an expletive element at LF and therefore does not receive any interpretation. In other words, the configuration at LF is in fact the *?Deude que tiñas dezeoito anos, (ti) chocachelos coches. Here the result of the determiner incorporation is pragmatically odd, since a single crashing that lasted 10 or more years is quite implausible. In sum, my argument here is that when the determiner that carries the quantity following: information of the internal argument is not in the specifier of AgrO, iteration is impossible, given that iteration is related to the ability to partition event and that is (71) AgrOP dependent on certain agreement features being in a certain relation with the features of the Agr' [[[Di+a]V]Agr] VP complement. Before we move on to the two other constructions I want to analyze, a few observations must be made about the data, since the compositional nature of aspect and In the structure above, the D element that carries the information that the cardinality of the different positions adverbs can appear can easily lead to confusion. the object is specified is not in the specifier of AgrO. My assumption here will be that being so deeply embedded, the D is too far from the Agr head. The determiner cannot be used to combine with the verb and partition the event into subevents that are discretely identifiable and, therefore, allow iteration. Notice that it is crucial for this analysis that 87 Independent evidence for such a proposal comes from the Finnish cases. As discussed in chapter 2, partitive objects never raise to AgrO. Iteration is in those cases blocked, (Heinämäki 1984) spite of the cardinality of the object be specified. iteration is partition of the event into subevents, and my assumption here is that it has to 164 165 3.3.3. A few observations about the data8 8 c. La tocaba a la sonata hasta las 12 todos los días pues tenía que practicar para el concierto. So far I have used eventive verbs in the past perfective tense and I have been careful in it (I) played a the sonata until 12 every day because I had to practice to the not using verbs like beat in English which have inherent iterated meanings, in order not to concert make things confusing, since they will allow a reading that is similar to the one I ruled out 'I used to play the sonata until 12 every day because I had to practice for the concert.' above. In the following I will show some cases where habitual readings are possible and should be kept apart from the phenomenon just described, since they involve either an IP level aspect or a CP level aspect. In these cases some sort of quantification over situations is involved and quantification over the VP, i.e., quantification over the event is possible. I will in turn analyze (i) the present and imperfective past tenses, which allow habitual readings and (ii) plural subjects which allow the partitioning of situations.89 However, there is clear evidence for distinguishing habitual readings from VP iteration, since it is possible to have both iteration and habitual readings at the same time in simple sentences. Thus in English we can say: (74) a. John played the sonata repeatedly for an hour every day. 3 . 3 . 3 . 1 Habitual readings b. John use to play the sonata for five minutes every day. c. 'John plays the sonata for five minutes every day.' The present tense and the past imperfective allow a habitual reading of the event that provides quantification over the whole situation. This gives the impression that in the following sentences VP iteration is possible: (73) a. Lo toco/tocaba a Juan frecuentemente. it (I) touch/used to touch a Juan frequently 'I touch/used to touch Juan frequently.' b. Lo choca al coche frecuentemente (porque no ve nada). In Spanish both the present and the imperfective reading allow habitual readings and in this case the whole situation is taken to repeat a number of times so that it can be considered habitual. This is not to say, however, that the VP has been iterated, i.e., subdivided into subparts. 3 . 3 . 3 . 2 Plural subjects Another case that gives rise to confusion is when we have a plural in subject position: it (he) hits a the car frequently (because (he) not see anything) 'He hits the car frequently (because he doesn't see anything) (75) La tocaron a la sonata por 15 minutos. it (they) played the sonata for 15 minutes 'They played the sonata for 15 minutes.' 88 I thank Margarita Suñer for most of the facts discussed in this section, which made me realize that I needed to clarify the distinction between a VP aspect and an upper level aspect involving the subject and other aspectual markings on the verb. 89 I am using event and situation as to indicate the VP aspect level and the IP or CP aspect level, respectively. Nothing deep hinge on these terms. 166 The examples below, from English, show the role played by the subject in the aspect calculus: 167 configuration that blocks an iterative reading is met. Note that the position the clitic is in is (76) Three men crossed the street in two minutes. (77) Three men hit the ball in two minutes. irrelevant for aspectual interpretation. As it stops at AgrO to check its case it leaves a copy. However, not being argumental, it will have to be eliminated because it cannot receive an interpretation at the interface. Here both events described are terminative and can be modified by in two minutes, This proposal allows us to show (contra Kayne 1994) that accusative clitic doubling for example. However, because the subject is three men, it can be true that they each and clitic left dislocation are not instances of the same phenomenon in Spanish. In clitic crossed the street in 2 minutes, or that they crossed it together. The same is true for the left dislocation constructions in Spanish, iteration is not blocked as shown in (78). If both second example above with hit. What is important is that a plural subject will allow accusative clitic doubling and clitic left dislocation were the same phenomenon with partitions of a situation but not partitions of the VP. Thus it is not true that each man identical LFs, we should not expect this distinction. However, such a distinction can be crosses one part of the street or that the ball was partially hit by each of the men. A bare accounted for if we assume that the DP in clitic left dislocation is base generated in a preplural subject will allow an undetermined amount of crossings of the street or hittings of sentential position. In such case we do not find the right configuration for iteration to be the ball (see chapter 6). blocked (since the clitic in this case is the only argument of the verb, sitting in the The conclusion we can draw from the facts discussed above is that in order to "see" specifier of AgrO and there is no D-incorporation at that point.) lack of iteration of the VP it is the best to use a singular subject, a tense/aspect marker that disallows generic/habitual readings and finally be sure that the adverbial is modifying the (78) a. A los autos, los choqué frecuentemente.90 VP aspect. Moreover we have to choose verbs that do care about the internal properties of (Cordoba Spanish) The cars, I hit them frequently the complement. For example a verb like push (or read) does not, since both push a cart b. A la sonata, la toqué frecuentemente. for an hour and push carts for an hour are acceptable. The sonata, I played it frequently So far I have shown how the case of both DPs is checked at AgrOP: the doubled DP checks its case via head movement and the clitic checks its case by movement to Spec of (Cordoba Spanish) 3.4 Participial Absolutes and have+agreeing participles AgrO. The terminative interpretation of the VP will be derived by the fact that the Di is sitting in the head of AgrO at LF and the lack of iteration is related to the fact that the In the previous sections I have proposed that accusative clitic doubling constructions are complement is not in a close relation with the V+Agr complex. an instance of an identificational small clause, with a pleonastic DP as its subject and a DP headed by a D-determiner as its complement. Such a proposal accounts not only for the 3.3.4. Distinguishing clitic left dislocation and accusative clitic restrictions on the determiners that can appear in the doubled position but also forces the doubling The proposal above makes one very clear prediction: not all phenomena that looks like accusative clitic doubling will block iteration. It will depend on whether at LF the 168 90 There is an enormous amount of variation in the use of the preposition a in these constructions, which is outside the scope of this work. 169 incorporation of the D-determiner into the verb. The incorporation has the consequence of (80) Busca-*a/ ela na geladeira always blocking iterative readings of the predicate VP. (Spoken Brazilian Portuguese) get her-ACC/ she-NOM in the fridge In this section I will show that participial absolutes in Spanish and Brazilian 'Get it in the fridge' Portuguese and have+agreeing participles in Brazilian Portuguese should be treated similarly, since they display the same properties of accusative clitic doubling. 3.4.1. Participial Absolutes and Determiner Restrictions Given that I will be dealing with Brazilian Portuguese determiners, it is necessary first This section is divided as follows: first I outline the basic properties of participial to show that the D-determiners in Brazilian Portuguese pattern with Spanish Dabsolutes and show, based on Brazilian Portuguese facts, that the determiner restrictions determiners and not with French or English D-determiners. The data that supports this on the DP internal arguments of participial absolutes cannot be explained by claiming claim is given in (79): these DPs are subjects. Then I move on to discuss have +participle constructions, which (79) a. o de Paris (Brazilian Portuguese) have two forms: one with agreement in the participle and one without agreement. Determiner restrictions on the internal argument will only hold in the agreeing 'the (one) from Paris' construction. b. todos os de Paris 'all the (ones) from Paris' 3 . 4 . 1 . 1 Basic Facts Participial absolutes share with other participial absolutes the following basic c. *todo de Paris properties: *'all (ones) from Paris' (79) shows that the same set of determiners that license a null element in Spanish also license a null element in Brazilian Portuguese. As we have seen before, only strong determiners (in Torrego's sense) can license null elements and move as heads. Thus the assumption I will be making is that the determiners in Brazilian Portuguese are strong: they can license null anaphora and can also move as heads to check the case of the DP. Brazilian Portuguese does not have accusative clitic doubling for independent (81) Properties of participial absolute constructions a. absolute constructions are not lexically selected by the predicate of the clause they modify; b. they have a lexical DP which corresponds thematically to the internal argument of the predicate, coreferent or not with a NP in the main clause; c. they adopt an adverbial interpretation which does not correlate with any overt complementizer. reasons, the most important being the fact that the third person accusative clitics do not (adapted from Hernanz 1991) exist anymore in the spoken language, being substituted by the nominative strong pronoun as shown in example (80). 170 171 Participial absolutes are formed by a past participle and an overt DP, with which the participle agrees in gender and number. Examples from Spanish and Brazilian Portuguese are given below: impossible with ergative verbs, it follows that they lack AgrS which would be required to check case of the external argument. Finally, they cannot appear with complementizers or negation. (See Hernanz (1991) for a similar treatment of the same facts in Spanish.) These facts motivate the structure in (84), which I will assume from here on. (82) a. Convocada la reunion, María se levantó. (Spanish) called-FEM.SG the-FEM.SG meeting-FEM.SG, Maria stood up (84) AgrOP Agr' 'With the meeting called, , Maria stood up.' Agr b. Resolvidos os/ estes problemas, poderemos sair. (Brazilian Portuguese) VP V DP solved-MASC.PL the/these-MASC.PL problems-MASC.PL, (we) will be 3 . 4 . 1 . 2 Determiner Restrictions able to leave 'With the problems solved, we can leave' The DP that agrees with the participle corresponds to the internal argument. Unaccusatives (83a) and transitive verbs can appear in participial absolutes but Participial absolutes in Spanish and Brazilian Portuguese91 exhibit the same constraints on DP complements observed for accusative clitic doubling in Argentinean Spanish: bare plurals and todo+NP are unacceptable (85b,d), while definites and todo+DP (85a,c) are perfectly acceptable. intransitive verbs cannot as in (83b). (83) a. Chegados os convidados, podemos servir a sopa. (Brazilian Portuguese) arrived-MASC.PL the-MASC.PL guests-MASC.PL, (we) can serve the soup Once the guests (have) arrived, we can serve the soup. b. *Telefonado, ficaríamos mais tranqüilos. called-MASC.SG, (we) would be more relaxed Once he (had) called, we would be more relaxed. (85) a. Comprada a passagem, não deve haver problema. (Brazilian Portuguese) bought-FEM.SG the-FEM.SG ticket-FEM.SG, not should have problem 'With the ticket bought, there should be no problem.' b. *Resolvidos problemas, podemos viajar. solved-MASC.PL problems-MASC.PL, (we) can travel. 'With problems solved, we can travel.' There is no evidence to postulate a full clausal structure for participial absolutes in Brazilian Portuguese and Spanish. First, since the constructions lack tense (the tense of the participial absolute depends on the tense of the main clause (see Ambar, (1988) for example)) there is no grounds to posit a TP projection. Second, because they are 172 91 The same facts hold for Argentinian Spanish. Of course, speakers who accept indefinites in accusative clitic doubling also seem to accept indefinites in both constructions that I am about to discuss, especially with a partitive reading. 173 internal argument as moved from its internal position to a derived subject position, or at c . Compradas todas as passagens, não deve haver problema. bought-FEM.PL all-FEM.PL the-FEM.PL tickets-FEM.PL, not should have least to a position where incorporation to the verb is impossible. For her, bare plurals in problem Italian are only allowed when incorporation to the verb is possible. Vinet (1989) also notes what she calls a "specificity effect" in absolutes in French. 'With all the tickets bought, there should be no problem.' According to her, only specific nouns phrases can appear in participial absolutes, a fact d. *Comprada toda passagem, não deve haver problema. she associates with the fact that these noun phrases are subjects of small clauses. bought-FEM.SG every-FEM.SG ticket-FEM.SG, not should have problem All things being equal, Belletti's analysis makes the prediction that bare plurals should 'With every ticket bought, there should be no problem.' be allowed in participial absolute constructions, if a language allows bare plurals in any Although the accusative clitic doubling constraints have been widely noted in the literature, the same constraints in participial absolutes have passed almost unnoticed. One of the two studies that mention constraints on bare plurals in participial absolutes is Belletti's (1990) study of Italian. Belletti notes that bare plurals in participial absolute subject position, since presumably in those languages bare plurals do not need to incorporate into the verb. Here, however, is where the Brazilian Portuguese data become important. In Brazilian Portuguese, unlike Spanish and Italian, bare plurals are freely allowed in both subject and object position, as illustrated in (87). Even so, they cannot appear in participial absolutes, as we have seen in (85b). constructions are unacceptable in Italian: (86) a. Arrivata Maria, la festa cominciò. arrived M., the party started (Italian) (p. 83 ex. (122)) (Brazilian Portuguese) children fed fish-PL in the lake 'Children fed fish in the lake.' 'Once Maria (had) arrived, the party started.' b. Separatistas vão ser presos daqui por diante. b. *Cadute pietre, la strada è rimasta bloccata. fell rocks the road was blocked (87) a. Crianças alimentaram peixes no lago. (p. 142 fn. 18) Separatists will be arrested from now on. 'With the fallen rocks, the road was blocked.' If Brazilian Portuguese allows bare plurals in subject position and nonetheless does c. *Letti romanzi, uno si sente meglio. not allow bare plurals in participial absolutes, subjecthood cannot be the straightforward read novels, one feels better explanation for the lack of bare plurals (and the other ruled out noun phrases) in "Once novels (are) read, one feels better.' participial absolutes (at least in Brazilian Portuguese).92 Given that bare plurals cannot appear in subject position in Italian (nor in Spanish), Belletti takes the constraint on bare plurals (86b,c) as extra support for an analysis of the 174 92 There are other reasons, as Margarita Suñer (p.c.) points out, to believe that participial absolutes in Italian should be kept apart from the participial absolutes in Spanish and Portuguese. First, Italian allows clitics with the participial absolutes; second, it also allow intransitive verbs with pro subjects. Finally the have+past participle in Italian can be modified by adverbials like yesterday but not in Spanish or Portuguese. (See Schmitt 1996 for more discussion.) 175 A final argument against subjecthood as the explanation for lack of bare plurals in f. Las salas y el cuarto fueron arreglados these constructions comes from DP conjunction again where the pattern of agreement the-FEM.PL living room-FEM.PL and the-MASC.SG bedroom-MASC.SG under government shows up again (contrast (88a) and (88c) for Brazilian Portuguese and are tidied up-MASC.PL (88d) and (88f) for Spanish:93 I would like to propose that the constraints on the types of DPs in Spanish and (88) a. Arrumadas as salas e o quarto, ... (Brazilian Portuguese) tidied up-FEM.PL the-FEM.PL living-room-FEM.PL and the-MASC.SG bedroom-MASC.SG Brazilian Portuguese participial absolutes are to be treated identically to the constraints on the types of DPs in accusative clitic doubling. To do this, I will turn to the have+participle construction in Brazilian Portuguese, which, unlike the absolute constructions, need not show agreement. b. *Arrumados as salas e o quarto tidied up-MASC.PL the-FEM.PL living-room and the-MASC.SG bedroom- 3.4.2. Have+agreeing Participles and Determiner Restrictions MASC .SG In (89) the participle does not agree with the DP object. Here bare plurals and todo+NP c. As salas e o quarto estão arrumados are acceptable (the examples are from Brazilian Portuguese) in object position. the-FEM.PL living room-FEM.PL and the bedroom-MASC.SG are tidied upMASC .PL (89) a. O João tinha resolvido problemas. d. Arregladas las salas y el cuarto,... (Spanish) (Brazilian Portuguese) João had solved problems. tidied up-FEM.PL the-FEM.PL living-room-FEM.PL and the-MASC.SG b. O João tinha negociado toda terra que comprara no ano anterior. bedroom-MASC.SG João had negotiated all land that (he) had bought the year before. e. Arreglados las salas y el cuarto,... tidied up-MASC.PL the-FEM.PL living-room-FEM.PL and the-MASC.SG c. O João tinha feito todas as camas. João had made all the beds. bedroom-MASC.SG When the participle agrees in gender and number with the internal argument, however, the same constraints on the DPs that can appear in accusative clitic doubling and 93 Another common argument for the subjecthood is the case of the DP complement: nominative. Nominative case follows from the theory that says that accusative is only possible if nominative is assigned. Given that the participial absolute has only one Agr and only one case checker the V, the only case available is nominative. In fact that the DP complement is a nominative follows more from the theory than from empirical evidence. Only pronouns show case markings. Accusative clitics are independently ruled out on the assumption that clitics have to adjoin to some higher Agr position, which doesn't exist. With respect to nominative pronouns, first and second person pronouns are unacceptable and third person nominative pronouns are only marginally acceptable. I believe that the reason may be related to the discourse function of these clauses and/or prosodic facts. 176 participial absolutes arise again. Bare plurals and todo+NP are impossible: 177 (90) a. O João tinha resolvidos os /todos os problemas. (Brazilian Portuguese) the João had solved-MASC.PL the-MASC.PL/ all-MASC.PL the-MASC.PL proposing here. It follows then that we should find that iterative readings are also blocked in these constructions. problems-MASC.PL 3 . 4 . 2 . 1 Lack of Iteration in Absolutes and have+agreeing participles 'João had solved the/all the/ problems.' Studies on participial absolutes in Spanish (Bosque (1990), Hernanz (1991)) and b. *O João tinha resolvidos problemas. European Portuguese (Ambar (1988)) mention that these clauses are terminative. The the João had solved-MASC.PL problems-MASC.PL example in (92) shows the same aspectual behavior for Brazilian Portuguese: 'João had solved problems.' c. *O João tinha feita toda cama. João had done every-FEM.SG bed-FEM.SG (92) Lidos os artigos em uma hora, sobra tempo para uma caminhada. With the papers read in an hour, there is still time for a walk. (Brazilian Portuguese) 'João had made every bed.' The difference between (89) and (90) is the same as the difference between the doubled and non-doubled objects in Spanish. Similarly, the determiner restrictions only hold on the agreeing form of the participle, which is the form that appears in participial absolutes. In addition, we again find the same agreement patterns in conjoined structures. (91) a. Maria tinha arrumadas as salas e o quarto Maria had tidied up-FEM.PL the-FEM.PL living-room and the-MASC.SG What has not been noted is that participial absolutes are not only terminative but also do not allow iterative readings, just as we have seen for accusative clitic doubling. Example (93), also from Brazilian Portuguese, shows that iteration is not possible with the adverbials for an hour or frequently. (93) a. #Lido o artigo por uma hora, chegamos a seguinte conclusão: era horrível. read the articles for an hour, (we) arrived at the following conclusion 'With the articles read for an hour, (we) arrived at the following bedroom conclusion: it was horrible' b. *Maria tinha arrumados as salas e o quarto94 Maria had tidied up-MASC.PL the-FEM.PL living-room and the-MASC.SG b. *Aumentado o imposto sobre cigarros freqüentemente, a população parou de fumar. bedroom increased the tax over cigarettes frequently, the population stopped of It is unlikely that these similarities are merely coincidental. Rather, they provide evidence for the unified analysis involving identificational small clauses that I am smoking 'With the tax over cigarettes frequently increased, the population stopped smoking.' 94 Because these constructions are much more productive in the written language, some people might be inclined to believe that the agreement in (b) is acceptable. However, if we are careful we can identify a clear pause between the participle and the DP complement, when agreement is with both conjuncts. 178 (Brazilian Portuguese) 179 The same facts hold with have+agreeing participles. First, consider (94a), where there is no agreement. The adverbial for x hours or until x time allows an iterative reading 3.4.3. Implementing the identificational small clause for agreeing participles of the predicate, since play the sonata is a terminative VP. When the participle agrees Both agreeing participles and have+agreeing participles share the same properties of with the object, however, the adverb for x hours produces an unacceptable result, as accusative clitic doubling that I have argued are tied to the identificational small clause illustrated in (94b). In other words, there is an incompatibility between the agreement in structure of the internal argument. In order to implement the same analysis we will have the participle and partition of the VP into subevents. Agreement blocks the iteration of the to explain the lack of the a in both participial absolutes and have+agreeing participles, as event. exemplified below: (94) a. O João tinha tocado a sonata por uma hora, quando Maria chegou. (96) a. *Convocada a la reunion, Maria se levantó. The João had played the sonata for an hour, when Maria arrived called a the meeting, Maria stood up 'João had played the sonata for an hour, when Maria arrived.' 'With the meeting called, Maria stood up.' b. *O João tinha tocada a sonata por uma hora, quando Maria chegou. b. *O João tinha abertos aos portões. The João had played-FEM.SG the-FEM.SG sonata-FEM.SG for an hour The João had opened-MASC.PL a the-MASC.PL gate-MASC.PL when Maria arrived 'João had opened the gates.' 'João had played the sonata for an hour, when Maria arrived.' (Brazilian Portuguese) The structure for accusative clitic doubling given above is repeated here in (97): The same contrast is illustrated below with the adverb repeatedly; while the adverb (97) repeatedly is perfectly acceptable if the participle does not show agreement, it becomes unacceptable if the participle agrees with the object. αP clitic (95) a. A Maria tinha aberto a porta repetidamente. α' DP* a DP^ (Brazilian Portuguese) D NP 'Maria had opened the door frequently.' b. *A Maria tinha aberta a porta repetidamente. the Maria had opened-FEM.SG the-FEM.SG door-FEM.SG frequently For the complement of participial absolutes and have+agreeing participles, the structure I propose is the following: 'Maria had opened the door frequently.' 180 181 αP (98) a. DP* is a null DP because the head has the necessary φ-features to identify it. In other words, a α' null expletive will correlate with the presence of agreement features. Overt expletives will Di +AGR DP^ correlate with the lack of agreement features. Now we are able to answer why the Di NP preposition a never occurs with participial absolutes. If the preposition is present, the DP as camas in the specifier of the αP has to be overt; if it is overt, it needs to move to AgrO and from b. as camas there it needs to move as a clitic to guarantee that its strong PF features are checked. This the beds movement is impossible in participial absolutes because there is no site for the clitic to In both accusative clitic doubling and the agreeing participles, the D-determiner, being strong, raises to the functional head a, and from there the complex raises to check for case. The only modification we need to make, when we compare the structure above with attach to. (Recall there is no TP or AgrSP.) Moreover, if the preposition a is present, agreement will not show up in the participle since the morphological features will not be there.97 The derivation in the participial absolutes is then the following: the one given for the accusative clitic doubling construction, is to allow for the head of the small clause to be the preposition a in accusative clitic doubling or morphological features that can check for case once incorporated into the verb at LF. Again the intuition here is that the distinction between an a head of accusative clitic doubling and a morphological head should mirror the pro-drop phenomenon in languages where it is restricted to positions with overt agreement. In languages like Irish (McCloskey 1985), for example, if the agreement is not overt, a lexical subject must be construction where the DP appears before the regular participle is unacceptable (iib) and the irregular form is acceptable as illustrated in (iic). Examples (ib) and (iic) are probably adjectival. (i) present. If the agreement is overt, then the subject cannot to be lexical. In the case of the accusative clitic doubling, the specifier of the a phrase has to be filled with a lexical expletive, since a has no φ-features.95 In the case of the participial absolutes and the have+agreeing participles96, the expletive in the specifier of the a phrase 95 This is not to say that every a is identical. There is an a that precedes every [+human] complement; the a that appears with demoted subjects of causative constructions and there is an a that appears in complex predicates with subjects appearing to the right. The latter is the most likely to create confusion with have+agreeing participles. A full discussion of these constructions is beyond the scope of this thesis. 96 The constructions I am discussing here should not be confused with another construction with ter in which the DP appears before the agreeing participle. In these constructions the DP does not need to be headed by a D-determiner as (ia) shows. In (ia) the bare plural is perfectly acceptable. Moreover, there are restrictions on what participles can appear after the DP they agree with. An example is given in (ii). If the DP appears after the noun the regular participle of aprontar is perfectly acceptable (iia). However, the 182 a. Maria tinha as roupas passadas Maria had the-FEM.PL clothes-FEM.PL ironed-FEM.PL 'Maria had the clothes ironed' b. Maria tinha roupas passadas Maria had clothes-FEM.PL ironed-FEM.PL 'Maria had ironed clothes' (ii) a. Maria tinha aprontadas as roupas Maria had ready-FEM.PL the-FEM.PL clothes-FEM.PL 'Maria had ready the clothes' b. *Maria tinha as roupas aprontadas Maria had the-FEM.PL clothes-FEM.PL ready-FEM.PL C. Maria tinha as roupas prontas Maria had the-FEM.PL clothes-FEM.PL ready-FEM.PL 'Maria had the clothes ready' 97 Margarita Suñer suggests that agreement and a can actually co-occur, which is not predicted by my analysis provided the a and the agreement marker share the same position. Again, it is very clear that in her dialect the a is not necessarily playing the same role. If the objects are inanimate there can be no a . 183 (99) (101) AgrOP VP DP* Agr' ter Agr+[[Di +AGR]feitas]j VP DP* αP [[Di +AGR]feitas]j DP* AgrO Agr+[D+Agri +feitas] α' Di +AGR Agr' VP αP [D+AGRi]+feitas DP^ DP* Di NP as camas α' D+Agri DP D b. feitas as camas made-FEM.PL the-FEM.PL beds-FEM.PL In both agreeing participles iteration is blocked because the D is incorporated into the (with) the beds made verb at LF. The same structure accounts for have+agreeing participles. Let's first examine the non-agreeing participles. For the participle constructions with have without agreement, 3 . 5 . Conclusion In this chapter I have argued that the internal argument in accusative clitic doubling, participial absolutes and have+agreeing participles is in fact an instance of an the structure is roughly the following: identificational small clause. By assuming that identificational small clauses have (100) VP pleonastic DPs as subjects (overt in the first construction and empty in the other two ter AgrO constructions), it is possible to account for the fact that only one of the elements in Agr' accusative clitic doubling in fact receives a theta role from the verb. The analysis also VP accounts for the determiner restrictions in all three constructions. Moreover, the structure feito DP proposed for the internal argument forces a strong determiner in the complement of as camas identificational small clause, since only strong determiners can move as heads to check have made the-FEM.PL beds-FEM.PL for case. The structure derived from the D-incorporation into the verb at AgrO blocks The participle raises to AgrO to check its features and the object raises to spec of AgrO to check case. A terminative reading will then obtain and iteration is possible. The structure for the agreeing participles with have is given below: iterative readings of the VP predicates. The analysis of accusative clitic doubling also shows that the internal structure of direct object of the verb is crucial to define what will be at AgrO when aspect is calculated. In the next chapter we will see more evidence of the internal structure of the complement affecting the aspectual interpretations. 184 185 Chapter 4 Determiner transparency: relatives, adjectives, demonstratives* (2) a. Chomsky wrote that book #for years / in 6 days b. Chomsky wrote those books for years/ in 2 years Introduction Here, while (2a) only allows a terminative reading stretched for a period of two years, I have claimed in the previous three chapters that eventive verbs require a specified quantity of their object to be in AgrO in order for a VP to be interpreted as terminative. Definite determiners will always give a specified quantity of the object, and so should always force terminative readings. In this chapter I will explore a set of apparent counterexamples to the claim that definite determiners always induce terminative readings. In these cases, I will argue that the determiner of the object is "transparent" for aspect calculation, and develop a general theory of determiner transparency based on observations first made for relative clauses by Vergnaud (1985 and Kuroda (1968 that (2b) allows a durative reading that has roughly the reading that Chomsky wrote books of that sort for years. This suggests that there is a difference between demonstratives and definite determiners, which seems related to the fact that we can point to a single book and say something like I hate those books (e.g. books that are not well-bound and whose pages get lost). Notice as well, that we cannot say I hate that book, with the same reading. The example in (3) shows, however, that it is not the demonstrative per se that is allowing the durative reading in (2b) since a definite determiner with a relative clause shows the same effect:98 extends to certain kinds of adjectives and to demonstratives. The analysis will account for the aspectual facts, and, in addition, provide a way to deal with Antecedent Contained (3) a. Chomsky wrote the book that revolutionized the field #for years / in 6 days Deletion (ACD). b. Chomsky wrote the books that revolutionized the field for years/ in 2 years Consider first the data in (1): (1) a. Chomsky wrote the book #for years / in 6 days b. Chomsky wrote the books #for years/ in 2 years While in (3a) only a terminative reading is possible (i.e. a reading in which Chomsky takes years and years writing that book), a durative reading is available in (3b), where In (1a), the complement of the verb is the book and a terminative reading is the only Chomsky wrote books of the sort that revolutionized the field for years. This durative possible reading, as the unacceptability of the adverbial for years shows. This is in reading is possible in spite of the fact that there is a definite determiner which in (1b) accordance with the idea that an eventive verb and an internal argument with its quantity forced a terminative reading. specified will produce a terminative reading. Also, in (1b), the complement is the books and a terminative reading is the only reading available. Now consider (2): In this chapter I want to pursue the idea that the determiner in (2) an (3) is "transparent". This amounts to saying that (3b) can behave like a bare plural for the purposes of calculating aspect, thus providing independent evidence for the claim, * Parts of this chapter were conceived and presented in Schmitt (1995b) and Schmitt (1995c), although the ideas there have undergone significant revision (and improvement). 186 9 8 Unless otherwise stated I am interested here in the readings where the adverbial is a modifier of the verb phrase of the matrix clause and not as a modifier of the VP in the relative. 187 defended on entirely different grounds in Kayne (1994, that the and books in (3b) do not Higginbotham (1985), can license a definite determiner that is transparent with respect to form a constituent. There are two main questions to be answered: the noun. This will allow us to unify, in section 3, the cases of (i) relative clauses, (ii) wrong adjectives and (iii) demonstratives, in their ability to license transparent (i) under what conditions can the determiner be made transparent? (ii) why are both terminative and durative readings available with plurals (2b, 3b) but unavailable with count singulars (2a, 3a)? determiners. In section 4 I show how the behavior of "bare noun" complements in Brazilian Portuguese argues against a purely semantic account of the aspectual differences, and provides evidence for the determiner transparency analysis. The chapter is divided as follows. In section 1 I make a general proposal for the Finally, in section 5 I will return to relative clauses and argue that determiner relative clauses' ability to license durative readings based on Vergnaud (1985, Kuroda transparency is exactly what is needed to circumvent the infinite regress problem in (1968 and more recently Kayne (199499. Then, in section 2, I move on to a discussion of Antecedent Contained Deletion structures. a class of adjectives that also allows durative readings, as exemplified in (4) and contrast them with those in (5) which force terminative readings: Throughout this chapter I will deal mainly with English and Brazilian Portuguese, which in most cases behave alike. Brazilian Portuguese will be particularly useful when dealing with bare noun complements and with adjectives, since adjectives in Brazilian (4) a. Mary wrote the wrong book for #years/ in two years b. Mary wrote the wrong books for years/ in two years (5) a. Mary wrote the difficult book #for years/ in two years Portuguese, as in many other Romance languages, show agreement with the head noun they modify and appear to the right of the noun. 4.1 Relative Clauses b. Mary wrote the difficult books #for years/ in two years100 The question to be answered here is what accounts for the durative readings in (6c), and A third general question thus arises: what do relative clauses, demonstratives and the lack of durative readings in (6a) and (6b): adjectives like wrong (but not like difficult) have in common that allows them to license a (6) a. Chomsky wrote the books #for years/ in two years transparent determiner? b. Chomsky wrote the book that revolutionized the field #for years / in 6 The discussion of what distinguishes (4b) and (5b) will motivate the proposal I will make that only elements that have an independent <R> position, in the sense of days c. Chomsky wrote the books that revolutionized the field for years/ in 2 years 99 I will not be assuming Kayne's theory in the argumentation that follows. Thus, the arguments presented here are entirely independent of the Linear Correspondence Axiom (LCA) that drives Kayne's analysis. Should the LCA be correct, most of what I will say here will be consistent with it. I differ from Kayne's exposition of the relative clauses in certain important details which are empirically well justified, but I will not argue explicitly against his more programmatic proposal. 100 A durative reading is acceptable if we use difficult contrastively: the difficult books were Mary's job and the easy ones were John's. 188 I will pursue the idea that the determiner is not a constituent with book or books in the (b) and (c) examples above and that only book or books raise to AgrO to check case, 189 deriving terminative and durative readings, respectively. The terminative reading in (6c) This structure has in common with Vergnaud (1985) and more recently Kayne will be dependent on the relative clause being able to provide an external boundary to the (1994), that the relative clause, and not the head noun, is the complement of the D. I will event. This section is divided in two parts: first, I will present the structure I am take this complement to be an AgrP, part of the extended projection (used descriptively proposing for relative clauses and show how the basic mechanism will derive the lack of here, in the sense of Grimshaw 1991; see below for more careful discussion.) of the CP durative readings in singular count nouns and the possibility of durative readings with that modifies the NumP head. Here we can assume that the head CP raises to Agr to plurals. Second, I will deepen slightly the analysis of relative clauses in order to account check features against the NumP head of the relative and then raises covertly to D. for the terminative readings in (6c). So that the discussion does not get too messy, I will Further support for this idea will come from the discussion of adjectives in §2. make some preliminary assumptions about the structure in this section that I will not The NumP books is generated adjoined to the CP (as if it were an external argument, motivate explicitly. Motivation for the structure will become clear when I discuss the getting a theta role via predication)101 , instead of being a copy of the operator as adjectives in the next section. proposed by Vergnaud and recently implemented by Kayne as movement to the specifier 4 . 1 . 1 Basic Structure for Relative Clauses and its Effect on the Aspectual of the DP operator.102 This forces both the operator of the relative clause and the NumP Interpretation: durative readings head of the relative to check their case independently103, as various languages with overt case markings (Russian, for example) show to be necessary. In this section I will center my attention on the relation between the head of the relative In the structure in (7), the NumP head of the relative clause raises to the specifier of clause and the relative clause per se. Agr in order to check for the nominal features of C+Agr, and there it enters a Spec-head The structure I will assume in order to account for the durative readings of definites relation with the relative clause head that. The C+Agr complex head raises to D, and with relative clauses is given in (7). thereby licenses the definite determiner (for reasons that will become clear below when (7) we discuss adjectives). DP the From the specifier of Agr, the NumP raises in order to get case in the specifier of AgrP NumP AgrO in the cases we are interested here, just as has been proposed for Exceptional Case Agr' books that+Agr Marking constructions (Chomsky 1991, Lasnik 1993 etc.) The result is then the CP tbooks following104: CP OP C' C tthat IP tOP revolutionized the field 101 See Chapter 6 for discussion. 102 I am assuming the copy theory as discussed in Munn (1994a) to create the operator chain. See section 5. 103 As I will argue below, only nominal extended projections need case. 104 Below we will see why the movement proposed does not happen overtly at least in the languages we are discussing. 190 191 (8) (ii) movement of N to D is covert, movement to AgrO is also covert and triggered by AgrOP NumP[books] case. In this case, again we have no problems, since the complex head [N+D] created Agr' V+Agr after spell-out will trigger the movement to AgrO for case reasons also after spell-out; VP (subj) (iii) the apparently problematic case is the following: movement to AgrO is overt but V' tv N to D movement happens after spell-out. If the trigger is the case of N, why the D is DP C+Agr+D the pied-piped? The DP is pied-piped for PF reasons since the determiner cannot be left AgrP tNumP stranded. However, although probably correct, PF pied-piping does not explain how the Agr' C+Agr CP movement is triggered, since before raising of N to D, D, per se, does not need case, according to my assumptions. We could assume that any strong feature inside a The proposal I am making relies on two assumptions: (i) only nominal projections need case. More specifically only NumPs, QPs or DPs that are part of the extended projection can trigger the movement of as much material as necessary for PF convergence (as in Chomsky 1994). projection (in the sense of Grimshaw 1991) of nouns need to have their case checked. I I find Chomsky's (1994 proposal highly unconstrained at this point, and until a will take an extended projection of a [+N] head-chain to be a head chain whose tail is a clearer picture of the visibility of features is empirically discussed, I would like to suggest [+N –V] element. In other words, only nouns need case. The need for case comes from that the overt movement to AgrO is not required by case. The reason to doubt of the the head noun that moves into Num0, Q0 or D0. If D takes C as its complement, then D correctness of the case reason for movement is that it forces us to assume that case can be does not require case, since C is plausibly [+V] and Cs do not require case in general; (ii) weak or strong within the same language, since in general the overt movement to AgrO is the determiner that is part of a CP projection does not need to have its case checked "optional" (see Mahajan 1990 for Hindi; De Hoop 1992 for Dutch; and others). because the [C+Agr] chain that moves covertly to D does not have a [+N –V] tail. Only If we do not accept case as the trigger for the DP movement to AgrO, and, instead, the NumP head of the relative does, since D+[C+Agr] is not a nominal extended we assume that the DPs that raise to AgrO check for some optionally added feature, we projection105. avoid problem (iii). This feature can be checked in AgrO but is not obligatorily present. Given these two assumptions, there are a few cases that we need to consider: Being a feature that can be added or not to D would account for the fact that, in general, (i) movement of N to D is overt and movement to AgrO is also overt and triggered by overt movement to AgrO is "optional" (as the case of German and Dutch for example). case. In this case we have no problems, since the complex head [N+D] will trigger the movement for case reasons; Treating overt movement to AgrO as triggered by something other than case seems to be preferable, given that it would avoid the need to always have a set of weak and strong case features in the same language for the "same" heads. In other words, AgrOs are not 105 Agr is not a [+N] element per se, since it only activates the features of the head that moves into it. In other words, Agr is a bundle of features. Some of the features become activated or not depending on what moves to Agr0. If a N moves to Agr, the nominal features of Agr become active and Agr can is [+N], but if a verb moves to AgrO0, then verbal features become active and the complex [Agr+V] is [+V] and can continue to move to check tense. 192 193 only positions where case features get checked but also a place that can attract other features.106 At AgrO then, we have a NumP checking its case. Durative readings for the matrix clause will be immediately available since at LF we have a NumP plural with its quantity A second question arises with respect to subject relative clauses. If the head of the unspecified: relative needs case, and the DP does not, then why do subject relative clauses pied-pipe, (9) a. Chomsky wrote [the books that revolutionized the field] for years i.e., why the whole DP moves to AgrS. Here we need to assume that movement to b. The maid washed [the towels that Mary used] for years subject position is needed to satisfy the Extended Projection Principle, as Lasnik (1993) c. Bill added [the footnotes his advisor hated] for years argues. If the EPP is dependent on D features, then movement of the head of the relative d. John drew [the caricatures that won prizes] for years will not suffice to check the EPP, and consequently subject relative clauses will be forced to pied-pipe.107 If, however, instead of a plural, we have a singular NumP, the singular counts as A third question is related to the fact that the DP that is theta marked by the verb is not one, and the result is either an odd stretched reading or an iterative reading characteristic the NumP that checks its case against the verb in AgrO. Such a dissociation between case of terminative predicates, as the examples in (10) show. A schematic tree is given in and theta marking is, however, quite clear in the minimalist program: case assignment (11). happens in the checking domain, and belongs to the functional structure of the grammar, while theta marking is an independent process that happens in the complement domain (I (10) a. #Chomsky wrote [the book that revolutionized the field] for years will have more to say about thematic roles in section 2).108 b. #The maid washed [the towel that Mary used] for years For the moment what is important is the assumption that the determiner is not a c. #Bill added [the footnote his advisor hated] for years constituent with the NumP, and that the NumP head of the relative is raised from the d. #John drew [the caricature that won prizes] for years operator of the relative clause so that it can check its case independently. The operator gets case by being part of a chain of an element that has checked its case features, and the head of the relative raises to check case outside the DP. (11) AgrO (a) booki Agr wrote+Agr VP v' 106 Another possibility is to assume that PF case and LF case are distinct (as proposed, for example by Nunes 1993). Thus the PF case reasons for movement are to get rid of a strong feature in the Agr that the D has to have its case checked, and that the LF case in the N will be eliminated by having the D+N case checked against a verbal head at LF. This raises the question of why N case is an LF case and D case is a PF case. For this reason I will assume that overt movement to AgrO is not necessarily triggered by case. 107 There are a number of interesting implications of this analysis which would take us too far from the present discussion here. One in particular is the idea that the relative clause moves to Spec TP to satisfy the EPP and then the head raises at LF to satisfy case in AgrS. I leave this for further research. DP the AgrP [ti]NumP Agr 108 Lebeaux (1988) also explores the dissociation between case and thematic structure in the acquisition process. 194 195 The proposal I made in chapter 1 says that if the element that is in AgrO has its (15) a. The maid discarded trash for years cardinality specified, the result is a terminative reading. The prediction is then that if the b. #The maid discarded the trash for years NumP plural has its cardinality specified, the result is terminative. This is exactly the c. The maid discarded [the trash that Mary produced] for years case, as the examples in (12) show, and will follow directly, if we assume that numerals in their cardinal readings are generated as specifiers of NumP or adjoined to NumP. Under our analysis this is expected since mass nouns and bare plurals in AgrO produce durative readings. Singular nouns and quantized objects will produce terminative (12) a. #Chomsky wrote [the two books that revolutionized the field] for years readings.109 b. #The maid washed [the two towels that Mary used] for years c. #Bill added [the two footnotes his advisor hated] for years 4 . 1 . 