* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download `Word syntax` and semantic principles
Focus (linguistics) wikipedia , lookup
Transformational grammar wikipedia , lookup
Lojban grammar wikipedia , lookup
Ojibwe grammar wikipedia , lookup
Distributed morphology wikipedia , lookup
Old Irish grammar wikipedia , lookup
Dependency grammar wikipedia , lookup
Meaning (philosophy of language) wikipedia , lookup
Preposition and postposition wikipedia , lookup
Word-sense disambiguation wikipedia , lookup
Macedonian grammar wikipedia , lookup
Musical syntax wikipedia , lookup
Untranslatability wikipedia , lookup
Esperanto grammar wikipedia , lookup
Old English grammar wikipedia , lookup
Junction Grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Contraction (grammar) wikipedia , lookup
Semantic holism wikipedia , lookup
Agglutination wikipedia , lookup
Polish grammar wikipedia , lookup
Yiddish grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Serbo-Croatian grammar wikipedia , lookup
Symbol grounding problem wikipedia , lookup
Cognitive semantics wikipedia , lookup
Compound (linguistics) wikipedia , lookup
Morphology (linguistics) wikipedia , lookup
Pipil grammar wikipedia , lookup
‘Word syntax’ and semantic principles* Gisbert Fanselow 1. INTRODUCTORY REMARKS The development of generative grammar has led to a significant reinter pretation of central notions in the study of human languages. The very notion of ‘language1 itself is now considered to be an epiphenomen of the interaction o f a number of independent task-specific modules of the hum an mind, am ong them being ‘formal competence* or Universal Gram mar (UG), the conceptual system (responsible for part of semantics, but not necessarily in a task-specific fashion), pragmatic competence, and so on (cf. Chomsky, 1980, 1986a). Consequently, for each regularity observable in natural languages we have to identify the mental module responsible for it, since the organi zational architecture of the hum an mind will not necessarily mirror the traditional divisions of linguistics. E.g., semantic facts concerning scope or binding appear to be reducible to UG, i.e. syntax, (cf. Hornstein, 1984; R einhart, 1983; May, 1985), while regularities in the domain of word semantics presumably are a consequence of principles of our conceptual system (cf., e.g., Jackendoff, 1983). The localization o f rules of word formation among the modules of the hum an mind has always been a central issue of generative research in this domain. One tradition, beginning with Lees (1960), considers word form ation (or at least part of it) to form an integral part of syntax, i.e., to be governed by general or word formation specific modules of UG (cf., e.g., Fabb, 1984; Pesetsky, 1985; or Roeper, 1987, among many others). The foundations for the other mainstream in word formation research were laid by Chomsky (1970), who argues that the formation of complex words should not be accounted for by syntactic rules and principles in a strict sense. In this paper, I will propose a specific version of the lexicalist hypothesis. While truly syntactic aspects of word formation are considered to be rather Author’s address: Allgemeine Sprachwissenschaft, Universität Passau, Postfach 2540, D- 8390 Passau, FR Germany. 96 G. Fanselow trivial in nature, and henceforth leamable, it will be argued that apparently syntactic facts such as ‘argument inheritance or restrictions on productivity reduce to properties of the human conceptual system, i.e., the set of rules and principles governing semantic representations and conceptual struc tures. The argument is organized as follows. A brief consideration of the relation between word formation and UG in section 2 is followed by an outline of certain central principles and regularities that can be attributed to the conceptual system. Section 4 then investigates the nature o f these conceptual principles as they bear on issues of word formation, with special emphasis on a) argument inheritance in German and Dutch nominalizations and b) constraints on productivity. 2. WORD FORMATION AND UNIVERSAL GRAMMAR In this section I will argue that no fully convincing examples of regularities in word formation exist that can be explained in terms o f principles of UG as developed by Chomsky (1981), (1986a), (1986b). Early generative approaches to word formation were completely embedded in the trans formational model of phrasal syntax. Disregarding the issue of their adequacy on a descriptive level1, three major problems may be identified: (i) (ii) (iii) arguments for a transformational approach involving comple mentation are invalid since the relevant data reduce to X-bar- theory (cf. Chomsky, 1970); quite a number of arguments in favour of word form ation as syntax referred to semantic or pragmatic facts (cf. Motsch, 1970); with the development of interpretive