2 More evidence for determiner transparency d. #John drew [the two caricatures that won prizes] for years In this section I would like to present some more evidence for determiner transparency, (13) and along with it, provide some empirical and conceptual motivation for the idea that the AgrOP two booksi definite determiner takes the CP of the relative clause as its complement, both Agr' wrote+Agr VP V' DP the AgrP ti Agr' The analysis also predicts that if the NumP is a mass noun, a durative reading will again be available. This prediction is also borne out as the contrasts in (14) and (15) show. (14) a. Bill wrote junkmail for years b. #Bill wrote the junkmail for years c. Bill wrote [the junkmail that Sam asked for] for years 109 The only cases in which a singular noun with a relative clause seems to allow a durative reading is the cases in which canonical type readings are available, and that is easily true for apples and cars but not so easy for bananas (at least for American speakers). This is because in order to get a durative reading of a singular noun, we have to be interpreting the singular noun as the name of a kind. In such case, we know there are a potentially infinite number of tokens of the same element. The examples below illustrate the point. Thus in (a) while convertibles, compacts, etc. are standard types of cars, clean or new are not and durative readings become basically impossible, if the intended reading is that the car Mary dreamed of is a new car. (i) a. John sold [the car Mary dreamed of] for years = convertible or a Ferrari ≠ clean, new cars b. #John wrote [the letter that Mary dreamed of] for years ? love letter ≠ long c. John ate [the apple Mary liked] for years = Macintosh ≠ green ones d. #John ate the [banana Mary liked] for years ? apple-banana ≠ yellow with a touch of green In (b) it is very hard to establish a durative reading, since it is hard to establish a canonical type of letter whose tokens could have been written for years. A durative reading is marginally acceptable with a reading in which we are talking about love letters but it would be impossible to have a durative reading with the letter that Mary dreamed with being a long letter. The contrast between apple and banana is given in (c) and (d) respectively. While it is easy to establish that we are talking about a type of apples, Macintosh or Granny Smith, the same is not true for banana for American speakers. I will have more to say regular overt type readings as opposed to the canonical types described above in chapter 5. For now I will assume that the durative reading in those cases is due to this ability of certain nouns to be in place of brand/type names. 196 197 semantically and syntactically. To do this I will first explain Kuroda's intuition about (16) a. what relative clauses are. Throughout this chapter I will be following here the steps of DP the Kuroda (1968 and Vergnaud (1985) quite explicitly. b. DP AgrP AgrP NumP Agr NumP Agr According to Kuroda, relative clauses are the opposite of a problem that coordinated Agr+that CP objects pose with respect to semantic role assignment. In the case of relative clauses we Agr+that operator [+/-definite] CP operator [+def/+specific] seem to have two theta roles for one noun phrase (one from the relative clause and one from the matrix clause) while, in the case of coordinated objects, we have more than one The internal structure of the relative can force the operator to be definite, indefinite or head noun for a single thematic role. If we assume that in relative clauses there is one specific. In the following I exemplify cases in which the operator must be interpreted as noun phrase for two thematic roles, and thematic roles cannot be assigned twice, then we indefinite and cases in which both the operator and the head of the relative must be have to assume that there are two noun phrases that are coreferent. Given that in general indefinite which will force the presence of an overt determiner. In other cases the operator we cannot introduce an indefinite noun phrase twice in the same discourse, then one of must be definite and an overt determiner can optionally show up in the position before the the noun phrases must be anaphoric to the other. Thus every definite restrictive relative head of the relative clause. clause involves, at a certain level, an indefinite version of it. (The man I met includes a 4 . 1 . 2 . 1 Evidence for indefinite operators in Relative Clauses form a man that I met.) At the same time, an " indefinite" relative clause such as a man I met must contain a definite operator (DEF) somewhere so that relativization can take The examples below, which were used to argue for the separability of the definite determiner from the head of the relative (in a raising analysis of relative clauses) show place. Kuroda's basic intuition is that definitization is part of the very nature of a relative clause. I will interpret this operation of definitizing as requiring two things: first that the C of the relative clause move to a D element to check D features (the D can be empty or filled with a definite determiner) and second, that there be either an overt that the definite determiner that appears with the head of the relative cannot possibly come from the relative clause operator, since, in these cases, only indefinites are acceptable. The cases below are discussed in Vergnaud (1985. The first case involves there constructions. determiner/quantifier in D or the operator of the relative clause be definite or specific. (17) a. The books there were on the table Schematic trees are given in (16). b. Every book there was on the table c. *A book there was on the table d. *Fewer than three books there were on the table 198 199 (18) a. *There were the books on the table The last two cases I will present exemplify not only the fact that the definite cannot b. *There was every book on the table come from the operator of the relative clause but cannot also come from the matrix clause. c. There was a book on the table (20) a. Mary weighs forty-five kilos d. There were fewer than three books on the table b. *Mary weighs forty-five kilos Susan would love to weigh The contrast between (17a,b) and (17c,d), first noted by Carlson (1977b), shows that, although definites and universal quantifiers cannot appear in simple there constructions, they can, unlike indefinites, easily appear before the head of a relative clause of a there construction. In other words, it seems that the definite determiner or the c. Mary weighs the forty-five kilos Susan would love to weigh (21) a. John loved sincerity b. *John hated sincerity that Mary loved c. John hated the sincerity that Mary loved universal quantifier cannot be a part of the operator of the relative clause. The unacceptability of an indefinite in (17c,d) can be only explained if we assume The examples above illustrate that the definite determiner cannot possibly come from with Vergnaud and Kuroda that relativization is impossible if a definite is not provided the operator of the relative. In both case the definite does not come from a definite relative either by the operator or by some extra definite element. We have seen already that the clause head either, supporting the idea that the head of the relative clause is not a operator of the relative clause cannot be definite since there can be no definites or specific constituent with the definite determiner. Notice that in all these cases the lack of a DPs inside there constructions. If a definite is not provided elsewhere we end up with determiner renders the sentence unacceptable which is illustrated in the (b) examples two indefinites that are supposed to be coreferential and the result is unacceptable. above. Since there is no source for the definite in either the matrix or the relative clause, The second case discussed by Vergnaud is given below. Here again the definite the sentence is uninterpretable. A definite determiner or a strong quantifier like every can element cannot come from the operator of the relative, given the unacceptability of (19b). be generated in the D position of the complement (or in a Q position raising to a D Again the indefinite in (d) is not acceptable, since in those cases there is no source for a position) of the matrix verb licensing the relative clause110. At AgrOP, in both the relative definite inside the relative clause. Headway is an idiom and, as such, cannot be specific clause and the matrix clause we have forty-five kilos and sincerity and no problem arises or definite. Thus the definite/specific element has to be provided by an outside element. for the interpretation of these sentences. (19) a. John made headway b. *John made the headway c. The headway John made was amazing d. *Headway John made was amazing e. *Some/little headway John made was amazing 200 110 With respect to the behavior of many and few in relative clauses of there constructions the judgements are not as clear but speakers tend to consider them not very good, when compared to (ii): (i) a. ?*Few books there were on sale are good b. ?*Many books there were on sale are good (ii) a. The few books there were on sale are good b. The many books there were on sale are good Assuming that (i) is unacceptable and (ii) is acceptable, it seems that many and few are adjoined to NumP. 201 The next set of cases illustrate the fact that sometimes the operator chain must contain c. *(Unë) i pash (disa) vajza a definite. I pro saw some girls I saw a girl 4 . 1 . 2 . 2 Evidence for definite operators in relative clauses The discussion here is based on the cases discussed by Morgan (1972) for Albanian and English. In Albanian, the coreferential NP in the relative clause is definite when the But the clitic pronouns can occur in relative clauses with indefinite NPs as heads as shown in (24): head NP is indefinite. Evidence that the coreferential NP in the relative clause is definite comes from the behavior of a clitic pronoun. There is a set of clitic pronouns that show (24) a. disa pika që i kam lënë të errëtë up immediately preceding the verb when there is a definite direct object (even when it may some points that-ACC pro have left PRT dark have been deleted). 'Some points which I have left dark b. një luftë pa shpresë, të cilën e kishte lënë si të humbur (22) a. (Unë) e pash I pro saw a war without hope, which pro had left as PRT lost I saw him/her/it a war without hope which she had given up as lost b. (Unë) e pash vajzén The examples above show that the operator of the relative clause must be definite in I pro saw girl-the order to license the clitic. Similar cases are found in English. Intensive reflexives, for I saw the girl example, can only appear with definites as the contrast between (25a) and (25b) shows. However, (25c) shows that, in spite of the relative head being indefinite, the intensive c. (Unë) i pash reflexive is perfectly acceptable. I pro saw I saw them (25) a. The plumber himself reads Sartre b. *A plumber himself reads Sartre (23) a. (Unë) pash një vajza c. A plumber who himself reads Sartre said that I should take a look at it I saw a girl Another example given by Morgan is related to tough constructions. Tough-movement b. *(Unë) e pash një vajza I pro saw a girl cannot apply to true indefinites as shown in (26b). But it can apply in relatives with I saw a girl indefinite heads as in (26c). 202 203 (26) a. The man is hard to argue with (28) a. Chomsky wrote [the books that revolutionized the field] in two years b. *A man is hard to argue with (*Existential) b. The maid washed [the towels that Mary used] in two years c. A man who is hard to argue with wrote this paper c. Bill added [the footnotes his advisor hated] in two years d. John drew [the caricatures that won prizes] in two years Finally, certain adjectives cannot be predicated of true indefinites. But they can occur in relatives with indefinite heads. Given that only the NumP is raising to AgrOP at LF, one might expect that terminative readings would not be possible. In this section I will show that the relative (27) a. The man is quite tall clause is able to impose an external boundary on the durative reading, and that this b. *A man is quite tall c. A man who is quite tall is standing on my hair possibility is predicted, given that durative readings are always the default (i.e. unmarked) case for aspect. I will argue that this external boundary can be provided by The examples above show that indefinite headed relative clauses require a definite the relative clause itself and is strongly dependent on internal properties of the CP.112 operator.111 We have already seen that durative predicates can be bounded by external boundaries. We have seen two sources of the definiteness required by the relative clause. Either it For example, if we add to (29a) the adverb for two hours, the result is a terminative comes from the operator in the relative clause in which case an indefinite head is possible predicate (29b), as the addition of repeatedly (which as we have seen requires discrete or it comes from an external determiner, required since there is no independent source for subevents) shows in (29c): the definite, assuming restrictive relatives modify NumPs. In both cases, it is the relative (29) a. Mary ran clause that is either supplying or requiring the definite determiner, thus providing both b. Mary ran for an hour conceptual and empirical motivation that the determiner takes the (Agr)CP of the relative c. Mary ran for an hour repeatedly clause as a complement. To put it simply, it is always possible to bound a durative event, but in order to 4 . 1 . 3 Terminative Readings with Plurals "unbound" a terminative event, it is necessary to force iteration of the verbal predicate. The sentences exemplified in (9) have, as we have said before, a terminative reading in To a durative predicate we can add an external boundary. To a terminative predicate we addition to a durative reading. This is shown by the fact that the same examples are also need to force the repetition of the subevent. This fact is easily derivable from a theory that compatible with the adverbial in two years. takes durative predicates as the default case and terminative predicates as the marked case. 111 Morgan's examples are all , however, cases of subject relatives with subject gaps. For some reason it is hard to force a definite in object position. I have no account for this, and Morgan does not mention it. 204 This section basically opens a can of worms with respect to relative clauses and I believe that a finer grained analysis is required than what I will present here. For example I will ignore tense, mood and focus effects in the relative clauses. I will also ignore different types of relative operators. 112 205 The durative interpretations of the VP whose verb has a relative clause as its the matrix VP. It is not necessary that the object operator to have its quantity specified. complement are not the result of iteration. This is important, because it shows that, in All that is necessary is that the relative be interpreted as having clear temporal or spatial contrast to (29c), the durative readings are not derived by "unbounding" a terminative boundaries, creating a specific context for the complement as a whole. event, i.e., by multiplying bounded subevents, but rather the terminative reading is the There is a scale with respect to the elements inside relatives that are more likely to result of applying an external boundary to a durative predicate. If we assumed the allow the relative clause to act as an external boundary. This scale is partially pragmatic opposite, i.e., that the derived reading is the durative reading, we would expect durativity and partially syntactic: if the relative clause is basically a generic or habitual sentence, then to be the result of iteration. it is unlikely to serve as an external boundary for the durative VP. If it is stative it also 4 . 1 . 3 . 1 The bounding nature of the relative clause tends not to act as an external boundary. Other markings on the verb will disfavor the relative to serve as an external boundary. For example, the progressive marker disfavors Consider the following examples. the relative clause as an external bound for the upper VP: (30) a. John erased the ads that [Mary posted in two hours] in three days (31) John [erased [the ads that Mary was posting] for three days] b. #John erased the ads that [Mary posted in two hours] for three days c. John erased the ads that [Mary posted for two hours] in three hours d. ?John erased the ads that [Mary posted for two hours] for three days Another example of these nuances comes from the distinction between the perfective and imperfective inside the relative clause in Brazilian Portuguese. While the perfective can be translated roughly as the simple past, the imperfective is related to an action in the In English in very few cases can we guarantee having a definite operator inside the relative clause. However, it is easy to force specified quantities or unspecified quantities in the object of the relative by the addition of adverbials. In (30a) and (30b) we are forced past that can continue at the moment of the speech. Not surprisingly the latter will basically force durative readings on the matrix VP (for more on the perfective, see chapter 6). to have a specified amount of ads inside the relative clause as the adverb in two hours shows. Now, in (30c) and (30d) we do not have a specified amount of ads but we have (32) a. O Pedro [rasgou [os anúncios que a Maria colocou no jornal]] #por três an extrinsic time boundary that tells us that there are as many ads as there can be in two anos hours. The underlined adverbs are meant to be modifying the matrix predicate. If we The Pedro [tore [the ads that the Maria put-PERF in the newspaper]] for three years examine their behavior, we can see that while the in three hours is perfectly acceptable, Pedro tore up the ads that Maria put in the newspaper for three years for three days, which indicates duration, tends to force stretched readings or repetitive readings. The generalization we reach at this point is that it seems to be enough to have a bounded event in the relative clause in order to use the relative as an external boundary for 206 207 b. O Pedro [rasgou [ os anúncios que a Maria colocava no jornal]] por três that passed yesterday's exam which provides a spatio-temporal boundary for interpreting students.113 anos The Pedro [tore [the ads that the Maria put-IMP in the newspaper]] for Now consider the following: three years Pedro tore up the ads that Maria used put in the newspaper for three years (35) a. The professor failed the students that came from poor countries for years b. #The professor failed the students that passed yesterday's exam for years Now consider the case with stative verbs inside the relative clause: While in (35a) a durative reading is very easy if we interpret the relative as providing (33) a. John erased [the ads that [ Mary hates]] for three days/ in three days b. John erased [the ads that [ Mary hated]] for three days c. John erased [the ads that [ were small]] in three days a permanent property for students, in (35b) a durative reading of the matrix is almost impossible. The adverb for years forces the professor to fail the students repeatedly. In (35a) a terminative reading is also possible provided we interpret the relative clause as defining a closed set of students. Although it seems that durative readings are much easier to obtain, terminative It should be noted that the variation is not in the matrix clause, but in the relative readings are also possible. This is expected if we consider that the NumP head is related clause. Once the relative is interpreted as providing an external bound, a bounded reading to the head of the relative clause and that the relative clause can provide an external bound will be available for the durative AgrO. If the relative clause is interpreted as identifying a for the event in the matrix clause. Some relatives will be better than others at supplying a closed set in time or space, then it will provide us a way to delimit the matrix event. On such a boundary. the other hand, if the relative can just describe a property without establishing a spatioThe effect of the relative clause can be observed quite strikingly in the following temporal bound then the result is durative because the relative clause cannot be used as an examples. external bound to the event. (34) a. Students that come from poor countries are hardworking 4 . 1 . 3 . 2 Summary b. Students that passed yesterday's exam are hardworking The problem posed in the beginning of this section was what could account for the In (34) the bare plural will have a generic interpretation in (34a) and an existential interpretation in (34b) and we can see that the only difference is the "type" of the relative clause. In (34a) the indefinite is not bound in time, since we are predicating come from poor countries to students with roughly the meaning students from poor countries. This is an example of what Kleiber (1987) calls a non-specifying relative as opposed to students 208 fact that definite determiners with relative clauses allow durative readings of the matrix clause. The proposal I argued for can be summarized as follows: 113 "Ils n'impliquent aucune localisation externe, aucun point de référence spatio-temporel, ou, si l'on préfère, leur sens ne requiert pas le recours au hic et nunc de l'énonciation. (...) Ce sont des prédicats internes dans la mesure où ils constituent des propriétés que ne se localisent précisément que par rapport à l'object auquel ils s'appliquent." Kleiber (1987:37) [They don't imply any external localization, no spatiotemporal reference point or, if one prefers, their sense does not require us to resort to the here and now of enunciation. These are internal predicates in as much as they are properties whose position is dependent on the object to which they apply.] 209 (36) again illustrates the fact that definite determiners force terminative readings of the (i) (ii) the determiner is not a constituent with the noun phrase that follows it. Instead predicate. In (37a), where the book is singular, a durative reading is also unavailable. In the determiner is licensed by the relative clause. (37b), however, a durative reading is available in spite of the definite determiner. It seems the head of the relative clause is a NumP and that is what raises to the matrix then that the adjective is having an effect similar to the effect found with the relative clause AgrO in order to check case, producing durative readings if plural or clauses in the preceding section. This effect is clearly is not a mere product of mass singular and terminative readings if count singular or if plural preceded by modification since in (38b) a terminative reading of the predicate is again obligatory. a numeral; In this section I will provide an account for the durative readings in (37b) in terms of (iii) the relative clause left behind can act as an external boundary allowing determiner transparency and I will discuss the distinction between adjectives like wrong terminative readings of the matrix VP. The ability to get terminative readings is and adjectives like difficult. In giving an account of the adjectives, I will also provide dependent on factors mainly internal to the relative clause. more motivation for the mechanism assumed in the previous section to license the relative clause, but left unexplained. Note that this analysis crucially relies on the fact that the definite determiner is left The basic idea is the following. Definite determiners are special in the sense that they stranded below and durative readings are the default, being able to be bound if an external require a referential head to license them. Thus, following Higginbotham (1985), the boundary presents itself. definite determiner in the phrase the man, must theta-bind a referential argument in the NP 4.2 The Wrong and the Difficult Adjectives man in order to be licensed. I will argue that certain adjectives are able to bind a definite determiner. Then I propose a structure that will account for the durativity of plurals with Having now sketched out the basic ideas of how determiner transparency works, we can wrong but not with difficult. and extend this analysis to Brazilian Portuguese. move on to examine other cases where it also seems to hold. Consider the following examples: 4 . 2 . 1 Adapting Higginbotham (1985) to minimalism (36) a. Mary wrote the book #for years/ in two years b. Mary wrote the books #for years/ in two years (37) a. Mary wrote the wrong book #for years/ ?in two years b. Mary wrote the wrong books for years/ ?in two years (38) a. Mary wrote the difficult book #for years/ in two years b. Mary wrote the difficult books #for years/ in two years 210 Higginbotham (1985) departs from the assumption that predicates start out unsaturated and that all arguments are saturated. The Theta Criterion then is to be stated as in (39). (39) a. Every thematic position is discharged b. If X discharges a theta-role in Y, then it discharges only one. There are four ways to saturate a predicate, i.e., there are four modes of thematic discharge, described briefly in (40). 211 domain of lexical heads and coincides pretty much with what Chomsky (1995) refers to (40) Modes of Thematic Discharge as theta marking. We can implement theta-binding by means of head movement of a lexical head a. theta-marking: this is the case in which a predicate V, for example, thetamarks its internal argument (the nominal phrase under V'); through the functional structure of its extended projection.114 Theta-binding is the result of a lexical head moving as high as it can in the functional structure, to a Num, Q or D, if b. theta binding: this is the case in which a determiner or quantifier theta nominal, or to a T, or C if verbal. We could say that the functional elements are LF binds an open position in a nominal; affixes that need to attach to a lexical head in order to be interpreted. The lexical elements c. theta identification: this is the case in which one open position from the will move to D and T to check features. adjective and one from the noun merge into a single open position; Only theta marking is a relation that happens in the complement domain. All the other d. autonymous theta-marking: is the case in which an adjective, for example, would have a thematic position which can be used as an attribute theta relations of Higginbotham are in fact, for the minimalist program, checking relations that occur in the checking domain. Theta binding is a head-head relation between a lexical interpreting the adjective relative to a context. For Higginbotham, this is discharged by theta marking of the phrase marker of the noun itself. and a functional element and theta identification is a relation between two XPs. The motivation for theta identification and autonymous theta marking is given by Examples are given below: Higginbotham as the distinction between red flower and big ant. In the first case we get a reading as in (42a) and in the second case a reading as in (42b): (41) a. VP<1*, 2> b. DP<*1> V<1,2> DP D NP <1> eat the cake the cake theta marking (42) a. x is a flower and x is red b. x is a ant and x is big (for an ant) theta-binding DeGraff and Mandelbaum (1993, however, argue that (42b) does not actually c. NP <1> AP<1> red d. NP <1> NP<1,2*> AP<1,2*> flower toy theta-identification NP <1> instantiate autonymous theta-marking, but rather identification relative to an extensional class. They reserve autonymous theta marking to cases such as (43): gun autonymous theta marking The distinction between theta marking and theta binding is easy to identify and be put to work in minimalist terms. We can assume that theta-marking happens in the internal 114 212 I am using the term extended projection merely descriptively here. 213 we apply this test to the adjectives we are considering, we can see that wrong and same (43) a. John is an old friend b. John is a future president pattern as heads and difficult and long as XP modifiers. Thus while we can say (44a), c. John is a good dancer (44b) does not mean that it is the wrong book, rather it means that the book says something wrong. (44c) is also not acceptable: If we consider the meaning of (43a), where John is a long-time friend, we cannot paraphrase it as John is old for a friend. However, old is still relative even in its normal (44) a. the book is difficult/long meaning: an old fruit fly can be hours old, while an old rock can be millions of years old. b. *the book/pencil is wrong The difference between these two is that old rock is compared using the extension of rock c. *the book is same while old friend is not compared to the extension of friend, under the "long-time" reading Given the contrast in (44) I will assume that wrong and same are heads that take but with some property of being a friend. arguments within the DP, and that difficult and long are true adjoined modifiers. Evidence DeGraff and Mandelbaum argue that autonymous theta marking then is merely theta marking of the NP by the adjective as a lexical head.115 I will adopt their proposal and for this comes from the fact that to use wrong and same in predicative constructions, we need to add a definite determiner and one. show that it is this class of adjectives that have the potential to license a definite determiner. I believe that what distinguishes autonymous theta marking from regular theta marking is the fact that regular theta marking requires a DP to be the argument but (45) a. that book/pencil is the wrong one b. that book is the same one autonymous theta marking does not require the argument to be a DP. I examine roughly the same facts from Brazilian Portuguese since this is a language This suggests that same and wrong have two arguments. One of the arguments can be where adjectives agree with the head noun they modify. The differences between English theta bound by the definite determiner116 and the other enters into a theta marking relation and Brazilian Portuguese can be accommodated under the same analysis. with the complement. Same and wrong also seem to require a complement, one. Adjectives like red, long etc., on the other hand, can appear in predicative position 4 . 2 . 2 The Difference Between Wrong and Difficult without a determiner and a complement. Higginbotham shows that adjectives that enter a theta identification with the head noun Given this distinction I want to propose that red enters a theta identification relation can usually appear as predicates, and adjectives that can autonymously theta mark the with the head noun. Long, difficult and red are adjoined to NPs (see Munn 1995a and head noun cannot appear in predicative positions. Bernstein (1993) independently uses references there), as exemplified in (46). Wrong, on the other hand, enters a theta binding the unacceptability of an adjective in predicative position as a test for a "head" adjective. If 115 A careful reading of Higginbotham seems to show that this was what he had in mind for these cases (cf. Higginbotham 1985:566 fn. 19) 214 116 The notion of theta binding is really one that should be something "done" by the noun. Thus it is not the determiner that gives reference to the noun but the noun that gives reference to the determiner. For this reason I will reverse binder/bindee relation in theta-binding: the noun theta-binds the determiner. Obviously this is no longer a case of binding, but this is implicit in reanalyzing theta binding as incorporation. See §2.5 for more discussion. 215 relation with the definite determiner and an autonymous theta marking relation with the (49) a. the same story head noun. A schematic tree is given in (47): b. *every same story c. every wrong story (46) a. red <1> b. NP <1> AP<1> (50) The student brought the wrong two blue pencils NP<1> (51) (47) a. wrong <R, 1> b. DP the DP<*R> the AP wrong AP<R,*1> wrong<R,1> NumP two NP<*1> Num pencil every, as shown in (49). I will also assume that wrong can actually select for NumPs, NP AP NP blue pencils Wrong is higher than long/difficult/blue117 but lower than same.118 as shown in (48). I will assume that same is in a quantifier position, since it cannot co-occur with Num' That wrong can take NumPs can be seen from the examples of one substitution and overt numerals: assuming numerals are specifiers of NumP phrases (see (50). Given that, tentatively I will propose the structure in (51): (52) a. the wrong one b. the wrong two pencils (48) a. the same wrong pencil b. *the wrong same red pencil The full structure of the wrong two blue pencils is then the following: c. *the red same wrong pencil d. *the same red wrong pencil e. *the wrong red same apple f. *the red wrong same apple 117 Of course there is an internal ordering with respect to the adjectival true modifiers. I will not discuss this ordering since it is not relevant for the point I want to make. (see Sproat and Shih 1987). 118 Not all speakers like same and wrong together. 216 217 (53) theta properties. While the former theta marks a complement and is able to license a DP the definite determiner, the latter are just modifiers. Wrong, being a head, takes a NumP AP wrong complement, which is too far from the definite to theta bind it. The structure created NumP two forces the NumP to check its case by raising from the specifier of an Agr projection to a Num' blue case checking position. A question arises with respect to languages whose adjectives pencils appear to the right of the noun. In the next section I deal with the counterpart of wrong in What is important so far is that the definite determiner is not theta bound by the head noun Brazilian Portuguese. but by the adjective, which entails that, at LF, wrong will ultimately incorporate into the determiner. (54) 4 . 2 . 3 Adjectives on the right: the case of Brazilian Portuguese DP wrong+the The analysis that wrong is a head appears to be, at first sight, highly problematic for AP twrong languages like Brazilian Portuguese, where the adjectives appear to the right. I will show NumP that in fact the Brazilian Portuguese facts support the analysis. Before doing so, I will Wrong autonymously theta marks the NumP, but it does not check its case. In order for the NumP to receive case it must raise to a case checking specifier position. Since wrong agrees in φ-features with the NumP, I will assume that above wrong there is an have to present some basic facts about adjective placement in Brazilian Portuguese. First, adjectives tend to appear to the right of the noun. Second, wrong must be in the rightmost position, as the contrast in (56) shows. Agr projection. Wrong moves there in order to get to the D. The NumP raises to the (56) a. o cavalo branco errado specifier of Agr to check its features with the adjective. From there, at LF, it moves to a the-MASC.SG horse-MASC.SG white-MASC.SG wrong.MASC.SG case checking position. Evidence for the Agr position will be given in §2.3 below. b. *o cavalo errado branco (55) DP the horse wrong white [wrongi+agr]j+the AgrP NumPk Agr' Adjectives like white usually appear after the noun. The internal ordering of the tj AP ti NumPk adjectives that can appear in predicative positions is not clear (see 58) and for sake of the discussion I will assume that they are generated left adjoined to the NP.119 two pencils Summarizing, the analysis I have presented here for the adjectives like wrong as opposed to difficult, long, etc. distinguishes the two classes of adjectives in terms of their 218 119 I tend to believe that they are in a mirror image with the English adjectives, which is in fact a problem for Bernstein's proposal that adjectives in Romance are left adjoined. It may be possible that both options are realized in BP. See Sproat and Shih 1987 for an interesting discussion on the cross-linguistic variation found in adjective ordering. I will come back to his proposal in the next chapter. 219 b. #o cavalo é mesmo (57) a. o cavalo é branco the horse is same the horse is white (59) becomes acceptable if a definite determiner is added to the predicate. Recall from b. o cavalo é/está doente chapter 3 that BP licenses a null complement position after the definite determiner. the horse is sick c. a maçã está madura (65) a. este cavalo é o errado the apple is ripe this horse is the wrong (one) d. a maçã é verde b. este cavalo é o mesmo the apple is green this horse is the same (one) (58) a. o cavalo branco velho There is, however, a difference between wrong and errado. While wrong is always the horse white old prenominal like other adjectives in English, errado is always postnominal. Same is prenominal and higher than wrong in English. The Brazilian Portuguese counterpart is, b. ?o cavalo velho branco however, prenominal. the horse old white The example (60) illustrates that both wrong and same can appear in the same DP: c. o cavalo branco doente the horse white sick (60) ?o mesmo cavalo errado the same horse wrong d. ?o cavalo doente branco the horse sick white As in English, same cannot co-occur with the counterpart of todo 'every': Just as wrong and same in English cannot appear in predicative positions unless they theta bind a definite determiner and have an internal argument and a definite determiner, in (61) *todo mesmo cavalo every same horse Brazilian Portuguese, the same elements cannot appear in predicative positions. For English I assumed that both same and wrong were heads and that same would be (59) a. #o cavalo é errado higher. If we adopt the same exact structure for Brazilian Portuguese, we do not get the the horse is wrong right word order as shown in (62a,b). We want (63b) rather than (63a): 220 221 (62) a. DP b. (65) a. as duas revistas erradas DP the two journals wrong as AP erradas a NumP AP mesma duas Num' b. as mesmas duas revistas erradas120 AP errrada the same two journals wrong NumP NP NP revistas revistas The analysis proposed makes the prediction that wrong should be the last adjective to (63) the right, which is in fact true, as the examples below show:121 a. *as erradas duas revistas (66) a. o cavalo branco errado the wrong two journals the-MASC.SG horse-MASC.SG white-MASC.SG wrong.MASC.SG b. as duas revistas erradas 'the wrong white horse' the two journals wrong b. *o cavalo errado branco the horse wrong white If we are to assume the same structure for Brazilian Portuguese as we did for English, it must be the case then that the complement of wrong can move to some higher position above it. I will assume that this is exactly what happens. Wrong not only has two arguments but in Brazilian Portuguese has strong agreement features that need to be checked overtly in the specifier of the Agr projection above it. Same, on the other hand, always appears prenominally right after the determiner, which as in English is obligatory. The structure is given in (64b) (64) a. NumPi Now that we have two different structures for wrong and difficult, we can go back to the aspectual differences and explain the ability of wrong to allow for determiner transparency. In section 1 of this chapter I proposed that determiner transparency was the result of having the determiner take something other than the noun that seems to be its DP as 4 . 2 . 4 Accounting for the aspectual role of wrong b. AgrP as Agr' NumPj tNumP property. Given our implementation of Higginbotham's theory, this means that the NumP QP mesmas duas revistas erradas AP tA complement as its complement. The structure I proposed above for wrong has exactly this DP is not the element that theta binds the definite determiner. The definite determiner is theta- AgrP Agr' duas revistas erradas AP tA tNumP 120 This analysis probably means that two wrong journals will end up in the specifier of an Agr projection above same at LF. Same and different, which I have not dealt with here, are also interesting in that same appears prenominally and different always postnominally. This is true under the interpretation we are interested in here, i.e., the interpretation that that one is the wrong white horse, not that the horse did something wrong. 121 222 223 bound in the case of the relative clauses by the head of the relative that moves to it. In the b. case of wrong it is bound by movement of wrong to it. AgrO = Durative sonatas Agr' The NumP argument of wrong moves to check its case at AgrO. If the NumP is play+agr VP singular or if the NumP is plural with its quantity specified, only terminative readings will obtain. If the NumP is a bare plural or mass noun, the result is a durative predicate. In the c. AgrO = Terminative (iteration is possible) latter case, wrong can only serve an external bound and force a terminative reading, if two sonatas Agr used in a contrastive form, in which case a discourse boundary can be provided for the play+agr sentence.122 VP Examples and respective trees are given below: d. (67) a. John played the wrong sonata #for years AgrO = Durative music b. John played the wrong sonatas for years play+agr Agr VP c. John played the wrong two sonatas #for years d. John played the wrong music for years In sum, the same account that was given for the relative clauses can be extended to the (68) a. AgrO = Terminative sonata (iteration is possible) wrong type adjectives.123 The NumP raises to the Agr position inside the DP and from there it moves to the specifier of AgrO. The movement of NumP leaving the determiner Agr stranded is possible because the determiner is not in a theta binding relation with it. played+agr VP Instead, the adjective or the relative clause is. 4 . 2 . 5 What is so special about definite determiners? At this point we can now ask the question of why the NumP in (69) below cannot move The fact that wrong can be used in to contrast the amount of time that someone takes to do something wrong as opposed to something right is reminiscent of the contrast described by Depraetere (ms.) and reproduced below: 122 (i) leaving the determiner stranded and consequently allowing a durative reading. When did you live in Paris? I lived in Paris in 1920. #In fact I still live there (ii) Where did you liven in 1920? I lived in Paris in 1920. In fact I still live there If in 1920 is focused this adverbial imposes a boundary on the whole event. Exactly how this is to be explained is not clear, but it seems to me that whatever explains (i) and (ii) will explain the bounding effect of focused wrong. 224 Same will allow the same set of readings that wrong does in the case it is bound by a quantifier. Otherwise only terminative readings will arise. 123 225 (69) a. John watched the movies #for years of the (b) examples, which show that headway, forty kilos and sincerity cannot appear alone with a definite determiner. b. * AgrO NumP[movies] (70) a. John made headway Agr' watched+Agr b. *John made the headway VP c. *Headway John made was amazing V tV DP the tNumP d. The headway John made was amazing e. *Some/little headway John made was amazing (71) a. Mary weighs forty-five kilos The reason is very simple. If the complex head NumP movies does not raise to D, the D features are left unchecked and the result is uninterpretable since at LF there is a definite determiner that has not been theta bound. More generally, it is impossible to move an XP b. *Mary weighs the forty-five kilos c. *Mary weighs forty-five kilos Susan would love to weigh d. Mary weighs the forty-five kilos Susan would love to weigh whose head is part of an extended projection without carting along all of the extended projection, because there will be always something left behind unchecked. This case is important in understanding the fact noted in fn. 19 above, that it is the (72) a. John loved sincerity b. *John loved the sincerity noun that licenses the definite determiner and not the other way around. We can consider c. *John hated sincerity that Mary loved definite determiners as a marked option in the grammar. Because they are anaphoric, they d. John hated the sincerity that Mary loved need to be licensed by something that is "referential" in the sense of Jackendoff (1985, The data above clearly implicate "referentiality" as a requirement on licensing the i.e. something that is ostensibly demonstrable. (See chapter 5 for a more careful definite determiner. Idioms and measure phrases are commonly taken to be nondiscussion of this.) In the case of (69), nothing at all licenses the definite determiner, referential, and in keeping with Jackendoff's intuition, they are not ostensibly because the NumP has moved as a phrase to AgrO. The unlicensed D therefore has demonstrable. Abstract nouns seem to be a similar case124. Consequently, they do not unchecked features and the derivation does not converge. license a definite determiner as the (b) examples show. Interestingly, measure phrases This view of the dependence of the definite determiner can now explain two other appear to be able to license a definite determiner, as shown in (73a) (again, see chapter 5 cases of obligatory determiner transparency noted in §1.2 above. The data is repeated below. Recall that we argued, following Vergnaud (1985 and Carlson (1977b that idioms and measure phrases provide evidence that the determiner and the head of a relative clause do not form a constituent. The argument was based on the unacceptability 226 124 Obviously the term abstract is far too vague a term for what characterizes the difference between, for example, the theorem and *the kindness. We need to say that the former is demonstrable in some way. I will leave this question open. 227 for more careful discussion of this case). Abstract nouns also at first sight seem to (75) a. #Peter watched that movie for 3 years (only iterative readings) license a definite determiner when they take a complement (224b). b. Peter watched those movies for 3 years/ in two hours c. #Peter watched those two movies for 3 years (only iterative readings) (73) a. the kilo of meat d. Peter watched that junk for 3 years/ in two hours b. John loves the sincerity of young children While (75a) allows only terminative readings, the plural with demonstratives allow Before we move on to a discussion on demonstratives, it is important to note the durative and terminative readings (75b). Again, if the cardinality of the plural is specified, similarities between relative clauses and wrong type adjectives: they have in common the ability to bind a definite determiner and license an argument via theta marking125, besides the only possible reading is a terminative reading (75c) and yet again if the noun is mass, a durative reading will be available. exhibiting the same range of aspectual interpretations. Also the adjective wrong and the If the structure I proposed in the previous sections to account for the identical restrictive relatives behave alike in predicative positions: aspectual properties is correct, then the internal structure of a DP with a demonstrative (74) must be quite similar. In this section I will provide some motivation for construing a. *A pencil is wrong b. *A pencil is that is on the table demonstrative DPs as having something other than the head noun as the binder for the determiner. The reason these sentences are unacceptable lies on the fact that the subject of the This section is divided as follows. First I present some intuitions that need to be small clause is plausibly the NumP, and NumPs, not having any D-features, cannot captured about demonstratives. Then I introduce a structure that is able to capture the satisfy the Extended Projection Principle. basic facts in English with respect to phrase structure and aspectual readings. 4.3 4.3.1 Demonstratives I now return to examples of the sort noted in (2). Consider the following: Demonstratives as Disguised Definite Descriptions There are at least two ways of viewing demonstratives. One is to assume that they are individual constants as in Kaplan (1977). The other way is to view them as disguised definite descriptions. Here I will assume the second route and follow Bennett's (1977) intuition. The basic intuition is that when we say this house we are actually saying the house here and that house is the house there. Demonstratives require demonstration, typically a pointing that makes clear which place is intended. According to Bennett, 125 I am actually assuming that the head of the relative is licensed as an external argument by predication with the CP. This entails that predication is also a form of theta marking for external arguments, at least with respect to nouns. (more on that in chapter 6). 228 229 however, only places can actually be demonstrated. Here and there are then the only true (77) a. Il y avait une fois un ogre qui ne se nourrissait que de chair fraîche. #Un demonstrative pronouns.126 beau jour, cet ogre-là décida de changer de régime When a speaker uses this or that and demonstrates an object in a place, "the context b. Once upon a time there was an ogre that would only eat.... ...#this here usually makes the intended demonstratum so salient that the audience grasps the intended ogre decided to change his diet referent and thus the intended assertion." (Bennett 1977) Thus in a sense every demonstrative expression has its reference dependent on the context. The noun house that c. Era uma vez um ogre que só comia ...#esse/este ogre aqui decidiu mudar de dieta. accompanies this house is not the element that is providing reference for the DP. Rather it is the here i.e., the pointing (the demonstratum) that is providing the reference for it. This pointing can be an actual pointing, or it can be made explicit in the discourse by the Relative clauses can also provide the place for the pointing. In its restrictive reading, the relative clause cannot co-occur with here as illustrated in (78): addition of here, as in (76a); là in French (76b) and aqui in Brazilian Portuguese. (78) a. ?*This here man we talked about (76) a. This here man (Dialectal) b. Cet homme là (*restrictive reading) b. *?Cet homme là que nous avons rencontrés c. *?Esse homem aqui que nós encontramos c. Esse homem aqui The point that should be clear from the discussion of demonstrative phrases, is that Also, discourse anaphora will provide a place for the pointing if there is no explicit the demonstratum is some element other than the head noun, and that the demonstrative here. Evidence for the complementarity between here and discourse anaphora comes from phrase is a disguised definite description. Definite descriptions are assumed to carry a the fact that if here is present, discourse anaphora is not possible. This observation, due uniqueness presupposition. According to Kleiber (1989), however, rather than carrying a to Tasmowski-De Ryck (1990) is illustrated in (77) for French, English and Brazilian uniqueness presupposition, the demonstrative asserts this uniqueness. When we say this Portuguese. man we are picking a member of the class of men. Such an observation led Kleiber (1989) to assume that the demonstrative is equivalent to an attributive structure of the type this is a man. Thus, in a sense, there is an "indefinite" in a Demonstrative phrase, just like there was an indefinite of sorts in relative clauses and in the wrong phrases. However, we should not assume that demonstratives are true indefinites since they can appear with definites in various languages. Moreover, the demonstrative is not banned in contexts where true 126 Here can be an indexical besides being a demonstrative Indexicals are like I and you expressions. They do not require a demonstration to establish the referent. In I am here, here does not depend on a demonstration: it is the here of the context. Thus in I love this city, i.e., I love the city here, the here is the here of the context. There, on the other hand, is only demonstrative. 230 231 indefinites are definitely unacceptable. For example, in individual-level predicates take to be a DP (i.e. a pronoun) raises then to check its D features with the D head where indefinites are unacceptable (79a) while demonstratives are perfectly acceptable (79b): it can get its case checked and agrees in proximity and phi features with the LOC+Agr complex, which I assume checks case just as in the English prenominal possessive. The (79) *A man is intelligent net result is that the demonstrative agrees with the locative element in terms of This man is intelligent proximity127 and with the NumP in number.128 The structure above captures the intuition that the noun phrase is not the 4 . 3 . 