semantic theories like Montague Grammar, these facts have found a more adequate account in terms of semantic and pragmatic regularities (cf., e.g., Dowty, 1979; Fanselow, 1981,1985a; Hoeksema, 1985); since a highly structured theory of Universal G ram m ar has been formulated in the past two decades, the mere existence of some apparently formal-syntactic processes in word form ation does not demonstrate that word formation is part of syntax; rather, it would have to be shown how regularities of word form ation reduce to principles of Universal Grammar, or how arguably independent principles of word syntax interact with further modules of UG. Let us concentrate on the last point and examine some recent proposals relating principles of UG to word formation. Studies such as Hohle (1982), (1985), Lieber (1980), (1983), Selkirk (1982), Toman (1983), and Williams ‘Word Syntax’ and Semantic Principles 97 (1981) have dem onstrated that the process generating compounds reduces to a trivial statem ent like (1) plus a language-particular definition of the notion head. (1) X — Y Z where X,Y and Z are words. (2) In [x Y Z ], Y is the head of X (French, Italian...) In [x Y Z ], Z is the head of X (German, English...) Furtherm ore, it has been argued that derivational processes can also be captured by (1), with variables Y or Z refering to (dependent) morphemes as well (cf. the references given above).2 Selkirk (1982), (1984) and Toman (1983) - among many others - propose to relate (1) to the X-bar-scheme (3) of phrasal syntax, but it must be kept in mind th at an extension of (3) to word syntax requires a significant modification o f the content of X- bar-theory. (3) Xj ... X1..., where i^ j a n d ... is a sequence of maximal projections. In (3), non-head elements have to be maximal projections, and this does not hold true for word structures normally and perhaps even generally (but cf. Hoeksema, 1985 for a contrary view). Furthermore, no interesting configurational structuring comparable to the specifier-adjunct-complement distinction can be observed in word structures. Thus, while it is certainly possible to find a technical formulation unifying (1) and (3), this does not seem to carry much explanatory force: the X-bar-scheme of phrasal syntax generates a complex structure for which it is quite unclear how it might be learned by children and which therefore is a plausible candidate for a principle of UG. (1), on the other hand, just describes the ‘primitive’ process of concatenating two morphematic items, a regularity that is much too trivial to invoke a separate principle,3 since it can be learned in a very direct way.4 A more promising parallelism between word and phrasal structure might be seen in the shared existence of a head category in both kinds of structure. In other words, the features of a category /3 dominating a i , . a p p e a r to be determined by the features of one of the ax in both syntax and word form ation. Again, however, the situation is less clear than it might appear at first glance. Following Tom an (1987), let us first introduce a distinction between what one might call ‘categorial features’ (like lexical classes) and ‘selectional features* (like argument structure). For trivial reasons, both simple and complex words have to belong to lexical classes accepted by UG, i.e., a complex word m ust possess a fully licensed categorial specification. More 98 G. Fartselow often than not, this categorial specification is indeed derived from the features of one of its parts, which may be either a word (as in compounds or some prefix structures in English or German) or an affix. Heads do not have to be located on the same side of the branching in syntax and word structure (cf. Lieber, 1980), and it might even be the case that head location is different for various subtypes of affixes (cf. Lieber, 1980) and between derivational structures and compounds. Obviously, then, the pro cesses of head identification in syntax and word form ation cannot be identified in a simple way. Furthermore, it is far from clear whether or not complex words always possess heads determining their categorical features, cf. French compounds of the compte-gouttes type and ‘zero-derivations’ like English to blanket or German weissen ‘to whiten’.5Obviously, phonologically zero head affixes might be postulated here, but in a sense this move appears to be empirically vacuous since no theory of licensing conditions for empty affixes is in sight. In particular, it does not seem to be the case that principles governing empty categories in syntax apply to empty affixes. It is im portant to stress this difference between syntax and word formation: empty categories in syntax help to predict a huge set of empirical facts while empty affixes in word formation would just have to be postulated in order to save some apparent generalization of