2 A transparent determiner structure for demonstratives demonstratum, since it is the LOC that raises to D and not the noun phrase. It also The structure I will be proposing for the demonstrative is an attempt to capture the maintains the analysis of Szabolcsi (1994) and Uriagereka (1988) among others that definite/indefinite nature of demonstratives and the similarities with the relative clause and demonstratives are modifiers in Spec DP and captures the agreement facts of wrong in terms of aspectual interpretation. demonstratives in English. Now we have exactly the same structure for demonstrative phrases as we have for the The structure at Spell-Out for (76a) is given in (80): relative clause and for the adjectives like wrong. Consequently the same analysis we used (80) DP this to account for the aspectual properties of DP relative clauses and DP wrong will hold for D' D the demonstrative cases, illustrated in (75) . All that raises to AgrO is the NumP. If it is a AgrP count singular or a quantized plural, a terminative reading will arise. If it is a mass noun Agr' or a (bare) plural NumP, then the result is durative. Terminative readings in the latter here PPLOC tthis cases will be available if the demonstratum can be interpreted as an external boundary. P' there NumP man 4.4 Bare Nouns in Brazilian Portuguese and Aspect The analysis proposed so far groups together bare plurals129 and mass nouns in that both Here can be an overt locative element or a null locative anaphoric pronominal element force durative readings in relative clauses or by themselves, i.e., with or without which I will notate for the discussion as LOC. It raises overtly at least to Agr and probably modification. Singular count nouns and plurals with specified quantities, on the other from there to D. In (80), the NumP man is generated as a complement of a locative phrase hand, force terminative readings with or without relative clauses. Since in English both which theta-marks it as in the adjective cases above. At LF the NumP raises to AgrP to Proximity, etc., all these elements should be interpreted rather abstractly. have the agreement features on the LOC+Agr complex checked and from there the NumP 127 moves there to AgrO to check case. The LOC+Agr enters a Spec-head agreement relation 128 It is interesting to note that both prenominal possessives and demonstratives are the only elements that license null complement anaphora in English DPs, lending some support to giving them a parallel treatment. with the NumP, being able then to check its phi features. The demonstrative, which I will 129 232 Bare plural is used here to mean plural heads no specified for quantity. 233 mass nouns and plurals can be seen as names of kinds, it is possible to think that this (82) a. Eu escrevi carta por muitos anos alone would account for the durative readings in the relative clause, provided we adopt an I wrote letter for many years analysis for relative clauses in which the determiner is not a constituent with the noun. 'I wrote letters for many years' Under this reasoning, we might expect that in a language where bare singular count nouns can appear by themselves with a kind-like interpretation, singular count nouns as heads of relatives would also allow durative readings. I will test this hypothesis on b. Eu dei CD/ disco/ fita por muitos anos I gave CD/ record/ tape for many years 'I gave CDs/records/tapes for many years' Brazilian Portuguese. Besides a set of a singular and a plural indefinite, exemplified in (81a) and (81b), c. Eu toquei sonata por muitos anos respectively, Brazilian Portuguese, unlike the other Romance languages, allows bare I played sonata for many years plurals and bare singular count nouns in argument positions. The bare plurals and bare 'I played sonatas for many years' singular count nouns can, in general, appear in both subject position and object position, Now consider cases of relative clauses, shown in (83). According to our analysis, in as exemplified in (81c,d). (83a), for example, the head of the relative clause is carta 'letter', a singular count noun. (81) a. Eu comprei um livro. Given (82), we might expect that a durative reading should be possible; however, in (83), 'I bought a book.' only terminative readings are possible; durative readings do not arise. b. Eu comprei uns livros. (83) a. Eu escrevi a carta que o Pedro queria #por muitos anos/ em cinco minutos. 'I bought some books.' I wrote letter that Pedro wanted for many years/ in five minutes c. Crianças comem balas. b. Eu dei o CD que ele queria #por muitos anos. 'Children eat candies.' I gave the CD that he wanted for many years d. Criança come bala. c. Eu toquei a sonata que era mais complicada #por muitos anos. Child eats candy. I played the sonata that was more complicated for many years 'Children eat candy.' In other words, in the relative clauses, the bare count noun behaves more like a The bare singular count nouns, just like the bare plurals and the mass nouns, allow singular indefinite than as a bare plural or mass noun, as the contrasts in (84) illustrate. durative readings of the VP predicate, as illustrated in (82). 234 235 (84) quantized. It must therefore be the case that a bare noun in an argument position is not a a. ??Eu matei um coelho por três anos. I killed a rabbit for three years. (iteration is pragmatically odd) NumP, and I will devote the rest of the section to showing that, in fact, argumental bare nouns are best analyzed as DPs which select for NPs rather than NumPs. b. Eu matei coelho por três anos. I killed rabbit for three years 4.4.1 Heads of definite relative clauses are NumPs and not DPs or NPs 'I killed rabbits for three years.' (durative reading is OK) Suppose (for the moment without argument) that bare nouns are either NPs or DPs c. ??Eu matei o coelho que a Maria queria por três anos. 'I killed the rabbit that the Maria wanted for three years.' (iteration is pragmatically odd) without number information, and it is the lack of number that allows them to induce durative readings in argument positions. In the previous sections I argued that the definite determiner is special in the sense that it needs to have some referential element head raising to it at LF in order to license it. I also argued that N, C, locatives and certain Because the only way to interpret the sentence with the indefinite is to have iteration (i.e. kill a rabbit many times), (84a) has a pragmatically odd reading. (153b), on the other adjectives could license a definite determiner, because they are referential. Now consider the following DP basic structure: hand, is perfectly acceptable, under a reading that kill rabbits is something I did for three years. In (84c), on the other hand, the same oddness of (84a) arises again. (85) DP the The contrast shown above raises at least two questions: NumP [friends+Num]Num NP (i) why do singular count nouns that are heads of relative clauses and bare count tfriends nouns in argument positions behave differently, but mass nouns and bare plurals behave alike for matters of aspectual interpretation? The N head raises to the NumP head and from there the complex head (ii) what are bare noun arguments in Brazilian Portuguese: are they quantificational elements or are they like bare plurals? If the latter option is taken, the question is what is the difference between a bare singular, on one hand, and a bare plural and a bare mass noun, on the other hand. [friends+Num]Num raises to the definite determiner licensing it. However, because head raising is an adjunction process (see Chomsky 1993) what head-raises to D is not strictly N, but a Number projection. Thus, what is actually licensing the definite determiner is a combination of Number+N. It follows from this that N by itself cannot license a definite In what follows I will first begin with an answer to question (i) that depends on certain assumptions about the syntax of bare nouns, and show that, for feature checking reasons, the head of a relative clause with a definite determiner must be a NumP; it cannot determiner. Suppose we generalize this to the other cases discussed above: C, locatives and some adjectives alone cannot license a definite determiner but must bear number features to do so. be a DP or an NP. From this it follows that the lack of durative readings in relative In all of those cases it is a complex of a X+Number that is able to license a definite clauses with singular heads comes from the fact that singular NumPs are interpreted as determiner. In the case of the relative clause, for example, since the C head itself does 236 237 not bear number features, it must enter a specifier head agreement with a NumP in order lack of durative readings with singular heads on the assumption that the bare count noun to successfully license the definite determiner. This is mediated by the Agr projection lack number, and a NumP with a singular noun is interpreted as quantized. In the rest of above CP in the relative clause. Raising the NumP to Spec AgrP activates the Nominal the section I will provide evidence for treating bare count nouns as DPs without NumPs. features on the Agr, including Number, and the C+Agr that raises to D is therefore able to The similarities between bare nouns, mass nouns and bare plurals will follow from the license the definite determiner. lack of overt quantity information. Now suppose the head of the relative clause is an NP. As long as the NP does not In 4.1 I present the main evidence for treating bare count nouns as DPs with no have number features, it will not activate those features on the Agr and the C+Agr will not number information. In 4.2 I address the similarities between bare nouns, bare mass be able to license the definite determiner. The same argument will hold if the head of the nouns and bare plurals. They will form a clear contrast with the singular indefinites. I will relative clause is a DP with no number features. also discuss similarities between bare count nouns and bare mass nouns as opposed to Note that it does not follow from this analysis that the head of every relative clause bare plurals. This section is not to be seen as a complete study of the distinction between must be a NumP. Provided no definite determiner is to be licensed, then in principle no bare singular count nouns, bare plurals in Brazilian Portuguese, but rather as an argument problem arises because number features are not required. The example below shows that against a purely semantic account for the behavior of bare plurals and bare mass nouns in relatives with bare noun heads are perfectly acceptable and force a durative reading on the relative clauses. VP predicate when in complement position. 4 . 4 . 2 Evidence for Lack of Number in Bare Count Nouns (86) Eu comprei caderno que estava em liquidação por muitos anos We have seen that bare mass nouns and bare plurals behave identically inside relative I bought notebook that was on sale for many years clauses and by themselves, but singular count nouns display an odd behavior. Inside In sum, we cannot have the NP to license a relative clause with an overt determiner because that will create a situation where the C+Agr complex lacks number features and only a C+Agr that has checked features against a NumP can license a definite determiner. We cannot have a DP with no Number features as the head of a relative clause that has an relative clauses they act as indefinites and by themselves they act as bare plurals or mass nouns. In this section I will provide two pieces of evidence for the lack of number in bare count nouns, and their presence in bare plurals and bare masses. The first one is the behavior of conjoined bare nouns preceded by a definite determiner, and the second is the number features of cross-sentential anaphoric pronouns. The behavior of conjoined bare overt determiner for the same reason. Since bare plurals and bare mass nouns can be heads of definite relative clauses then they must be NumPs. The fact that their quantity is unspecified will produce durative nouns without a definite determiner will argue for bare count nouns as being DPs with empty Ds. readings. I have presented an argument that the heads of definite relative clauses must be NumPs in order to license the definite determiner. This can be used to account for the 238 239 4 . 4 . 2 . 1 Conjoined Bare Count Nouns: Evidence for DPs with no NumP Consider the contrasts between the (c) examples in (87) and (88) with the (a) and (b) relative. Crucially, this is not possible in the case of the bare plurals or the mass terms (the (a) and (b) examples.)130 The fact that conjoined singular count nouns can be interpreted as having identity of reference lends support to the idea that Number is necessary (but not sufficient) for the examples. referential interpretation of NPs.131 We can attribute the ambiguity in (87c and 88c) to (87) a. Os professores e alunos votaram em Clinton nas últimas eleições. 'The professors and students voted-PL for Clinton in the last election.' the possibility of either conjoining two NPs under one Num projection or two NumPs. Thus the structure of o amigo e parente can be either the structure in (89a) or (89b).132,133 b. O leite e açúcar devem ser bem misturados. 'The milk and sugar must be well mixed-PL' c. O amigo e parente votou na última eleição. The professor and student voted-SG in the last elections (89) a. DP D o DP NumP NumP amigo (88) a. Ele encontrou os amigos e parentes no aeroporto. b. D o BP B e NumP Num NP NumP NP parente amigo He met the friends and relatives at the airport. BP B e NP parente b. Ele sempre busca o café e leite no supermercado. The lack of identity of reference with bare plurals will follow directly from the fact He always picks the coffee and milk in the supermarket. that they bear number, and thus can only be realized with the structure of (89a). c. Ele encontrou o amigo e parente no aeroporto. He met the friend and relative in the airport. Examples (87c) and (88c), with singular count nouns inside a DP allow an interpretation in which the referent of the NP is the same. Thus we can interpret (87c) as being the professor who was also a student voted in the last election; similarly, o amigo e parente in (88c) can be interpreted as meaning the person who was both a friend and a 130 It may be possible to interpret conjoined NPs with a single determiner as having different referents (i.e. o amigo e parente could be different people.) This becomes clearer as the properties that the nouns denote become contradictory, e.g. o homem e cavalo são inteligentes 'the man and horse are intelligent' can mean that the man and the horse are intelligent, not that the centaur is intelligent. What is important for the present discussion is the inability of plural and mass NPs to take on the identical referent interpretation. 131 This observation is certainly not a new one; it can also be invoked to account for the lack of morphological number inside compounds or other derived words (rat catcher vs. *rats catcher). While standard analyses of these facts (e.g. Kiparsky 1982) have used level ordering to explain them, any theory that treats inflectional morphology as present in the syntax would derive the same effects. 132 The fact that the determiner will always agree with the first conjunct follows from the fact that the D governs the first conjunct. See Munn (1993; 1994b; 1995b). 133 I am assuming, following Munn (1992; 1993) that movement out of the first conjunct to license the D in ($89) is possible and does not violate the Coordinate Structure Constraint. See Munn (1992, 1993) for details. 240 241 Since bare count nouns do not bear morphological number, an overt manifestation of The only structure that is compatible with the interpretation of (91) in which amigo number is required when the determiner is null: an indefinite singular (a, for example) or and parente refer to different sets is given below. Of course, since (91) is ambiguous in the numeral one. Schematic trees are given below:134 the same way that (89) is, an alternative structure with conjoined NPs under a single D is also possible. (90) a. NumP pl+flores b. NP NumP uma NumP' (92) DP sg+flor NP DP D flores 'flowers' uma flor 'a flower' NP amigo BP e DP D NP parente The data from conjoined NPs show that conjoined singular NPs can either be NPs or NumPs, and in the former case, identity of reference obtains. We can now use this fact to show that bare nouns in argument positions are DPs with no NumP rather than simply amigo e parente friend and relative bare NPs. Consider (91). Because the D is empty, no NumP is necessary to license it. The empty D is enough (91) Eu encontrei amigo e parente no aeroporto. I met friend and relative at the airport 'I met friends and relatives at the airport.' to guarantee that the second conjunct is not just a property. The basic proposal is then that bare count nouns are DPs with empty Ds. By empty D I mean a D that at LF has no number information. We know it is not a singular NumP because it is not interpreted as such and because we have no overt marking of singular. Here it is not necessary that the friend and the relative are the same person. The So far I have not said much about mass nouns. However, given that conjoined bare difference between (91) and (88c) is the lack of an overt definite determiner. If bare mass nouns and bare plurals when preceded by a definite determiner do not allow the nouns were simply NPs, we would expect (91) to force the interpretation where I met reading in which they refer to the same entity, it must be the case that mass nouns have those people who were both friends and relatives. This fact provides us with some number information even though overt morphological marking does not show up on them evidence that it is not the case that bare count nouns are always just NP denoting in English or Portuguese. Delfitto and Schroten (1991, however, confronted with the properties. However, given that we still read (91) as having one or more friends and one same dilemma, show (citing Hall 1968 and Marcantonio 1978) that many Ibero- and or more relatives at the airport is evidence that we do not have a NumP in those cases. Italo-Romance languages show morphological mass/count distinctions. I will not repeat their data here, but will provide some additional evidence in support of the idea that mass 134 If a determiner or a quantifier head takes the NumP singular count as a complement or if some element that has referential properties takes the NumP as a complement and has overt N, the indefinite singular marking is deleted. That is why we do not say: the a man or some a man or a wrong person.. 242 noun DPs must have NumPs. 243 Bare plurals and mass nouns can conjoin with or without a shared definite determiner as shown in (93). Following Munn (1993 and references there, we can use conjoinability same properties. In both English and Brazilian Portuguese they are unacceptable, as illustrated below: a test for identity of semantic category. The fact that mass nouns and bare plurals are conjoinable therefore shows they are of the same semantic category. More importantly, (96) a. *The judge and lawyers however, the fact that the conjuncts are reversible shows that they are also of the same b. *O juiz e advogados syntactic category. c. *The lawyers and judge (≠people who are both lawyers and judges) d. *Os advogados e juiz (93) a. milk and bananas Recall that the interpretation of identity of reference for conjoined singular nouns was b. bananas and milk c. the milk and bananas dependent on their being NPs. We can take the lack of conjoinability in (96) to show that d. the bananas and milk NP and NumP are distinct semantic categories, the former being (roughly) a property and the latter a (roughly) a set of individuals. (94) a. Leite e bananas/ bananas e leite estão mais baratos. Milk and bananas/ bananas and milk are cheaper. b. Eu comprei bananas e leite/ leite e bananas. I bought bananas and milk/ milk and bananas. In Brazilian Portuguese, however, there is an interesting difference. It is possible to conjoin a bare singular noun with a bare plural. As in (96) when we do so, we never obtain a reading where the two conjuncts refer to the same person. The fact that the two are conjoinable, however, shows that they must be of the same semantic category. This shows that a bare noun in an argument position cannot be simply It seems then that the structure for the acceptable coordinated bare plurals and bare mass nouns is the following with D, overt or not. an NP, or else the unacceptability of (96) would be unexplained. Since we have already argued that it cannot be a NumP, the only possibility is the structure proposed in (92) in which both are DPs with empty Ds. This is in keeping with the interpretation of (97) which disallows the identity of reference reading, but allows the singular bare noun to (95) DP refer to one or more individuals. NumP NumP BP and NumP On the other hand, it is not possible to conjoin a bare count noun with a bare plural and obtain a reading where there is only one individual or a set of individuals sharing the 244 245 (97) a Promotores e advogados receberam presentes ilícitos. Prosecutors and lawyers received illicit presents b. Advogados e promotores compraram uma testemunha Lawyers and prosecutors bought a witness One of the arguments Carlson uses to show that the existential and/or the generic reading is not a property of the bare plural itself or the bare mass noun is the behavior of anaphoric pronouns. Existential bare plurals and generic bare plurals can act as antecedents for pronouns. These pronouns in turn, depending on the context, can receive an existential reading or a generic reading: In sum, the conclusion we can reach so far is that Brazilian Portuguese bare count nouns are DPs with empty Ds and no number information. Both bare plurals and mass nouns, (98) a. Mary hates raccoons because they stole her sweet corn, so she now hates them with a passion. on the other hand, have number information. b. Raccoons stole May's sweet corn, so she now hates them with a passion. Additional evidence for lack of number in bare count noun DPs comes from crossc. My brother thinks that snakes are nasty creatures, but that hasn't stopped sentential anaphora, a test used for Carlson (1977a) to show that bare plurals are not inherently existential or generic. me from having them as pets. d. I've had snakes for pets my whole life, but my brother still thinks they are nasty creatures. 4 . 4 . 2 . 2 Anaphora Carlson's analysis for bare plurals is that they are names of kinds. Kinds serve to tie In (98a) raccoons receives a generic reading, the anaphoric subject pronoun receives together objects or stages of objects. The interpretation of the bare plural will depend on an existential reading and the object pronoun receives a generic reading. In (98b) the predicate. The existential reading arises whenever there is an existential claim being raccoons receive an existential reading, and the pronoun has a generic reading. (c) and (d) made about one or more of its stages. In other words, the existential quantifier is illustrate the same alternations. associated with the predicate and not with the bare plural. Predicates that force the In Brazilian Portuguese, pronouns in subject position can be null or overt. They are existential reading of the bare plural are happenings and apply to stages of things via a obligatorily null when the pronoun is a bound variable; otherwise they tend to be overt. I function that Carlson calls the Realization function. The realization function makes will concentrate my attention on the subject position, because in object position matters 'available" entities of a lower level from those of a higher level. Predicates that force the are more complicated. In object position beside a nominative pronoun (subject to some universal reading of the bare plural are characteristics: be intelligent, weigh etc. register restrictions and preferences), a null object (which does not correspond to a null Characteristics of things are predicates that apply directly to individuals (which is pronominal element) is possible.135 whatever ties together a series of stages). The generic reading of children run is obtained via a generic operator that makes available predicates from a lower level to those of a higher level. 135 The accusative pronoun is not used in the spoken language in Brazilian Portuguese as we have seen before. 246 247 (99) a. Maria detesta coelhos porque eles /?*ø roubaram suas cenouras. Maria hates rabbits because they//??ø stole her carrots. A singular null or singular overt pronoun is unacceptable as anaphoric to the bare singular count noun. The overt plural pronoun, on the other hand, is possible, as illustrated in (101). b. Coelhos roubaram as cenouras da Maria; por isso agora eles fazem parte da sua lista de inimigos. Rabbits stole Maria's carrots; that is why now they are part of her enemies list. (101) a. Maria detesta coelho porque *ø/eles roubaram suas cenouras. Agora ela detesta eles de coração. Maria hates rabbit because *ø/they ate her carrots. Now she hates them with a passion. c. Eu tive cobras como animais de estimação durante toda a minha vida, mas meu irmão ainda acha que elas/??ø são criaturas nojentas. b. Coelho sempre rouba as cenouras da Maria, por isso agora ela detesta eles I've had snakes for pets my whole life, but my brother still thinks they are de coração. nasty creatures. Rabbit always steals the carrots of the Mary, for this reason now she hates them with a passion. The data below shows that the null pronoun is not acceptable but the plural overt c. Meu irmão acha que cobra é criatura muito nojenta, mas isto não me impediu de ter ??ø/elas como animais de estimação. pronoun is perfect. Now let's consider the bare singular noun: My brother thinks that snake is a very grosse criature, but this didn't stop me from have ??ø/them as pets. (100) a. Maria detesta coelho porque *ø/*ele roubou suas cenouras. Maria hates rabbit because *ø /it stole her carrots. b. Coelho sempre rouba as cenouras da Maria, por isso agora *ø/*ele faz parte da sua lista de inimigos. d. Eu tive cobra como animal de estimação durante toda a minha vida, mas meu irmão ainda acha que elas/??ø são criaturas nojentas. I had snake as a pet all my life, but my brother still thinks that they/??ø are nesty creatures. Rabbits stole Maria's carrots; that is why now *ø/*it is part of her enemies There is one case in which the singular count noun can serve as an antecedent for a list. singular pronoun. Such case is illustrated in (102). Existential constructions allow a c. Eu tive cobra como animal de estimação durante toda a minha vida, mas singular or a plural pronoun to be anaphoric to the bare count noun. meu irmão ainda acha que ela/ø é uma criatura nojenta. I've had snake for pets my whole life, but my brother still thinks *it/ø is nasty creatures. (102) a. Tem criança na sala. E ela está/ elas estão ouvindo. There is child in the room. And she is/they are listening. 248 249 The fact that the bare count noun can serve as the antecedent for a plural or a singular b. Eu vi criança na sala. E ela estava / elas estavam ouvindo. I saw child in the room. And she was/ they were listening. pronoun in existential constructions and a plural in the other cases is highly suggestive that the bare count noun is not singular but unspecified, i.e., it has no number Mass terms behave slightly differently in cross-sentential anaphora. They require singular pronouns and in my dialect of Brazilian Portuguese can be null or overt. The information. The mass noun, on the other hand, counts as singular from the syntactic point of view. plural pronoun is completely unacceptable. From the discussion above, it seems that bare count nouns allow the same range of interpretations that the bare plural and the bare mass noun allow. In the next section I will (103) a. Maria detesta arroz porque ø/?ele tem muitas calorias. Agora ela só come ?ele quando não tem escolha. show that all the bare nouns behave similarly based on their uniform behavior in matters of opacity and scope. For that I will be following Carlson (1977a) quite closely. Maria hates rice because ø/?it has many calories. Now she only eats it when she doesn't have a choice. 4 . 4 . 3 Similarities between Bare Nouns and Bare Plurals b. Café sempre custa caro, por isso agora ela detesta *ele de coração. Carlson (1977a) starts his discussion on bare plurals by showing that (i) they do not Coffee always costs a lot, that is why now she hates it with a passion. pattern as the plural counterpart of the indefinite a ; and (b) they do not pattern as if they c. Meu irmão acha que sabão é anti-ecológico, mas isto não me impediu de achar que ø/?ele deve ser comprado. had either a null generic and/or existential quantifier associated with the NP itself. This is done by examining bare plurals in opacity-inducing contexts and bare plurals in matters of My brother thinks that soap is anti-ecological, but that didn't stop me from scope. In this section I compare the behavior of bare nouns and bare plurals in Brazilian thinking that it should be bought. Portuguese for opacity and scope effects. 4 . 4 . 3 . 1 Opacity and Scope136 d. Eu tive cimento na garagem por muitos anos, mas agora ø/?ele não pode ser usado porque ø/?ele está molhado. In opacity-inducing contexts, the bare plural cannot take scope over an opacity inducing I've had cement in the garage for many years, but now it cannot be used because it is wet. verb. The bare count noun behaves like the bare plural, while the indefinite singular can have both narrow and wide scope readings. Consider the following examples: Summarizing, bare count nouns, bare plurals and bare mass nouns behave partially alike in cross-sentential anaphora, all three bare nouns allow existential and generic pronouns to corefer with it. Bare plurals serve as antecedents for plural pronouns, and mass nouns serve as antecedents for singular pronouns. 136 250 All the examples below are adapted from Carlson (1977). 251 (104) Pedro quer encontrar um policial. In matters of scope, the examples below illustrate the fact that while the singular Pedro wants to meet a policeman. indefinite can but need not take scope over negation, the bare plural and the bare count a. (∃x) policeman(x) & Pedro wants (Pedro meet x) (transparent reading) b. Pedro want ((∃x) policeman(x) & (Pedro meet x)) (opaque reading) (105) a. Pedro quer encontrar policiais. (opaque reading only) Pedro wants to meet policemen. b. Pedro quer encontrar policial. noun have to stay under the scope of negation: (107) João não viu uma mancha no chão. João didn't see a spot on the floor. a. João saw no spots on the floor. (narrow scope reading) b. There is a spot João didn't see. (wide scope reading) (108) a. João não viu manchas no chão. (narrow scope reading only) (opaque reading only) Pedro wants to meet policeman. João didn't see spots on the floor. (106) a. Pedro está procurando um livro de culinária. Pedro is looking for a cooking book. (opaque & transparent readings) b. João não viu mancha no chão (narrow scope reading only) João didn't see spot on the floor. b. Pedro está procurando livros de culinária. Pedro is looking for cooking books. (opaque reading only) The examples below show the same point. The bare count noun, just like the bare plural, seems to be forced to have narrow scope, although the indefinite has both wide and c. Pedro está procurando livro de culinária. narrow scope. In (109) the bare noun and the bare plural must have narrow scope but the Pedro is looking for cooking book. (opaque reading only) indefinite can have a wide scope reading. (104) has two readings. In the transparent reading (the term is borrowed by Carlson (109) a. Todo mundo leu um livro sobre girafas. (narrow and wide scope reading) from Quine 1960), there is some existing policeman that Pedro wants to meet. In the Everyone read a book on giraffes. opaque reading, Pedro wants to meet some policeman or other. In (105), the transparent reading (i.e., in the reading where the existential quantifier is outside the scope of the b. Todo mundo leu livros/livro sobre girafas opacity inducing verb) is absent both for the bare count noun and for the bare plural. Everyone read books/book on giraffes (narrow scope reading only) There is no ambiguity. The same fact is illustrated in (106). With respect to opacity inducing effects, the bare count noun and the bare plural are behaving alike. In (110), only (110a) has a reading in which there is a specific dog that was seen in the garden, and only (111a) has a reading in which there is a specific hot-dog stand where the majority of students ate. 252 253 (110) a. Pedro viu um cachorro no seu jardim às 3, às 4 e às 5 da tarde. Pedro saw a dog in his garden at 3, 4 and 5 in the afternoon. The second point of Carlson's arguments is that bare plurals could not be the surface form of a generic quantifier + NP and/or an existential quantifier +NP, because this would predict a multiplicity of readings that is in fact disallowed. For example, if there b. Pedro viu cachorros no seu jardim às 3, às 4 e às 5 da tarde. Pedro saw dogs in his garden at 3, 4 and 5 in the afternoon. were an existential quantifier associated with the bare noun, in the examples below, existential readings should be possible, contrary to fact. c. Pedro viu cachorro no seu jardim às 3, às 4 e às 5 da tarde. (112) a. Cachorros gostam de gente. Pedro saw dog in his garden at 3, 4 and 5 in the afternoon. Dogs like people. (111) a. A maioria dos estudantes comeu numa carrocinha de cachorro quente b. Cachorro gosta de gente. The majority of the students ate at a hot-dog cart. Dog likes people. b. A maioria dos estudantes comeu em carrocinhas de cachorro quente 'Dogs like people.' The majority of the students ate at hot-dog cart. (113) a. O João detesta crianças. c. A maioria dos estudantes comeu em carrocinha de cachorro quente The João hates children. The majority of the students ate at hot-dog cart. b. O João detesta criança. So far bare count nouns and bare plurals are patterning alike, i.e., they fail consistently to allow the range of interpretations that the indefinite allows in opacity inducing contexts and in terms of scope. Thus, as Carlson points out, the bare plural cannot be taken as the plural of the indefinite, and similarly the bare count noun cannot be The João hates child 'The João hates children.' (114) a. Elefantes são facilmente domesticáveis. Elephants are quite easily trained. taken (perhaps unsurprisingly) as an indefinite subject. In the next section I will show that to propose that the bare count noun has a null generic and/or an existential quantifier associated with it is not going to help to explain its behavior, in the same way as Carlson has shown that null quantifiers will not help to b. Elefante é facilmente domesticável. elephant are quite easily trained 'Elephants are quite easily trained.' account or explain the behavior of the bare plural. Now consider the converse case, i.e., contexts that only allow an existential reading 4 . 4 . 3 . 2 Lack of quantificational readings 254 of the bare plural. If there were a null generic operator associated with the bare plural or 255 the bare noun, generic readings should be possible. Yet again only the existential reading is possible. (118) a. Texans are often tall. b. Many Texans are tall. c. #all Texans/most Texans are often tall (115) a. Eu notei crianças no ônibus. I noticed children in the bus. The same facts hold for bare count nouns and bare plurals in Brazilian Portuguese, as illustrated by (119) and (122). The interpretation of (a) is given and (b) and (c) illustrates b. Eu notei criança no ônibus. the oddity of the adverbial with a quantificational subject and an predicate that ranges over I noticed child in the bus. individuals: (116) a. Eu acho que vi livros espalhados pelo chão. I think that (I) saw books spread around on the floor. (119) a. Brasileiro é freqüentemente baixo. Brazilian is often short. b. Eu acho que vi livro espalhado pelo chão. I think (I) saw book spread on the floor. b. Many Brazilians are short. c. *Todo brasileiro/ algum brasileiro é freqüentemente baixo (117) a. Tem criança na sala. all brazilian/some brazilian is often short There is child in the room b. Tem crianças na sala In sum, positing a generic and/or an existential null quantifier in the bare noun cases There are children in the room. is as awkward and unrewarding as postulating it to the bare plurals since postulating null quantifiers would predict ambiguities should arise where they do not, as Carlson points One could argue that maybe the predicate can select for the generic or the existential out for the bare plural cases. Thus, it seems that the bare nouns in Brazilian Portuguese quantifier, depending on the context. However, there is another case pointed out by are amenable by the same analysis of bare plurals, namely, their interpretation depends on Carlson that I believe strongly argues against the generic quantifier. The example is given the predicate. below. In (118a) the adverbial often is quantifying over the subject NP of the sentence as It should be pointed out that mass terms, with respect to the properties discussed the paraphrase in (118b) suggests. If there were a quantifier associated with the NP above (i.e., opacity and scope), behave alike, as the examples below from Carlson Texans (all or most, for example), the adverb cannot quantify over the NP as (118c) (1977a) illustrate: illustrates. 256 257 (120) a. Bill believes that furniture is in Belle's attic. (opaque reading) 4 . 4 . 4 Differences between the Bare Count Nouns, Bare Mass Nouns and Bare Plurals b. Everyone saw furniture yesterday. (narrow scope only) c. Rain fell over the city. (existential reading) In the following I will discuss two cases in which bare count nouns and bare mass nouns (generic reading) behave alike and distinctly from bare plurals. The goal here is to show that there are d. Milk is white. differences between bare plurals and bare count nouns and also to show that while bare Given the similarities between bare count nouns, bare plurals and bare mass nouns, we have two options with respect to their internal structure: either they are structurally identical, i.e., they are both DPs with empty Ds and this accounts for the semantic similarities or they don't have the same structure but share clear semantic properties. plurals require always a NumP to check its morphology, bare mass nouns can be generated with or without morphological number information. Only in the second case they will require a NumP. In other words bare plurals are unambiguously NumPs but bare mass nouns are ambiguous between a NP and a NumP. Suppose we assume that all bare arguments have an empty D. The D would provide a variable that can receive a generic or an existential reading, depending on the predicate 4.4.4.1 Bare Noun and Mass Noun Restrictions that attaches to it. This would leave us with the cross-linguistic problem of accounting for Here we will see that the bare mass noun and the bare count nouns are not as freely the lack of bare count nouns in English. One possibility is to say that what licenses an generated in subject position as the bare plurals are. empty D in English is Number features. Lack of number will disallow a bare count noun If the verb is in the simple present, or the imperfective past in Brazilian Portuguese to have an empty D and thus to appear in an argument position in English. Brazilian (which roughly corresponds to a non-punctual, habitual past), the generic reading of the Portuguese, on the other hand, can license empty Ds and thus count nouns can appear bare noun and of the bare plural is the only possible reading. bare in argument positions. In Spanish empty Ds are only possible if there is a quantifier like todo binding the empty D, as we have seen in chapter 3. Alternatively, we could say that bare nouns are DPs with empty Ds, but bare plurals are NumPs and bare mass nouns are NumPs. The similarities in terms of scope and (121) a. Criança faz muito barulho. child makes a lot of noise 'Children make a lot of noise.' opacity come from the lack of specified quantity information. If there is no quantity information we can't say how many and this may be enough to account for the narrow scope only readings; the ability to be interpreted as generic or existential comes from the lack of an overt D or Q. At this point it is very hard to distinguish between both b. Crianças fazem muito barulho. children make a lot of noise 'Children make a lot of noise.' hypotheses and I will leave the matter open. What is important is that in all three cases, in spite of the fact that the noun appears to be bare, there is at least one functional projection above the NP, so that it can function as an argument. 258 259 c. Ele disse que amigo tinha direiro à uma explicação. he said that friend had-IMP right to an explanation b. Eu acho que vi livro espalhado pelo chão I think (I) say book spread on the floor 'He said that friends had rights to an explanation.' Sometimes a generic reading for the bare plural in subject position is impossible because d. Ele disse que amigos tinham direiro à uma explicação. the predicate makes explicit reference to stages. In those cases the bare plural is acceptable he said that friends had-IMP right to an explanation with an existential reading, but the bare noun, when explicit reference to stages is made is 'He said that friends had rights to an explanation.' quite odd This is the case of bare count nouns in subject position of predicative constructions with the copula estar (see chapter 6), for example, as exemplified in (125a). As we have seen bare count nouns can easily have existential readings in there This oddness is similar to the one one finds with mass nouns, as illustrated in (125b) for constructions and as subjects of small clauses that are complements of perception verbs. English and Brazilian Portuguese. Examples were given in (115), (116) and (117) and are repeated in (122) and (123) and (124) below: (125) a. ??Flor está no vaso. flower is in the vase (122) a. Tem criança na sala. b. ??Leite está na geladeira. There is child in the room 'Milk is in the fridge.' b. Tem crianças na sala There are children in the room. Although the examples in (125) are by themselves quite odd, they become fully acceptable if they are part of a list reading. For example, (125b) is fully acceptable in a (123) a. Eu notei crianças no ônibus context where the host tells the guest to help himself in the morning and goes around in I noticed children in the bus the kitchen showing where things are: milk is in the fridge, coffee is in the cabinet, sugar b. Eu notei criança no ônibus is in the bottom shelf, etc. The same is true for the cases with the bare count noun, as the I noticed child in the bus glosses below illustrate: (124) a. Eu acho que vi livros espalhados pelo chão (126) a. ??Flor está no vaso. I think that (I) saw books spread around on the floor Flower is in the vase 'flower is in the vase, plant is in the pot...' 260 261 Note that negation and adverbials like always immediately make available a b. Leite está na geladeira. contrastive reading and the bare noun is more natural in those cases. Milk is in the fridge 'milk is in the fridge, coffee is the cabinet...' (129) a. Mulher não esteve discutindo política. Woman was discussing the fight If the bare noun is the subject of a perfective past eventive verb (roughly a punctual 'woman was not discussing politics' past), a generic interpretation of the subject is odd, unless again the list reading becomes available, as illustrated by the glosses. b. Greve sempre foi considerada ilegal pelo governo. Strike was always considered illegal by the government (127) a. Mulher esteve discutindo política. woman was discussing the fight (130) a. Homem não chegou tarde. 'Woman was discussing the fight, man was discussing soccer, etc.' Man didn't arrive late 'Men didn't arrive late, women did.' b. Greve foi considerada ilegal pelo governo. strike was considered illegal by the government b. Homem sempre chegou tarde. 'Strike was considered illegal by the government, drug dealing was Man always arrived late considered legal, etc.' Similar examples with negation can be construed for the bare mass nouns. c. Homem chegou tarde. man arrived (131) Snow didn't melt fast today. man arrived, woman left Note that the bare count noun in the examples (125) and (127) can receive a regular In similar cases the mass noun is also less acceptable both in Brazilian Portuguese and English, unless the list reading becomes available. existential interpretation, if the subject is embedded in a there construction, or if the subject is postposed, as is possible for subjects of unaccusative verbs. Here we do not need to set a list reading. Examples are given in (132): (128) Estava tão quente ontem que plástico derreteu, etc. It was so hot yesterday that plastic melted, etc. vs. ??It was so hot yesterday that snow melted. (132) a. Tem flor no vaso. there is flower in the vase 'There is a flower/ are flowers in the vase.' 262 263 b. Tinha mulher discutindo a briga ontem. With respect to object positions, again the bare noun is less acceptable in a context there was woman discussing the fight where specific reference is made to individuals, unless it can be interpreted as a member 'There was a woman/were women discussing the fight.' of a list. c. Tinha greve considerada ilegal pelo governo. (133) a. Alice conhece músicos pessoalmente. there were strike considered illegal by the government Alice knows musicians personally 'there were strikes considered illegal by the government.' 'Alice personally knows musicians.' d. Chegou homem. b. Alice conhece músico pessoalmente (não ator) arrived man Alice knows musician personally (not actor) 'There arrived a man.' 'Alice knows musicians personally not actors' In sum, with respect to subject positions, the facts can be summarized as follows: (i) if the tense/aspect of the eventive verb allows generic readings, the bare noun will be interpreted as generic; The only acceptable reading for (133b) is a contrastive focus reading, as the glosses above show. If negation is added or some modality is added, again the acceptability improves (iii) if the aspect of the eventive verb is incompatible with the generic operator (i.e. the easily. verb is in the perfective, for example), the bare noun is acceptable with if a list reading is allowed or if some operator can take scope over it; (ii) if the bare noun is in a there construction, or is the subject of a small clause under a perception verb existential readings are readily available. The generalization that emerges from (i) to (iii) is that the bare noun and the mass (134) a. Alice não conhece músico pessoalmente. Alice doesn't know any musician personally. b. Alice não pode conhecer músico pessoalmente. Alice can't know musician personally nouns, unlike the bare plurals, are not as freely licensed. They seem to need to be bound by a generic operator or by some other element in the sentence. Of course a much more The presence of plural morphology seems to be crucial to define the acceptability of detailed study is necessary here to understand exactly what is going on with the bare the bare noun in stage-level predicates that refer specifically to stages. The lack of this count nouns and the list reading, which I believe is related to the ability to establish some quantity information and discrete partition seems to be crucial in those cases. Although I contrastive focus (in the sense of Rochemont, 1986 and Rooth 1992). All I want to show don't have an account for the properties above, it seems to me that it should be treated on is that bare nouns, just like the mass nouns, seem to depend on something extra to be a par with the Homogeneity Constraint, discussed for bare mass nouns. acceptable in normal subject position. 264 265 In the literature on bare mass nouns, it is a common observation that bare mass nouns can only be subjects of homogeneous predicates (see Higginbotham 1994 and references (137) a. Tem sorvete de 30 sabores There is ice-cream in 30 flavors. there). The Homogeneity Constraint, as it has been called, would explain the unacceptability of (135a) and the acceptability of (135b) for Brazilian Portuguese and b. *Sorvete é de 30 sabores Ice-cream is in 30 flavors. English respectively. I do not have an explanation for the distinct behavior in existential constructions but it (135) a. *Ouro pesa duas gramas. is probably linked to the ability of the singular count bare noun to be the antecedent for a *Gold weighs 2 grams. singular pronoun, as we have seen in (102). b. Ouro é caro. 4.4.5 Summary Gold is expensive. Suppose we take the generic operator to be an operator that creates homogeneous predicates since they will force a number of subcases that are frequent enough to apply to the kind and not only to specific stages of a kind. Notice that here we are definitely not operating at the VP level but at the IP or even the CP. One way of thinking of the similarity between the bare mass nouns and the bare count nouns is that they both require homogeneous predicates and that explicit reference to stages will disallow the "operation of generalization" to apply. Thus in a sentence like the following, we can only get a reading in which children in general weighs 50 kilos. In this section I have provided evidence that bare nouns in Brazilian Portuguese are best analyzed as DPs with no number information. This, combined with the fact that the head of a relative clause with a definite determiner must be a NumP allows us to explain why Brazilian Portuguese does not allow durative readings with singular headed relative clauses despite the fact that bare nouns by themselves produce durative readings in object position. In the final section of this chapter I will provide an independent argument for determiner transparency based not on aspect, but on Antecedent Contained Deletion. 4 . 5 Antecedent Contained Deletion as an argument for Determiner Transparency137 In the previous sections, I presented a theory for determiner transparency that was able to (136) Criança pesa 50 quilos account for the aspectual properties of apparently definite DPs. In this section I want to Child weighs 50 kilos present an independent argument in support of the theory of determiner transparency. I 'Children weigh 50 kilos.' will argue that determiner transparency is exactly what is needed to avoid the infinite In existential constructions or under perception verbs the Homogeneity Constraint regress problem posed by Antecedent Contained Deletion (ACD) facts in a theory that doesn't seem to hold. 137 This section is based on the discussion in Schmitt 1995c, but with a more elaborated structure for the relative clauses as motivated by the preceding sections. 266 267 dispenses with Quantifier Raising (QR), at least as proposed by May (1985. This section IP, the VP is free to be copied after QR because it only contains the verb and the trace of also has the goal to make a bit more precise the assumptions I am making about the copy the raised NP, and no infinite regress arises, as shown in (139). theory. The section is divided as follows: in 5.1 I review the basic reasons to prefer a solution (138) a. John kissed John's mother and Sally did [VP e ] too. b. John kissed everyone that Sally did to the infinite regress problem in ACDs that relies on A-movement to AgrO as opposed to a QR solution which relies on A'-movement, and the basic arguments for copies being (139) a. [IP [everyone [that [Sally did [VP e]]]]i Bill kissed ti]] necessary even in A-movement operations. In 5.2, I make a proposal to account for ACD b. [IP everyone [that [Sally did [VP kiss ti]]]]i [IP Bill kissed ti]] constructions in terms of a parallelism requirement between the AgrO in the relative clause and the AgrO in the matrix clause. In 5.3 I will support this proposal, based on facts that May's solution crucially relied on traces not being copies. Under the assumptions of have been left unaccounted for in previous theories: pied-piping, coordinated objects, Chomsky (1993 QR (which forms Operator-variable structures) is subject to the certain quantifier restrictions and certain idioms. Preference Principle Minimize Operator, which, under one of its interpretations, forces the relative clause to be interpreted in its base position. If this assumption is correct, then 4 . 5 . 1 On the Necessity of Copies the infinite regress problem arises again, as exemplified in (140b) (deleted material is In order to eliminate the need for reconstruction, Chomsky (1993 reintroduces the copy striked out). No such problem arises if the Preference Principle does not need to apply. theory of movement in which the trace of movement is a copy of the moved element, However, there are other problems with the QR approach. deleted by a principle of the PF component. The PF deletion part of the operation is to be (140) John ate everything that Bill did subsumed under other cases of ellipsis that require a parallelism of some kind (PARR) a. John [everything that Bill did] John ate t between the 'antecedent' and the deleted material. At LF, the copy remains, providing the material for 'reconstruction', a reflex of the formation of operator-variable constructions. b. John [everything that Bill did] John ate [everything that Bill did]TR In light of these principles I argue for an account of Antecedent Contained Deletion Hornstein (1994; 1995) and Lasnik (1993, using minimalist assumptions, argue that focused on ensuring recoverability of a deleted VP, instead of providing an antecedent for ACD is better analyzed as the result of object raising to AgrOP, an instance of A- a base generated empty VP as in May (1985) and Hornstein (1994, 1995). movement as opposed to an instance of A'-movement. Thus for Hornstein movement to Antecedent Contained Deletion has long been an argument for LF raising of quantified expressions (May 1985). The basic idea is that, although the VP gap in (138a) is identical AgrO, obligatory for case reasons, will create the necessary empty VP, just as QR did. Thus May's intuition is kept pretty much intact, as illustrated in (141): to kiss John's mother, in (138b) one cannot copy the VP as it appears at S-structure because the result would be kissed everyone that Sally did [kiss everyone that Sally did e], yielding an infinite regress. Since QR moves a quantified NP at LF and adjoins it to 268 269 (141) a. John bought everything that you did [e] b. Johnj [T [AgrO'' [everything that you did [e]]i [AgrO [VP tj buy ti]]] solution for the infinite regress problem in ACDs. Being an instance of a required Amovement it does not leave copies according to Chomsky (1993) and therefore no infinite regress problem arises. The idea that A-movement does not leave copies is adopted in The AgrO solution has a few advantages over the QR solution. First it solves the Hornstein's (1994 paper. However it can be shown that there are empirical reasons to Boundedness Restriction on QR pointed out by Baltin (1987 and illustrated in (142) believe that A-movement leaves copies as has been pointed out by Hornstein (1995) below. In (142) the reading in which the VP is think that Fred read is impossible. The following Aoun (1981) and Lebeaux (1987; 1994). Lebeaux's (1994) argument is based only possible reading is the one in which the head of the VP is read. on the ambiguity in (144): Under a QR/wh solution, it is unclear why the whole QP had to stay in the downstairs clause. Under an AgrO movement account for ACDs, this fact follows naturally from the assumption that the movement of the DP happens obligatorily for case reasons. Once the (144) Two women seem t to be expected to t to dance with every senator (ambiguous) wide scope of every senator possible DP has case, then it has no reasons to move any further, and therefore it does not move: (144) can mean that there are two women who dance with every senator and it can (142) Who thought that Fred read how many of the books that Bill did (Ok with read but not with think) also mean that for every senator there are two women who dance with him. Given the ambiguity in (144), it must be the case that the quantifiers are in the same clause by the time they are interpreted. Rather than moving every senator to sentence initial position, The AgrO solution for ACDs also explains why subjects of tensed clauses cannot Lebeaux argues that in (144) two women must reconstruct into the scope of every provide a site for ACD (143a), but subjects of ECM clauses can (143b). In tensed senator. The evidence comes from the interaction between reflexives and quantifier clauses, the subject has case and therefore doesn't need to move. The subject of interpretation. When a reflexive traps two women in the upstairs clause, the inverse uninflected complements, however, raises to AgrO of the upper verb to check case. In scope reading is unavailable. The unambiguousness of (145) shows that the wide scope such case, a site for ACD is provided. reading in (144) must be due to reconstruction rather than scoping the universal quantifier to the matrix clause. (143) a. *John believed that everyone you did was a genius b. John believed everyone you did to be a genius (145) Two women seem to each other t to be expected t to dance with every senator wide scope of every senator impossible Second, if QR is to be maintained in the minimalist framework, the movement must be forced by some morphological feature that needs to be checked in some higher (146) shows that it is not upward movement of every senator that is crucial in the position. In other words, if QR is to be maintained, we need an extra landing site and preceding example, but rather the possibility of reconstruction of two women down. In extra features for the QP. Movement to AgrO seems to be a good alternative to the QR 270 271 (146) only one reading is allowed, namely, the reading where two women takes scope 4.5.2 ACD and identity of AgrOs over every senator. In this section, I depart from Munn's (1994a treatment of relative clauses under the copy (146) Maryi seems to two women ti to be expected ti to dance with every theory and, based on that, I formulate a proposal for how to deal with PF deletion and senator. LF recoverability in ACD constructions. Munn (1994a argues that traces of relative clauses must contain a copy of the operator Given Lebeaux's (1994 argument, and interpreting 'reconstruction' to mean 'leaves a that moves, in order to account for reconstruction effects in relative clauses. Evidence for copy', it becomes obvious that both A'-movement and A-movement solutions for ACD this comes from (147a). Principle A reconstruction is obligatory in (147a) since LF must be compatible with copies and yet must circumvent the infinite regress problem. The cliticization of the reflexive applies from the trace site. For cases like (147b), however, if problem is how we can avoid the infinite regress and still adopt the principle that there are we follow the principle of minimize operator, it is necessary to delete everything but the copies. Hornstein's (1995) proposal assumes both the need for copies in A-movement operator and keep a copy of picture of Bill in the trace site, which should induce a operations and movement to AgrO as the means to avoid the infinite regress. The basic Principle C violation. To account for the acceptability of (147b) Munn proposes that, idea is that A-movement leaves copies but is not subject to the preference principle unlike wh-moved elements, however, both the operator and the trace site copy can be motivated by A'-chains. Rather, any member of an A-chain can be deleted, and all but deleted since they can be recovered by the head of the relative clause and thus no Principle one must be. In addition to this requirement, Hornstein makes two more assumptions: C effect arises.138 first, he assumes Diesing's (1992) mapping hypothesis, according to which strong determiners have to be interpreted outside VP. Second, he also assumes that subjects in (147) a. the picture of himself that Bill likes ACD constructions must always have wide scope, since in order for ACD to obtain the b. the picture of Billi that hei likes139 subject cannot be interpreted inside the VP shell. The VP shell must be empty of its c. [ [the picture of Bill]H [CP [ which picture of Bill]OP that he likes [which picture of Bill ]TR] ] subject, which will always force a wide scope reading of it with respect to the object. Both May's and Hornstein's solutions are based on the assumption that ACD must provide an antecedent for an empty VP. In keeping with minimalist assumptions I will depart from the assumption that the problem in ACDs is not providing an antecedent for an empty VP in as much as ensuring recoverability of a deleted VP. The advantages of this proposal over Hornstein's (1995) will become evident in the third section. 138 For ease of discussion, I will maintain his examples and his use of H to refer to the head of the relative; OP to refer to the Operator position, and TR to refer to the trace site. 139 Norbert Hornstein pointed out that the following examples are unacceptable, which seem to put Munn's proposal in jeopardy: (i) a. *my interest in Bill which he likes b. *the destruction of the city which it deplored Although I don't have an explanation for the contrast between ($147) and (i) above, it should be noted that the cases in (i) are cases of appositive relatives. 272 273 In (148), a genitive clause, however, the operator is different from the head of the very simple: we need identical AgrOs at LF in order to be able to delete AgrOP in the relative clause. In this case, the operator cannot delete and a Principle C violation will lower clause. By identical AgrOs I mean that the head of the extended projection of the arise: specifier of AgrO has to be identical to the head of the specifier of the downstairs AgrO and the Vs have to be identical: (148) a. *The man whosei father hei saw b. [The man] [whose father]OP he saw [whose father]TR (150) LF configuration for recoverability of PF deletion of AgrOP A PF deletion of an AgrOP is recoverable if at LF there is an AgrOP For relative clauses, recoverability is the criterion for deletion of an operator and its copy in the trace site. Chomsky and Lasnik (1991) assume that ellipsis involves PF which instantiates the same Spec-head relation as the AgrOP that was deleted. deletion of elements of a phrase marker under identity. If we assume that in simple cases Conceptually, deletion, as opposed to insertion, seems more compatible with the bare of ACD, the structure is (149b), the question is what allows the deletion of 'VP' and its syntax proposal of Chomsky (1994). The idea of filling a VP later seems to be at odds recoverability at LF. In other words, what element is identical so that deletion is allowed? with the idea that one does not insert complements in the course of the derivation, if this (149) a. John read everything Bill did b. John read [[everything]H [everything]OP [that Bill did read [everything]]TR]DP mechanism of interpreting elided material is to be syntactic at all. What I am using as an antecedent for the AgrO constituent deleted at PF is not itself a constituent at LF. It contains (i) the complex chain V+Agr and (ii) the object that has to have its case checked in the specifier of AgrO. Since I am assuming that ellipsis involves If ACD deletion requires identity between the VPs, something has to be done because in the matrix clause we have everything that Bill read everything as the complement of PF deletion of structure rather than copying some constituent, it is not crucial that the antecedent for the deletion be a constituent at LF or PF. read and in the relative clause we have everything as the complement of read. In order to Note that at PF the whole AgrO has to be deleted and not just part of it. This explains get the right result, i.e., in order for the deleted site to be identical to its antecedent in the why the following ACD construction, pointed out by Norbert Hornstein, is unacceptable: matrix, it must be the case that the relative clause is not part of what needs to be identical in order for the deletion to proceed. Given our discussion of relative clauses above, only the head of the relative clause raises to AgrOP. Once we assume that the head of the (151) a. John expected everyone that I expected to win to lose b. *John expected everyone that I did to win to lose relative clause is the element that raises to AgrOP, we can assume that deletion of AgrOP can occur in the relative clause just in case both the head of AgrOP and the head of the DP in the specifier of AgrOP in the relative clause are identical to the head of the AgrOP and the head of the DP in the specifier of AgrOP in the matrix clause. The proposal is thus The AgrO inside the relative clause has to be deleted and that will include to win. In other words, at PF the whole AgrO inside the relative clause is marked for deletion and has to be deleted and that proceeds independently of the LF recoverability. The sentence will be interpretable at LF, however, if the AgrOs are identical. 274 275 It is far from clear what exactly drives PF deletion of material that is not a copy of 4.5.3 Empirical Support some moved element and under what conditions material can be deleted (see Chomsky In this section I discuss five cases that cannot be accounted for using the analysis in either 1995:125-6 for discussion). What we could say is that elements are marked for deletion May (1985) or Hornstein (1995). Restrictions on wh-operators and pied-piped relative before spell-out and that the derivation receives an interpretation if the elements marked clauses and constraints on ACDs with coordinated objects will be discussed first, since for the deletion have identical counterparts at LF. they do not require any further discussion on what the internal structure of relative clauses A question arises at this point, since we are assuming here that the AgrOs both of the is. The last three cases (quantifier restrictions, idioms and focal stress) will support the relative clause and the matrix clause have to be identical for the derivation to converge. analysis of relative clauses argued for in this chapter. Although some of analysis should However, the elided material has arguably not raised to AgrO by the time deletion has be taken as tentative in its mechanics, it will serve to provide support for the main applied. hypothesis of this section, namely that ACD requires identity of AgrOs. A possible answer to this question is to say that deletion can only occur if the elements have checked its features. This would imply that both the verb and the 4 . 5 . 3 . 1 Pied-piping complement have raised to the specifier of AgrO before spell-out, which would force Carlson (1977b) noted the following contrast: AgrO to have strong features in these cases. Although overt raising of the verb to AgrO encounters some support (see Johnson 1991), the overt raising of the complement is (152) a. Dulles suspected everyone that / Op Angleton did b. *Dulles suspected everyone who Angleton did highly unlikely. c. Dulles suspected everyone who Angleton suspected What seems to be going on is that deletion occurs freely on elements marked for deletion. The elements marked for deletion have to have at LF an identical counterpart. That something like that is necessary is clear from any gapping procedure or noun complement deletion. My suggestion here is that only certain configurations are visible to (152b) contrasts minimally with (152c). Fiengo and May (1994) suggest that the difference arises because wh-phrases must bind traces that are overtly identifiable as wh- the process that requires parallelism of structures and those are Agr projections in the case traces, unlike lexically empty operators, which are not so restricted. A way of interpreting of ACD and noun complement anaphora. their suggestion is to assume that wh-operators cannot be base generated in the operator If ACD requires identity between the AgrOs of the relative clause and the main clause, position but empty operators can. This line of reasoning would have to be used also for the prediction is clear: ACD should not be possible if the head of the relative clause that Hornstein, given that there is no way of explaining the contrast just by movement to raises to AgrOP in the matrix clause is not identical to the DP in the AgrOP inside the AgrO. relative clause. In the next section I will show cases in which this prediction is fulfilled If we assume that null operators contain copies of the head of the relative clause, as and which cannot be accounted for by QR or movement to AgrO of the whole Munn (1994a) argues, then the difference between (152a) and (152b) follows naturally. complement, without special stipulations. 276 277 In (152b) the operator who is different from the head of the relative clause, a non wh- Consider first the contrast between (156a) and (156b). element. Consequently the AgrOs will not be identical and ACD is not allowed.140 The same explanation will hold for cases of pied piped relative clauses as in (153a). Following Munn (1994a, the head of the relative clause contains only the theory while the (156) a. *John watched [the movies by Buñuel and [the excellent videos [that Fred did]DP] b. John watched the movies by Buñuel and the videos that Fred watched operator inside the relative clause contains in which theory. Again here the head of the relative clause is not string identical to the operator chain, as illustrated in (153b). c. John watched [[the videos and movies] [that Bill did]]DP If the relative clause containing the ACD site is embedded and modifies only the second (153) a. John believes the theory in which Bill believes conjunct as in (156a), ACD is impossible, although relativization of only the second b. John believes [[ the theory]H [in which theory ]OP [Bill believes [in which theory]TR]DP conjunct is perfectly possible as shown in (156b). If the relative clause modifies both conjuncts, ACD is allowed as shown in (156c). It is not clear how either an AgrO Our analysis correctly predicts that ACD should be impossible in these cases: (154a) analysis or a QR analysis of ACDs would account for the contrast in (156), since in both contrasts minimally with (154b) where the head and the trace in the elided VP are cases the whole coordinated DP will be moved in order to avoid the infinite regress. The identical. unacceptability of (156a) cannot be subsumed to the locality effects observed in ACD constructions such as (157a). In (157a) the problem is that the operator is too far from its (154) a. *John believes the theory in which Bill does trace, where ACD occurs or not, as the unacceptability of (157b) shows. b. John believes in the theory that Bill does (157) a. *Dulles suspected everyone that Angleton made the claim that Philby did A similar example is found in (155). In (155), the head of the relative clause contains only the book, while the trace inside the relative clause contains three chapters of which b. *Dulles suspected everyone that Angleton made the claim that Philby suspected book and ACD is impossible. Under our proposal no extra assumptions need to be made to account for the contrast (155) *John read the book three chapters of which Bill did in (156). In (156a) the head of the relative clause contains only the second conjunct, John ... [AgrO the book [Agr' read+Agr [VP ...[CP three chapters of the book ... Bill ... [AgrO three chapters of which [Agr' read +Agr [VP... while the object of the matrix verb is the first conjunct. The AgrOs not being identical, ACD is impossible. In (156b), the head of the relative clause is the conjoined phrase which raises to AgrOP and is identical to the trace in the ACD site. Thus ACD is possible. 4 . 5 . 3 . 2 Coordination Note, however, that if the ACD is dependent on the first conjunct the result is perfectly acceptable. 140 As pointed out to me by Howard Lasnik this entails that the appositive ACDs are in fact different from regular ACDs. 278 279 (158) a. John watched [ [the excellent videos [that Fred did]DP and the movies by Buñuel] This is predicted under our proposal since in that case the AgrO of the matrix clause will contain the first conjunct and the AgrO of the relative clause will also contain only the first conjunct. (161) a. Mary read all the papers that Bill did b. Mary read at least three papers that Bill did In Hornstein's (1995) proposal, copies of A-movement can be deleted freely as long as they obey a version of Diesing's (1992) mapping hypothesis which says that strong determiners have to be interpreted outside VP. In ACD constructions, however, in order 4 . 5 . 3 . 3 Quantifier restrictions for a VP to be copied in the elided site, it is crucial that the VP is emptied of its subject, as The contrast in (159) is not predicted by May (1985) or Hornstein (1994; 1995): the example in (162b) shows (the empty subject position marked tj). However, if the subject is not deleted from the VP, the copy will contain the subject. This subject, (159) a. John read at least three books that Bill did however, is different from the VP external subject. This is illustrated in (162c), where the b. *John read fewer than three books that Bill did external subject of the relative clause is you but the VP to be copied still has John as its c. John read fewer than three books that Bill read subject. The result cannot be interpreted. For May it would be necessary to say that at least three books undergoes quantifier (162) a. John bought everything that you did [e] raising but fewer that three does not. The problem with this approach is that, as has been b. Johnj [T [AgrO'' [everything that you did [e]]i [AgrO [VP tj buy ti]]] noted by Beghelli (1993), both at least three and fewer than three cannot enter scope c. [T [AgrO'' [everything that you did [e]]i [AgrO [VP John buy ti]]] interactions as exemplified in (160). Following Beghelli (1993) and references there, I Under Hornstein's (1995) proposal, a way to distinguish (159a) from (159b) is then take the lack of scope interactions to be evidence that the quantifiers in question do not to assume that fewer than three is not a strong quantifier and therefore must be interpreted undergo QR, contra May (1985). Given that they do license ACD, QR must not be the inside VP, and thus the incompatibility with ACDs. This assumption, however, makes determinant for ACD unless a special proviso for distinguishing quantifier scope and the prediction that fewer than three should not be able to appear in subject position in ACD is made. ACD constructions either, since, according to Hornstein, in ACD constructions the VP (160) a. Two students read all the papers *'for all the papers, each of them was read by two students' b. Two students read at least three papers *there are at least three papers and each of them was read by two students' 280 must be empty of its subject so that it can be copied in the elided VP. Unfortunately this prediction is not borne out as the examples below show: (163) a. Fewer than three students read every book that Bill did b. John read every book than fewer than three students did 281 Given that the subject fewer than three students must be interpreted in AgrS, i.e., something about what is the head and what is the operator of the relative clause, then it outside VP, then the question is why the same DP cannot be interpreted that way in seems that we do not have identical AgrOs, and therefore, contrary to the facts, ACD AgrO. In other words, something special will have to be said about fewer than three should not be possible with every and the. books in object position. Both a QR analysis and an AgrO analysis seem unable to account for the quantifier restrictions in a principled way. The core idea of my proposal is that for ACD to receive an interpretation, we need identity of AgrOs. Assuming the verbs are the same, let's assume that identity here means Determiner transparency is exactly what we need order to show that in fact AgrOs are identical in the case of the, every, all at least but not in the case of fewer than three and the indefinite a, which also produces unacceptable ACDs. According to the discussion in §1, relative clauses have the following structure: that the head of both the XP operator and the XP head of the relative clause have the same (164) a. DP content. A way to explain the contrast in (159) would be to say that, in order to interpret the/every AgrP (159a), it is necessary that Bill reads (at least) three books. Thus in (159a) both the head NumPHj and the operator of the relative clause are identical: in both cases we have three books. man AgrP Agr CP Identity of AgrOs will obtain and ACD will be possible. Now consider the fewer than NumPHj three case in (159b). Here ACD is not possible. If the head of the relative clause has CP man Op that I met t fewer than three books, and the operator has fewer than three books, we meet the identity condition but we derive the interpretation that John read fewer than three books and Bill The head of the relative clause raises to check case in the specifier of AgrO. In the read fewer than three books, which is not what a possible interpretation. In order to give relative clause the operator (that we know to be identical to the relative clause head since it (159b) an interpretation it would be necessary that John reads fewer than three books of can be deleted) checks its case at the specifier of AgrO. The determiner is left behind, (the) books Bill read. However to derive this interpretation the operator chain must being licensed by the relative clause. contain books and not something like fewer than three books. But then identity will not hold between the AgrOs and ACD is predictably unacceptable. It converges but it cannot I will assume that all, every, the, and also many, at least, and few (in their strong readings) are generated outside the NumP, giving a structure as in (165): receive an interpretation. Although a reasoning of this type might work for the two cases above, it is not clear that it would work for other quantifiers. Why should John read every book Bill did be an acceptable ACD, if the reading we obtain is that Bill read a number of books and John (165) AgrO[man-SGi [Agr[saw+agr] [VP saw [DP the AgrP[ti ... [AgrO[man-SG [Agr [saw+agr] ... Now consider (166) and (167): read every one of them? In fact the same problem arises with simple definites in ACD constructions. In John read the books Bill did, the interpretation we seem to obtain here is that Bill read some books and John read them. If we take the paraphrases to show 282 283 (166) a. John read many books that Mary did b. AgrO[book-SGi [Agr read+agr] [VP saw [DP many... AgrP[ti ...[CP... [AgrO[book-SG [Agr read+agr] I am assuming here that indefinite relative clauses (i.e., relative clauses that do not have an overt Determiner or quantifier in the head of D) are only acceptable under a partitive reading, which requires a definite in the operator. However, we cannot generate a as a quantifier or a D since it is just the spell out of singular count NumP head. (167) a. John read the many books that Bill did b. AgrOP[[many books]i [Agr read+agr] [VP saw... AgrP[ti ...[CP... [AgrOP[many book-SG [Agr read+agr] 4 . 5 . 3 . 4 Idioms Consider again the paradigm discussed above and repeated in (170), which has been The contrast in the interpretation of (166) and (167) can also explained in the same noted by Schachter (1973), Carlson (1977b) and has been extensively studied by way. The interpretation in (166), (for the speakers who find many acceptable in ACDs), Vergnaud (1985). The basic fact is that headway is an idiom that cannot be preceded by a is roughly that Mary read a certain amount of books and John read many of them. In this strong determiner (170b) unless it is modified by a relative clause as in (170c).141 case, many is generated in Q and raises to D, where it has roughly the interpretation of Headway alone or preceded by an indefinite cannot be modified by a relative clause, as 'many-of-the'. In (167) the interpretation is that Bill read many books and John read all shown in (170d) and (170e). of them. Here the definite determiner the is generated in D. The NumP head of the relative clause has many books and the operator of the relative clause also has many books and (170) a. John made headway b. *John made the headway ACD is possible. c. The headway John made was amazing Now consider (168). d. *Headway John made was amazing (168) a. *John read a book Mary did e. *Some/little headway John made was amazing b. AgrO[book-SG [Agrread+agr] ... CP[ ... [AgrO[ the book-SG [Agr read+agr] (169) a. *John read fewer than three books that Bill did b. AgrOP[fewer than three book-SG [Agr read+agr] ... CP[ ... [AgrOP[ the (three) book-SG [Agr read+agr] f. John made the headway Bill did (171) a. John weighs thirty kilos b. *John weighs thirty kilos Bill does c. John weighs the thirty kilos Bill does For ACDs, however, the requirement is identity of the heads, and identity is not met What is generated inside the relative clause is a NumP that roughly means amount of as shown in (168). In the operator position we have a definite DP and in the Spec of the headway, given that that is the reading we obtain. The present analysis predicts that matrix clause AgrO, a NumP. AgrOs are not identical and the result is an unacceptable ACD construction. 141 Neither demonstratives nor adjectives like wrong, are acceptable with headway . I believe the reason for this is semantic. 284 285 headway can only appear with a relative clause, if a definite determiner is generated in the (173) a. John bought that book that Bill did head of the DP to which the C of the relative clause raises. This is because headway b. *John bought the wrong book that Bill did cannot be generated as a definite DP as the object of the relative clause nor can it be made specific. Thus we need an external source for a definite or specific element. Now we have two options: either headway both in the embedded clause and in the (174) *John bought the wrong book that Bill bought 4 . 5 . 3 . 5 Focal Stress matrix clause do not raise to AgrO, since idioms are predicates and as such under certain assumptions do not require case or headway in both cases moves to AgrO where it will still be adjacent to the verb satisfying its idiom requirement. The last case I want to discuss has to do with the elements that can receive focal stress in ACD. The relevant paradigm is given below: Given that a terminative reading is possible with headway in the examples below, and (175) a. John bought every BOOK that Bill bought given the assumption we are making that noun phrase projections need case, I will assume that headway does move to AgrO where it can be adjacent to the verb.142 In b. John bought EVERY book that Bill bought c. John BOUGHT every book that Bill bought (172a) we have some headway in the specifier of AgrO. Thus a terminative reading will obtain. In (172b), we have headway at AgrO and a durative reading is available. A (176) a. ?John bought every BOOK that Bill did terminative reading is also possible if we use the relative clause as an external boundary, b. John bought EVERY book that Bill did as in the previous cases. c. *John BOUGHT every book that Bill did (172) a. John made some headway in three days b. John made the headway Bill made in three days/ for years In (175) it is possible to stress the verb, the head noun or the determiner. In the ACD cases in (176), however, the only possible element within the VP that can be stressed is EVERY. This follows naturally from the analysis of determiner transparency and the Given that our analysis parallels relative clauses demonstratives and adjectives like proposal that the deleted material has to have an identical counterpart at AgrO. In (176a) wrong, the prediction is that demonstratives should also allow ACD since the NumP can book will be marked for focus in the matrix AgrO but not in the relative clause AgrO, be moved to the Spec of AgrO leaving the demonstrative behind. This prediction is borne since presumably it is not possible to mark something for focal stress and deletion at the out as (173a) shows. Now if we take (173b), the result is unacceptable. Although I don't same time. In (176b) no problem arises because (i) EVERY is not a constituent with have an explanation for its unacceptability it should be noted that (174) is not any better, book; (ii) EVERY does not get to AgrO and therefore need not to have an identical which suggests that something else is at stake. counterpart at AgrO. In (176c), on the other hand, again the verb raises to AgrO in the matrix clause, and being marked for focus cannot have a deleted identical counterpart at the relative clause AgrO. 142 With Vergnaud I am assuming that headway is the idiom and not make headway. 286 287 4 . 5 . 4 Summary Chapter 5 A note on types, measures and possessives The proposal that in restrictive relative clauses the determiner/quantifier is left behind and the head of the relative clause raises to check case can be made useful in solving the Introduction infinite regress problem of ACDs. I departed from the idea that in ACD the problem is to In the previous chapter I provided a theory of determiner transparency that accounted for ensure the recoverability of a deleted VP, instead of providing an antecedent for a base the aspectual interpretations of certain DP complements. This chapter is an extension to generated empty VP as in May (1985) and Hornstein (1994, 1995). I proposed that PF Chapter 4. Here I will briefly examine three other types of complements and their effect deletion of an AgrOP is possible, if there is an AgrOP at LF which instantiates the same on the VP aspectual interpretation: overt type expressions, measure phrases and Spec-head relation as the AgrOP to be deleted. This proposal completely dissociates ACD possessives. The discussion of all three cases provides further independent support for constructions from quantifier scope and has the advantage of accounting for various the theory of determiner transparency and for the proposal that the internal structure of the restrictions on ACDs that were left unaccounted for in previous theories. It also provides VP complement is crucial to define the aspect of the VP. surprising confirming evidence for determiner transparency that is entirely independently motivated. The main results of chapter 4 were that (i) definite determiners needed to be licensed by a combination of an element with an <R> position plus number; and (ii) elements other than Ns could provide the <R>. Determiner transparency effects thus showed up because the NumP, not licensing the definite determiner itself, would be forced to raise to AgrO to get case, yielding the durative interpretations if it was non-quantized. The first part of this chapter will provide further evidence for the licensing proposal for definite determiners made in chapter 4, by examining two nominal elements, type expressions and measure phrases, that can never license a definite determiner by themselves and consequently require a determiner transparency configuration to appear after a definite determiner. We will see that the former always yield durative readings even when quantized, while the latter do not. The second part of the chapter uses the aspectual interpretations of possessive phrases to argue that they too implicate an analysis involving determiner transparency. The first case I want to discuss is the case of DPs with overt type expressions, which will provide an argument for the proposal that only elements that have an <R> can license 288 289 definite determiners. The examples below illustrate the two main properties of type expressions I will deal with in this section: Given that in (2) the cardinality of type is specified, the reading should be terminative. The question is then why the result is not terminative but durative. In other words, why are type expressions behaving as non-count? (1) a. *Peter wrote the types of articles Section 1 examines the properties of overt type expressions. I will claim that the b. Peter wrote [the two types of articles Mary liked] for years/ #in three hours. inability of type expressions to force terminative readings even when their cardinality is specified, as in the examples above, is partially related to the fact that type is a defective c. Peter wrote those two types of articles for years. noun unable to bind a definite determiner on its own, but that this alone cannot explain its d. Peter wrote the wrong two types of articles for years. properties. First, (1a) illustrates that a definite determiner is unacceptable with an overt type Following Carlson (1977a) I will assume that type is semantically a modifier of the expression. Unless a relative clause (1b), a demonstrative (1c) or an adjective like wrong noun and I will provide a structure for type complements that can account for its main (1d) appears with it, the definite determiner is unacceptable. Second, as the acceptability properties. Here again data from Brazilian Portuguese will be useful in the discussion of of the for x time adverbial shows in (1b,c,d), VP predicates with overt type expressions the structure proposed. have the property of always allowing durative readings, in spite of the fact that the Finally I contrast the cases in (1) with (3) below, and I propose a structure for (1) that accounts for its aspectual properties. I show that (1) and (3) cannot be derived from each cardinality of type is specified (by two in the above examples). According to the analysis of relative clauses, demonstratives and wrong type other. adjectives proposed in the previous chapter, the durative readings in (1b,c,d), at first (3) a. Mary wrote an article of that type in an hour/ #for hours sight, constitute a problem. The expectation is that we would have a structure in which b. Mary wrote two articles of the type nobody publishes in an hour/#for two types of articles will be at AgrO by the time aspect is calculated as in (2): (2) c. Mary wrote two articles of the wrong type/ in an hour/#for hours AgrO [two types of articles]NumP hours Agr Section 2 deals with measure phrases in English and Brazilian Portuguese. In contrast wrote VP to type phrases, measure phrases force terminative readings whenever they have their V' cardinality specified: DP the AgrP NumPt (4) a. Mary wrote two piles of memos in five minutes/ #for five minutes ... b. Mary wrote piles of memos for years/ ??in five minutes that Mary liked 290 291 There are two goals here. The first goal is to show that measure phrases are quite cardinality of the possessed noun. The -s/of and the possessor noun seem to play no role different both syntactically and semantically from type expressions and should be treated in the aspect calculations in spite of the uniqueness presupposition of possessives (see differently (contra Higginbotham 1994). The differences will account for their distinct Barker 1991, 1994 and references there). aspectual role. I will relate kilo measures with for x time adverbials rather than with type The analysis I will propose will follow the intuition pursued in Kayne (1994) that expressions. The intuition I want to pursue is that if John peeled two kilos of potatoes, relative clauses and possessive constructions share a fair amount of similarities, i.e., that then the interpretation is that John did two hours of potato peeling, just as in John ran for the possessed noun does not form a constituent with the definite in (5c) and (5d). The two hours, we have an interpretation that John did two hours of running. structure I will propose, however, will depart from Kayne's analysis in not assuming that The second goal is to show that referentiality is not what counts for aspectual (6), called by Quirk et al. (1972) a double genitive, is to be derived from (5) by terminative interpretations. Thus, although measure phrases are like type expressions in movement of two problems, just as the difference between (1) and (3) is not to be derived not being able to license a definite determiner, the fact that they yield terminative readings by movement. I will argue that (6) is to be treated as a small clause construction headed when quantized, will provide an argument against using referentiality as an explanation by of. for the presence or absence of durative readings. This chapter should be seen as an exercise in letting the theory determine the structure Finally, in section 3, I extend the determiner transparency analysis to possessives in and as a demonstration of the role of the internal argument properties on the VP aspect English and Brazilian Portuguese. As in the previous chapter, the aspectual interpretations interpretation and further evidence for the determiner transparency analysis. Needless to of these expressions, which pattern like the constructions discussed in chapter 4, will be say, the discussion here is to be taken as a preliminary excursion into the syntax of these used as a starting point for describing their structure. nominal expressions. (5) a. Mary solved John's two problems #for hours/ in an hour b 5.1 Overt Type Expressions Mary solved John's problems for hours/ in an hour In this section I will discuss the internal structure of type expressions with the goal of c. Mary wrote the reports of the research project for 5 years/ in 1 month explaining their properties and their behavior in constructing the VP aspect. The section is d. Mary wrote the two reports of the research project #for 1 month/in 1 year divided as follows. First I present the basic semantic intuitions for type expressions and (6) a. Mary solved two problems of John's #for hours/ in an hour b. *Mary solved problems of John's provide an intuitive account for the durative readings we have seen in (1). I then provide a structure for type complements that takes into account the semantic intuition and their syntactic properties and discuss the contrast between type of books and books of two Both in (5) and (6), if the cardinality of the possessed noun is specified, then the reading is terminative. Otherwise, a durative and sometimes a terminative reading is types. I then discuss the same constructions in Brazilian Portuguese and end with a discussion of their aspectual properties. available. In other words, what seems to count for the aspectual interpretation is the 292 293 5 . 1 . 1 The Basic Semantic Intuitions verb, type expressions behave more like mass nouns. (10a) contrasts with (10b) in that (10b) is odd. Consider the following cases: (10) a. Gold cleans itself. (7) Peter wrote [the two types of articles Mary liked] for years/ #in three b. #Gold cleaned itself. hours c. Women killed themselves yesterday. As we have seen, type expressions always allow durative readings, in spite of their cardinality being specified. In other words, type expressions seem to be behaving like d. #That type of woman killed herself yesterday. e. That type of woman kills herself. bare plurals or mass nouns. The similarity between overt type expressions and bare plurals has been independently noted in Carlson (1977a). Carlson notes that this kind of animal behaves very much like a bare plural, in that bare plurals and kind expressions are unambiguous: they denote only the name of a kind. The existential readings the terms receive in (9) are the result of quantification over stages introduced by the meaning of a stage-level As examples (10a) and (10b) show, a mass noun is not always an acceptable subject for a predicate that overtly refers to a stage. In the same way, type expressions are sometimes equally odd as subjects of stage-level predicates which make clear reference to a stage, as in (10d). The problem with (10d) is not that type expressions cannot bind a reflexive because with a generic reading the sentence is acceptable, as (10e) shows.143 To account for the properties of type expressions, I will follow Carlson's claim that predicate: type is semantically a modifier, i.e., an expression of the type CN/CN, and the intuition I want to pursue is that it functions syntactically as an "auxiliary" noun much like an (8) a. Sparrows are common. b. This kind of bird is common. auxiliary verb does in the verbal system. In other words, type is part of the extended projection of the noun (however we want to account formally for extended projections). (9) a. Sparrows are sitting on my lawn. The intuition that type expressions are auxiliary nouns, and therefore parasitic on b. This kind of bird is sitting on my lawn. another noun in the same way as auxiliary verbs are parasitic on verbs, relies heavily on Thus every sparrow in (8) and (9) is a realization of the kind sparrows and every realization of the sparrow-kind is a sparrow. However, it is not the case that bare plurals and type expressions have exactly the Jackendoff's (1985) discussion of types and tokens. According to Jackendoff, types have no reference because they do not correspond directly to experience nor can they be understood as a measure on the noun as kilos are. For Jackendoff, the reference of a same distribution. In (8) and (9), type expressions and bare plurals seem to display the same distribution. However, when reference is overtly made to stages, depending on the 143 When mass nouns are also acceptable with stage-level predicates, type expressions are also acceptable: (i) Paper flew in the wind yesterday. (ii) That type of paper flew in the wind yesterday. 294 295 linguistic expression is not the "real" object but rather the object in the projected world. only alternative, because type, even when preceded by a numeral, does not tell us about Referring expressions of natural language will be just those expressions that map onto how much of the object has been used up to build a PATH. projectable expressions of conceptual structure. Type expressions have no projection, and In sum, the basic intuition is that (i) type expressions are non-referential and phrases expressing types are non-referring. In other words, if we consider a token dependent on a noun and (ii) they pattern as mass nouns in that we cannot individuate concept as the representation of the thing being categorized, a type concept is the discrete tokens from them, that will be relevant in the PATH construction. In providing a representation of the category itself. A type concept is the information that the organism structure for type expressions below I will pay attention to these properties. creates and stores when it learns a category. Type, kind, and sort expressions create new categories, for which there are an infinite number of tokens available. The referential part of an expression such as a type of bird is 5 . 1 . 2 Structure 5 . 1 . 2 . 1 Basic Syntactic Properties provided by the noun bird. Type expressions can thus be seen as turning a noun into a The following examples from English and Brazilian Portuguese, respectively, "mass expression". illustrate the basic properties of type: the (a) examples show the inability of type to appear It is this property of type that accounts for its inability to license terminative readings with definites; the (c) examples show that type cannot appear without a complement. of an event. Consider the expression the type of person I like. Such a type may be quite These two properties reflect the discussion in the previous section. idiosyncratic but it can include an unbounded number of persons that correspond to the type I like. Type then never denotes. (11) a. *John bought the type of pencil. Now consider a type expression that reaches AgrO for aspectual calculation. Even if person, the noun that is providing the referential part in type of person, had quantity b. John bought the pencil. c. *John bought the/a type. information (and it is not clear whether it does), it would anyhow be unavailable because of its being too embedded within the noun phrase. The result is then that type is all the [verb+Agr] can see at AgrO. Given that we cannot count tokens from the number of types, we cannot use the two types of persons I like to measure any event.1 4 4 (12) a. *O João comprou o tipo de lápis. the João bought the type of pencil b. O João comprou o lápis. Consequently the result is durative, since type is neither a referential expression, nor a the João bought the pencil measure (that partitions and quantizes the noun), nor something specified in the 'João bought the pencil.' discourse. After all, it introduces new types.145 Thus durative readings are always the 144 Note that mass nouns can be preceded by numerals, yet never denote quantities of tokens, but rather types, as in John drank two waters for years. So "counting" a mass is not the same as quantizing it. 145 Some noun phrases however are easily interpreted as types, because we know them as belonging to canonical types. Thus when we say I like that apple, I am usually referring to a canonical type of apple. It could not mean an over-ripe apple. Thus it will be lexically dependent and context dependent. 296 Overt type expression, however, allow creation of new categories. Consequently the type of apple that Bill likes could be a Granny Smith but could also an over-ripe apple. This difference between canonical types (which appear unsignaled) and overt type expressions has been pointed out to me by Norbert Hornstein. 297 c. *O João comprou o/um tipo. c. O João escreveu o tipo de artigo errado. the João bought the/a type the João wrote the type of article wrong 'João wrote the wrong type of article.' However, if we add a relative clause, an adjective like wrong or a demonstrative, then the determiner is again acceptable. Examples from English and Brazilian Portuguese are d. O João escreveu aquele tipo de artigo. The João wrote that type of article given below:146 'João wrote that type of article.' (13) a. *John wrote the type of article. b. John wrote the type of article that will never get published. We have seen that relative clauses, adjectives like wrong and demonstratives are able c. John wrote the wrong type of article. to license definite determiners. Given the fact that type cannot appear with a definite d. John wrote that type of article. determiner unless accompanied by one of these modifiers, it is quite reasonable to conclude here that type is not able to bind a definite determiner. In the article, article has (14) a. *O João escreveu o tipo de artigo. the João wrote the type of article features that will bind the. Type, on the other hand, has no <R> and thus cannot license a definite determiner. I will assume this to be characteristic of a defective noun. b. O João escreveu o tipo de artigo que nunca vai ser publicado. The second property of type expressions that we need to discuss in order to propose a the João wrote the type of article that will never get published structure is the fact type is parasitic on some other nominal element for its reference. 'João wrote the type of article that will never get published.' Semantically it is clear that type takes a noun and produces another noun element. However, syntactically type appears as a transitive noun requiring a complement. The question, in other words, is what is the relation between type and its complement. We can think of type as being a modifier of the noun, in which case it enters a process of 146 What I will say for type of expressions is similar to what one can say for manner and kind of expressions. Although I will not deal with them, it should be noted that adjectives like mere, utter and virtual. also inhibit a noun to license a definite determiner as the examples below from Jackendoff (1972:54) show: (i) a. He is a mere boy. b. *He is the mere boy. c. He is the mere boy we expected him to be. (ii) a. The play turned out to be an utter disaster. b. *The play turned out to be the utter disaster. c. The play turned out to be the utter disaster I predicted. (iii) a. The war caused a virtual collapse of the economy. b. ?The war caused the virtual collapse of the economy. c. ?The war caused the virtual collapse of the economy that had been predicted. 