feature determination from empirical counter examples. To this, one might add the problem of additional word formational processes mentioned in note 4. Viewed from a universal perspective, all that can be said about categorial features of a complex word is that these can be derived from categorial features of one of its components, but do not have to. Since a ‘free’ shift in categorial value is forbidden in phrasal syntax, principles determining categorial features in syntax and word formation do not appear to stem from a common base. One central point needs to be emphasized: the leading principle of theory construction in generative grammar, the issue of learnability, is relevant for word formation as well. Thus, while we might very well identify some generalizations about categorial feature determination in German, or IndoEuropean complex words, or of natural language in general, this in itself does not suffice to reify the generalization as a principle of the hum an mind. In addition - as in phrasal syntax - it would have to be shown that a child could not possibly learn about the categorial feature deter mination of a given class of complex words. Obviously, however, the child does have massive positive evidence here, because the complex word in question is used as the head of some maximal projection occupying a determined slot in a phrasal branching. Syntactic principles identify the categorial features of this maximal projection, and the X-bar-scheme (3) enforces an identification of these features with the features of the complex word. In languages with a rich morphological system like German, case ' Word Syntax’ and Semantic Principles 99 and inflectional endings will also tell the child if the complex word he/ she encounters is a noun, a verb or an adjective. This is in sharp contrast to the situation the child is faced with in phrasal syntax, where ‘logical problems’ of the acquisition of certain regularities abound. To sum up, the necessity of a theory of the determination of categorial features as a submodule of the hum an mind has yet to be demonstrated for word formation. ‘Selectional features’ appear to pose a problem in this respect. This is in particular true o f rules governing the argument structure of complex words, which are usually formulated as some process of ‘argument inheritance’: in a structure like driver o f a van, the NP a van seems to function as the direct object argument of the verb drive, i.e., the complex word driver appears to have inherited the argument structure of its nonhead com ponent. Categorial typifications of inherited arguments appear to follow from principles independent of word formation (e.g., the object argument of drive cannot be realized as an object NP since a noun like driver will not assign case in English). It does not appear to be possible to predict argument inheritance properties of a given affix from its morphophonological class, cf. Randall (1987). At first glance, it might be claimed that all one has to do is to state inheritance properties for each affix, but a brief consideration of some G erm an data proves this assumption to be false. Like in English, G erm an agent nouns formed with an er-suffix generally appear to inherit the object argum ent of the underlying verb; cf. Fahr-er des Lastwagens ‘driver of the van’.6 While in general this holds also for newly coined agent nouns, there are quite a number of verbs not allowing for inheritance of arguments, e.g. lösen ‘to solve’. A phrase like *der Los-er dieses Rätsels ‘the solver of this riddle* is ungrammatical. Since data like these constitute negative evidence, a learning problem comparable to the situation in phrasal syntax will arise that calls for a principled solution. The main body of section 4 will be devoted to an account of argument inheritance in terms of semantic regularities. In a recent paper, Pesetsky (1985) argued for an application of both Move a and Binding Theory in complex words. Again this would constitute an argument for integrating word formation into syntax proper, but Booij (1987) and Hoeksema (1987) have argued that Pesetsky’s approach does not really account for argument structure facts in complex words. Applying Move a to affixes might still, however, represent a solution to well-known problems with scope. E.g., the morphological structure of unhappier, i.e. \un\happi-er\\ seems to be in conflict with semantics: interpreting this structure in a compositional fashion, the resulting meaning appears to be ‘not being more happy’ instead of the correct ‘being more (not happy)’. Here, affix movement might map the word’s S-structure to an LF such 100 G. Fanselow as [[un-[happi-ti]] en], in which the comparative affix er would correctly have scope over the negation morpheme un. Similar processes of reanalysis might be necessary for well-known riddles such as transformational