298 theta identification with the head noun, or as an element that enters a theta marking relation with the noun it modifies. The two rough structures are given below: (15) a. NP NP type XP of pencil N XP type of pencil In (15a) type is an XP. It is a modifier of pencil and enters a process of theta identification with the head noun, which means that the noun and the modifier combine 299 and share an open position. If red flower roughly means that x is red and flower, then First, a definite determiner is unacceptable under type as illustrated in (17a). Thus it is (15a) should mean that x is a type and x is a pencil. However, not only is this not unlikely to be a DP. Second, it can be modified by adjectives that adjoin to NPs (17b,c) intuitively what type of pencil means, but also it is not clear why pencil could provide the (see Bernstein 1993 and Munn 1995a) but not by adjectives like wrong (17d). Third, <R> and license the definite determiner in this case. According to what we have seen, pencil cannot be quantized, as illustrated in (17e, f), or substituted by one as in pencil can bind a definite determiner but type cannot. Thus I will assume that type enters (17g).148 a theta marking relation with of pencil, which blocks raising of the referential noun to the What is of? I will assume that of is theta bound by the nominal head that raises to it at LF, as the tree below illustrates: D. (16) (18) NP type PP of NP type PP NP of pencil NP/NumP pencil Now we need to know what is pencil: a DP, a NumP or a NP? I believe it has to be Now we need to show that type is actually a modifier of sorts and I will show this by an NP or possibly a NumP with no quantity specifications.147 The examples in (17) examining restrictions on the types of adjectives that can modify type and the complement support this conclusion. of of. 5 . 1 . 2 . 2 Adjectival Restrictions on type expressions (17) a. *John bought that type of the pencil. b. John bought that type of red pencil. The argument I will present here is that if type modifies the noun phrase directly, then d. John bought that type of women's clothing. it will obey cognitive ordering restrictions with respect to other modifiers. The following d. *John bought that type of wrong pencil. discussion draws heavily on Sproat and Shih's (1988) analysis of adjective ordering facts e. *John bought that type of two/many pencils. in various languages. Sproat and Shih argue that many kinds of adjective ordering facts f. *John bought that type of every pencil. are cognitively rather than syntactically based. They also show that adjectives can modify g. *John bought that type of one. at (roughly) the NP level or the DP level, and that cognitive ordering constraints hold only of the NP level adjectives. Striking evidence for this idea comes from Chinese 147 It seems that the lack of quantity information is due to a semantic restriction that disallows an object to realize two different subkinds of a noun. In other words, three types of three dogs is not acceptable, because, as Carlson points out, (ia) follows from (ib). (i) a. Two kinds of dogs are in the next room. b. At least two dogs are in the next room. 300 148 Example ($17g) has an acceptable interpretation in which one is clearly not the complement of type, as evidenced by the acceptability of a definite determiner before one in (i) in contrast with (ii). I will ignore such a reading here. (i) (ii) I bought that type of the one and this type of the other *That type of the car 301 which allows adjectival modifiers to appear with or without the particle DE which also As for the ability to appear in prenominal position in French, Spanish and marks relative clauses. Adjectives marked with DE do not show ordering restrictions, Portuguese, again, the adjectives that can appear with type sometimes can and sometimes while adjectives without DE show the same ordering restrictions as English. We can use cannot appear in prenominal position, as the examples from Portuguese show: this idea to test whether type is an NP modifier (in which case it will show ordering (21) a. Esse tipo de planta novo é fantástico. effects with respect to other adjectives. This new type of plant is fantastic. Consider the following: b. Pedro comprou dois tipos de armas caros. (19) a. John bought that new/good type of red pencils. Pedro bought two types of weapons expensive. b. John bought that old type of women's clothing. 'Pedro bought two expensive types of weapons.' c. *John bought that red type of pencils d. *John bought that chocolate type of ice-cream (22) a. A nova planta é linda. The new plant is beautiful. Type can be modified by some adjectives but not others. Thus while good, new, old (in the sense of "former"), are perfectly acceptable with type (19a,b), color adjectives and b. *Caras armas Expensive weapons substances are not. It is not exactly clear how to account for the differences between new and good, on the one hand, and chocolate and black, on the other hand, from a syntactic point of view. (23) a. A planta é nova. The plant is new. There are two tests that are commonly used to distinguish two types of adjective: the ability/inability to appear in predicative positions and the ability/inability to appear prenominally in languages such as French, Spanish or Portuguese. Neither test, b. A arma é cara. The weapon is expensive. however, can shed any light to the distinction between the adjectives that are acceptable I would like to suggest here that the restrictions on the modification of type is partially with type expressions and the adjectives that are not. Both the acceptable and the non(at least) a semantic fact. acceptable adjective modifiers of type can appear in predicative constructions: In their discussion of the nature of the cognitive basis of adjective ordering, Sproat and Shih suggest that the relative ordering correlates with the amount of comparison (20) a. The pencil is red. necessary to establish whether the various types of description are appropriate to the b. The pencil is new. particular object being described. The more comparison is necessary the further away from the noun the adjective is: 302 303 (27) a. *red type of table vs. type of red table (24) a. humungous red peach (*red humungous peach) b. ??large type of table vs. type of large table b. wonderful red ball (*red wonderful ball) c. ??round type of table vs. type of round table d. *six-inch type of ruler vs. type of six-inch ruler The more apparent the quality, the closer it is to the noun. Size, in principle, will be e. *Chinese type of food vs. type of Chinese food further from the noun than color as (25a) shows, unless the size is absolute (25b). f. interesting type of person (25) b. *purple humungous pen; humungous purple pen g. new type of Chinese restaurant a. ?purple six-inch pen; six-inch purple pen It seems then that we can treat type as a direct modifier of the noun that has to be Good and ugly will involve largely matters of opinion, which would makes us expect them to be further from the head noun. The prediction seems to be correct, as the examples below show. placed between clearly intensional modifiers and predicative modifiers. 5 . 1 . 2 . 3 The structure The structure below is an attempt to capture the intuitions discussed above: (26) a. an ugly blue sofa (28) b. *a blue ugly sofa NumP one Num' typei If type is itself an auxiliary noun syntactically, but semantically a modifier, it makes sense that other modifiers will be placed in a certain position with respect to type. If it creates a new category of a common noun, then, only adjectives that are not "helping" to establish the reference to the expression are allowed to modify type of car, for instance. In other words, while the head noun can be modified by "more absolute" qualities, once we establish a new type, only "opinion-dependent" adjectives will be allowed to modify it. NP typei of red PPj NP pencil Type theta marks (in Higginbotham's sense) the of phrase. The intuition here is that in the same way that have "selects" for a participle and the progressive be "selects" for an -ing form, type can select for an of-phrase. The NP red pencil has case features to check. Its head moves to of and checks its case with it. Red pencil will move no further given that it has no D features to check and has satisfied its case needs.149 149 If the complement of of is in fact a NumP rather than an NP then the head raises through NumP to of. I will ignore this point here as it will have no effect on the analysis. 304 305 Type cannot license a definite determiner since it does not have and <R>.150 It will (31) DP D move to NumP to check number features. D is either not there or will remain empty. The AgrP two books whole complex moves to have its case checked. Agr of If type of NP is modified by a relative clause, the result is the following: DP the AgrCP (29) a. DP type (of books) Agr' the AgrP that+Agr CP two types of cars Agr' t type (of books) CP that+agr CP tthat IP that Bill likes Mary likes b. the two types of cars that Bill likes The relative clause is formed as in the previous chapter. Type of books is in the Spec The NumP two types of cars will raise to AgrO, in order to check case. 5 . 1 . 2 . 4 Analyzing two books of the type Mary likes of the Agr where it checks agreement with that+Agr, and from there at LF type incorporates into of, where it has its case checked. Of books is erased at PF under recoverability with respect to the NumP books that is in the specifier of the of partitive Now consider the following sentences: head. (30) a. *John bought two books of the type I am assuming here that type in (30) is actually of the same semantic type of type in b. John bought two books of the type that Mary likes. the types of pencils I like; in other words, type in (30) has a "complement" that is c. John bought books of two types. anaphoric to books. I do not believe that books of two types is generated by movement of books from the Note again the need for modification on type in order for it to appear after a definite complement position to the specifier of an Agr projection and the reasons are twofold: determiner, illustrated by the contrast between (30a) and (30b).The structure I will under the assumptions I am making this would amount to a left-branch extraction from propose is the following: the specifier of the AgrP of the relative clause. The NP books would have to raise out of type of books: 150 It has number, however, and it can appear with every, which means that it can move up to the Q position. 306 307 (32) * DP (33) a. Esses tipos de pessoa são muito complicados. D NumP two this-M.PL type-M.PL of person-F.SG are very complicated-M.PL Num' Num 'These types of people are very complicated.' PP booksi b. *essa tipo de pessoa P' of this-F.SG type-M.SG of person-F.SG DP 'this type of person' the AgrCP NumPj type Agr (34) a. Eu examinei os tipos errados de escola. PP that+agr CP I examined the-M.PL type-M.PL wrong-M.PL of school-F.PL of booksi NumPj CP tthat 'I examined the wrong types of school.' IP b. Eu examinei os tipos de escola errados. Mary likes I examined the-M.PL types-M.PL of school-F.SG wrong-F.SG The second argument is theory internal. There is no reason for books to raise, since it 'I examined the wrong types of schools. ' has its case already checked. Later I will also show that similar arguments hold (contra c. *Eu examinei os tipos de escola errada. Kayne 1994) for possessive constructions of the type friends of John's. I examined the-M.PL types-M.PL of school-F.SG wrong-F.SG 5 . 1 . 2 . 5 Type Expressions in Brazilian Portuguese Before we move to the aspectual interpretations of the constructions above, I would like to draw attention to one assumption I have been making, and provide some evidence for it, namely that the relative clause, the demonstrative and adjectives like wrong are attached to type of NP and not to the NP itself. The evidence comes from Brazilian Portuguese adjectives and demonstratives, which will exhibit agreement with type. 'I examined the types of wrong schools.' The contrast between (34a) and (34c) shows that wrong can appear in two positions: after type or after type of school. While (34a) is acceptable, (34b) is not, since the adjective that theta binds the definite agrees with escola. The examples below illustrate the possibility of having adjectives modifying both nouns: The examples below illustrate the fact that agreement is related to type and not to the head noun. 308 309 (35) a. O Pedro encontrou o tipo rígido de escola urbana que procurava. The Pedro found the type-M.SG rigid-M.SG of school-F.SG. urban.F.SG that he was looking for. b. DP os AgrP NumP 'Pedro found the rigid type of urban school he was looking for.' Agr' dois tipos (de escola) erradosi AP ti b. O João vendeu muito o novo tipo de mochila forrada. NumPj (dois tipos) de escola The João sold a lot the new-M.SG. type-M.SG of backpack-F.SG linenF .SG 'João sold the new type of lined backpack a lot.' Evidence for this being a possible strategy in Portuguese but not in English comes from the fact that in wh-movement the of phrase can be stranded in Brazilian Portuguese The structure we proposed before to account for the English type expressions is able but not in English as (37b) illustrates. to account for (34b) with the additional overt movement to the specifier of the Agr, as illustrated in (36a). To account for the adjective order in (34a) I will assume the same (37) a. Que tipo de computador ele comprou? Which type of computer he bought structure and the workings of the copy theory. I will assume here (as in chapter 4) that A-movement is subject to the copy theory. 'Which type of computer did he buy?' Thus it is possible to either move the whole NumP and erase the copy downstairs, or to b. Que tipo ele comprou de computador? move the whole NumP and erase from the tail of the NumP chain dois tipos, leaving Which type he bought of computer? behind de escolas and erasing from the head of the chain the de escolas part. The two possibilities are illustrated below:151 (36) a. Assuming the copy theory for question movement, the possibility of stranding de computador follows naturally. This is not an option for English, and I believe the reason DP os lies in the fact that of tends to cliticize onto the left. Thus, a PF restriction will disallow AgrP NumPi dois tipos de escola the deletion. Needless to say, however, more has to be said with respect to recoverability. Agr' errados AP (errados) (NumP)i I will not pursue these issues here. Now we can go back to the aspectual properties of type constructions. 5 . 1 . 3 The Aspectual Properties of type expressions The paradigm introduced in the beginning of this chapter is given in (38): 151 The parentheses mark deleted copies. 310 311 (38) a. John wrote [the type of article nobody liked] for years/ ??in two hours b. John wrote [an article of the type nobody liked] #for years/in two days While the first example always allows a durative reading, the second example is 1.1. If the head is a regular NP, then, depending on the features of the noun phrase, a durative or a terminative reading will be possible. 5 . 1 . 4 Summary clearly terminative. The following cases are parallel to (38b) and illustrate that a From the discussion above the following points should be clear: First, not every noun terminative reading is obligatory if the cardinality is specified and a durative reading is can license a definite determiner: non-referential elements do not license definite elements. obligatory if the cardinality is not specified in cases like (39b,c): This provides an argument for the proposal made in the previous chapter that we need (39) a. John wrote two letters of the wrong type in three minutes/#for years <R> and Number to license a definite determiner. Type has number but no <R>. If we b. John wrote letters of the wrong type for years/#in five minutes add a relative clause, type expressions can provide the number and the CP of the relative c. John wrote junk of the wrong type for years/ #in five minutes clause can provide the <R>.152 Second, for the aspect calculus, type always forces durative readings. There are at This is expected given the structure I have proposed. Two letters, and an article will least two possible explanations for this fact. First, we could tie the durative readings to force terminative readings given that the head that raises to AgrO for Case has and <R> the non-referential character of the type expression. Alternatively we can think that the and its quantity is specified. On the other hand, a bare plural and a mass noun will force issue is not referentiality for aspect calculus, but partition. Given the parallel behavior of durative readings, as expected. Examples are given in (39b) and (39c). type expressions and mass nouns with certain stage-level predicates (in cases where a In (40a), on the other hand, type is heading the projection, and a durative reading is type expression is the subject of an overtly specified stage), then we could say that always available, as we discussed above, because type does not give us any information because type of expressions disallow discrete partitioning of the object they disallow about the cardinality of the referential elements. terminative readings. In the next section I will show that the latter hypothesis is more plausible. (40) a. John wrote one type of books for years. b. John wrote two types of books for years. c. John wrote the two types of books Bill liked for years. Third, type imposes restrictions on adjectival modification as expected if it is also a direct modifier in the sense of Sproat and Shih (1988). Only "intensional" adjectives can appear modifying type expressions. The discussion above provides a way to understand the aspectual interpretations that the complements with type expressions allow. If type is the head of the nominal element that reaches AgrO, the result is a durative reading for the reasons discussed in section 312 152 We could make a parallel between the definite determiner and tense. We could say that tense is not licensed only by [+V] elements. Instead, tense is licensed by [+V] elements plus <R>, i.e., information from the DP. If this is correct, then the EPP can be taken to be a requirement on licensing tense, just like <R> plus number is the requisite to license a definite determiner. 313 5.2 the extent that (42a) is acceptable, it has an implicit, discourse linked reading like "there is Measure Phrases something of which John bought two types." In section 1 I raised the issue of whether type was forcing durative readings because it could not bind an <R> or whether the reason had to do with the fact that type blocked information about the cardinality of the tokens. In this section I hope to show that terminativity is not clearly tied to referentiality and that what counts is whether some expressions can serve as a measure or not. Moreover I want to establish a difference between type expressions and measure expressions. This section is divided as follows: It has been suggested in the literature (e.g. Cinque 1989 and others) that measures are inherently non-referential and act as predicates on verbs like weigh. This seems true given that kilo per se cannot bind a definite determiner. Suppose this is true. Then it must be the case that referentiality has nothing to do with terminativity, given that measure phrases with the right verb can force terminative readings. Examples are given below: first I will describe some differences between type and kilo expressions. Then I propose a tentative analysis and finally I discuss the interactions of measures and definite determiners. (43) a. John ran three miles in five minutes/#for two hours b. John flew 200 miles in 30 minutes. c. *John ran the three miles in five minutes 5 . 2 . 1 Main Differences between Type and Kilo In the cases above a definite determiner will not be acceptable, as illustrated in (43c). Consider (41): The only way (43c) can be made acceptable is if we are talking about some specific stretch of a road or the name of a sports competition. Cliticization in Spanish is also (41) a. John weighs fifty kilos. unacceptable with measures. Examples from are given below: b. *John weighs the fifty kilos c. John weighs the fifty kilos Bill would love to weigh. (44) a. Corrí 2 millas. (42) a. #John bought two types (I) ran 2 miles. b. *John bought the types 'I ran 2 miles. c. John bought the types of books Mary likes. b. *Las corrí. As we have discussed before, measure phrases do not accept a definite determiner, (I) them ran just as type does not (42b). Like type expressions, under a relative clause, measure I ran them.' phrases can appear with a definite determiner, as illustrated in (41c) and (42c). However, unlike type, measures do not require a complement to be acceptable, as (41a) shows. To However, if the measure has what seems to be a complement, then a definite is possible and these expressions can appear as complements of transitive verbs and as complements of intransitive verbs. In those cases, as (46) shows, cliticization is possible: 314 315 (50) a. John bought two types of meat. (45) a. John ate the two kilos of potatoes in a single meal. b. John bought meat of two types. b. The three meters of material were not enough for the dress. c. Bill swam the five miles of the channel too slowly. The unacceptability of meat of two kilos and the acceptability of meat of two types again shows that they are different. In the previous section I suggested that meat of two (46) a. Comí los 2 kilos de patatas. (I) ate the 2 kilos of potatoes. types was in fact meat of two types of meat and that the second instance of meat was deleted because it could be recovered. The unacceptability of (49b) suggests that either it b. Los comí. is impossible to recover meat or there is nothing to be recovered because the structure is (I) them ate. quite different. In Brazilian Portuguese it is possible to say (51). However, in this case the measure It seems that we are facing another case of parasitic reference. However, in this case, is not measuring anything. The interpretation is that I found a type of flour packages and the definite can be licensed without modification of the whole complex by a relative clause they are two kilo ones. Given the interpretation we get, it seems that we have a different or some other D-binder. All that is necessary is that it has "a complement." This is structure in (49b) when we compare it with (50b). different from type, which cannot license a definite determiner under any circumstance. There are also more differences. Numerals are acceptable in the "complement" of of in kilos but not in types. Plus, as (47) shows, types can follow kilos but not the other way around (48). (51) Eu só encontrei farinha de dois quilos. I only found flour of two kilos 'I only found flour in two-kilo packages.' (47) a. Mary bought two kilos of two fish. b. John brought two liters of two types of cognacs. (48) a. *Mary brought two types of two movies Further evidence for a semantic and syntactic distinction between kilos and types comes from the behavior of kilos as a modifier of the VP and the inability of type to do that in the same way. Consider (52): b. *Mary wrote two types of two cook books. (52) a. John ran for three hours. Another difference between type and kilo expressions is given in (49): b. John did three hours of running. c. John peeled two bags of potatoes. (49) a. John bought two kilos of meat. d. John did two bags of potato peeling. b. *John bought meat of two kilos 316 317 (52a) can be paraphrased with (52b) and (52c) can be paraphrased as (52d). (52d) indicates that the amount of bags is also a way to provide an external measure for the VP, interval of three hours. The adverbial gives another dimension in which we can fit an event. in the same way as for x time adverbials provide a measure for the PATH when the cardinality of the object is not specified. When we examine (53), however, it becomes 5 . 2 . 2 Tentative Structure and Some Predictions clear that type does not have the same ability of adding a measure to the VP. (53b) is not a Various analyses have been proposed for the structure of measure phrases. I will not be paraphrase for (53a). able to do justice to them here. (see Selkirk, 1977; Jackendoff 1977; Mallén 1992 and references there). Here I will just take the observations in section 2.1 and tentatively (53) a. John peeled two types of potatoes. propose that two kilos is a modifier of an of phrase; in other words, measures modify b. #John did two types of potato peeling homogeneous XPs. The of checks the case of the nominal element under it. To the extent that these paraphrases tell us something, they tell us that measure (55) phrases can act much more like VP modifier adverbials than type expressions. Type AgrP two kilos expressions are not as detachable from the noun as kilo is. The similarity between AgrP potatoes Agr' measures and for x time adverbials can be captured if two bags, two kilos etc. are also of+Agr PP modifiers, in this case external modifiers (as in Sproat and Shih 1988): (54) a. XP two kilos b. XP XP XP of potatoes tof tpotatoes At LF the complement of the of moves to the specifier of the AgrP to check case. The for x time run We can think of the of as acting just like the imperfective forms in Polish. Remember that the imperfective forced the objects (if quantized), to be consumed or produced in a measure phrase is generated adjoined to the AgrP projection as probably adverbials such as for an hour can be generated adjoined to AgrO. Two kilos enters a theta identification with the of in an Agr projection and from there the whole projection moves to AgrO to check case. Alternatively, only two kilos moves to AgrO to check case. lump. And this is actually the interpretation we get. Thus for example if Mary bought The result is a measure at AgrO and consequently a terminative reading of the VP. two kilos of two fish for a party, we do not get a reading in which each fish weighs two Moreover, because it modifies a homogeneous expression, and it has to move to AgrO to kilos, nor is it necessary that each fish weigh exactly one kilo. Instead, we get the same check its case, this measure will be also in the checking domain of the V+Agr. The tree reading we get for adverbial modifiers: i.e. that there must be as much of the two fish as it below illustrate both possibilities of movement for Case. Either only two kilos moves or is necessary to reach the amount two kilos. This reading is thus similar to the fact that in the whole DP moves: Mary ate apples for three hours there must be as many apples as is needed to fill the time 318 319 (56) a. AgrO two kilos b. Let's consider (57a) and (57b) first. (58) is the overt structure. (59) shows its LF. Of AgrO Agr' DP raises to Agr and the DP the meat raises to the specifier of the AgrP, where the meat Agr' V+Agr VP two kilos of potatoes V+Agr VP checks its case and enters a specifier head relation with the of. From there the complex [of+Agr] raises to check its D features with the definite. In the next section I discuss the interaction of measures and definite determiners. But before I do that let me summarize where we are so far: (58) DP (i) the measure has no <R> and behaves like a modifier that enters a theta- D' identification relation with the of phrase. The measure phrase is not taking a complement. the This explains why measures can only appear with determiners if there is an of phrase, AgrP two kilos since when there is an of phrase there is an <R> that can be used to license a definite AgrP AgrP determiner (see below). PP (ii) of forces homogeneous interpretations on its complement. In English and of NumP Portuguese this is the work of the of head and the de head. If we were to generate two meat kilos as the complement of, the result is that the measure does not measure anything. It is (59) DP the complement of of and is theta bound by it. As the complement of of, it has to be D' interpreted as a lump. [of+Agr]+the AgrP (iii) the aspectual properties of these constructions follow naturally. Terminative two kilos AgrP readings are the norm, if a measure that has a specified number with it moves to AgrO. t[of+Agr] AgrP meat 5 . 2 . 3 Measures and Definites PP tof In this section I examine the paradigm in (57): NumP meat (57) a. John bought the two kilos of meat. We can see from the structure given in (60) that if, instead of a NumP, as the b. *John bought the two kilos of the meat complement of of, we have a DP with a definite determiner, (57b) will be unacceptable. c. John bought the two kilos of the meat that was fresh. The DP will raise to the specifier of Agr and will enter a specifier head agreement with the d. John ate the packages of the meat complex of+Agr. In this case the complex of+Agr will check its D features with the specifier and will have no reasons to move to D. The definite in this case will not get 320 321 licensed because nothing incorporates into it. The complex [of+Agr] will have all of its features checked and therefore it will not need to move. If package is a measure, our prediction is that the sentence in (61c) should be unacceptable. While in both (61a) and (61b) John eats meat in the amount of one This predicts that if the complement of the partitive is a DP with a relative clause, two package, in (61c) we get a pragmatically odd reading in which John ate the package itself. definites will be acceptable, since in the specifier o the AgrP there is only a NumP. (57c) This is due to the fact that package is ambiguous between a measure and a regular noun shows this prediction to be correct and the LF structure is given in (60) phrase and cannot be interpreted in (61c) as a measure, due to the lack of a way to license the definite determiner. Its only interpretation, then is as a regular noun with a (60) DP complement theta marked (meat) by the it. D' Summarizing, we have seen in this section that measures behave as modifiers and not [of+agr]+the AgrP as theta marking elements. They are unable to license definite determiners because they two kilos AgrP are modifiers. They modify homogeneous expressions and act as measures on the event. [meat]NumP AgrP They force terminative readings on the predicate if preceded by quantity information. t[of+Agr] PP tof DP 5.3 Possessives and Partitives [that+Agr]+the AgrP tmeat Agr t[that+Agr] CP was fresh In this section I want to present the sketch of an analysis of possessive constructions that accounts for their aspectual input on the VP aspect, extending the empirical coverage of the determiner transparency analysis proposed in Chapter 4. It will have the advantage of treating possessive constructions in English and Portuguese in a similar fashion, although Here the definite can be licensed because of, not having a definite in its specifier, can and, in fact, must raise to D in order to check features. The reason there is not a definite determiner in the specifier of the of+Agr phrase is that the definite is left stranded since it belongs to the relative clause and need not move. Finally consider the following: Brazilian Portuguese does not have a prenominal possessive construction with a full NP possessor. The basic idea is that the of/s possessives have structural similarities but different morphological properties. This section is divided as follows: first I deal with the possessor 's and of in simple possessive constructions of the type John's books and the books of the library, and I also deal with the Brazilian Portuguese counterpart de, which (61) a. John ate a package of meat displays the same aspectual properties. Then I move to a discussion of a more complex b. John ate a package of the meat possessive construction—some books of John's —which I will argue to be a partitive c. John ate the package of the meat construction. Finally I discuss the Brazilian Portuguese counterpart of the contrast 322 323 between my books and books of mine, and I will use the facts to support the structure (63) a. a shadow of John proposed for the English case. b. John's shadow 5 . 3 . 1 John's books and the books of the library have a similar structure While both (63a) and (63b) can mean "the shadow of John's body", only (63b) can In this section I deal with the possessive s and the possessive of, focusing mainly on their also mean "some shadow he is doing with his fingers onto the wall". According to Barker aspectual properties. (1991), this follows from the fact that in (63a), the of possessive, is only possible with Consider the following examples: relational nouns. In other words, the of possessive depends on the ability of a possessed (62) a. Mary solved John's problems in 5 minutes/for 5 years noun to lexically establish a relation with the other noun. The notion of relational noun is, however, not very clear. Although we have no b. Mary solved John's problem in 5 minutes/#for 5 years c. Mary solved John's two problems in 5 minutes/#for 5 years problems in saying that brother, captain etc. are relational nouns, car and report are much less clearly relational nouns. Yet the brother of John not very acceptable, while the report While for 5 years is perfectly acceptable in (62a), it only allows an oddly stretched of the committee is, yet these are on both sides of the relational noun split. Thus I will reading in (62b) and (62c). To account for the fact that neither the possessor John nor the assume that there is no syntactic difference between relational and non-relational nouns. It -s seem to play any role in the aspectual interpretation of the VP, in spite of the fact that is clear, though, that of establishes a narrower set of possible relations than the -s. This possessive constructions carry a uniqueness presupposition just like definite descriptions leads us to the second assumption. do,153 it must be the case that what is visible for the aspect calculus at AgrO is only the cardinality of the possessed noun. The second assumption I will be making is that of can license a definite determiner but -s cannot. Rather, -s inherits the definiteness from the possessor head as the example Before I present the structure for possessive constructions, three not very standard below suggests. assumptions I will be making should be clear. First I will assume that the -s and of (64) a. *There is John's book on the table. possessives are structurally identical, in the sense that they are both heads of a projection b. There is a student's coat in the office. that takes the possessor as its complement and the possessed noun as its specifier. The differences between the affixal s and the more prepositional looking element of lie in their While (64a) is clearly unacceptable, (64b) is possible. The only difference is the morphological properties. To assume that they are structurally identical is, at first sight, an odd assumption to definiteness of the possessor nominal. The third assumption (as I have been assuming so far) is that the specifier of DP (or make if we consider the following contrast: of AgrDP) is an L-related position154 and not an operator position, contra Kayne (1994). 153 For example, John's problem is felicitous just in case there is at most one maximally salient problem possessed by John. 324 154 An L-related position is roughly that of an A-position in pre-Minimalist terms. 325 With these three not very standard assumptions in mind (but see Uriagereka 1992 for (66) DP a very similar proposal), we can inspect the basic structure below. The NumP problems D is generated adjoined to the of/s phrase which can be filled by an of or a -s head. The of/s element theta marks the possessor which is the complement of the of/s head. Above the AgrP John PP there is an Agr projection and a DP projection as in (65b).155 Agr' -s+Agr PP NumP (65) a. PP PP NumP problems ts PP problems of/s The -s raises overtly to Agr and, being strong triggers overt movement of the DP possessor, which moves to the specifier of Agr to check its case against the [-s+Agr] John b. complex. S and the possessor element enter a specifier head agreement and the head -s DP D tJohn will be interpreted as definite if it has entered a specifier head relation with a definite AgrP possessor. Otherwise it is interpreted as indefinite, explaining then the contrast in (64), Agr Agr repeated here as (67): PP NumP problems PP of/s DP John I will now consider in turn the s and the of possibilities. (67) a. *There is John's book on the table. b. There is a student's coat in the office. At LF, the complex [-s+Agr] raises to D to check agreement features with the NumP (the possessed noun) that raises to the specifier of DP. 5 . 3 . 1 . 1 -s Possessives Evidence that the possessive head agrees also with the possessed noun comes from In the schematic tree above we have the overt structure of the possessive s construction. possessive pronouns in the Romance languages. These pronouns agree in gender and number with the possessed noun and in person with the possessor (see 3.3). (68) a. meus amigos my-M.PL friend-M.PL 'my friends' 155 Given that possessives can license null anaphora, there is probably an AgrD projection above DP, if the proposal made in chapter 3 is correct. I will for the time being just assume a DP for ease of discussion. 326 327 (70) b. minhas amigas DP D my-F.PL friend-F.PL the 'my friends' AgrP Agr' The NumP in the Spec of PP in (66) raises from there to the specifier of DP (or Agr AgrDP) and then raises from there to AgrO to check its case. Since-s has no <R> it PP NumP consequently cannot license a definite determiner. A schematic LF structure is given in reports PP of (69): DP the research project (69) If the head is of, of moves covertly to Agr, and the research project, at LF, moves to AgrO the specifier of Agr to check case with [of+Agr]. [of+Agr] moves to D to check D NumP[problems] Agr features and the NumP reports raises to check the agreement features of [[of+Agr]+D] V+Agr VP subj and from there to AgrO to check case. The LF structure is given below: V vt DP tNumP (71) AgrO D NumP[reports] [sj+Agr]+D AgrP Johni Agr V+Agr VP Agr' subj [sj+Agr] PP tNumP vt PP tj V ti 5 . 3 . 1 . 2 of Possessives DP tNumP D [ofj+Agr]+the AgrP Johni Now let's consider the of constructions. The following tree illustrates the overt Agr [ofj+Agr] structure of possessive of constructions. PP tNumP PP tj ti The difference between of and -s is that of can license a definite determiner but -s cannot. The reason is that -s checks its D-features against the element that moves to its 328 329 specifier but of has to move to D for that. The net result is the following: the possessed out to be incorrect, the point here was to show that we have independent reasons to noun is in AgrO by the time aspect is calculated, just as in the case of relative clauses. believe that the possessed expression is detachable from the rest of the possessive phrase. The structure I have given for of/s allows us to end up with the same configuration for possessives at LF, which seems to be a positive result given that they do not differ in 5 . 3 . 1 . 3 Brazilian Portuguese Possessive d e meaning except for the selection restriction differences. In both cases we end up with a In Brazilian Portuguese, only the de 'of' version possessive constructions exists with NumP (possessed noun) at AgrO. If NumP is mass or if NumP is a plural with no a full NP possessor, and we find the same effects on the aspectual interpretation, i.e., the specified quantity, durative readings will arise; if NumP is a singular count noun or a possessed noun is responsible for the aspectual interpretations, independent of the plural with its quantity specified, the result is a terminative reading, as the examples definite determiner: below illustrate: (74) a. O vizinho matou os cachorros da Maria por três anos. (72) a. Mary solved John's problems for years. (Ele matava sempre que ela comprava, até que ela decidiu processá-lo.) b. Mary bought John's milk for years 'The neighbor killed the dogs of-the Maria for three years. c. Mary solved John's two problems #for years '(He killed it every time she bought one, until she decided to sue him.)' d. Mary solved John's problem #for years b. O Pedro comprou a massa de pizza do supermercado por três anos. (73) a. Mary wrote the reports of the research project for years The Pedro bought the pizza dough of- the supermarket for three years b. Mary wrote the junkmail of the consulate for years 'Pedro bought the pizza dough from the supermarket for three years.' c. Mary wrote the two reports of the research project #for years d. Mary wrote the report of the research project #for years c. A polícia matou o amigo da Maria #por três anos. The police killed the friend of-the Maria for three years The proposal I made here for possessives is by the no means the only way we can get 'The police killed Mary's friend for three years.' the detachability of the possessed noun. Various other possibilities could be explored and a discussion of each of them would take us too far afield. For example, we could argue d. A polícia matou dois amigos da Maria #por três anos. that both s and of take small clauses as complements (see Uriagereka 1993). Alternatively The police killed two friends of-the Maria for three years we can produce two different structures for the two possessive constructions. It seems to 'The police killed Mary's two friends for three years.' me very unclear how to decide between these two hypothesis given that we are far from understanding the role of elements like of and s, what are their actual morphological needs and what are their conditions for interpretation. Thus, if the implementation turns 330 In sum, the structure proposed, of course, pending some more independent motivation, is able to account for possessive constructions in both Brazilian Portuguese 331 and English. I will come back to Brazilian Portuguese in section 3.3. It should be noted Schachter and Partee (1973), for example, the possessive NP is underlyingly prenominal, that this proposal is very similar to Kayne (1994), as we will see below, although not then it is postposed and finally of is inserted for case marking. identical in its details. In the next set of cases I will depart from Kayne more radically. More recently (and in the same vein), Kayne (1994) has proposed an analysis of the so-called double genitive constructions, where (75) is derived from the prenominal 5 . 3 . 2 Two problems of John's is not John's two problems In this section I follow Barker (1994) and I argue that constructions such as in (75a) are partitive constructions. Moreover I argue that (75a) is not to be derived from John's two problems. Consider the following cases of possessives in English: possessor. His analysis is based on work by Szabolcsi (1983) on Hungarian possessive constructions. In Hungarian, possessors are prenominal. If the possessor is in the nominative case, the possessor is preceded by a definite article (i.e., the John's articles). When the D head is null, thus indefinite, the possessor moves to the specifier of D where it gets dative case (75) a. Mary solved two problems of John's #for hours/ in an hour b. ?Mary solved problems of John's. c. *Mary solved the two problems of John's. and from there it moves out of the DP. Kayne proposes that in English the prenominal possessor is also preceded by a D head. The difference is that the D head must be null. Thus as the structure in (77) shows, there are two specifier positions: the specifier of D and the specifier of s. Contrary to (75a) shows that, again, what matters for calculating aspect is the cardinality of the possessed noun. What is puzzling is the relative unacceptability of (75b) and the what I have assumed, Kayne proposes that the specifier of D is an operator position, as Szabolcsi did for Hungarian. unacceptability of (75c), which reminds us of one of the properties of partitive (77) a. D0 [John ['s two problems] constructions (Jackendoff 1977) illustrated below with regular as complements and with b. two problemsi [D of] John's [ei] possessive noun phrases as complements of of. (76) a. I met one of the singers. Kayne interprets this movement in terms of case and proposes an analysis for (75a) b. I met one of John's friends. that derives (77b) from (77a). The incompatibility of (77a) with a definite determiner is c. *I met the two of the singers. stipulated and the incompatibility of (77b) with the definite determiner is derived from the d. *I met the one of John's friends. fact that of is a D element. In other words, for Kayne the lack of a definite is a syntactic fact. The examples in (76) show that two definites are not possible in a partitive construction. Various analysis have been proposed for the construction in (75a) that derive it from an underlyingly prenominal possessive construction. In Stockwell, I would like to note three properties of (75a) that suggest that (77b) is not a tenable analysis for it. The first two reasons are empirical, and the third one is more theoretical. First, (75a) obeys the Partitive Constraint and has an anti-uniqueness presupposition as Partitive Constructions do. As in partitive constructions, the set denoted by the 332 333 complement of of has to be bigger than the specified in the two problems part. In other should be as good as in (79a). If, as Kayne argues, the movement to the specifier of D is words, (75a) does not mean that Mary solved John's two problems. It has a partitive A' movement, reconstruction should be possible, and the sentence in (79a) should be meaning: (75a) means that Mary solved two problems of John's set of problems and not acceptable on a par with the perfectly acceptable (79c). If, as I am proposing, specifier of that Mary solved two problems of John's two problems. (This is akin to the oddity of D is an A position, reconstruction should be also possible on a par with (79d). two problems of two problems). The requirement of proper partitivity explains the The consequence of these two observations is that we cannot assume that two pictures relative unacceptability of (75b) since the bare plural problems is not easily interpreted as of himself has raised from a position internal to John's, as proposed in Kayne (1994). a proper part of John's problems, and the unacceptability of (75c) is explained in the Theoretically, the movement analysis also violates the principle of compositionality and same way. It also explains why the following example is odd: it is hard to imagine a arguably the theta-criterion. Assuming the movement analysis, we would be forced to situation where John would have more than one mother.156 semantically evaluate problems in problems of John's twice, since problems will contribute a predicate to the semantic interpretation in two different places.157 (78) a. John's mother This is very different from what we have been doing so far. We have been checking b. *a mother of John's the thematic relations in one position and features in another position, as necessary in the minimalist program. We are not forcing a NumP to be an argument of different heads, Moreover, as in partitive constructions, there is an anti-uniqueness presupposition: which would be the case in the raising analysis. any two problems of the set of John's problems will allow the expression two problems Given these observations, we have to assume that two problems is not generated as of John's to be felicitous. part of the possessive phrase that includes John's. The second empirical reason is associated with (79). (80) (79) a. Mary took Johni's two pictures of himselfi. DP D b. *Mary took two pictures of himselfi of Johni's. NumP c. Which pictures of himselfi did Johni take? problems d. Pictures of himselfi seem to Johni to be nice. PartP PartP of DPi John's Although (79a) is perfectly acceptable, (79b) is not. A raising analysis, as sketched in Of in this construction is a partitive head. It takes as its complement DPi. The of Kayne (77b), will not produce the desired results, given that, if we can base-generate two partitive phrase has the NumP in its specifier. The Partitive Constraint will disallow the pictures of himself in a position c-commanded by John, then binding of the anaphor 156 For how to deal with all pictures of John's which seem to violate proper partitivity, see Barker (1994) 334 157 This fact has been noted by Barker (1994), although he didn't see any problem with it and the adoption of the raising analysis, associating such possibility to control structures. But control structures are exactly different from raising in the respect that in a control structure there is an independent element that receives an anaphoric interpretation rather than a single element being interpreted twice. 335 definite determiner that heads the projection to be overt in these constructions. However the definite can reappear if licensed by something else other than the N that takes the partitive phrase as its complement: (83) DP D PartP problems Part of (81) a. Mary solved the two problems of John's that were worrying him. b. Mary brought the wrong books of John's. b. Mary brought those books of John's. DP pro D D' AgrP John Agr '-s PP DP i in (80) is a regular possessive construction with a null possessed nominal tpro element. I will assume this element to be pro, although the exact properties of it are far P' ts from clear: tJohn The -s moves overtly to Agr, where it checks is agreement features and John moves (82) DP to the specifier of Agr to check its case and agreement with the s head. Whether pro D AgrP moves overtly or not is immaterial. For consistency, though, I will assume that it moves Agr overtly, on the assumption that null pronominal categories can only be licensed in a Agr PP checking position, given that they depend on certain agreement features. The null pro P' pronominal element can reach a checking position by moving to the specifier of an Agr s John projection as illustrated above or by incorporating into a head that can license it. In 3.3 we Overtly we have the following: will see some evidence for the latter option in Brazilian Portuguese. The complete structure of two problems of John's will have the following LF: 336 337 (84) complement. What distinguishes the of in type or kilo expressions would be the fact that DP D NumP problems the former is a part of a complement of type and the latter is part of a complement of a PartP verb in case it is in the object position. Not much seems to distinguish the partitive from PartP of the possessive of, which is not to say that these constructions will not behave differently DPi prok due to the fact that they have different semantic properties. D' I do not see at this moment what is the right approach: collapse the of's into a single AgrP Johni lexical item and work out the differences in terms of their possibilities of combination, or Agr sj+Agr assume that they are different items. In fact, given that learning will have to happen PP tk anyway, it seems easier not to try to derive their properties purely from its combinatorial PP sj ti possibilities. Summarizing, in this section I have argued that two books of John's is not derived Assuming that two problems of John's is akin to partitive constructions allows us to from a regular possessive and should be treated instead as a partitive. Although the relate the unacceptability of (85a) with the unacceptability of (85b): ultimate technical implementation of the structures above is still very primitive and (85) a. *Mary solved two problem's of a man's/ some man's. b. *Mary solved two of some problems. requires independent motivation, the point here is that by inspecting the interactions of nominal expressions with verbs we can uncover various important properties of the internal structure of noun phrases. At this point a question arises: how many of's are there? We have seen of in type of In this section I also suggested that the possessive constructions in Brazilian car. We also have seen of in kilo of flour. Now we also have a possessive and a partitive Portuguese are identical to the English ones, with the exception that Brazilian Portuguese of. In other words we have seen four uses of of. Are they all the same? At first sight it does not have prenominal full NP possessors. In the next section I will discuss the seems that they should all be collapsed into one. However, the analysis I have provided counterpart of a problem of John's and the difference between my problems and argues against collapsing all the of's. In type of car, of is part of the extended projection problems of mine in Brazilian Portuguese. of the noun car, since type is a defective noun. In two books of John's, of is the head of a small clause, which would allow us to think that it theta marks its complement, unlike of in type of or kilo of which are theta bound. The possessive also theta-marks its 5 . 