gram marian and constitute the second domain in word form ation where syntactic explanations seem to be called for. Again, we will sketch a non-syntactic approach in section 4. With one major exception, further subtheories of U G like the theory of barrierhood (cf. Chomsky, 1986b), control and binding theory do not even have a chance of application in complex words, since the relevant structures or items do not show up.7 On the other hand, 0-theory has been claimed to be of some importance for derivation and composition (cf., e.g., Boase-Beier & Toman, 1986), but it seems to be obvious* that the 0-criterion itself does not hold for elements internal to complex words directly. Thus, in a German nominalization such as Vermiet-ung ‘renting’ none of the underlying verb’s 0-roles has to be satisfied.8 A final domain9 of syntactic explanation in word form ation appears to consist in a set of constraints on productivity both in composition and derivation. E.g., there are no N + P compounds in G erm an, and verbal composition is quite restricted. Some of these constraints might involve reference to non-trivial syntactic distinctions like ergativity: er-affixation seems to be restricted to non-ergative verbs in German and English. Since the notion of ergativity is definable in terms of UG only, any account of the ungrammatically of forms like *Ankomm-er ‘arriver’ would place at least er-derivation into syntax proper. Constraints on productivity will consequently constitute the third topic of section 4. Summing up the discussion presented so far, we may conclude that there are no compelling arguments in favour of considering word formation to be part of UG. Rather, a number of data would render this approach quite problematic. Nevertheless, we are able to identify three major problems for a non-syntactic theory of word formation: (i) (ii) (iii) the process of argument inheritance; restructuring as an account of affixal scope; constraints on productivity. I shall develop a semantic approach to these issues in section 4. Before proceeding, however, a few remarks on the structure o f a semantic theory are in order. 3. OUTLINE OF A SEMANTIC THEORY OF WORD FORMATION The primary goal of any theory of word formation is to explain the hum an ‘Word Syntax ' and Semantic Principles 101 capability of understanding an unlimited number of new complex words. In the eighties, beginning with the major contribution by Dowty (1979), quite a num ber o f rule systems for semantic interpretation of complex words have been developed in the Montague Grammar framework, such as Fanselow (1981), Hoeksema (1985), Moortgat (1983), (1985), to cite just a few. While these systems appear to be more or less descriptively correct insofar as they ascribe correct interpretations to complex words, they do so in a conceptually problematic manner.10 According to this ‘standard’ approach, for each syntactic class of complex words such as [ nN + N ] or [NV + er ] and for each interpretive class (agentive vs. instrumental), a specific semantic operation is designed. Thus, a rule like (4) might be responsible for the correct interpretation oiD orf-burgermeister ‘village m ayor’, but it is clearly inapplicable to words like Dampf-boot ‘steam b o at’, Kiisten-stadt ‘coast town’ and many more. (4) If a is a compound of the form 06, and if 8 is a relational noun with the meaning 8 \ and if ft is a common noun with the interpretation j3’, then a ’, the interpretation of a, is: the property of being an x such that x is the 6’ of some /3’.11 An elaborate analysis of German N + N compounds requires some twentyfive rules of this kind, and it is obvious that one will arrive at quite a considerable num ber of different interpretive processes in an attempt to cover all kinds of compounds and derivatives of German and further languages. It is this very fact that renders the approach conceptually untenable. As H andw erker (1985) has pointed out, the rule system just consists of a listing of possible semantic options, but does not give a principled answer to the question why certain complex words receive just the meanings they have. This objection, however, holds of the standard Montague grammar approach to phrasal semantics as well. Here, each specific syntactic branching is connected to a specific interpretive rule, too. Obviously, this is incompatible with the results of modern syntactic theory, since there are no specific rules for the AP 4* N’ and V + NP branchings which have to be interpreted by conjunction and functional application, respec tively; all phrasal branchings are generated by the X-bar-scheme. Con sequently, interpretive processes may not be closely linked to syntactic rules in the way proposed by Montague (1973) as semantic differences are most often not directly mirrored in syntax. There are a few proposals of how this difficulty might be overcome (cf. e.g. type driven translation developed by Klein & Sag, 1985 for GPSG). According to the approach originally developed in Fanselow (1985a), a restricted set o f general interpretive principles is part of conceptual 102 G. Fanselow competence and is applicable to any kind of phrasal or word-internal branching.12It will include operations such as conjunction of two properties, applying a function to its argument, inducing a contextually prominent relation, closing an argument place of some function with some contextually prominent object or by existential quantification, and two or three opera tions in addition. Interpreting constructions such as green apple or conductor-composer is quite simple: one just has to apply the conjunction operation. Phrases like loves Mary and words such as Reagan-fan are interpreted by functional application. It is important to note that an application of other semantic processes is blocked in these cases for general reasons. Thus, both green and apple denote properties, i.e., technically speaking, one-place functions from objects (of whatever ontological status) to truth values (or whatever the reader may prefer), and obviously, a function of this sort accepts objects as arguments only, not functions. Consequently, functional application cannot be used in the process of interpreting green apple. Similar con siderations hold for loves Mary. Taking Mary to denote some object, con junction will be inoperative since only semantic entities of the same type can be conjoined. Thus, in general, type diversifications for denotata of words and phrases will exclude incorrect applications of interpretive processes without any further stipulation. It appears to be unnecessary to stipulate any connection between types of syntactic structures and semantic operations. A system freely applying primitive semantic operations to any kind of structure presupposes, however, an independently m otivated correlation between syntactic categories and semantic types, both in the sense of Montague (1973) and certain further suggestions by Jackendoff (1983). Nouns, e.g., generally denote one- or two-place properties (cf. man and brother, respectively). In a complex word like Reagan-biographer, biographer denotes a one-place-property of objects, and Reagan refers to some person. In principle, it would be possible to apply the function expressed by biographer to the object denoted by Reagan in the system just sketched, but this will yield either a truth value or the proposition that Reagan is a biographer as a possible meaning for the compound. Clearly, the proposition that Reagan is a biographer is not among the possible meanings of Reagan-biographer. Obviously, however, this undesirable instance of functional application violates the meaning type correlation of nouns with properties for the simple fact that it ascribes a propositional interpretation to a noun. Generally, the operation of semantic processes seems to be governed by a principle requiring that the result of their application may not violate correlations between categories and types of meaning that are established independently. The derivation of the correct interpretation of Reagan-biographer illus W o rd Syntax’ and Semantic Principles 103 trates yet another aspect of the semantic system proposed in Fanselow (1985a). Reagan-biographer refers to those persons who are a) biographers and who b) described the life of Ronald Reagan. This interpretation can be arrived at by an iterative application of different semantic processes to the same structure. Thus, first the process of inducing a prominent relation13 (viz. ‘writing about the life of sb.’) from the meaning of biographer will operate. This two-place-relation is then applied to the meaning of Reagan, yielding the one-place-property ‘describing the life of Reagan’ which now may be conjoined with the one-place-property expressed by biographer. In Fanselow (1985a), I tried to show that, in fact, interpretive options of compounds in German and further languages can be exhaustively predicted by a system of the kind just outlined, but for reasons of space, I cannot go into this m atter any further.14 Derivations pattern with compounds quite generally. E.g., stereotypical relations are induced in German er-nouns such as Lyrik-er ‘lyrician’ (= sb. who writes lyrics), Gewerkschaft-er ‘unionist’ (=sb. who works for the unions). The induced relation will subsequently be applied to the quantified meaning of the non-head-part of the derivative. Branching affixal structures thus will be interpreted by processes independently motivated for phrases and compounds. This implies that affixes are considered to be meaningbearing elements themselves, and not just morphological reflexes of specific semantic operations. The German er-affix seems to express the property of being a concrete object (which may be