3 . 3 Brazilian Portuguese possessive pronouns In this section I want to discuss the paradigm in (86) below: arguments, which suggests that the last two (partitives and possessives) should be collapsed as in Uriagereka (1993) and the first two should be collapsed (of in type and of in kilo phrases). Thus we have one of that is theta-bound and one of that theta-marks a 338 339 (86) a. Meus dois amigos vieram à festa. b. Eu comprei todos os de bolinha. my-M.PL two-M.PL friend-M.PL came to the party I bought all the pro with polka dots 'My two friends came to the party.' 'I bought all the ones with polka dots.' b. Dois amigos meus vieram à festa. two-M.PL friend-M.PL my-M.PL came to the party If we assume that John's is basically the pro of John, then the Brazilian Portuguese counterpart of two books of John's can be assumed to be (89a). 'Two friends of mine came to the party.' (89) a. dois livros dos do João The proposal I will make is that while the DP subject in (86a) corresponds to my two friends as the glosses indicate, the DP subject in (86b) correspond to two friends of mine. two books of-the pro of-the João 'two books of John's' In order to do that, let me start by discussing the counterpart of the partitive two pictures of John's in Brazilian Portuguese. The proposal I made in the previous section was that John's was a full possessive construction with an empty possessed nominal, recovered from the subject of the partitive construction. b. *um livro de alguns do João a book of some pro of-the João 'a book of John's two' Brazilian Portuguese, as we have seen, allows null elements to be licensed by The fact that (89b) is unacceptable does not come from the inability of alguns to definites and some indefinites. Examples are given in (87a) for definites and (87b) for license the null element as (87b) shows. If we consider (89a) as a partitive construction indefinites. From chapter 3 we know that not every quantifier can license null elements, like two books of John's with a null pro coindexed with books and being licensed by the as, for example, the contrast between todo and todos os is repeated in (88): definite determiner, we can understand the unacceptability of (89b) as related to the (87) a. Eu comprei os com bolinhas. constraints that involve partitive constructions. As we have seen, one of some books is not acceptable. Following the structure given in the previous section, we can see how the I bought the with polka dots 'I bought the ones with polka dots.' mechanism works to produce the overt structure in (89a). We start with the following: b. Eu comprei alguns com bolinhas. I bought some with polka dots. (88) a. Eu comprei *todo de bolinha. I bought every pro with polka dots 340 341 (90) In Brazilian Portuguese, a definite determiner can appear before a possessive pronoun DP os although that is not obligatory.158 AgrP PP pro (92) a. os meus livros PP P the-M.PL my-M.PL book-M.PL DP 'my books' de1 o Joãoi First, o incorporates into de1 (the number is just for ease of discussion) producing b. meus livros do, checking the case of the whole DP. Remember, the definite determiner in Brazilian my-M.PL book-M.PL Portuguese has the option of incorporating into another head to check case for the whole 'my books' DP. Pro has case and agreement features to check. It will move as a head to Agr and from If the possessive is postposed, i.e., appears after the noun, the definite is impossible. there it will incorporate into the definite determiner and move with it to check its case, by The following examples illustrate two types of cases where the possessive determiner can incorporating into de2 as the tree below exemplifies. (91) be postposed. DP D NumPk dois livros (93) a. Meu Deus, que tragédia! PartP My-M.SG God-M.SG, what a tragedy PartP de2+pro+os 'My God, what a tragedy!' DPi [[pro+agr]t+os]t AgrP [pro+agr]t prot b. Deus Meu, que tragédia! PP God-M.SG my-M.SG P 'God of mine, what a tragedy! do João (94) a. Meus dois amigos vieram à festa. If there is no overt D in the head of DPi, the pronominal element cannot incorporate and be recovered from the agreement with the definite and the result will be unacceptable my-M.PL two-M.PL friend-M.PL came to the party 'My two friends came to the party.' as the examples with todo showed above. I assume that this incorporation is overt. Now I want to consider the case of possessive pronouns in Brazilian Portuguese. 158 Analysis of the variation of the use of the possessive with and without the definite determiner show that Brazilian Portuguese alternates the use of the definite determiner almost in the same way throughout its history. (See Oliveira e Silva 1982) 342 343 b. Dois amigos meus vieram à festa. (96) DP AgrP two-M.PL friend-M.PL my-M.PL came to the party os 'Two friends of mine came to the party.' Agr meus In the first case (93) we have a vocative and, in these cases, whether the possessive PP dois amigos appears before or after the noun, a definite determiner is never possible, which, as P' tmeus Longobardi (1994) suggests, may indicate overt movement to D. Meu is possessive + [1person, singular]. After incorporation we have meu. Meu then The cases I will concentrate on in this section are the cases in (94). In (94) the incorporates into Agr to check its features as illustrated in (96). This is what we see. On possessive can also appear before or after the noun in argument positions. When the the way to LF, the possessor pronoun has agreement features and case to check and possessive appears prenominally, the definite determiner is possible, but not when the amigo, the possessed noun has case features to check. The possessed noun then head possessive appears after the noun, unless the definite determiner is licensed by a relative raises to meu and moves with it to D to check D features in meu. The whole complex will clause as in (95c): move to check its case. Evidence for the incorporation (which mimics agreement under government, as we (95) a. os meus dois amigos have seen in chapter 3 and 4) comes from the examples from coordinated possessed b. *os dois amigos meus c. os dois amigos meus que tu conheces I will argue that those cases do not involve movement of the noun to D as argued by Longobardi (1994) for Italian. Instead I will argue that dois amigos meus has the same structure and meaning of two friends of mine, which will explain the unacceptability of the definite determiner in (95b). nouns below. (97) a. os meus amigos e amigas the-M.PL my-M.PL friend-M.PL and friend-F.PL 'my male friends and female friends' b. as minhas amigas e amigos There are two main reasons to treat dois amigos meus as structurally different from meus dois amigos. The first one has to do with its syntax and the second one has to do the-F.SG my-F.PL friend-F.PL and friend-M.PL 'my female friends and male friends' with its meaning. Suppose that os meus dois amigos has the following structure: c. *os meus amigas e amigos the-M.PL my-M.PL friend-F.PL and friend-M.PL 344 345 In (97ab) we can see the agreement of the determiner and the possessive pronoun with the first conjunct. When first conjunct agreement fails the result is unacceptable. However, in such analysis we are left without an explanation for the unacceptability of the definite determiner when the possessive is postposed. This is not, however, the Now let's consider dois amigos meus. Assuming the structure in (96), it would be only problem with this analysis. Assuming such an analysis we should not find the easy to propose that for dois amigos meus all that is happening is that dois amigos moves differences in meaning between dois amigos meus e meus dois amigos, that we in fact overtly to the specifier of AgrP instead of incorporating at LF to the possessive pronoun. do. Consider first (100) and (101): The structure would then be the following: (100) a. Só dois amigos meus vieram à festa. (98) DP o only two friends of mine come to the party. AgrP dois amigos b. Só meus dois amigos vieram à festa. Agr meus only my two friends come to the party. PP tdois amigos P' (101) a. Dois amigos meus não vieram à festa. tmeus two friends of mine didn't come to the party. Agreement is not anymore via incorporation but via specifier head agreement. The prediction is then that the possessive head should agree not with the first conjunct as the b. Meus dois amigos não vieram à festa. my two friends didn't come to the party. examples in (97) do, but with both conjuncts. The prediction that agreement should occur with both conjuncts is borne out as the examples below show. In (65a) only two people came to the party. In (100b), besides my two friends other people came to the party. A similar contrast is illustrated in (101). (99) a. amigos e amigas meus The examples below illustrate the same differences in interpretation between my two friend-M.PL and friend-F.PL my-M.PL friends and two friends of mine. 'male friends and female friends of mine' (102) a. Only my two friends came to the party. b. *amigas e amigos minhas b. Only two friends of mine came to the party. friend-F.PL and friend-M.PL my-F.PL 'my female friends and male friends' In (102a) only two people came to the party. In (102b), besides my two friends other people came to the party. c. amigas e amigos meus friend-F.PL and friend-M.PL my-M.PL 346 347 Now consider the partitive construction two friends of mine in English. The (103) a. Two friends of mine didn't come to the party. difference between English and Portuguese is that mine is my one(s) and Brazilian b. My two friends didn't come to the party. Portuguese has a null pronominal form as the possessed nominal element as opposed to While in (103) other friends may have come to the party, the second one does not an overt element. The structure at LF corresponds to (105a) and (105b), respectively: allow the reading in which more friends may have come. The examples from Brazilian (105) a. DP b. DP Portuguese show exactly the same difference. D Agr D Agr If dois amigos meus is like friends of mine, as its meaning requires, then the lack of a proi+meus Agr definite determiner in this construction follows from the Partitive Constraint. my PP tpro In the following I will assume that to be true and I will propose that Brazilian one PP P' P' tmeus Portuguese amigos meus has the same structure as friends of mine. tmy If meu is followed by a noun then we have (104a), which corresponds to the English my+one = mine my friend in (104b). The respective Spell-Out structures are given in (104c,d): Now we can consider the full structure: (104) a. my friend (106) a. friends of mine b. meu amigo c. DP d. b. amigos meus DP Agr D Agr c. Agr DP Agr my PP friend meu P' D PP amigo tmy NumP P' friends amigos PartP PartP of DP= mine meus tmeu (106) illustrates another difference between English and Portuguese. In Portuguese The difference between English and Brazilian Portuguese is that my in English will the partitive de does not show up in these constructions. Meus will raise to the partitive raise up to D and therefore the definite determiner can never appear with the possessive, head to enter a specifier head agreement with the NumP problemas in (107) and the whole but in Brazilian Portuguese meu can license the definite determiner. It raises to Agr partitive expression will raise to check case. overtly and the head noun of the possessed noun incorporates at LF into it to check the The proposal above predicts the agreement of the possessive with both conjuncts agreement of the possessive. given that the agreement obtains in a Spec-head relation, as illustrated in (99). 348 349 case, the partitives, what matters is what is the subject of the small clause that takes the (107) DP possessive phrase as its complement. Needless to say that still a lot of questions remain D PartP open with respect to possessive constructions. NumP problemas PartP meusi DP= meus Moreover, the structure above also predicts that meus is always ambiguous between my and mine, which is a desired result, given what we have seen. Third, it predicts that the possessive in partitives is not part of the NumP. Thus, while we can say (108a) we cannot say (108b) with an indefinite non-specific reading. (108) a. um amigo meu b. ??um meu amigo 5 . 3 . 4 Summary In this section I discussed two different cases of possessives: the actual possessives and the so called double genitive possessives. First I dealt with the regular possessives and I provided a way to deal with s/of possessors. The similarities were captured by identical initial structure and the differences were captured by the fact that they have different selection requirements and different morphological properties. This allowed me to unify the treatment of possessives in English with the possessives in Brazilian Portuguese. The second construction, the double genitives, were analyzed as partitives given the semantic similarities with regular partitives. The distinction between the regular possessive constructions and the double genitive provided the basis for the analysis of possessive pronouns in Brazilian Portuguese, which are in the surface ambiguous between the my and mine interpretations. With respect to the aspectual properties of these constructions, in regular possessive constructions we find (again a case of) determiner transparency and what matters is the properties of the possessed noun. For the second 350 351 Chapter 6 Subjects and aspect: the case of ser and estar seems that the adverbials force or block the boundedness of a situation. In other languages, this can be the work of A-quantifiers and so called aspectual verbs. The general goal of this chapter is to show the interaction between these higher verbal Introduction elements and verbs with DP subjects. In the previous chapters I dealt with VP aspect properties. More specifically I was 6.1 interested in the role of the internal argument in the aspectual composition. The base line 6 . 1 . 1 The role of the subject Subjects and the Stage-Level/Individual-Level distinction was that the VP aspect was the result of the composition between the verb and the internal argument. The result of this composition could be modified by adverbials such as for an hour and repeatedly, creating bounded and unbounded events, respectively. The main point was that, at least at the VP level, there is no place for a durative/terminative feature, nor a special aspectual projection for aspect feature checking. What counts is what is at AgrO by the time the VP aspect is calculated. Here and there I mentioned that the aspect of a sentence was not to be reduced solely to the properties of the verb and its internal argument, and that other elements could also be layered on top of the VP aspect. This is a well-known descriptive fact, and led many authors to establish categorical distinctions between the boundedness that is dependent on the Aktionsart of the verb and its interaction with the complement and the higher (in the tree) aspectual morphology that also gives bounded/unbounded situations. The terminology varies enormously, and authors differ on whether they treat the boundedness of the VP and higher level boundedness as identical or not. To be more clear, if I say I ate the bagel, we know that, by the time I finish the bagel, the event is over. If, on the other hand, I say I ate bagels yesterday, the event is unbounded since we cannot establish a natural end point for it. However, the adverbial yesterday binds the situation as an According to our hypothesis about aspect, aspect is the way in which the events are temporally constituted. Events can undergo partitions that are dependent on the partitions that the participants of an event allow. We have seen this with respect to objects. Mass nouns, not having individual parts, cannot be used to partition the event into discrete individual subparts. Thus, a VP containing a mass object is not be able to be modified by repeatedly. An important role was played by the object in its interaction with the verb at AgrO. If we assume that participants of the event affect the temporal constitution of an event, then it must be the case that subjects that are participants of an event can also play a similar role provided certain conditions hold. In fact, if we consider the following examples we can see that subjects also interact in partitioning situations that include the subject. Events that include subjects I will call situations, just for ease of explanation, following Verkuyl (1993). A clear example (from Dowty 1989) of the role of the subject is given below: (1) a. The settler crossed the desert in one day/ #for three years b. The settlers crossed the desert in three weeks/ #for three weeks c. Settlers crossed the desert in three weeks/ for years "yesterday" situation. This situation does not hold anymore. Because today is not yesterday, we know that the situation is over. That the work of bounding the situation is Here a terminative reading for the VP is predicted given that cross is an eventive verb. not purely related to the fact that we have a past situation can be seen by the fact that John the desert is a [+SQA] object and that object raises to AgrO to check case and the verb always ate bagels is also a past event, but this time an unbounded one. In English it raises to AgrO on its way to tense. (1a) and (b) behave as predicted and (1c) also allows a 352 353 terminative reading, as the adverbial in three weeks suggests. However, in (c) a durative The PATH function, as we have seen, provides for a set of pairs <i,p>159 and the reading is also possible. Besides an iterative reading (i.e., the settlers crossing repeatedly subject is "going through" these pairs until the last pair. Thus if we have Mary lifted four the desert), there is also a durative reading in which an indefinite number of settlers kept tables, we establish a partition among the set of four tables in any way we want. Mary crossing the desert. The only difference between the (b) and the (c) sentences in (1) is the can have lifted one table at a time, or two or three or four. What is important though, is fact that the subject in (c) is a bare plural. Thus bare plural subjects, at least of verbs like that the information is cumulative. This implies that if she lifted at index 1 one table, and cross (which may be unaccusatives of some sort), show a similar pattern with the object then at index 2, two tables, the information of index 1 is preserved at index 2. While the cases. I assume here that the adverbial for years in (1) can be generated in two positions: PATH function connects the verb with its internal argument, another function, the function it either modifies the VP aspect, or it modifies a higher projection that includes the π, assigns to the members of the external argument denotation their own VP. subject. As Verkuyl (1995) puts it, the predication assumes a "passing through" because we The example below illustrates again the role of the subject, in this case a mass subject, in the aspect composition. can say things like Mary began to eat four sandwiches but before she arrived at the third one she stopped eating. If we have children ate two sandwiches, on the other hand, we do not know how (2) Water leaked through the ceiling for hours many children passed through the PATH. #A gallon of water leaked through the ceiling for two hours These facts sound familiar and suggest that at AgrS, at least, the subjects of certain verbs can also "measure" situations. The examples below show that regular agentive Given the facts above, we could then try to restate the same proposal for AgrO for AgrS which, mutatis mutandis, should have a similar role, i.e., it should have the function of allowing the complex V+T to enter a specifier-head agreement with the quantity information of the DP: subjects can also induce durative readings: (4) (3) a. Mary hit the nail for two hours If the VP is terminative, interpret the situation as bounded if a [+SQA] subject is in Spec AgrS and [+ADD TO] element is in its checking domain. b. Children hit the nail for hours As stated in (4), it is not the case that any subject will be able to bound the situation. The distinction between bare plurals and count nouns is playing a similar role here, and we can see that this is not due to some sort of partial unaccusativity. There can be in (3) as many hittings of the nail as there are children. Thus the situation is unbounded That will depend on what is the verbal material in AgrS (which was true for AgrO as well). But the point is that the we operate compositionally. First we build a PATH and then we use the subject as the domain of the PATH. This is again not to say that other because we do not know how many children participated in it. 159 354 In <i,p>, p stands for a place and i for indices. 355 elements, such as adverbials, can bound or unbound the situation on a higher level (for a relation with the subject that is different from the relation between subjects and example, at the CP level). predicates that have temporal information (roughly stage-level predicates). Once we have Whether there are or there aren't other verbal projections to which the matrix verb a solid description of temporally unspecified predicates, I will add temporal information moves to check morphology on its way to AgrS is immaterial to the point I am making via morphology, syntactic incorporation and adverbial modification so that the role these here. The point here is that the subject will participate in the situation in the way that the V elements play in the interpretation of subjects will become clear. (plus its A-modifiers) that are at AgrS by the time the "upper aspect" is calculated will This chapter is then a case study on two copula constructions in Brazilian Portuguese allow it to. If, as in Polish for the VP, certain morphological markers act as A-modifiers which, although they seem to be the grammaticalization of the stage-level and individual- for situations,160 the morphology may or may not impose certain readings on the subject level distinction, will turn out to be a strong argument against its grammaticalization in the in order to derive a situation that is compatible with the upper head-A-modifiers. syntax as proposed by Diesing (1992). In this chapter I will again claim that there is no such a thing as a durative or a terminative feature for the IP level. There are verbs, verb complexes and A-quantifiers 6 . 1 . 2 The stage level/individual level distinction that will force a certain interpretation of the subject at AgrS, understood here, as before, The semantic distinction between stage-level predicates and individual-level predicates as the relation between the verbal elements and the nominal elements. Also there are was first proposed by Carlson (1977a) to account for the different interpretations of the different types of subjects that will play a similar role as nominal elements played in the bare plural subjects. Subjects of individual-level predicates bare plural receive a generic object position. interpretation and subjects of stage-level predicates bare plural receive an existential I will take, however, the opposite route I took in the previous chapters. In the interpretation. previous chapters I kept the verbs pretty much constant and I played with different forms Kratzer (1989) and Diesing (1992) propose that the distinction between individual- of arguments. Here, first I will keep the DP subject as constant as possible and I will play level and stage-level predicates is to be derived in the syntax. With respect to copula with the verbal material. Then I examine the behavior of different subject interpretations constructions, Diesing follows Stump (1985) and proposes that English has (at least) two depending on what the verbal elements that it will combine with are. copula verbs: BE1, which takes stage-level predicates and BE2, which takes individual- In order to do this I will examine the interpretation of subjects in two types copula constructions in Brazilian Portuguese: predicates that have no temporal information (ser predicates) and predicates that have temporal information (estar predicates). Predicates that do not have temporal information roughly correspond to copula constructions of what level predicates. She also suggests that ser and estar copula verbs in Portuguese and Spanish in fact correspond to BE1 and BE2, respectively. At first sight, the correlation between ser and estar and individual-level and stagelevel predicates seems straightforward. Not only does ser+predicate denote permanent has been called individual-level predicates (but see section 1). These predicates will enter 160 A way to distinguish A-quantifiers in the VP level than in the IP level is to make the latter dependent on tense, as the perfective and imperfective distinction in Brazilian Portuguese and Spanish which is only present in past tenses. 356 357 properties and estar+predicate denote transitory properties, but more importantly the accounts are empirically and theoretically untenable. A purely semantic account, subject interpretations seem to pattern as expected by her theory.161 however, as proposed by Carlson (1977a) will be shown to require us to multiply the number of lexical entries for ser and estar. (5) a. Bombeiros são/*estão altruístas. Firemen are altruistic. Instead, I move towards a more traditional account and propose that the distinction (Individual level predicate) for temporal information and estar is specified for temporal information. In other words, b. Bombeiros *são/estão disponíveis. Firemen are available. between ser and estar is a reflex of aspectual properties of these verbs: ser is unspecified (Stage-level predicate) ser is neither plus nor minus [ADD TO]. Estar is [–ADD TO], and thus has temporal properties. This distinction accounts for the two verbs' distributional and selectional c. #Um homem é altruísta. (IL) A man is altruistic. properties, including the ACT BE (Partee 1977; Williams 1984) interpretation, without the need for the multiplication of lexical items. An investigation of the properties of the d. Um homem está disponível. (SL) predicates taken by ser and estar, respectively, points toward a distinction in the argument taking properties of adjectives. My proposal is that estar is an aspectual verb that takes A man is available. [+V] predicates and ser is a true copula and takes [+N] predicates. In [+N] predicates, the Thus, in (5a), an individual-level predicate, the only possible interpretation for the arguments do not take direct part in the aspect composition, i.e., we do not have the bare plural is generic. In (5b), a stage-level predicate, the bare plural receives an configuration to build a PATH between the head element and the internal argument. This is existential interpretation and more marginally a generic interpretation. In (5c) the the case of pure ser predicates, tough constructions (which not surprisingly appear with existential indefinite reading is impossible and the only possible reading is that of a the copula ser), multiple subjects in Japanese and many genitive constructions. The partitive (i.e., one man of the group is altruistic); and (5d) allows an indefinite and a arguments taken by predicates that are [+V] may take direct part in the inner aspectual partitive reading. composition and therefore are spatio-temporally situated. Therefore, they can, depending In this chapter I first examine a syntactic treatment of the distinctions between ser and estar copula constructions in terms of the stage-level and individual-level distinctions as on what is in the head of AgrS, impose partitions on the situation that will allow as to define a situation as bounded or unbounded. proposed by Diesing (1992). Based on this examination, as well as data on from English The advantage of distinguishing the argument taking properties of different types of and Dutch (cf. de Hoop 1992), I argue against the syntactic accounts for distinguishing predicates is that such a distinction allows us to reinterpret the lack of existential readings ser and estar in terms of individual-level and stage-level predicates. I show that such for subjects of certain predicative constructions and allows us to begin to explain the following apparently unrelated facts: (i) why estar predicative constructions can take 161 The examples presented in this chapter are all from Brazilian Portuguese, although with relatively minor lexical variation the adjective selection of ser and estar, hold for Spanish and European Portuguese. One important difference between Spanish and Br. Portuguese is that only the latter allows bare plurals in subject position. 358 purpose clauses; (ii) why ser displays the opposite behavior; (iii) why absolute 359 constructions (without the copula), secondary predicates and reduced relatives can only 6 . 2 . 1 The VP/IP split hypothesis appear with adjectives that are temporally specified. Bare plurals of stage-level predicates (SLPs) allow existential and generic readings and The rest of the chapter is divided as follows: in the second section I argue against the individual-level predicates (ILPs) allow only generic readings of the bare plural. In syntactic accounts for the interpretation of the bare plural subjects of stage-level and Discourse Representation Theory (DRT) (Kamp 1981; Heim 1982), this amounts to individual-level predicates. In the third and fourth sections, I develop an alternative saying that individual-level predicate bare plural subjects cannot appear in the nuclear account based on aspectual properties of the verbs ser and estar. The "aspectual account" scope at the level of logical form, where existential closure applies, but instead they must can be extended to explain the interpretation of the subjects of ser and estar predicative appear in the restrictive clause where they can be bound by a generic operator. Stageconstructions. In the last section, I deal with the selection of ser and estar and propose an level predicate subjects, on the other hand, can appear in the nuclear scope or in the account for the distribution of ser and estar adjectives in contexts such as secondary restrictive clause and receive existential and generic readings respectively. With respect predicates, reduced relatives and absolute constructions. I propose that ser takes nominal to indefinites, roughly the indefinite reading appears in the nuclear scope, and the partitive elements and estar verbal elements. Thus, whatever partition the subject imposes on reading, in the restrictive clause. ser+predicate, the subject cannot be the domain of a function that goes to a PATH because Adopting a form of DRT, Kratzer (1989) and Diesing (1990; 1992) argue that the there is no PATH, i.e., there are no pairs of indices and positions. The only relation to ability or inability of an element to appear in the nuclear scope and/or in the restrictive time is through tense. In other words, subjects of plain ser predicates establish relations clause is to be derived from its position in the syntax. In other words, there is a direct that are closer to regular nominal modification. In verbal predicates with estar, the mapping between the syntactic and semantic representations and this mapping takes the subjects cannot force bounded situations because we are dealing with a [-ADD TO] verb following form: and subjects will play a role in the aspectual composition. (6) 6.2 Against the VP/IP split hypothesis The basic motivation for a syntactic account of individual-level and stage-level predicates Mapping hypothesis Material inside VP is mapped into the nuclear scope. Material outside VP is mapped in to the restrictive clause. is the VP/IP split hypothesis. In this section first I outline the VP/IP split hypothesis and Thus, although both subjects of individual-level predicates and stage-level predicates then I explore ser and estar predicative constructions in the light of Kratzer and Diesing's can appear outside VP, only the stage-level predicate subjects can appear either inside or syntactic accounts. The main purpose of this section is to show that these accounts cannot outside VP at the relevant level for interpretation.162 Individual-level predicates, on the handle the properties of ser and estar predicative constructions. other hand, are always interpreted outside VP. 162 The level in which the binding of the bare plural variable applies differs from language to language according to the Earliness Principle (Pesetsky 1989). In English where subjects are always in the Spec IP, the relevant level for binding is LF, but in German, where subjects can appear inside or 360 361 6 . 2 . 2 Against two different D-structures c. When(ever) Mary speaks French, she speaks it well. Alwaysl [ (speaks (Mary, French, l )] [speaks well (Mary, French, l )] For Kratzer, the argument structures of ILPs and SLPs are different in that stage-level predicates have an extra argument—a Davidsonian argument—for spatio-temporal If ser+predicate is an ILP, then, provided there is no indefinite, the sentence should relations. This argument is marked as the external argument (in Williams' 1981 terms). be ungrammatical under when(ever). If estar+predicate is a SLP, the prediction is that the Assuming external arguments are mapped outside VP, every time the Davidsonian sentence would be grammatical even if there is no indefinite, given the availability of the argument is present, the next argument will occupy the specifier of VP. This is what Davidsonian argument. The following examples support this idea. happens with subjects of stage-level predicates. The Davidsonian argument is an implicit argument of the verb and functions as a variable that only shows up if there is something (8) a. *Sempre que Maria é feliz, ... to bind it. When(ever) Maria is happy ... One of the pieces of evidence for the existence of this implicit argument is the b. Sempre que um homem é feliz... behavior of when(ever) clauses with SLP and ILPs. When(ever) clauses restrict a When(ever) a man is happy... temporal operator "always" which binds a variable in the restrictive clause. The variable can be provided either by a Davidsonian argument or by an indefinite (see Heim 1982). (9) Thus (7a) is unacceptable because the predicate is an ILP and therefore has no Sempre que Maria está feliz, ... When(ever) Maria is happy... Davidsonian argument for the operator to bind. In (7b), the indefinite, and in (7c), the Davidsonian argument, provide the variable and the result is an acceptable sentence. However, there is a group of constructions with ser that do not meet the expectations: (i) some adjectival predicates (cruel, rude, mean, etc.); and (ii) pseudo-equatives with (7) a. *When(ever) Mary knows French, she knows it well. proper nouns. Although there is no indefinite in (10) and (11) the sentences are perfect. *Always [knows (Mary, French)] [knows well (Mary, French)] (10) Sempre que Maria é rude/cruel/gentil, ela é mesmo rude/cruel/gentil. b. When(ever) a Moroccan knows French, she knows it well. When(ever) Maria is rude/cruel/gentle, she is really rude/ cruel/ gentle. Alwaysx [ (Moroccan (x) & knows (x, French)] [knows well (x, French)] (11) outside the VP the relevant level is S-structure. (See below and de Hoop 1992 for problems for this proposal.) Since Kratzer and Diesing's proposals were written assuming framework of Chomsky (1981, 1986), criticizing them for being non-minimalist is obviously unfair. It is worth noting, however, that, to the extent their explanations of the empirical facts are based on different levels of representation they pose serious problems for those wishing to adapt them to minimalism, particularly because Spell-Out is not an analog to S-structure since it reflects only the result of movements due to strong morphological checking requirements. (Although see Hornstein 1995 for a version of the Mapping Hypothesis that is in principle compatible with Minimalism.) 362 Sempre que Clark Kent é Superman, a cidade está segura. When (ever) Clark Kent is Superman, the city is safe.163 163 I owe this example to Juan Uriagereka. 363 In order to explain these facts it would be necessary to postulate Davidsonian Also, Diesing points out some mismatches in the interpretation of the bare plural and arguments in (10) and (11) (but not in (8a)). But if there is a Davidsonian argument in of the predicate, namely psych predicates (be anxious, for example) and verbs like belong these cases, then the generalization that ser+predicate is an ILP and estar+predicate is a to. According to her, these stative predicates behave ambiguously. The psych predicates SLP is not valid anymore. Arguably, in (10) and (11), ser+predicate is behaving as a "might intuitively seem to be stage-level predicates in that they are transitory states. SLP and in (8) as an ILP. Another possibility is that ser+predicate can also be interpreted However, applying the tests described above [bare plural interpretation, there contexts, as a stage-level predicate; but then why can it not do the same for feliz 'happy'? We etc. CS] places them in the category of individual-level predicates" (Diesing 1990:67). In cannot appeal to lexical ambiguity of ser, because if there were two verbs ser, one for the belong to case, although the predicate denotes individual-level properties, it "shows stage-level predicates and one for individual-level predicates, then we run into two many of the syntactic and semantic properties of stage-level predicates." (1990:76) For additional empirical problems: (i) why adjectives like cruel cannot appear with estar in example, belong to is permitted in there-constructions. such constructions and (ii) why the bare plural is still generic in (12). If (12) were a To overcome these empirical problems, Diesing 1990,1992) proposes a two-way distinction between SLPs and ILPs. The former are treated as instances of raising and the stage-level predicate, an existential reading should be available. latter as instances of control. In both cases there is a subject base-generated inside VP. In (12) Policiais são rudes/ cruéis SLPs the subject raises from Spec VP position to Spec IP because Infl is not a theta Policemen are rude/cruel... (generic interpretation) assigner. In ILPs, a PRO is generated inside VP, and the overt subject is generated in Spec IP, being theta marked by Infl. In sum, some predicates that appear with ser can pass the "stage-levelhood" test, although they do not appear in other stage-level contexts.164 The two different Infls are held responsible for the interpretation of the bare plural. The permanent/transitory readings are related to the presence or not of the Davidsonian argument as proposed by Kratzer. Diesing's proposal is summarized in (13): 6 . 2 . 3 Against two different Infls Diesing 1992) rejects Kratzer's account for the interpretation of the bare plural subjects in (13) stage-level and individual-level predicates. As evidence against Kratzer's proposal, +Θ Infl –Θ Infl (CONTROL) (RAISING) Diesing points out that assuming individual-level predicate subjects are not generated inside VP, quantifier floating facts will have to be explained in another way (as discussed + Davidsonian argument be anxious eat in Bonet 1991.) In other words, quantifiers should not be allowed to be left behind in – Davidsonian argument know belong sentences like the pigs are all stout, if the subject is generated in Spec IP. It is important to note that this two-way distinction completely undermines Carlson's 164 More mismatches between the interpretation of predicates and the interpretation of subjects have been noted by Stowell (1991a). For more arguments against this hypothesis, see de Hoop (1989,1992). 364 (1977a) correlation between the bare plural interpretation and the level of the predicate. 365 For Carlson, the interpretation of the bare plural is dependent on the interpretation of the seems that the problem has been just relocated, because to say that the external argument predicate. is licensed by Inflection and is coindexed with the PRO external argument inside VP There are problems, however, with implementing the two Infl proposal for ser/estar. amounts to saying that know has three arguments, two of which are external. Leaving the issue of the Davidsonian arguments aside, and leaving aside the discussion of Furthermore, going back to ser predicative constructions, in the cases in which the theoretical consequences of assuming that Infl assigns a theta role to the subject of ser+predicate appears to have a Davidsonian argument (be cruel, for example), it would ILP, let us test ser/estar with respect to the Infl hypothesis. be necessary to create another lexical entry for ser. In short, if we treat ser as a control Given the behavior of the bare plural (example (5) above) the hypothesis is that ser takes a theta assigning Infl and estar takes a non-theta assigning Infl.165 Thus ser is a control predicate and estar is a raising predicate: verb, we can't account for its raising properties. If we treat it as a raising verb, then we don't have an explanation for the lack of existential readings for the subject. For estar, the raising hypothesis would work: we can account for the existential interpretation of the bare plural subject in these constructions, given that the lexical (14) a. [IP NPi I0 [VP PRO ser intelligent]] b. [IP NPi I0 [VP ti estar available]] The immediate problem with this proposal is that copula constructions with ser subject is generated inside VP and can be reconstructed back at LF166 (example (5)) and for the possibility of having expletive subjects: (16) a. Está tarde. display properties of raising constructions and not of control structures. They take (It) is late expletive subjects and the verb seems to play no role in the selection of the subjects of the 'It is late.' small clauses they take. Under Diesing's analysis, however, if ser is a raising verb, there b. Está certo que a gente vá ao Congresso. is no way to account for the bare plural interpretation. (It) is certain that we will go to the Conference. 'It is certain that we will go to the Conference.' (15) a. É tarde. (It) is late. In languages that allow subjects to appear inside or outside VP at S-structure, b. É claro que vamos ter um aumento. It is clear that we will have a pay raise. according to Diesing, the binding of the indefinite happens at S-structure. In those languages, more problems arise for the VP/IP split hypothesis. De Hoop (1992) provides evidence that the mapping hypothesis is unable to account for the interpretations of other More importantly, the control hypothesis was designed to explain why the generic interpretation is the only possible interpretation for the bare plural subjects. However, it 165 We ignore here the possibility that ser and estar are generated in Infl, in view of the work done by Zagona (1988) on auxiliaries in Spanish. 366 166 at LF. I am assuming here that Portuguese is like English, where the binding of the bare plural happens 367 indefinites in subject position in Dutch and German, which according to Diesing and for the subject readings) seem unable to derive the facts related to the subject Kratzer's accounts are bound by an operator at S-structure. interpretations of stage-level and individual level predicates. Consider the following cases with other indefinites: In the remainder of this section I want to point out that the VP/IP split hypothesis is also unable to account for the distribution of individual-level and stage-level predicates in (17) weil zwei Wildshweine ja doch intelligent sind. (German) general. At the same time I want to show that Carlson's account of predicative since two wild boars 'indeed' intelligent are constructions in purely semantic grounds results in the need to postulate more than one 'since two wild boars are indeed intelligent.' lexical entry for ser and estar. As Diesing (1990,1992) herself notes, the VP/IP split hypothesis does not account (18) weil ja doch zwei Wildshweine intelligent sind. since 'indeed' two wild boars intelligent are naturally for the following facts: (i) why small clauses under consider do not exhibit 'since indeed two wild boars are intelligent.' existential readings although other ECM and for clauses seem to allow both interpretations depending on the predicate they take; (ii) why with clauses are generally Diesing assumes that the adverb ja doch indicates whether the subject is in a VPinternal or a VP-external position. Given that, the prediction is that (18) should be unacceptable, given that the subject should not be generated inside VP. But it is fully bad with individual-level predicates but do not permit generic readings. The problem is not trivial under the VP/IP split hypothesis. Consider the following facts (from Diesing 1992 fn. 17): acceptable. Furthermore, the interpretation of zwei Wildshweine is of a partitive (a reading obtained in the restrictive clause). Here the only way to account for this fact is to say that the subject is VP external after all and the adverbial is adjoined to IP, as Diesing (1992) actually does. The Dutch cases, however, do not seem to allow any way out. Here the subject appears VP internally and the interpretation is of a partitive. Again, these facts (20) a. I consider firemen available. (generic only) b. I consider firemen intelligent. (generic only) (21) a. I believe firemen to be available. (both generic and existential) b. I believe violists to be intelligent. (generic only) are unexpected. (22) a. With firemen available, we are well protected... (19) Els zegt dat er twee eenhoorns intelligent zijn. (Dutch) b. *With firemen intelligent, we have nothing to fear. Els says that there two unicorns intelligent are 'Els says that two (of the) unicorns are intelligent.' If we take small clauses to be clausal, e.g. to be of the form [IP NP [AP t A0]] for stage-level predicates and [IP NP [AP PRO A0]] for individual-level predicates and In sum, both Kratzer's account (which adds an external argument to the stage-level therefore allow some version of the VP/IP split hypothesis, we can account for (21) but predicates in order to account for the subject interpretations) and Diesing's account then it is not clear (i) why with clauses should not allow a generic reading (22a) and (ii) (which adds an external argument—PRO—to the individual-level predicates to account why the consider small clauses should disallow existential readings in (20b). 368 369 We could suppose, on the other hand, that there is a structural difference between adjunct small clauses and complement small clauses as it has been proposed in the c. *Há metalúrgicos estando ansiosos/ em greve. There are steelworkers being anxious/ on strike. literature (Pesetsky 1982 and others). In this case, the generic reading is not available to with clauses because they are phrasal categories (as opposed to clausal) and genericity is Ser, on the other hand, can also combine with some adjectival predicates and appear probably a clausal (perhaps CP) property. This would be in the spirit of the VP/IP in these contexts. In these cases, a bare plural subject will receive an existential hypothesis. However, if we explain the lack of generic readings in terms of the lack of IP interpretation and be interpreted as having some control or agency over the subject. This in small clauses, then it is unclear how to rule out the existential readings in the consider is equivalent to what has been called ACT BE (Partee 1977, Williams 1984 and others). small clauses as in (20a). According to Williams (1984) the ACT BE interpretation corresponds to the main verb BE Moreover, as shown by Raposo and Uriagereka (1992), in Italian and Spanish, for as opposed to the auxiliary BE which disallows such an interpretation. example, consider small clauses are subject to restructuring. In this case, the small clause is part of the VP. Why is an existential reading not available for the subject of the small clause?167 I will come back to these problems in the section 5. 6 . 2 . 4 Ser and estar and the stage-level contexts (24) a. Eu vi Maria ser cruel (para com os gatos). I saw Maria be cruel (to the cats). b. Maria está sendo cruel. Maria is being cruel. Ser and estar pose more problems for the distribution of stage-level and individual-level predicates. Carlson (1977a) noted that there are contexts where a SLP can occur but an ILP cannot, namely existential constructions, progressives and perception verbs (but see c. Há mulheres sendo cruéis. There are women being cruel. de Hoop's facts above). Although estar passed all tests for stage-levelhood in Kratzer and Two questions immediately arise: (i) why can ser but not estar have the ACT BE Diesing's terms, it cannot appear in any of the typical stage-level contexts. interpretation, and (ii) why can estar not appear in the so-called stage-level contexts. In (23) a. ??Eu vi metalúrgicos estarem ansiosos/ em greve.168 I saw steelworkers be (3ppl) anxious/ on a strike. an attempt to maintain Diesing's proposal at least for estar, we could answer (i) by saying that there is an ACT ser, but not an ACT estar. Thus, nothing but a lexical idiosyncrasy would be at stake. For (ii), one could say that stative verbs are in general disallowed in b. *Metalúrgicos estão estando ansiosos/ em greve. Steelworkers are being anxious/ on a strike. these contexts and estar is a stative verb.169 This is partially correct, but a more refined notion of stative verb will be needed to explain why other stative verbs (sit, stand, lie, and others) are possible in these contexts. Examples (25 a,b) from Dowty 1979:173) 167 For other arguments see Raposo and Uriagereka (1992). 168 This reading improves if see is not used as a verb of physical perception, but as a verb of mental perception. 370 169 Curiously, this is exactly the history of estar. It derives from the Latin estare, which was a locative verb like stand. 371 illustrate the progressive with lie and sit, and examples (25 c,d) illustrate the fact that 6.3 An aspectual analysis these stative verbs can appear in existential constructions and embedded under perception The main goal of this section is to reduce most of the properties of ser and estar verbs, respectively: predicative constructions to their aspectual properties. (25) a. The socks are lying under the bed. 6 . 3 . 