conjoined with a verbal meaning in words like Schlaf-er, ‘sleeper’). We will go into some more details below.15 4. WORD SYNTAX AND SEMANTIC PRINCIPLES 4.1. Argument inheritance I: the case o f German er-nouns In Fanselow (1985a), I tried to show how the semantic system just outlined will make non-trivial predictions in the domain of apparently syntactic restrictions on compounding. In this section, it will be argued that the same holds for derivation as well, in particular, options for argument inheritance will be shown to derive from semantic facts. Structures such as (5) are often cited as evidence for the claim that nominalizations formed by the agentive/instrumental er-affix in Dutch, English or Germ an allow for argument inheritance. (5) a. b. c. der Fahr-er des Lastwagen ‘the driver of the van’ der Käuf-er des Mantels ‘the buyer of the coat’ der Les-er des Buches ‘the reader of the book’ 104 G. Fanselow The genitive complements of the complex noun seem to function as the object argument of the underlying verb in (5a>(5c). Consequently, it appears as if at least part of the argument structure of the underlying verb would have percolated up to the nominal node dominating the whole complex word. At least for German, however, this line of argum entation is quite problematic for a number of reasons. In the first place, it will not account for the fact that argument inheritance appears to be blocked for the instrumental reading of er-nouns in general,16 although Oh (1985) and Booij (1986) proved agentive and instrumental readings of V + er to be due to the very same syntactic and semantic process. This is illustrated by the examples given in (6), which would only be grammatical with an agentive reading of the er-noun. (6) *ein Kopier-er von Aufsätzen ‘a copy machine of articles’ *ein Druck-er von Graphiken ‘a printer of graphics’ ♦der Misch-er von Beton ‘the mixer of concrete’ *ein Transport-er von 01 ‘a transporter of oil’ *ein Öffn-er der Konservendosen ‘an opener of cans’ Furthermore, the verbs underlying the nominalizations given in (5) possess an additional intransitive reading as well: (7) a. b. c. Hans fährt selber Hans drives himself ‘Hans is driving himself Hans kauft selten im Warenhaus Hans buys seldom in-the department store ‘Hans seldomly does his shopping in departm ent stores’ Hans liest nur am Abend Hans reads only at-the evening ‘Hans reads in the evening only* Olsen (1986) has presented some evidence demonstrating that the nom i nalizations in (5) are, indeed, derived from the intransitive versions of the verbs in question. Quite a number of German transitive verbs do not allow for an intransitive reading. Examples would be hemmen ‘to inhibit’, lösen ‘to solve’, verletzen ‘to violate’, schlucken ‘to swallow’ (in one reading), or befahren ‘to drive in*. Crucially, these verbs generally do not enter the type of construction exemplified by (5).17 (8) *der Hemm-er des Appetits ‘the inhibitor of appetite’ *er Lös-er des Problems ‘the solver of the problem* W ord Syntax’ and Semantic Principles 105 *der Verletz-er der Grenze ‘the violator of the border’ *der Schluck-er der Tabletten ‘the swallower of the pills’ *der Befahr-er der Autobahn ‘the driver of the motorway* Examples like (8) seem to suggest that - for some mysterious reason er does not attach to transitive verbs in German obligatorily, but data like (9) prove this assumption to be false. (9) ein ein ein ein ein Appetit-hemmer ‘an appetite-inhibitor’ Problem-löser ‘a problem solver* Grenz-verletzer ‘a border violator’ Tabletten-schlucker ‘a pill swallower’ A utobahn-befahrer ‘a motorway driver’ Similarly, examples such as (6) become grammatical if the verbal argument is expressed within the nominalization: (10) ein ein ein ein ein Aufsatz-kopierer ‘an article copier’ G rafik-drucker ‘a graphics printer’ Beton-mischer ‘a concrete mixer’ Ö l-transporter ‘an oil transporter’ Konservendosen-öffner ‘a can opener’ Obviously, then, German er cannot combine with a verb if one of its (non-subject) argum ent places is still unfilled; this will restrict er-derivation to those verbs which either are basically intransitive or allow for an intransitive reading, on the one hand, and N +V combinations, on the other. The picture that emerges from these considerations is, then, in close line with the suggestions made by Lieber (1983): we assume a structure [[N+V]-£r] for the words given in (9) and (10), and it also appears to be the case that verbal arguments must be satisfied within the complex word itself, and not by some phrase external to it (cf. (6) and (8)). Booij (1987) has raised some objections to Lieber’s approach which partially will also affect the one to be developed here. I will turn to his point concerning the non-productivity of N+V forms in section 4.3 below, and concentrate here on the issue of the correct interpretation of phrases