1 Reanalyzing the tests for Davidsonian arguments b. Your glass is sitting near the edge of the table. c. There are socks lying under the bed. We have been using adverbials such as for an hour, in an hour, as tests to determine the d. I saw the woman standing in line. VP aspect. They can also modify the IP as the example below, repeated from the introduction of this chapter. Another way to account for these facts is to assume that estar is in fact a function that takes a SLP and forms an ILP. This is the semantic approach taken by Carlson (1977a). He proposes that there are in fact four verbs BE, i.e., four functions BE: BE1 combines (26) a. Water leaked through John's ceiling for six months. b. *A gallon of water leaked through John's ceiling for six months. with stage-level predicates to produce individual-level predicates. BE2 combines with individual-level predicates and produces individual-level predicates. BE 3 shifts an individual-level property to the stage-level, and BE4 combines with stage-level properties and produces stage-level predicates. Thus, there should be two estar verbs: one combines with stage-level predicates and generates stage-level predicates, and the other estar The adverbial for an hour adds a temporal boundary for the predicate and induces a repetitive meaning or an extension of the event in time (27) a. John hit the ball. b. John hit the ball for an hour. (iterative event) combines with stage-level predicates and produces individual-level predicates in certain contexts. With respect to ser, at least two verbs ser have to be postulated: one for the cases that pass Kratzer's test for stage-level predicates an one for the other adjectives that (28) a. John ran. b. John ran for an hour. (non-iterative event) do not pass that test. A third verb ser may be added: the auxiliary for the passive which can take both individual-level and stage-level participles. In sum, the only way to accommodate the syntactic and semantic facts is to proliferate the number of lexical entries for these verbs and determine which of them can appear in Now consider the test Kratzer uses for stage-level hood. There is an interesting similarity between (26) and (29). The type of the complement of the verb changes the behavior of the predicate in these tests. each context in a rather ad hoc way. 372 373 (29) a. *When(ever) Mary knows French, she makes sure... plural is generic and they are unacceptable in there constructions. The same predicates that b. When(ever) Mary knows the answer, she makes sure... are grammatical in when(ever) contexts, are ungrammatical in there constructions (32c) c. When(ever) Mary likes the dessert, she eats more than she should. and the interpretation of the bare plural is generic. There are two possible ways to account for these facts. Either the when(ever) test is Thus, although a verb like know is considered an individual-level predicate and thus not quite a test for a Davidsonian argument, or the predicates in (32) are ambiguous, i.e., should have no Davidsonian argument, this verb can appear in when(ever) clauses there are two lexical entries: a stage-level and an individual-level one—one with and one depending on the complement of the verb. In Kratzer's terms, the facts in (29) are without a Davidsonian argument. As noted before, the lexical ambiguity explanation for unexpected: (29a), (b) and (c) do not have an indefinite in subject position (i.e., in the the behavior of these predicates cannot be correct, because these predicates are not restrictive clause). Given that these sentences are perfect, the only possibility for (b) and genuinely ambiguous between a stage-level reading and an individual-level reading. If (c) is to suppose that the definite object, but not French can move at LF to escape the they were lexically ambiguous, they should have the properties of both types of nuclear scope and provide the variable necessary to bind the operator in order to avoid predicates with respect to all the linguistic phenomena that seems to be sensitive to the vacuous quantification. However, some other explanation for the cases in (10) and (11) distinction, but they do not have them. In other words, if they can have a Davidsonian (repeated here as (30) and (31)) and the case in (32) is needed, because in none of these argument, they should be fine in (32c). The stage-level possibility, however, is blocked. cases is there an indefinite or a definite that could be interpreted as an indefinite. In this (See also Condoravdi 1992 for similar observations.) case, we need a completely different explanation for the unacceptability of (29)). If, on the other hand, the when(ever) test is not quite testing for stage-levelhood but instead it is quantifying over situations, then we could say that when(ever) is looking for (30) Sempre que Maria é rude/cruel/gentil, ela é mesmo ... When(ever) Maria is rude/cruel/gentle, she is really ... (31) Sempre que Clark Kent é Superman, a cidade está segura. When(ever) Clark Kent is Superman, the city is safe. situations that can be quantified by a temporal adverbial and thus induce multiple event readings. Moreover, the ability to quantify over a situation depend on having situations and on the pragmatics. In other words, hit the nail can be repeated but kill Mary cannot. Being an unselective binder when(ever) can quantify over any element of a situation. If pragmatically it is possible to have multiple individual situations, then we can explain the (32) a. Whenever John is serious/decisive..., his proposal is usually accepted. b. Politicians are serious/decisive/... (only generic reading). apparent problem above. As noted by de Hoop 1992, any predicate that cannot be repeated sounds odd in these constructions, as exemplified in (33): c. *There are politicians serious/decisive/... (33) a. Whenever Bertolucci loads the camera, he ruins the film In (32a) serious, and decisive behave like stage-level predicates, providing a b. #Whenver Bertolucci makes the movie, the Academy ignores him. Davidsonian argument to bind the operator. In (32b) and (32c), on the other hand, these predicates behave strictly like individual-level predicates: the interpretation of the bare 374 375 Despite the fact that make is an eventive verb, and should supply an argument for whenever to bind, (33b) is unacceptable. On the other hand, (33a), which can describe as providing independent evidence for the analysis of determiner transparency presented in chapter 4. an iterative event, is quite fine. In other words, what I am proposing is that when(ever) The real question then, is what is the difference between ser cruel 'be cruel' but not is playing a similar role to other temporal adverbials and operators in the sense that (i) ser inteligente 'be intelligent'. Why can only the former be interpreted in an iterative way certain aspectual selection restrictions must be met (when(ever) can only appear with without the addition of external boundaries?170 I will postpone this discussion right situations that can be iterated over time), and (ii) the adverbial interacts with the clause now. For the moment it suffices to say that cruel and happy have different argument forcing the repetitive meaning. taking properties. In any case, we cannot solve the problem of how to distinguish cruel from happy by stipulating that cruel is a stage-level predicate, because cruel is not good in Now consider (34). any other stage level contexts. (34) a. Whenever Bertolucci makes the wrong movie, the Academy honors him. b. #Whenever Bertolucci makes the lousy movie, the Academy honors him. (35) a. Whenever Bertolucci makes the movies that Bill likes, Bill goes to see them. b. #Whenever Bertolucci makes the movies that are in the library, Bill goes to see them. The second test used by Kratzer involves locative modifiers in German. According to Kratzer, stage-level predicates with locatives present two readings. Either (i) the spatial or temporal expression can modify the restricting predicate of the quantifier or (ii) the spatial or temporal expression can modify the main predicate of the sentence, because there is an event argument to be modified in these cases. Sentences with individual-level predicates, on the other hand, allow only modification of the restrictor of the quantifier, because there is no event variable. This distinction also apparently holds for the same sentences in The examples above show that whenever is sensitive to the determiner transparency Brazilian Portuguese, as long as there is a fair intonation break for interpretation (b).171 effects discussed in chapters 4. The contrast in (34) shows exactly the difference between wrong and difficult discussed in chapter 4. Similarly (35) shows the same effect with a (36) relative clause. If whenever is an unselective binder that can bind an indefinite in its porque quase todos os refugiados nesta cidade pereceram. a. since almost all of the refugees in this city perished. scope, then determiner transparency effects raise the indefinite out of the scope of the 'of the refugees in the city almost all died.' definite determiner, making it susceptible to being bound by whenever. Since (33) b. since almost all the refugees perished in this city. shows that this cannot be the whole story (unless we posit an implicit indefinite in load 'of the refugees who perished, almost all did in the city.' (not necessarily an unreasonable move if load means "load with film")) more remains to be said about how whenever operates. It is not my intention to provide a full analysis of whenever here, but the data above are suggestive, and point to the essential correctness of De Hoop's observation, as well 170 For example, it seems to me that when(ever) Mary is happy in a place, she has to move is perfectly acceptable. In Mary knows French, knowing French is an intrinsic property of Mary and Mary cannot be used to partition sub-situations of knowing French. 171 376 The same need for a fair intonation break is found in English. 377 (37) porque quase todos os cisnes na Austrália são negros. (40) Since almost all African Americans are black. since almost all swans in Australia are black. 'But not vis-a-vis the standards of certain African countries, where they 'of the swans in Australia almost all are black.' The examples below illustrate the behavior of almost with ser and estar, respectively: would be taken as light.' If this is correct, the interpretations have nothing to do with presence or lack of events, and this test cannot be applied for distinguishing adjectival predicative (38) porque quase todos os serviços no Brasil são essenciais. constructions with respect to stage-level or individual-level interpretations. a. since almost all the [services in Brazil] are essential. The use of locatives shows another distinction between ser and estar constructions. 'of the services in Brazil, almost all are essential. Consider the following sentences: b. since almost all the services are essential in Brazil. 'of the services almost all are essential in Brazil.' (41) a. Os livros estão encadernados na biblioteca. The books are bound in the library. (39) porque quase todos os serviços no Brasil estão paralisados. 'The books are bound in the library. They are not unbound any more.' a. since almost all the [services in Brazil] are paralyzed. 'of the services in Brazil, almost all are paralyzed.' b. Os livros são encadernados na biblioteca. The books are bound in the library. (as opposed to unbound) b. since almost all the services are paralyzed in Brazil. 'The books in the library are bound.' 'of the services almost all of them are paralyzed in Brazil.' As noted by Lemos (1987), in (41b) the adverb cannot modify the predicate; it is The examples (38) and (39) show that both types of predicates allow ambiguities. specifying the normal (established or inferred) property of the books. In the estar Therefore this piece of evidence cannot be used to argue for an extra element in stage- construction, on the other hand, the adverb can in fact modify the predicate. Why can't level predicates, unless the locative itself is creating the extra argument necessary for a the predicate be modified in (41b)? stage-level interpretation. But if this is so, the question is why be black in (37) allows Locative adverbials can be associated with (i) "aspectual"172 properties of the VP; (ii) only one interpretation and be essential in (39) allows two interpretations. When we "aspectual" IP properties (the so called perfective and imperfective markings for examine (40), it becomes clear that we can reinterpret the alleged ambiguity in the stage- example); (iii) other adverbs like when, tense and modals. level predicates by adopting the hypothesis that some adjectives can be further relativized, but others are odd, unless we establish a context in which it makes sense to relativize them. In (37) black cannot be further modified, but in (40), it can. 378 172 Aspectual here is again not to be interpreted as a specific node for aspect but as the place where the verb- internal temporal information is matched with nominal information. One cannot be in a place without being in a time. 379 Suppose ser lacks intrinsic aspectual properties (i.e. it does not establish roles for the composition. In this case durative or terminative readings cannot be derived, because their participants in the event, because there is no event) at the VP level: it is neither a state nor properties are the result of the interaction between properties of the verb and of the nouns an event. It is a dummy verb. Thus, unless tense or an adverbial like when is present, a and the verbal properties are missing in the verb. Ser is exactly this type of verb. The VP headed by ser has no temporal information, locative adverbial can only modify a default modality operator that gives this law-like i.e., it doesn't have any [ADD TO] specification. Other verbs have semantic content and reading. In (41a), ser is the verb and the tense is the present tense. Given that the present therefore, being verbal have temporal information. They will be either [+ADD TO] or [– tense is probably not a tense in Portuguese,173 there is no temporal structure to be ADD TO]. Among these verbs there is a class of verbs that seem to add to a situation an modified. In these cases, some modality is set by default and the locative can be extra temporal dimension. These are the so called aspectual verbs that have been interpreted as modifying this normal situation. The important point here is that no other described in the more traditional literature as verbs of initiation (start, begin), cessation interpretation beside the modality related one is available for the locative adverbial in (stop, finish) and continuation (keep, continue). Such a classification is, however, at (41b). In (41a), on the other hand, the locative can modify the predicate. odds with what we have been arguing aspect is about, i.e., participant related partitions of In short, the discussion above suggests that the apparent distinctions between ser and an event. estar in terms of absence or presence of a Davidsonian argument should be reanalyzed in However, there is another way of thinking about these verbs that is more akin to what terms of aspect. At this point, I abandon the attempt to distinguish ser and estar as the I have been arguing so far.174 We can think of these elements as quantificational in lexical counterparts of the stage-level and individual-level distinctions in terms of +/- nature. In fact Löbner (1987) has proposed that they should be treated as relations Davidsonian arguments. between propositions and times. Ter Meulen (1990) makes a similar proposal. She treats these elements as relations (aspectual relations) between an antecedently determined 6 . 3 . 2 Ser /estar: an aspectual distinction reference time t as the restrictive term and the event-type E described by the complement In the discussion of ser and estar VPs, of course, the question is not whether the VP is in the nuclear scope. In other words aspectual verbs are to be interpreted as a generalized [durative] or [terminative], given that both verbs are not [+ADD TO] i.e., they do not quantifiers over (reference) times as exemplified roughly in (42). denote change in time. The question is what distinguishes them. The basic idea is that, before we distinguish verbs as [+ADD TO] or [–ADD TO], we need to distinguish whether (42) a. start (t, E) = ∃t E(t) the verb is marked as having aspectual specifications or not. When the verb doesn't have b. keep (t, E) = ∀t E(t) any semantic content it doesn't describe any eventuality. Therefore it doesn't describe This way of viewing aspectual verbs as generalized quantifiers allow us to establish anything. Thus it cannot assign any roles to its arguments as participants in the aspectual correlations between aspectual verbs. Thus Ter Meulen proposes that start corresponds to 173 Like English, the present tense in Brazilian Portuguese cannot be used to report here and now events. For an analysis of different interpretations of the present tense, see Zagona (1993). 380 174 This is not the only way in which aspectual verbs can be analyzed. See Verkuyl (1995). 381 an existential quantifier because it introduces a new event. Keep and continue, on the other hand, correspond to universal quantifiers since for all times after the reference time, (43) Maria está grávida Maria is pregnant the event holds. (Keep and continue have different presuppositions which correspond to a. If Mary was pregnant from May to January, then Mary was pregnant from event internal and event external presuppositions: if Mary kept coughing through the July to August concert, she may not have been coughing before it, but if Mary continued coughing b. If Mary is pregnant with a boy then Mary is pregnant through the concert, she must have been coughing before it.) Other verbs such as finish are interpreted as the internal negation of start and the external negation of continue.175 Assuming this hypothesis is to assume that estar selects for complements with temporal Other evidence that Ter Meulen presents for the quantificational nature of the aspectual specification. Thus the interpretations we get with estar as opposed ser have to do with verbs comes from their monotonicity behavior. If keep is to be treated as a universal the fact that estar has the reference time as its first argument and an eventuality as its quantifier like every, then it should be monotone decreasing in the first argument (if every second argument. Ser, on the other hand, is deprived of temporal information and thus woman sings then every beautiful woman sings) and increasing in the second argument gives the interpretation of permanent properties. The examples below illustrate this (if every woman sings an aria, then every woman sings). For keep, this works in the contrast: following way: If Mary kept coughing through the symphony, then Mary kept coughing through the first movement. Since the times of the first movement of the symphony are a (44) a. O João é casado /gordo... b. O João está casado/ gordo... subset of the times of the symphony, keep is monotone decreasing with respect to its first The João is married /fat... The João is married/ fat... 'João is a married man.' 'João is a man who got argument. If Mary kept coughing loudly, then Mary kept coughing. Since the set of events of coughing is a superset of the set of events of coughing loudly, keep is married/fat... monotone increasing in its second argument. Assuming this framework as a guideline we can ask what estar is. I would like to The following facts support the aspectual distinction between ser and estar. First, ser argue that it is the equivalent of keep quantifying over stative events, i.e. eventualities that is the verb used in pseudo-clefts and equative constructions, where the copula does not are states. play any role in terms of selection and does not add any semantic aspectual content to the Consider first estar's presuppositions. If Isadora está morta 'Isadora is dead', there is certainly no presupposition that she was dead before, just as keep doesn't presuppose that the event was going on before. appear in these constructions, as the examples below show: With respect to monotonicity, if we examine estar we can see that it acts as monotone increasing in the first argument and monotone (45) a. O João é/ *está o homem sentado ali. The João is the man sitting over there. decreasing in the second argument: 175 construction. In these cases, ser is just a tense carrier. Estar, on the other hand, cannot The external negation of start is not lexicalized in English. 382 383 b. O que o João é, é/*está incompetente. What the João is, is incompetent. Quase 'almost' is an adverb that can relativize an adjective or can modify an event or subparts of an event (see Pustejovsky (1989), Dowty (1979) for a discussion along these lines) that are not related to the quantity of the object. Thus, in this is almost important, it Second, the expression há uma hora 'it has been an hour' can only appear with estar. is the adjective that has been relativized. On the other hand, in John almost built a house, This fact shows that ser is incompatible with a durative interpretation with an initial point either (i) John didn't finish the house or (ii) John didn't start building the house. In the delimited. first case, the adverb can be seen to be modifying the result state, and, in the second case, the adverb is modifying the whole event of building a house. Notice that we do not find a (46) João *é/está feliz/doente/cruel há uma hora. reading in which John almost built two houses means that he built one of the houses. João is happy/sick/cruel it has been an hour. Thus the partition that is being established by almost is not a participant related partition. 'It has been an hour that John is happy/sick.' It is a partition that can "see" so to speak the beginning or the end of a situation, i.e. its start or end point. Notice that in John almost built houses we only get the beginning point Third, if we combine ser and estar predicative constructions with the so called modification since there is no end point that can be modified because the event is durative. aspectual verbs, the acceptability for ser varies considerably depending on the adjective, Now, consider the behavior of quase 'almost' with ser and estar, respectively: but with estar the interpretation doesn't vary according to the predicate it takes. In other words, ser behaves as if it were not there for the purposes of semantic selection. Only the (48) a. A Maria é quase bonita. predicate counts. For example, in (47) ser is acceptable with começar 'begin' depending The Maria is almost pretty. on the adjective but estar is always unacceptable, no matter what kind of adjective is used, 'Maria is not pretty. She is not ugly.' although the adjective is acceptable with both verbs. b. A Maria está quase bonita. (47) a. João começou a ser inadequado /*doente depois que bebeu demais. João began to be inadequate/sick after he drank too much. b. *João começou a estar inadequado /doente depois que bebeu demais. João began to be inadequate/sick after he drank too much. The Maria is almost pretty. 'Maria is not pretty yet.' Neither (48a) nor (48b) is ambiguous; in (48a) the adverb quase 'almost' can only relativize the adjective. There is no aspectual property to be modified. In (48b) it modifies only the reference point (Mary almost got to the point of being pretty). Up to this point I have presented some evidence to support the hypothesis that ser lacks temporal specifications. In the rest of this section, I present two more pieces of evidence to support the analysis above: the behavior of quase 'almost' and the ungrammaticality of estar in the progressive. 384 385 6 . 3 . 3 Accounting for the lack of estar in stage-level contexts In order to explain the ACT BE interpretation, three questions have to be addressed: (i) why can ser appear in the so called stage-level contexts; (ii) how is the ACT B E Estar, as noted in 1.4, is impossible under perception verbs and in existential interpretation obtained; (iii) why do only some adjectives allow the ACT BE interpretation. constructions (cf. (49) below). In this section I will address the first two of these questions. The third will be partially (49) a. ??Eu vi metalúrgicos estarem ansiosos/em greve. I saw steelworkers be anxious/ on a strike. b. *Tem metalúrgicos estando ansiosos/ em greve. There are steelworkers being anxious/ on a strike. addressed in section 5. 6.4.1 Explaining the ability of ser to appear in SL contexts The first question is why ser can appear in the so called stage-level contexts. The reasoning is very simple. Ser has no temporal specification. Therefore, no aspectual restriction can prevent it from appearing in the progressive or in existential constructions. In these cases, again the nature of estar as an aspectual verb that universally quantifies over states will provide the reason for the unacceptability of (49).In the next section I discuss why ser can appear in those contexts. 6.4 Aspectual composition: the ACT B E In both cases, ser has to combine with the -ndo '-ing' marker and the resulting interpretation is a unit that can be interpreted as an ongoing action. The underspecification of ser also accounts for the fact that it can appear embedded under perception verbs and when(ever) contexts. Being unmarked, ser is not ruled out in these contexts, but the only possible interpretation is temporal, because ser, in these cases, has Ser can appear in the so-called stage-level contexts (progressive, existential constructions acquired temporal specification, either via morphology or by being under the scope of an and embedded under perception verbs) with an ACT BE interpretation (example (18) A-quantifier. repeated here as (50), although estar cannot appear in these contexts. The idea here is that ser is like a dummy verbal element. If it is in the scope of some temporal element, it will be bound by it and acquire a temporal reading, as the schematic (50) a. Eu vi Maria ser cruel (para com os gatos). examples below illustrate: I saw Maria be cruel (to the cats). (51) a. b. Maria está sendo cruel. b. -ndo VP '-ing' ser Maria is being cruel. c. when(ever) . . . ser ver VP 'see' ser c. Há mulheres sendo cruéis. There are women being cruel. In order to answer the second question, i.e., why ser induces an A C T B E interpretation, it is important to note that the ACT BE interpretation is only possible when 386 387 the verb ser has acquired temporal information (52) or is under the scope of some the interpretation of the A-quantifier. An example of this is the perfective marker in temporal modifier (53): Portuguese and Spanish. Here we are going to see similar restrictions for subjects to the ones we found in Polish for objects. We will be able to see what happens to the (52) João foi inteligente. temporally unspecified copula when these elements attach to it. João be-PAST.PERF intelligent. 'João acted intelligently.' (53) In the remainder of this section I will describe three pieces of data that illustrate the restrictions the perfective marker imposes on subjects. Sempre que Maria é rude/cruel/gentil, ela é mesmo rude/cruel/gentil. 6 . 4 . 3 The Perfective/Imperfective distinction in Brazilian Portuguese When(ever) Maria is rude/cruel/gentle, she is really rude/ cruel/ gentle. 'When(ever) Mary acts in a rude way... Brazilian Portuguese and Spanish have two forms of past: a perfective past and an imperfective past. Roughly, the perfective past forces the situation to be interpreted as a This fact, combined with the lack of evidence for a principled distinction between main verb ser and auxiliary verb ser (see Zagona 1988) suggests that the ACT B E definitively past situation. The imperfective, on the other hand, denotes a situation that happened in the past and may still continue in the present. interpretation is compositional in some way and there is no reason to treat the ACT BE as a A lot of work has been done to describe and explain the pragmatics of the distinction different verb (main verb) from the regular verb BE (auxiliary verb), as proposed by between the perfective and imperfective past,176 since only the latter can enter sequence Williams (1984) for English. Before I discuss how the composition may be implemented, of tense effects, as exemplified below: I would like to first show how the perfective marker in (52) interferes with the (54) a. O Pedro disse que a Maria estava grávida. interpretation of subjects. The Pedro said that the Maria was-IMP pregnant 6 . 4 . 2 Aspectual composition b. O Pedro disse que a Maria esteve grávida. As we have seen, the properties of the element in subject position can alter the aspect of The Pedro said that the Maria was-PERF pregnant the situation. We have seen that bare plurals in subject position do alter the aspect of the situation by creating an unlimited number of situations while a singular proper noun will In other words only in (a) Mary can be pregnant at the moment John is talking. In (b) create a unique situation to which it is related (this is not to say that subevents cannot be Mary cannot be pregnant, by the time John is talking. She was pregnant before. produced by the addition of repeatedly, for example.) In all these subevents a subject is In this section, I am not concerned with sequence of tense effects. What I want to involved. show is that the perfective element (just like the Polish perfective for objects) constrains Clearly there are A-quantifiers that seem to only interfere with the interpretation of the types of elements that can appear in subject position as well as the interpretations the subjects by imposing a certain reading on the subject so that the AgrS is compatible with 176 388 See Sten (1973); Rigau (1981) and Lo Cascio (1986) for example. 389 subject can have. I will provide an analysis for these facts that makes use of the same b principles we used to account for the relation between the verb and its internal argument. Se comia demais naquele restaurante. SE ate-IMP too much in that restaurant 'One used to eat too much in that restaurant.' 6 . 4 . 3 . 1 Pronoun interpretation and the perfective/imperfective 'We used to eat too much in that restaurant.' distinction The examples below illustrate the behavior of the impersonal second person singular pronoun tu and the impersonal pronoun se in the present, past imperfective and past c Se comeu demais naquele restaurante. SE ate-PERF too much in that restaurant #'One ate too much in that restaurant.' perfective. 'We used to eat too much in that restaurant.' (55) a. Tu esperas horas por um ônibus no verão. You wait hours for a bus in the summer 'One waits hours for a bus in the summer.' 'You wait hours for a bus in the summer.' In the (a) and (b) examples of (55), two readings are available for the second person singular tu. Either tu corresponds to any person that happens to be waiting the bus or it corresponds to the addressee. In (c), on the other hand, tu must refer to the addressee. In (56) we see the possible interpretations for the pronoun se. In (a) and (b) the preferred b. Tu esperavas horas por um ônibus no verão. You waited-IMP hours for a bus in the summer 'One used to wait hours for a bus in the summer.' 'You used to wait hours for a bus in the summer.' c. Tu esperaste horas por um ônibus no verão. reading is one in which any person eats or used to eat too much in that restaurant. It can also mean that a group including the speaker eats or used to eat too much in that restaurant. In (c), the speaker must be included in the reference of se. The difference between (a-b) and (c), as noted by Hernanz (1991) for Spanish, is not a difference in terms of tense. The difference is due to the perfective and imperfective You waited-PERF hours for a bus in the summer distinction. The notion of integrated whole is somehow incompatible with a non #'One used to wait hours for a bus in the summer.' referential interpretation of the pronoun. 'You used to wait hours for a bus in the summer.' (56) a. Se come demais naquele restaurante. SE eat too much in that restaurant 6 . 4 . 3 . 2 Bare nouns and the perfective and imperfective distinction Now consider the interaction of bare nouns with the perfective and imperfective marker, that we have briefly discussed in chapter 4. 'One eats too much in that restaurant.' 'We eat too much in that restaurant.' 390 391 (57) a. Brasileiro é alegre, diz o Guia Turístico (59) a. O João conhecia a Maria Brazilian is happy, says the Tour-Guide The João knew-IMP the Maria 'João knew Maria' b. Brasileiro era alegre, diz o Guia Turístico Brazilian was-IMP happy says the Tour-Guide b. O João conheceu a Maria no Rio The João knew-PERF the Maria in Rio c. ??Brasileiro foi alegre, diz o Guia Turístico 'João met (for the first time) Maria in Rio' Brazilian was-PERF happy says the Tour-Guide (58) a. Criança pequena chora muito Child small cries a lot Although this alternation is not productive, it shows clearly that it affects the subject by forcing it to be agentive in this case. b. Criança pequena chorava muito quando era vacinada Child small cried-IMP too much, when it was vaccinated c. ??Criança chorou muito quando foi vacinada The final case I want to describe is the interaction between the perfective and the imperfective marker and ter 'have' in existential constructions. 6 . 4 . 3 . 4 Existential Constructions in the Perfective Child small cried-PERF too much, when it was vaccinated In spoken Brazilian Portuguese, the auxiliary verbs ter and estar are homophonous in As we have seen, bare nouns subjects have odd readings if the verb is in the perfective, which means that again we are seeing the perfective interacting with the the perfective as the paradigms below illustrate, because the first syllable of estar is dropped in the spoken language (and also in modern prose): interpretation of the subject. The interpretation of the object is left untouched by the (60) Paradigms of ter in the perfective presence of the perfective and imperfective. In fact almost all the examples in the previous tive had(1SG) chapters used the perfective (unless otherwise indicated) and there the interpretation could tiveste be durative or not depending on the object, and not on the imperfective/perfective distinction. 6 . 4 . 3 . 3 Stative/eventive verbs and the perfective/imperfective distinction had(2SG) teve had(3SG) tivemos had(1PL) tiveram had(3PL) The perfective can sometimes change a stative verb into an eventive verb. An example is given below with the verb conhecer 'know'. 392 393 (61) Paradigms of estar in the perfective (es)tive (63) a. Tem um copo/ uma criança aqui (Present ) Has-SG a glass/a child here was(1SG) 'There is a glass/a child here' (es)tiveste was(2SG) (es)teve b. Tinha um copo/uma criança aqui was(3SG) (Imperfective Past) Had-IMP.SG a glass/a child here (es)tivemos were(1PL) 'There was a glass/a child here' (es)tiveram were(3PL) c. *Teve um copo aqui (Perfective Past) Ter in existential constructions does not agree with the subject of the small clause and Had-PERF.SG a glass here can appear with event nominals such as party, meeting, etc., in the present, past 'There was a child here' perfective and past imperfective, as illustrated below: d. *Teve uma criança aqui (62) a. Tem uma festa/ festas aqui (Present ) Had-PERF.SG a child here Has-SG a party/festas here 'There was a glass/a child here' 'There is a party/parties here' An identical form of (d) is, however, acceptable, namely the estar construction with b. Tinha uma festa/ festas aqui (Imperfective Past) VS order.177 We know it is estar and not ter that can appear in (64) because: (i) the Had-SG a party/parties here subject does not need to be indefinite and (ii) the verb agrees with the subject: 'There is a party/parties here' c. Teve uma festa/festas aqui (64) (Perfective Past) Tiveram os filhos da Aninha aqui (Perfective/estar) Were-PERF.3PL the sons of Aninha here Had-SG a party/parties here Whatever the explanation for the lack of ter with noun phrases other than nouns The perfective however, is unacceptable with non-event denoting subjects while the present and past imperfective are acceptable with these types of subjects. denoting events, the point here is that it is the perfective that is creating this constraint on the subjects of existential constructions, which again forces us to treat the perfective and the subject in a quite close relation. 177 394 For some reason estar does not allow subjects that are event nominals. 395 I do not have any idea for why this restriction is operating. A related fact is that I also 6 . 4 . 4 Deriving the ACT BE don't know why only the perfective seems acceptable in the Verb Subject construction as Ser, being temporally unspecified, cannot create an aspectual domain of its own (i.e., it an answer to a question like (65) or (66). cannot select participants for the event and assign roles to them). Lacking its own (65) Quem telefonou? (Perfective) aspectual properties, the only options for ser are the following: Ser does not have aspectual morphology or tense morphology, and thus the VP ends Who called-PERF? up passing some durative tests by default.178 Adverbials such as for an hour however, a. Telefonou (foi) o Pedro will not be able to modify ser. Called-PERF.3SG was-PERF.3SG the Peter Ser can be generated with temporal properties via morphology or it can combine with 'It was Peter who called' them in the syntax. Since ser has no intrinsic temporal content it will not be modified by perfective or progressive A-quantifiers. It will just acquire the properties of the perfective b. Telefonaram foram os amigos da Cláudia and the progressive and will impose exactly the same restrictions. For the perfective, for Called-PERF.3PL was-PERF.3PL the friends of the Cláudia example, it will disallow impersonal subjects and bare nouns although in the present or in 'It was the friends of the Cláudia who called' the imperfective are perfectly fine as illustrated in (68): (66) a. Quem telefonava pro João todos os dias? (Imperfective) Who called-IMP João every day? 'Who used to call João every day?' (67) a. *Criança foi inteligente Child was intelligent b. *Se foi inteligente b. ??Telefonava (era) a Maria one was inteligente Called-IMP was the Maria Whatever exactly explains the constraints above, it is clear that the perfective is imposing an interpretation of the IP that constrains what can appear in subject position in this construction. Again what is interesting is that the direct object and the indirect object are completely independent of the perfective constraints. This is expected since by the (68) a. Criança era inteligente Criança era inteligente (no passado) Child was intelligent (in the past) 'In the past children were intelligent' time verb+tense are in AgrS, the object is too far away to be affected by the perfective/ imperfective distinction. Now we can come back to the question of what the ACT BE is. 396 178 Note that to say that ser predicative constructions may be interpreted as durative is very different from saying that ser has temporal information. If the sentence is tenseless, a modal operator is set as default and a generic reading becomes available. 397 b. Se era inteligente (naquela época) (69) a. John saw children be cruel with the cats one was intelligent (those days) AgrOP 'Those days people were smart' children Agr With the progressive, the subject tends to be interpreted as expressing some agentivity [see+be]+Agr VP and/or consist of successive phases following one another in time. If the adjective assigns V' an agentive role to the subject, the subject will get an agentive reading; if the adjective t see+be assigns no agentive role and do not allow an interpretation of successive phases following ... VP one another in time, then the result is unacceptable. That is the case of cruel and fat tbe respectively. AgrP tchildren Under perception verbs179 an ACT BE is also possible provided we can see an action. Agr' cruel In this case ser acquires its ACT interpretation by incorporating into the matrix verb.180 The incorporation is required because ser has to move because it ultimately has to check Notice that it is not the movement of the verb to the perception verb that makes the the same tense as the perception verb. The subject moves to that complex where it can subject be interpreted one way or another, it is that whatever role it received from the have its properties checked. The derivation is illustrated below: adjective that now can be interpreted temporally. What is important here is that this type of aspectual composition is only possible if an independent Tense projection doesn't intervene. For example, in (70a), Tense intervenes between the perception verb and ser . The ACT BE in this case disappears. Also, if, instead of a bare infinitive complement of the perception verb, we have an inflected infinitive as in (70b), the ACT BE reading is also unavailable. In this case, the perception verb is interpreted as a verb of intellectual perception. Given that the lack of compositionality between the perception verb and ser indicates the presence of an intervening tense, we can correlate the independent tense of the inflected infinitivals in Brazilian Portuguese with the ability of inflected infinitives to 179 Here I am interested in the physical perception interpretation. In the mental perception interpretation, the perceptual verb and the lower verb do not form an aspectual domain because Tense intervenes (see Rochette 1989 for a similar intuition). license overt subjects and case mark them. 180 Note also that the same generalization holds for causatives: adjectives like cruel, can also appear with ser under the causative fazer 'make' which supports the claim that ser can compose with the upper aspectual element, in transparent domains. 398 399 (70) a. Penso serem eles inteligentes (I) think be-3PL they intelligent-3PL interested in here is the lack of existential readings in individual level predicates, since generic readings are always available to any predicate. The idea here is that existential readings are to be linked to temporal information. 'I think they are intelligent.' Temporally specified predicates will always allow existential readings. Non-temporally b. Vi serem eles inteligentes specified predicates will not allow existential readings of the subject. This is not much (I) see-PERF be-3PL they intelligent-3PL different from Carlson's (1977a) proposal. The implementation of the intuition, though, 'I saw them as being intelligent.' will be different. I will make two basic assumptions here. The first assumption is that ser takes [+N] small clauses and estar takes [+V] small Summarizing: we can see that the ACT BE readings are dependent on the verb BE clauses. [+V] small clauses need to be tense marked (as in Guéron and Hoekstra 1988). acquiring some morphology that will allow a temporal interpretation of the adjective.181,182 And this basically answers the question of why we can say Mary is To be tense marked means either raising to a T to theta bind it directly either because some tense morphology on the verb needs to be checked or by being adjoined to some verbal being cruel much easier than Mary is being fat. Thus there isn't an ACT BE. All there is BE projection that gets tense marked. with some temporal information. Depending on the adjective an agentive reading is The second assumption is that Nouns and Verbs license their arguments in a different derived. Now we need to account for the existential vs. generic readings of subjects. way. For that I will rely on Dowty's (1989) proposal that nouns are to be distinguished 6.5 Subject interpretations from verbs by their argument licensing properties: the ordered method and the NeoDavidsonian method. This will account for the fact that arguments of ser predicative Here I examine the lack of existential readings for the subject in the ser predicative constructions do not need to be licensed in an ordered fashion. constructions in general and not the generic readings. I will assume that whatever In the following first I present some evidence for the [+N] and the [+V] distinction captures the generic readings in ser predicative constructions can also capture the generic readings in estar predicative constructions and I will not deal with them here.183 All I am between ser small clauses and estar small clauses and then I will try to explain the lack of existential readings in ser small clauses. The results of this proposal are then extended to 181 There is another context in which an ACT BE interpretation is available, namely imperatives of the type Sê/ seja gentil! 'Be kind!' This case seems to be similar to the adverbial modification one in the when(ever) contexts. In these cases the operator seems to be able to bind the predicate and derive a temporal interpretation of the subject. 182 Diesing (1992) proposes that in the ACT BE cases the adjectives have "a more adverbial rather than predicative function" (p.45)(...) "Individual-level predicates that cannot readily modify act (such as overweight and tall ) are much less acceptable in this context."(p.45) This observation is not in principle incompatible with what we proposed but it doesn't explain why (i) and (ii) mean very different things: some other constructions. Although I don't analyze them in detail they strongly suggest that the proposal is on the right track. 6 . 5 . 1 Evidence for [+N] and [+V] The basic idea I will pursue in this section is that adjectives taken by ser are [+N] and (i) (ii) John acts depressing John acts depressingly. adjectives taken by estar are [+V], and this distinction is to be related to their different and it remains to be explained why (iii) is quite unacceptable. (iii) *I saw John be depressingly. 183 aspectual and argument licensing properties. First, I describe some subcategorization I suspect though that they belong to the CP Agr node. 400 401 properties of ser and estar constructions that lead to a distinction in terms of [+N] and [+V] small clauses, and I propose a characterization of [+N] and [+V] predicates that can There is another construction with noun phrases that supports the assumption that estar cannot take NPs. Consider the following example: divide the adjectival predicates in two groups. Ser can be followed by bare, definite and indefinite noun phrases, some prepositional phrases and adjectives. Ser is also the auxiliary for the so-called verbal passive. Estar subcategorizes for adjectival phrases (including adjectival passives), prepositional phrases and is the auxiliary for the progressive. Estar cannot be followed by nouns. The examples (72) a. Pedro é cozinheiro Pedro is (a) cook. (That is his profession) b. *Pedro está cozinheiro. Pedro is cook. below illustrate the predicative use with noun phrases. c. Pedro está de cozinheiro. (71) Ele é/*está médico/ o médico/ um médico. He is doctor/ the doctor/ a doctor. Pedro is of cook. ((Pedro is playing the role of a cook today because the real one didn't come) Given the fact that ser can take NPs and assuming that the participle is a nominal A noun phrase is only acceptable with estar if a preposition in inserted. head, it can be argued that what characterizes ser is the ability to take [+N] elements. Estar, on the other hand, selects [+V] elements184 . Another case that shows that NPs are not accepted with estar is the possessive. A PP has to be introduced in order for a NP to be acceptable with estar: 184 In Brazilian Portuguese, when stressed, a definite or an indefinite (but never a bare noun) are sometimes allowed with estar. In this case, the meaning is something like "he is the only or the best doctor in some place" or "he is such a doctor". There are other cases in which the indefinite can be used without being stressed: (i) O jantar está uma maravilha. The dinner is a marvel. (ii) Bonita a Maria! Pretty the Maria 'How pretty (is) Maria' The book is with me. c. *O livro está meu. The book is mine. *Um mamífero, o ornitorrinco A mammal, the platypus d. *O livro é comigo. However, not all DPs are excluded in these constructions. The same DPs that are allowed in the estar constructions are allowed in these Degree Exclamative constructions as shown in (iii): (iii) The book is mine. b. O livro está comigo. In all these cases, there is an active Degree Phrase preceding the indefinite (see Abney 1987). Some evidence for this claim comes from certain Degree small clause exclamative constructions. Exclamatives Adjectival small clauses can appear in the order Predicate NP as shown in (ii). Indefinites are usually excluded, as shown in (iii). (i) (73) a. O livro é meu. Uma delícia/ uma maravilha o jantar! A delight/ a marvel the dinner 'How delicious/marvelous is the dinner!' 402 The book is with me.185 185 (d) is acceptable only when it has the reading 'The matter of the book is under my responsibility.' 403 Adjectives, however, are usually considered to be [+N,+V]. So, in principle, 6 . 5 . 2 The Ordered Argument Method and the Neo-Davidsonian Method adjectives should be able to appear with both verbs. As shown below, this is only In order to maintain the generalization above, I will develop an aspectual distinction partially true: (74), (75) and (76) exemplify adjectives that can occur only with ser; between [+N] and [+V] that can accommodate the adjective facts. Here I will rely on adjectives that can occur mostly with estar; and adjectives that can occur with both ser and Dowty's 1989 proposal. estar: According to his proposal, noun arguments are always optional and verb arguments are always obligatory. The first ones are licensed by the Ordered Argument Method, and (74) Adjectives with ser the second ones, by the Neo-Davidsonian method.187 We know, however, that característico (characteristic) benefactives, instrumentals and other verbal arguments are also optional in the syntax. cruel (cruel) Thus, a distinction purely in terms of obligatory versus optional arguments is not enough brasileiro (Brazilian) to separate arguments of [+N] and arguments of [+V] elements. Internal arguments in the circular (circular) VP level are not optional. In fact we can argue that they are always present in one form or (75) Adjectives with estar186 another. If they are not lexically present in the syntax, they are interpreted as [–SQA], as pronto (ready) is the case of null objects in Italian and French (see Rizzi 1986 for example) and grávida (pregnant) intransitive verbs. In this case, the interpretation of the verb phrase will be durative, the present (present) unmarked case. satisfeito (satisfied) Thus, verbal predicates always have arguments "projected" in the syntax in an ordered fashion. If we take this seriously, it must be the case that an internal argument of (76) Adjectives with ser or estar a verbal argument will always be projected in the same position and the verb will always gordo (fat) be theta bound by tense. (Note that given that estar selects for stative events, then feio (ugly) adjectives that assign Agent roles are automatically excluded.) This captures the intuition ansioso (anxious) 186 Marginally present, satisfied and dead can appear with ser. but the meaning of the adjective is atemporal. 404 187 The basic difference between these two methods is that the Ordered Argument method assumes that verbs denote relations of a fixed number of places. Thus, a predicate with fewer than its prescribed number of arguments doesn't have any well-defined interpretation. Nouns, on the other hand, license their arguments via conjunction, which in essence is the Neo-Davidsonian method. As Dowty himself notes, this distinction between nouns and verbs is too strong. Grimshaw (1991) challenges the verb/noun distinction and basically following an intuition in Zubizarreta (1987) proposes that there is a distinction between true arguments and complements. If the arguments are optional, then they are not real arguments but complements or possessors. If they are obligatory, then they are real arguments. Real arguments are thematic, i.e., they are licensed via argument structure. Other nouns do not have thematic arguments. Their complements are licensed independently directly via LCS or via predication. Hence, according to Grimshaw, the distinction between true arguments and complements is in fact made in terms of their syntactic behavior and the meaning of the noun. 405 that estar takes unaccusative adjectives, which appears every so often in the literature (see (77) a. John bought children's books Luján 1988, for an analysis and a review.) b. Women's dresses are on sale In sum, the idea is that estar selects for events (and not descriptions of events). Thus the head of the small clause taken by estar has to be tense marked (i.e. the head has to In other words, ser is defective in that it has no temporal specification. Thus a subject move to estar and with it move to T.) If eventive, the predicate licenses its arguments by of a ser+adjective construction is not an argument of a temporal predicate. In this case the ordered method. This implies that if is there is an internal argument it must be only a generic reading is available. realized. Other arguments will be realized in an ordered fashion. Consequently if the There are cases, however, as we have just seen, where ser acquires temporal small clause has a subject this subject must be the internal argument, given that regular information. In this case, the prediction is that the bare plural subject should allow intransitive verbs will be banned independently since estar does not take eventive verbs existential readings. The following examples of ser+adjective and an aspectual marker (except in the progressive, where -ing is arguably a stative head). show that an existential interpretation of the subject is indeed available in (78a-c) but not Once we assume this, we can say that bare plural subjects will be existential if they in (78d): get a theta role and at AgrS enter a specifier-head agreement relation with a tense marked head. This way the subjects can be interpreted as participants in an event. Nouns are different. Not only do they not necessarily license their arguments in the (78) a. Crianças estão sendo cruéis. Children are being cruel. (existential & generic readings) ordered fashion, but they also are not tense bound. Assuming that nouns license their b. Eu vi crianças ser mal-educadas. arguments via the Neo-Davidsonian method implies assuming that they don't need to be I saw children be uneducated. (existential & generic readings) all present in the syntax.188 We can say gift, gift from Mary or gift to Mary. Moreover, being nominal they do not need to be tense marked and as such will not allow an existential reading for the subject. Evidence for the lack of temporal interpretation of arguments on regular nouns comes from the interpretation of the bare plural arguments of nouns (of course this is not true of eventive nominals, which as the label says, do have c. Crianças foram cruéis. Children were-PERF cruel (yesterday). (existential & generic readings) d. Crianças são cruéis. Children are cruel. (generic reading) some temporal structure). Bare plurals inside regular noun phrases can never get an existential reading as the examples below show: In (78a-c), where what I will call "aspectual composition" has taken place, existential readings are available for the bare plural. Where no aspectual composition has taken 188 In English there is one exception brought up to me by Randy Hendrick: fondness of Mary. I have no explanation for this fact. I think that this case is to be assimilated to DegP complements. Grimshaw's proposal distinguishes process nominals (which need to assign the internal role in the syntax in order to assign an external argument) from result nominals (which do not need to assign their roles in the syntax). Very few nouns, however, meet this distinction. Therefore I will consider the first case the exception and not the norm. 406 place, only a generic reading is available for ser predicative construction subjects. Estar subjects, on the other hand, always allow an existential interpretation, since estar is an aspectual head and thus selects for a small clause that has temporal information; 407 therefore, the argument of the adjective is always thematically licensed because estar always selects for stative eventualities. b. João está/*é ansioso para terminar o artigo. João is anxious to finish the paper. From what we have said with respect to the interpretation of subjects, we are de facto assuming that subjects can be licensed in two different ways: as participants in the aspect For Jones (1985), Roberts (1986) and others, the binder of the PRO subject of the composition (always thematic arguments) or as arguments of a pure predication (thematic purpose clause has to be thematic. Assuming this analysis, we have to assume that while or non thematic arguments). the subject of ser ansioso is not thematic, the subject of estar ansioso is thematic. Predication can be obtained in various positions in the syntax. For example: base 6 . 5 . 3 . 2 Of John and John alternations generated topics, multiple nominative NPs in Japanese and subjects of tough constructions, are licensed via predication but they do not participate in the aspectual Consider the contrast of (80) and (81):189 composition of an eventuality. Not surprisingly, bare plurals in these positions receive (80) a. É generoso da parte do João ter comprado o presente por nós. generic readings, just like ser subjects. It is generous of John to have bought the present for us. 6 . 5 . 3 More evidence for subject differences b. *Está disponível da parte do João me levar á cidade. In this section I want to describe the distributional facts of ser and estar in purpose *It is available of John to drive me downtown. clauses, and some psych-predicate predicate alternations. I will not provide any analysis, as this would take us too far afield, but the distribution correlates well with what I have proposed, namely that the subjects of ser and estar are licensed in a different way and (81) a. João é generoso em ter comprado o presente por nós. John was generous to have bought the present for us. that, while participation in the aspectual composition has to occur in a certain b. O João está disponível para me levar à cidade configuration, pure predication can take place in various syntactic configurations. It will John is available to drive me downtown. also provide some indirect evidence for the nominal character of the ser predicates. Ser generoso 'be generous' is a predicate that licenses its subject via predication only 6 . 5 . 3 . 1 Purpose clauses and estar disponível 'be available' is a predicate that licenses its subject via the mediation Certain adjectives can appear with both verbs as exemplified in (79). However, when of an aspectual role. In (80a) John can appear as an adjunct but not in (80b). Independent a purpose clause is added, only the clause with estar is acceptable. of the type of adjunct clause that follows these constructions, the contrast seems to be that of John cannot be licensed as an adjunct when it is an aspectually licensed subject. As we (79) a. João está / é ansioso. João is anxious. 189 Examples like ($80a) and ($81a) have been studied by Stowell (1991b) who shows that in fact neither John nor the clause can possibly be internal arguments because extractions are not possible from any of these arguments. 408 409 suggested above, only arguments licensed via predication do not need to be in a unique b. Maria está *burra/*chata/*metida/grávida/pronta. syntactic configuration to be interpreted as arguments of a predicate. If the argument has a Maria is dumb /boring/nosy/pregnant/ready. theta role then it can be placed as a participant in an Event only if it ends up a the aspectual The distribution of the adjectives is not the only property of the consider cases shared composition it has to be at some point in a certain syntactic configuration (Spec AgrOP or Spec AgrSP) with the verb. by ser predicative constructions. The interpretation of the bare plural is another one. Bare The distinction between subjects licensed via participation in aspect composition and plurals in ser constructions and in the subject of consider small clauses receive only subjects licensed via pure predication allow us to predict not only the interpretation of the generic interpretations. This fact was a problem for Diesing and Kratzer's account. bare plurals but also explain the distribution of ser and estar in the above constructions. According to them the subject of the small clause counts as inside VP and therefore In the last part of this section I establish a parallel between ser small clauses and should display an existential reading. consider small clauses in one hand and estar small clauses and secondary predicates, absolutes, reduced relatives and free adjuncts. 6 . 5 . 4 Two types of small clauses: some speculations However, under the analysis proposed, if we associate the interpretation of the bare plural with the aspectual property of the subject, then it is reasonable to think that the subjects of ser and consider small clauses should share the same property, namely the lack of a way a thematic role that would allow it to participate in the aspectual There is a correlation between adjectives that can appear with ser and adjectives that appear in consider contexts. The same adjectives that can appear with ser can also appear composition. In other words, both the subjects of ser and consider small clauses license the subject via predication, just like nouns do. embedded under consider, as the examples below show.190 Of these adjectives, the ones that cannot appear with estar, cannot appear in circumstantial adjuncts. Ser can thus be said to take nominal small clauses. In the same way, we can assume that consider takes small clauses that are nominal, i.e., that do not have thematic subjects. It is important to note that in the cases where an estar adjective is possible under consider (82) a. Eu considero Maria burra/chata/metida/ grávida/*pronta. there is a law-like flavor to the predicate (as pointed out by Heycock 1991). I think that, I consider Mary dumb/boring/nosy/pregnant/ready. in these cases, the adjective is being used as a nominal adjective, i.e., as an element b. Eu encontrei Maria *burra/*chata/*metida/grávida/pronta. I met Mary dumb /boring/nosy/pregnant/ready. deprived of temporal properties that can combine with the properties of the NP and create an event. In fact, I think all verbal adjectives can be "nominalized" and taken as deprived from its aspectual properties if they are not tense linked. This entails that there is no such (83) a. Maria é burra/chata/metida/ *grávida/*pronta. a thing as "estar only" adjectives. All adjectives that can appear with estar are possible Maria is dumb/boring/nosy/pregnant/ready. with ser, although some are very odd to be taken atemporally. Ready and pregnant are some cases.191 On the other hand not all ser adjectives can appear with estar because not 190 The inverse of this correlation is not always true. 410 191 It is possible to say : 411 all of them carry temporal information. Particularly, denominal adjectives are very hard can incorporate into a case assigning head. Thus nominal adjuncts will be ruled out with estar. because they cannot be in a Spec-head relation with some case assigning head nor can The parallelism between consider and ser constructions can be stated in the following they incorporate onto a case checking head. Thus these nominal elements are ruled out. way: ser and consider take nominal complements. As for estar, we can say that it takes Verbal predicates, on the other hand, don't have case features that need to be checked verbal small clauses as complements. Now we can turn to secondary predicates, which and therefore will be related to some tense. We can say that just by being c-commanded usually denote temporary properties. The so-called individual-level predicates are not by some tense element is enough to license a [+V] adjunct. allowed in this position, as the examples below show: Two problems can be raised with respect to this proposal: the first one has to do with the with absolutes. Can't with case mark the small clause? The answer relies on the (84) I met Mary pregnant/*intelligent. properties of with in these constructions. It can be shown that with is a temporal operator (see Stump 1985).192 Therefore with takes [+V] predicates and not nominal predicates. The so-called individual-level predicates are not blocked only in secondary predicates Thus, although it can check the case of the subject of a small clause, the construction is but also in absolute constructions with with, free adjuncts and reduced relative clauses as ruled out with adjectives like intelligent because with doesn't take nominal predicates as the examples show: its complement. The second problem is related to the tense relation itself. Although we (85) a. With firemen available / *intelligent, the city is safe. can easily envision a simple way to check the tense of secondary predicates and free b. Available /*intelligent, John will help. adjuncts (by predicating them to verb phrases or even tense phrases), reduced relatives c. The only firemen available / *intelligent are ex-marines. are not clearly in a position to check for tense. In this case, we would have to weaken the "tense marking proposal" of Hoekstra and Guéron and propose that being verbal they will There are two possible routes that can be explored in order to solve the problem of be linked to whatever tense c-commands it.193. the distribution of adjectives in these constructions. The first one relies on the hypothesis Another route that can be explored in order to account for the unacceptability of that nominal small clauses need LF case, but verbal ones, don't. The second one relies on nominal reduced relatives, secondary predicates, and free adjuncts is to assume that these conditions in which PRO can be licensed. constructions have the structure of the form [PRO predicate], as has been often proposed Given the discussion in chapters 4 and 5, we could follow Hoekstra and Guéron's in the literature for secondary predicates. Assuming nominal elements cannot license PRO proposal that all [+N] categories need to be case marked at LF. In order to do that, we (see Williams 1985; Hestvik 1991), this would explain why nominal predicates are not may assume that nominal elements have to check case features in a Spec-head relation or 192 There is a distinction between absolute clauses with with and without it. The first ones can receive a conditional or a temporal interpretation (which suggests they are under tense), the second ones can only receive a factive reading. (i) A Maria é grávida de três meses. The Maria is pregnant of three months. with the meaning that she is a three-month old pregnant. 412 193 One could object that relative clauses do not participate in sequence of tense and that is because of its syntactic configuration. However, one cannot treat reduced relatives in the same way as regular relatives because in the first case they definitely have a tense projection on their own. 413 allowed in this construction. Note that with absolutes cannot have the form with PRO. there are no special aspectual projections that force a terminative or a durative reading. This would follow if with is an exceptional case marker and PRO cannot receive Aspectual verbs are to be interpreted as quantificational elements on the VP. I believe that accusative case. the perfective morphology in Brazilian Portuguese can be amenable under a similar A possible problem for the PRO account (pointed out to me by Juan Uriagereka) is analysis. What is important is that these elements interact with the subjects and like the Aquantifiers did for Polish force certain types of DP subjects to be in the specifier of the raised by sentences of the following type: Agreement Phrase that they target. (86) Ser espanhol é um problema. In a sense then the interpretation of subjects dependent on the predicates in two ways: 'To be Spanish is a problem.' first there is or there isn't a structural thematic relation between the subject and its complement. If the subject is an argument of a nominal predicate then only generic What is then the subject of to be Spanish if it is not an arbitrary PRO? And if it is an readings are available for the bare plural. If the subject is an argument of a verbal arbitrary PRO, where is it generated if not in the small clause? This shows that we have to predicate, generic and existential readings are available. Moreover, depending on what is loosen the proposal that PRO cannot be generated in nominal small clauses. In order to in the verbal predicate, certain interpretations will be forced on the indefinite SE, bare maintain this proposal, one possibility is to say that [+N] elements cannot assign nouns etc. Particularly we have seen noun it cannot be license by because depending on structural case (null case being one) but ser can. Thus, although, [PRO espanhol] is not the predicate the subject will have a thematic role or not. At the same time we have seen acceptable, [PRO ser t espanhol] is fine. Alternatively we can just say that there is no that the perfective marker behaves like an A-quantifier in that it seems to take only the PRO in (86) and that ser espanhol is a DP, which is not implausible given that we can subject and the upper verbal complex head in its scope. also say: (87) O ser espanhol não ajuda na fronteira. The be Spanish doesn't help in the border 'Being Spanish doesn't help at the border.' 6.6 Final Summary In this chapter I showed that a distinction between ser and estar in terms of aspect not only can account for most of the interpretation of bare plurals in predicative constructions, but also can be extended to account for selection and distributional properties of adjectival small clauses. Such an aspectual distinction is also amenable under an analysis in which 414 415 durative reading. Behind this assumption is the idea that to separate verbs into four Conclusion classes is unnecessary and misleading because culmination points are given by the Clearly a part of the semantics and a part of morphology, aspect has been in the footnotes cardinality of the complement and not by the verb itself. This is an important assumption, of syntax papers in generative grammar for quite a while. Whenever some principle did because most syntactic work on aspect tried to syntacticize Vendler's aspectual classes not work for a class of elements, a footnote would appear suggesting that some aspectual without realizing that what is important to create a culminating event, i.e. an event with an property was at stake. In other words, everything would work fine if it were not for the end point is the type of complement we have and not just the verb. In other words, for aspect. But under the label of aspect, almost anything could be included: case, this thesis, run and build belong to the same class and verbs are to be distinguished in morphology on the verb, adverbials, properties of complements, etc. in any order. This two main classes: states and events; ubiquity made aspect a very suspicious property and suggested that any study on the (iii) aspect is layered. We have a point where we interpret the VP aspect, a point were syntactic reflexes of aspect would have to start by making a cut, a deep cut on what could we integrate the VP information with the IP aspect (subject and maybe Tense) and count as aspect. One of the possible cuts is to examine the verb properties and attempt to probably a level where we integrate the IP information with the CP information (modality decompose them in the syntax. I didn't take that route for two reasons: the first one is that and genericity). it fails to uncover the role of the internal arguments in the aspectual interpretations; With these assumptions in hand, I proposed, based on the way the computational second, aspect to a large extent is to be related to the functional material. In this thesis my system is supposed to work according to the feature checking mechanisms of Chomsky focus was on the role of nominal elements on the aspectual composition and on the (1993), that the VP aspect was to be calculated at AgrO. To interpret the VP as checking relations. In this conclusion I would like to summarize the basic assumptions terminative it is necessary to have an eventive verb at AgrO and information about the and basic results of this enterprise and highlight what I consider the most relevant cardinality of the complement. With this very simple assumption I kept the verb constant insights. and played with different types of complements in various languages, using the aspectual I by started asking what was the minimum necessary for aspectual interpretation. From there I developed a proposal for the relevant structural conditions for terminative aspect interpretations that allowed me to test the internal structure of various types of interpretations of the VP as a test for what should be at AgrO by the time aspect was calculated. The hypothesis that aspect is calculated at AgrO associates aspect with the functional structure and not with thematic relations. AgrO is the locus for Case and φ-feature complements. Drawing heavily on the work on aspect of Verkuyl (1972; 1989; 1993; 1995). I checking. I find the separation between thematic properties and functional properties, explicitly assumed the following: (i) aspect is compositional: it is the combination of a outlined in Lebeaux (1988) and formalized in Chomsky (1993, 1995) crucial to verbal and a nominal feature; (ii) durative readings are the default for the VP aspect. understanding the VP aspect. Thus, it is not surprising that people have associated aspect Terminative is the marked case. There is only one way to produce terminative readings, with Case and morphological features; however, what is actually crucial is not Case per but any violation of the minimal requirements on terminative readings will result in a se or a morphological features per se, but the fact that in a checking relation there are 416 417 always two elements. In the case of VP features, we have verbal and nominal elements chunks of material happen. Thematic roles are interpreted within the lexical projections that can "see" each other. and from then on all the interpretation happens at points where elements from two Agr projections were used for two reasons: first, they are checking positions and they are convenient positions where we can guarantee that elements of different extended different extended projections enter a checking relation. I assumed that these points are Agr projections. projections with be able to match features. Second, they are positions that have only the These are the main points that I would like to be clear. Also related to what I have just content of the elements that move into them. This allows us to have a "generic" position said is that there is no parameter of aspect. The learnability problem posed by aspectual where different functions can relate two different elements. Thus there would be no properties is the same learnability problem posed by the learning of morphology (Case problem of having verb and object checking features and entering a relation X at one Agr etc.,) and we have to learn the meaning of quantificational material and where they are in projection and in another position have Verb+T and subject checking features and the tree. entering a relation Y. complementation. I showed that case and A-quantification will alter the aspectual Nothing else. Chapters 2, 3, 4 and 5 were case studies of verbal Chomsky (1995) argues against Agr projections on the grounds that they are interpretations because they can force or inhibit something from reaching AgrO. I also unnecessary. If it turns out that really we should not have agreement projections then the showed that complements are not merely DPs and that regular transitive verbs can take conclusions that will follow from this thesis do not need to be thrown away, because quite complex structures as their complements. The complexity of the structures end up whether it is AgrO or some other position, V and DP complements have to enter a altering what needs to have features checked and thus alters movement. For example, in checking relation and have to match features. Depending on what enters the checking chapter 4 it became clear that certain types of modification can actually alter what element relation a different interpretation will obtain. I used Agr projections because I actually needs to check case, thus altering the aspectual interpretation and creating what I called think that agreement relations are necessary to provide interpretation configurations but determiner transparency. Determiner transparency allowed the head noun in a relative nothing necessarily falls apart if we have to check this relation in some other position, as clause, for example, to escape from the scope of a definite, which explains its "non- long as the checking relation can exclude the subject and other elements. definite" behavior. In Chapter 5 I showed how much we can learn about noun phrases if What is important is that at certain positions certain interpretations obtain. But what we use aspectual interpretations as a test for what needs to check features. In Chapter 6, I moves is not determined by the possibility of getting or not a certain interpretation, but by tried to show that the role of subjects is similar to the role of objects but not identical, the requirements of the output conditions. The basic output condition is to eliminate given that the function that relates the VP to the subject is quite different from the function uninterpretable features. Move, in the computational system, eliminates uninterpretable that related the object with the verb. More has to be done with respect to the role of features by having them checked. Case on nominals has to be eliminated and Number and subjects and the role of the A-quantifiers that alter subject interpretations. In fact, Chapter Person has to be eliminated from verbs. 6 opens cans and cans of worms that it doesn't close. If there is some explanation for the What seems to come as a conclusion for this thesis is that interpretation proceeds EPP it has to lie on the relation between certain verbal elements and certain nominal derivationally and I will assume that there are summing up points where interpretation of features of the subject. And that statement already sounds familiar. My guess is that 418 419 "aspect" is at stake but this time not as the mysterious footnotes but as something we have already some idea about. Also, as Carlson (1977a) noted, genericity has to do with aspect References and needs to be dealt with more carefully. I believe we can also find relations between nominal and verbal elements at the CP level (or AgrCP to be consistent). These relations will include arbitrary pro interpretations, certain null arguments and certain adverbials. If the proposal in this thesis is in the right track, we should look at adverbs and the indirect arguments under this light to see how far we can be push the idea of interpretations occurring derivationally. There is plenty to do. But I will stop here for now. Abbott, Barbara (1995) Some Remarks on Specificity, Linguistic Inquiry 26, 341-346. Abney, S. (1987) The Noun Phrase and its Sentential Aspect. Doctoral dissertation, MIT, Cambridge. Ambar, Manuela (1988) Para uma Sintaxe da Inversão do Sujeito em Português, Doctoral dissertation, University of Lisbon. Aoun, Joseph. (1981) The Formal nature of Anaphoric Relations Doctoral dissertation, MIT, Cambridge. Arce, M. (1989) Semantic structure and syntactic function: the case of Spanish SE. Ph.D. dissertation, University of Colorado, Boulder. Bach, Emmon. (1986) The algebra of events, Linguistics and Philosophy, 9, 5-16. Bahlouhl, Maher and Wayne Harbert (1992) Agreement asymmetries in Arabic. Proceedings of WCCFL 11, 15-31. Baker, Mark (1988) Incorporation: a Theory of Grammatical Function Changing. Chicago: University of Chicago. Baltin, Mark. (1987) Do antecedent-contained deletion exist? Linguistic Inquiry 18, 579595. Barker, Chris (1991) Possessive Descriptions. Doctoral Dissertation, UCSC. Barker, Chris (1994) Rethinking double genitives as partitives: explaining an antiuniqueness implication. ms. University of Rochester. Beghelli, Felippo. (1993) A minimalist approach to quantifier scope. Proceedings of NELS 23. Belletti, Adriana (1990) Generalized Verb Movement: aspects of verb syntax. Torino: Rosenberg and Sellier. Ben-Shalom, Dorit. (1993) Object wide scope and semantic trees. Proceedings of SALT III. Benmamoun, Elabbas. (1992) Functional and Inflectional Morphology: problems of projection, representation and derivation, Doctoral dissertation, University of Southern California. Bennett, Michael (1978) Demonstratives and indexicals in Montague grammars. Synthese 39, 1-80. Bernstein, Judy (1993) Topics in the Syntax of Nominal Structures across Romance. Doctoral dissertation., CUNY. 420 421 Bobaljik, Jonathan David, and Dianne Jonas (1993) Subject positions and the role of TP. Paper presented at GLOW 16, Lund, April 1993. Chomsky, Noam (1993) A minimalist program for linguistic theory. In K. Hale and S.J. Keyser (eds.) The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, MIT Press, Cambridge MA, 1-52. Bonet, Eulalia. (1991) Postverbal subjects in Catalan. ms MIT. Chomsky, Noam (1994) Bare Syntax. ms., MIT. Borer, Hagit. (1993) The projection of arguments. UMass Occasional Papers in Linguistics 17,19-47. Bosque, Ignacio (1990) Sobre el aspecto en los adjectivos y en los participios' in Ignacio Bosque, (ed.) Tiempo y aspecto en español. Catedra, Madrid, 177-214. Bresnan, Joan. (1973) Syntax of the comparative construction in English. Linguistic Inquiry 4:275-343. Browning, Marguerite. (1987) Null Operator Constructions, doctoral dissertation., MIT Cambridge. Burzio, Luigi (1989) The Role of Antecedent in Anaphoric Relations paper presented at Second Princeton Workshop on Comparative Grammar, to appear in Robert Freidin (ed.) Current Issues in Comparative Grammar, Kluwer, Dordrecht. Burzio, Luigi (1991) The Morphological Basis of Anaphora, Journal of Linguistics 27, 81-105. Chomsky, Noam (1995) The Minimalist Program, MIT Press, Cambridge Mass. Chomsky, Noam and Howard Lasnik (1991) Principles and parameters theory. In J. Jacobs et al (eds.) Syntax: An International Handbook of Contemporary Research , Walter de Gruyter, Berlin. Clements, J. Clancy (1988) The semantics and pragmatics of the Spanish <copula+adjective> construction. Linguistics 26, 779-822. Comrie, Bernard. (1987) Aspect. Cambridge: Cambridge University Press. Condoravdi, Cleo. (1992) Individual-level predicates in conditional clauses. Presented at the LSA Meeting, Philadelphia PA. Davidson, Donald. (1967) The Logical Form of Action Sentences. In Davidson, Donald, Essays on Actions and Events, :New York: Oxford University Press. Butt, M. (1993) Object specificity and agreement in Hind/Urdu. CLS 29, 89-104. Delfitto, Denis and Jan Schroten (1991) Bare plurals and the number affix in DPs, ms. University of Utrecht. Also in Probus. Bybee, Joan L. Morphology; a study of the relation between meaning and form. John Benjamins, Amsterdam. . Depiante, Marcela. (1993) Clitic Doubling and Strong Noun Phrases, ms. University of Maryland, College Park. Campbell, Richard. (1992) The temporal syntax of small clauses. ms. Oakland University, Michigan. Depraerete, I. (1994) On the necessity of distinguishing between (un)boundedness and (a)telicity.Linguistics and Philosophy. Carlson, Greg. (1977a) Reference to Kinds in English. Doctoral dissertation UMass, Amherst. Diesing, M. (1988) Bare plurals and the stage/individual level contrast. In M. Krifka (ed.) Genericity in Natural language: Proceedings of the 1988 Tübingen Conference, SNS-Bericht 88-42. Tübingen,107-154. Carlson, Greg. (1977b) Amount Relatives. Language 53, 520-542. Carrier, Jill and Janet Randall (1992) The argument structure and syntactic structure of resultatives Linguistics and Inquiry 23:173-234. Diesing, Molly. (1990) The syntactic roots of semantic partition. Doctoral dissertation. UMass, Amherst. Diesing, Molly. (1992)Indefinites. Cambridge MA: MIT Press. Chesterman, A. (1991) On definiteness; a study with special reference to English and Finnish. Cambridge University Press, Cambridge. Dowty, David. (1979) Word meaning and Montague Grammar. Dordrecht: Reidel. Chierchia, Gennaro. (1993) Questions with quantifiers. Natural Language Semantics 2, 181. Dowty, David. (1989) Semantic content of thematic roles. In B. Partee; Gennaro Chierchia; R.Turner. Properties, types and meanings. Dordrecht, Kluwer, 69-130. Chomsky, Noam (1981) Lectures on Government and Binding Dordrecht, Foris. Dowty, David. (1991) Thematic proto-roles and argument selection. Language 67:547619. Chomsky, Noam (1986) Knowledge of language. New York: Praeger. Enç, Mürvet (1991) The Semantics of Specificity, Linguistic Inquiry 22, 1-26. Chomsky, Noam (1986b) Barriers Cambridge MA: MIT Press. Engdahl, Elisabet. (1986) Constituent questions. Dordrecht: Kluwer. Chomsky, Noam (1992) A minimalist program for syntactic theory. MIT Occasional Papers in Linguistics, 1. 422 Fiengo, Robert and Robert May (1994) Indices and Identity. Cambridge MA: MIT Press. 423 Filip, Hana. (1992) A-quantifiers in Czech. Proceedings of CLS . Fodor, J. and I. Sag (1982) Referential and quantificational indefinites Linguistic Inquiry 5, 355-98. . Grimshaw, Jane. (1990) Argument structure. Cambridge, MA: MIT Press. Grimshaw, Jane. (1991) Extended Projection, ms. Brandeis University. Guéron, Jacqueline and Teun Hoekstra (1988) in Anna Cardinaletti, Guglielmo Cinque and G. Giusti. Constituent Structure, Dordrecht, Foris. Holmberg, Anders et al. (1993) The structure of INFL and the Finite Clause in Finnish. In Anders Holmberg & Urpo Nikanne. Case and other Functional Categories in Finnish syntax. (ed) Berlin, Mouton de Gruyter 177-206. de Hoop, Helen (1992) Case configuration and NP interpretation. Doctoral dissertation. U. of Groningen. de Hoop, Helen (1989) Case assignment and Generalized Quantifiers Proceedings of NELS 19, 176-190. Hornstein, Norbert (1991) Expletives: a comparative study of English and Icelandic. ms.U.Maryland. Haïk, Isabelle. (1986) The Syntax of Operators, doctoral dissertation., MIT. Hale, Kenneth and Maria Bittner. (1994) The structural determination of case. ms. MIT and Rutgers University. Hale, Kenneth and S. J. Keyser (1993) The View from Building 20, MIT Press, Cambridge, Mass. Hall, R. A. Jr. (1968) 'Neuters', mass nouns and the ablative in Romance, Language 44, 480-486. Heggie, Lori (1988) The syntax of copular structures. Doctoral dissertation, University of Southern California. Hornstein, Norbert (1994) An argument for Minimalism: the Case of Antecedent Contained Deletion. Linguistic Inquiry 25, 455-480. Hornstein, Norbert (1995) The grammar of LF: from GB to Minimalism. Oxford: Basil Blackwell. Hornstein, Norbert and Amy Weinberg (1981) Case theory and preposition stranding, Linguistic Inquiry , 12: 55-91. Jackendoff, Ray (1972) Semantic Interpretation and Generative Grammar. Cambridge, Mass.: MIT Press. Jackendoff, Ray (1977) X' Syntax: a study of Phrase Structure, MIT Press, Cambridge. Heim, Irene. (1982) The semantics of definite and indefinite noun phrases. Ph.D. dissertation, U.Mass, Amherst. Jackendoff, Ray (1985) Semantics and Cognition. Cambridge MA: MIT Press. Heinämäki, Orvokki (1984) Aspect in Finnish, in Casper de Groot and Hannu Tommola (eds.) Aspect Bound, Foris, Dordrecht, 153-178. Jackendoff, Ray (1992) Semantic Structures. Cambridge MA: MIT Press. Hendrick, Randall (1991) The morpho-syntax of aspect, Lingua 85, 171-210. Jaeggli, Oswaldo A. (1986) Three issues in the theory of clitics: case, doubled NPs, and extraction in Hagit Borer, (ed.) The Syntax of Pronominal Clitics Syntax and Semantics, 19. Hernanz, Maria Lluisa. (1991) Spanish absolute constructions and aspect. Catalan Working Paper in Linguistics Universidad autonoma de Barcelona p75-128. Hestvik, Arild. (1991) Subjectless Binding Domains. Natural Language and Linguistic Theory 9:455-496. Heycock, Caroline. (1991) Layers of Predication, Doctoral dissertation, University of Pennsylvania. Johnson, Kyle (1991)Object Positions' Natural Language and Syntactic Theory 9 577636. Jones, Charles. (1985) Syntax and Thematics of Infinitival Adjuncts. UMass dissertation. Kamp, Hans. (1975) Two theories about Adjectives. In E. Keenan (ed) Formal Semantics of Natural Language. Cambridge: Cambridge University Press. . Higginbotham, James (1985) On semantics. Linguistic Inquiry 16, 547-593. Higginbotham, James. (1987) Indefiniteness and Predication, in Eric Reuland and Alice G.B. ter Meulen. The Representation of Indefiniteness. MIT Press, Cambridge Mass. Kamp, Hans. (1981) A Theory of Truth and Semantic Representation, In Groenendijk, T Janssen and M Stokhof (eds) Formal Methods in the Study of Language, 277-321. Amsterdam: Mathematical Centre. Higginbotham, James. (1994) Mass and Count Quantifiers. Linguistics and Philosophy, 17 447-480. Kangasmaa-Minn, Eeva (1984) Tense, Aspect and Aktionsart in Finno-Ugrian, in Casper de Groot and Hannu Tommola (eds) Aspect Bound, Foris, Dordrecht, 77-96. Higgins, F. Roger (1985) The pseudo-cleft construction in English. Garland, New York. Kaplan, D. (1977) Demonstratives In J. Almog et al (eds) Themes from Kaplan, New York: Oxford University Press, 1989. Hoekstra, Teun (1988) Small Clauses Results. Lingua, 74 101-139. 424 425 Kayne, Richard S. (1993) Toward a modular theory of auxiliary selection. Studia Linguistica 47. Link, G. (1983) The logical analysis of plurals and mass terms: a lattice theoretic approach, In Bäuerle et al. (eds) Meaning, Use and interpretation of Language, Berlin de Gruyter. Kayne, Richard S. (1994) The Antisymmetry of Syntax. Cambridge: MIT Press. Kiefer, Ferenc. (1994) Aspect and syntactic structure, in Ferenc Kiefer and Katalin É. Kiss (eds) The Syntactic Structure of Hungarian (Syntax and Semantics 27), Academic Press, New York. Liu, F-H. (1990) Scope Dependency in English and Chinese, Doctoral dissertation, UCLA. Longobardi, Giusseppe. (1994) Reference and Proper Names: a theory of N-movement in Syntax and Logical Form, Linguistic Inquiry, 25, 609-665. Kitagawa, Y. (1986) Subjects in Japanese and English, Doctoral dissertation, University of Massachusetts, Amherst. Lujan, Marta. (1981) The Spanish copulas as aspectual indicators. Lingua 54, 165-210. Kleiber, G. 1987. Relatives restrictives et relatives appositives: une opposition >>untrouvable<<? Tübingen, Niemeyer. Mahajan, Anoop Kumar (1990) The A/A' Distinction and Movement theory'Doctoral dissertation, MIT. Koopman, Hilda and Dominique Sportiche (1991) The Position of Subjects, Lingua 85,211-258. Maling, Joan. (1993) Of Nominative and Accusative: the Hierarchical Assignment of Grammatical Cases in Finnish. In Anders Holmberg & Urpo Nikanne (ed) Case and other Functional Categories in Finnish syntax. Berlin, Mouton de Gruyter 49-74. Kratzer, Angelika. (1989) Stage and individual level predicates. In Papers on Quantification, NSF Grant Report, Department of Linguistics U. Mass, Amherst. Manzini, Maria Rita. (1983) Restructuring and reanalysis. Doctoral dissertation, MIT. Krifka, Manfred. (1989) Nominal Reference, Temporal Constitution and Quantification in Event Semantics. IN R. Bartsch, J. van Benthem, and P van Emde Boas (eds.) Semantics and Contextual expression. Foris Dordrecht, 75-115. Kuroda, S.Y. (1968) English relativization and certain related problems. Language 44 244-266. Lakoff, George. (1970) Irregularity in Syntax. New York, Holt, Rinehart, Winston. Larson, Richard K. (1988) On the Double Object Construction Linguistic Inquiry 19: 335-391. Larson, Richard K. (1990) Double objects revisited: Reply to Jackendoff, Linguistic Inquiry 21, 589-632. Marcantonio, A. (1978) Una particularità dei nomi 'mass' e dei nomi 'count' nel dialetto abruzzese di Sulmona, Lingua e Contesto 5, 83-99. May, Robert. (1985) Logical form: its structure and derivation. Cambridge, MA: MIT Press. McCloskey, James. (1986) Inflection and conjunction in Modern Irish, Natural Language and Linguistic Theory 4, 245-281. McCloskey, James. (1990) A note on agreement and coordination in Old Irish, in S. Chung and J. Hankamer (eds.) A Festschrift for William Shipley, Syntax Research Center, UC Santa Cruz, 105-114. Meulen, Alice G.B. English Aspectual Verbs as Generalized Quantifiers, Proceedings of NELS 20 378-390. Lasnik, Howard.(1993) Lectures on minimalist syntax, ms., U.Conn. Lebeaux, David. (1983) A distributional difference between reciprocals and reflexives, Linguistic Inquiry 14, 723-730. Milsark, Gary (1977) Toward an Explanation of certain Peculiarities in the Existential Construction in English Linguistic Analysis, 3, 1-30. Lebeaux, David. (1987) Verb aspectual classes and the progressive. Ms. U.Mass. Milsark, Gary. (1985) Case theory and the grammar of finnish. Proceedings of NELS 15: 319-331. Lebeaux, David. (1988) Language Acquisition and the Form of the Grammar, doctoral dissertation., University of Massachusetts, Amherst. Mitchell, Erika. (1991) Evidence from Finnish for Pollock's theory of IP. Linguistic Inquiry, 22: 373-379. Lebeaux, David. (1994) Where does the binding theory apply? ms., University of Maryland. Mohammed, M. (1987) Nominative case, I-subjects and subject-verb agreement, in Proceedings of the Parasession on Agreement, Chicago Linguistics Society . Lehtinen, Meri (1963) Basic Course in Finnish, Bloomington IN: Indiana University Publications, Uralic and Altaic Series, Vol. 27. Moltmann, Fredericke. (1994) Parts and Wholes in Semantics. (ms) to appear with Oxford University Press. Lemos, Claudia.(1987) Ser and estar in Brazilian Portuguese; with particular reference to child language acquisition. Tübingen: Narr. Morgan, Jerry .L. Some aspects of relative clauses in English and Albanian. The Great Chicago Which Hunt. CLS p.63-72. 426 427 Munn, Alan. (1987) Coordinate structure and X-bar theory, McGill Working Papers in Linguistics, 4-1, 121-140. Pustejovsky, James. (1989) The geometry of events. In C.Tenny(ed.) Generative approaches to aspect. Cambridge, MA: MIT Lexicon Project. Munn, Alan. (1992) A null operator analysis of ATB gaps, The Linguistic Review 9, 126. Quine, W. V. O. (1960) Word and Object. Cambridge, MIT Press. Munn, Alan. (1993) Topics in the Syntax and Semantics of Coordinate Structures, Doctoral dissertation, University of Maryland, College Park. Munn, Alan. (1994a) A minimalist account of reconstruction asymmetries, Proceedings of NELS 24, 397-410. Quirk, Randolph. et al.(1985) A Comprehensive Grammar of the English Language. London, Longman. Raposo, Eduardo and Juan Uriagereka (1990) Long-distance Case Assignment, Linguistic Inquiry 21, 505-37. Munn, Alan (1994b) First conjunct agreement without government, paper presented at NELS 25, U. Pennsylvania. ms. Michigan State University. Raposo, Eduardo & Juan Uriagereka. (1993) Two Tpes of Small Clauses. talk presented at GLOW, Lund; to appear in A. Cardinaletti and G. Giusti (eds) Small Clauses, Academic Press, San Diego. Munn, Alan (1995a) The possessor that stayed close to home in V. Samiian (ed.) Proceedings of Western Conference on Linguistics (WECOL) 24, 1995. Rizzi, Luigi. (1982) Issues in Italian Syntax. Foris, Dordrecht. Munn, Alan (1995b) First Conjunct Agreement without Government: a reply to A, B, S. ms Michigan State University. Nishida, Chyo (1994) The Spanish reflexive clitic se as an aspectual class marker. Linguistics 32: 425-458. Rizzi, Luigi. (1986) 'Null objects in Italian and the Theory of pro', Linguistic Inquiry 17, 501-57. Roberts, Ian (1989) Compound Psych Adjectives and the Ergative Hypothesis, Proceedings of NELS 19, 1989. Roberts, Ian (1991) Excorporation and Minimality. Linguistic Inquiry 22, 209-218. Nunes, Jairo (1993) English Participle Constructions: Evidence for a [+PF,-LF]Case. In: Mason, C et al (eds) University of Maryland Working Papers in Linguistics. p66-79. Roberts, Ian (1993a) Verbs and Diachronic Syntax. Dordrecht: Kluwer. . Parsons, Terrence. (1990) Events in the semantics of English: a study of subatomics semantics. Cambridge, MA: MIT Press. Roberts, Ian (1993b) Agreement Clitics and Case. In Research Papers in Welsh Syntax 5, 92-106. Partee, Barbara. (1977) John is easy to please: In A. Zampolli (ed.) Linguistic structures processing. Amsterdam: North Holland. Rochemont, Michael.S. (1986)'Focus in Generative Grammar', Benjamins, Amsterdam. Partee, Barbara.(1993) 'Domains of Quantification and Semantic Typology', in Frances Ingemann (ed) Proceedings of the 1990 Mid-America Linguistics Conference, University of Kansas, 1991. Rochette, Anne (1988) Semantic and Syntactic Aspects of Romance Complementation, Ph D dissertation, MIT, Cambridge. Rooth, Mats., (1992), 'A theory of Focus Interpretation', Natural Language Semantics, 1, 75-116. Pesetsky, David. (1982) Paths and categories. Doctoral dissertation, MIT. Pesetsky, David. (1989) Language Particular Process and the Earliness Principle. ms., MIT. Ross, John Robert. (1967) Constraints on Variables in Syntax, Doctoral dissertation, MIT, Cambridge, Mass. Rothstein, Susan. (1983) The syntactic forms of predication. Doctoral dissertation,MIT. Piñon, Christopher. (1992) Nominal reference and the imperfective in Polish and English. Proceedings of NELS 23. Piñon, Christopher. (1993) Accumulation and aspectuality in Polish Proceedings of NELS 24, 1993. Pollock, Jean-Yves. (1989) 'Verb movement, Universal Grammar and the Structure of IP' Linguistic Inquiry, 20,365-424. Postal, Paul M and Geoffrey. K. Pullum. (1988) Expletive Noun Phrases in Subcategorized Positions Linguistic Inquiry 19, 635-670. Rothstein, Susan. (1991) Syntactic Licensing and Subcategorization. In S. Rothstein (ed.) Perspective on Phrase Structure: heads and licensing, Syntax and Semantics, 25, Academic Press. pp 139-158. Rothstein, Susan (1995) Pleonastics and the Interpretation of Pronouns Linguistic Inquiry 26, 499-529. Ruiz, R. N. (1963) Ser y estar: estudio sobre el sistema atributivo del español. Salamanca: Filosofia y Letras, (Vol. 23). Schachter, Paul. (1973) Focus and relativization, Language 49 19-46. 428 429 Schein, Barry. (1993) Plurals and events. Cambridge, MIT Press. Schmitt, Cristina. (1992a) Ser and estar: aspectual selection and composition, Proceedings of the Leiden Conference for Junior Linguists, Leiden 1991. Schmitt, Cristina. (1992b) Ser and estar: a matter of aspect, Proceedings of the Northeastern Linguistic Society (NELS) 22. Schmitt, Cristina. (1995a) Accusative Clitic Doubling, Participial Absolutes and have+agreeing Participles, Maryland Working Papers in Linguistics 2. Stump, Gregory T. (1985) The semantic variability of Absolute Constructions. Kuwer, Academic Publishers, Dordrecht. Suñer, Margarita. (1988) The role of Agreement in Clitic Doubled Constructions, Natural Language and Linguistic Theory 6,391-434. Szabolcsi, Anna (1983) The Possessor That Ran Away From Home, The Linguistic Review, 3, 89-102. Szabolsci, Anna. (1992) 'Subordination: articles and complementizers'. In Kenesei & Pleh (eds.) Approaches to Hungarian, 4 JATE, Szeged. Schmitt, Cristina (1995b) Types, Tokens, AgrO and Aspect, in V. Samiian (ed.)Proceedings of WECOL, 24 [also inProceedings of Eastern States Conference on Linguistics 94 (ESCOL)]. Szabolsci, Anna. (1993) Islands and Scope. Course notes, LSA Linguistic Institute, OSU, Columbus OH. Schmitt, Cristina. (1995c) Antecedent Contained Deletion meets the Copy Theory, Proceedings of North Eastern Linguistic Society (NELS) 25, 1994. Szabolcsi, Anna (1994) The Noun Phrase In: The Syntax of Hungarian , Syntax and Semantics 27 Academic Press, New York. . Schmitt, Cristina. (1996) On the necessity of QR as a syntactic rule Proceedings of CONSOLE 2, Tübingen. Tasmowski-De Ryck, Lorraine. (1990) Les démonstratifs français et roumains dans la phrase et dans le texte. Langages, 25, 82-99. Schmitt, Cristina. (to appear) Lack of iteration: Accusative Clitic Doubling, Participial Absolutes and have+agreeing Participles, Probus . Tenny, Carol. (1987) Grammaticalizing aspect and affectedness. Doctoral dissertation, MIT. Schot-Saikku, Päivu. (1993) What makes Finnish different? Remarks on a Sentence Type Theory of Finnish.Anders Holmberg & Urpo Nikanne. Case and other Functional Categories in Finnish syntax. (ed) Berlin, Mouton de Gruyter 207-224. Torrego, Esther. (1985) On Empty Categories in Nominals, ms. University of Massachusetts. Travis, Lisa de Mena. (1991) Inner aspect and the structure of VP. NELS 22. Selkirk, Elisabeth O.(1977) Some Remarks on Noun Phrase Structure' In Culicover, Wasow and Akmajian, Formal Syntax, Academic Press, New York 285-316. Uriagereka, Juan. (1988) On Government, Doctoral dissertation, U. Connecticut, Storrs. Seppänen, A. (1983) Finnish 'kaksi poikaa' Studia Linguistica 37:161-174. Uriagereka, Juan. (1991) Extraction parameters: a case study on underspecification. ms. UMCP. Simpson, Jane (1983) Discontinuous verbs and the interaction of morphology and syntax' in Michael Barlow et all (eds) Proceedings of the West Coast Conference on Formal Linguistics, v.2 Stanford, Stanford Linguistics Association, 275-86. Uriagereka, Juan (1993) Specificity and the Name Constraint, University of Maryland Working Papers, 1:121-43. Smith, Carlota S. (1991) The parameter of aspect. Dordrecht, Kluwer. Uriagereka, Juan (1995) Aspects of Clitic Placement in Western Romance, Linguistic Inquiry. . Spencer, Andrew. (1991) Morphological Theory, Blackwell, Cambridge. Sportiche, Dominique (1993) Clitic Constructions, ms. UCLA. Sproat, Richard and Chilin Shih (1988) Prenominal Adjectival Ordering in English and Mandarin Proceedings of NELS 18: 465-489. Vainikka, Anne. (1989) Deriving Syntactic Representations in Finnish, Ph.D. dissertation, U. Mass. Vainikka, Anne (1993) The three structural Cases in Finnish.In Anders Holmberg & Urpo Nikanne. Case and other Functional Categories in Finnish syntax. (ed) Berlin, Mouton de Gruyter 129-162. Stowell, Timothy. (1983) Subjects across categories. The Linguistic Review 2:285-312. Vendler, Zeno. (1967) Linguistics in Philosophy, Cornell University Press, Ithaca, NY. Stowell, Timothy. (1991a) The alignment of arguments in adjective phrases. In S. Rothstein (ed.) Perspectives on Phrase Structure: heads and licensing, Syntax and Semantics, 25, Academic Press. 105-138. Vergnaud, Jean-Roger. (1985) Dépendances et niveaux de répresentation en syntax Amsterdam: John Benjamins. Stowell, Timothy. (1991b) Small clause reestructuring. In R. Freidin. Parametric syntax. Cambridge: MIT Press. Vergnaud, Jean-Roger and Maria-Luisa Zubizarreta (1992) The Definite Determiner and the Inalienable Constructions in French and English. Linguistic Inquiry 23:595-652 . 430 431 Verkuyl , Henk J. (1972) On the compositional nature of aspects. Dordrecht: Kluwer. Verkuyl, Henk J. (1989) Aspectual classes and aspectual composition. Linguistics and Philosophy 12, 39-95. Verkuyl, Henk J. (1993) A theory of aspectuality. Cambridge: Cambridge University Press. Verkuyl, Henk J.(1995) Aspectualizers and Event StrucureOTS Working Papers Utrecht University. Vinet, Marie Thérèse (1989) Des petites propositions à valeur aspectuelle, Canadian Journal of Linguistics 34(2): 171-192. Watanabe, Akira (1993) AGR-based Theory and its Interaction with the A'-system, Doctoral dissertation, MIT, Cambridge. Wechsler, Stephen (1989) Accomplishments and the Prefix re- Proceedings of NELs 19:419-434. Wierzbicka, Anna. (1967) On the semantics of the verbal aspect in Polish. In: To honor Roman Jakobson: essays on the occasion of his seventieth birthday. Vol. 3. Mouton, The Hague 2231-2249. Williams, Edwin. (1981) Argument structure and morphology, The Linguistic Review 1, 81-114. Williams, Edwin. (1984) There-insertion, Linguistic Inquiry 15, 639-73. Williams, Edwin. (1985) PRO in NP, Natural Language and Linguistic Theory 3, 27795. Zagona, Karen. (1993) Perfectivity and Temporal Arguments. Paper presented at the 23rd Linguistic Symposium on Romance Languages, Northern Illinois University. Zagona, Karen. (1988) Verb phrase syntax; a parametric study of English and Spanish. Dordrecht: Kluwer. Zubizarreta, Maria Luisa (1987) Levels of Representation in the Lexicon and in the Syntax, Foris, Dordrecht. 432