such as driver o f a van or its Germ an counterpart Fahr-er eines Lastwagens where the N P /P P complement appears to function as the object argument of the underlying verbal relation. But while it seems clear that the relation between the noun and the genitive NP (in German) and the o f PP (in English) may be one of driving (among many other conceivable relations), it is by no means a priori true that this relation is in a strict sense the meaning o f the verb. 106 G. Fanselow There are quite a number of German nouns referring to persons performing a specific activity that are not formed by some productive process of German word formation, such 2ls Autor ‘author’. In constructions like Roman-Autor ‘novel author’, Autor des Romans or Autor vom Roman ‘author of the novel’, the relation between novel and author is the one of writing, as it would be with words like English writer and G erm an Verfasser, where the relation implicit in the noun happens to be expressed overtly in the morphosyntactic structure of the noun. Obviously, the correct interpretation of Autor des Romans can be computed by applying the semantic operation of inducing a prom inent relation from the meaning of one of the parts of the construction, as will be the case with Passagier des Flugzeugs ‘passenger of the plane’ for the relation of moving-in. There is no reason to expect this semantic process not to apply to constructions with morphologically complex nouns. In other words, the relation of ‘driving’ is one of the relations that will become prominent when the word driver is being used, and we may compute the correct interpretation of driver o f a van or Fahr-er eines Lastwagens with the help of the process of inducing a relation, but w ithout any recourse to some process of argument inheritance. It is im portant to note that, on the other hand, interpretive facts about Autor des Romans or Passagier des Flugzeugs could be handled syntactically (i.e., by some process of argument inheritance) just in case we would return to full-fledged Generative Semantics, a solution I think few people would like to argue for. Before we demonstrate how the behaviour of Germ an er-nouns derives from the semantic system proposed in section 3, a brief discussion of the contrast between (5) and (7) seems appropriate. Within certain limits that will yet have to be determined, non-derived agentive nouns allow for an NP-complementation interpreted with the relation implicit in the noun’s meaning, cf. the examples given in (11): (11) der Autor des Buches der Arzt von Maria der Pilot der 747 ‘the author of the book, having writ ten the book* ‘the physician o f Mary, treating M ary’ ‘the pilot of the 747, flying the plane* On the other hand, non-derived instrumentals rarely (if ever) allow for a similar type of NP-complementation: (12) *das Flugzeug der Briefe ‘the plane of the letters, transporting letters’ *der Stift des Artikels ‘the pen of the article, with which the article has been written’ In other words, options for inducing relations from instrument nouns are ‘Word Syntax’ and Semantic Principles 107 in general much more restricted than is the case for agentive nouns, a fact that implies the contrast between (5) and (6) without any reference to word form ation.18 The behaviour of German £/*-nominalizations with respect to argument inheritance can be derived from semantics, however, since it can be shown that there is no way to interpret NP as the object of the verb in [N V + er] + NP. There are two principled ways o f interpreting structures such as Schlaf-er ‘sleeper’. Since schlaf refers to the one-place-property of sleeping, er could either denote a one-place property itself, which has to be conjoined with schlaf s interpretation, or could refer to a function mapping verbal one-place properties to nominal ones. In this case, functional application would be appropriate. If this perspective is adopted, a problem will arise if er is to be combined with a transitive verb. Since transitive verbs denote two-place-properties, a functional reading of er could not apply to the verb’s meaning since the latter does not have the proper argument type (i.e. one-place-property) of the former function. If er itself refers to a property, conjunction will be blocked since it applies to elements of the same logical type only. The only way to combine the meanings of transitive verbs with er in the framework developed above is by first closing the object argument place of the verb. The result of this operation will be a one-place property, and the process o f interpretation will proceed as for Schlaf-er. But since the object argument place has already been closed, no NP or PP external to it can be interpreted as the verb’s object any more. To sum up, even if optionally any feature of the non-head (besides categorial specification) may percolate up to the node dominating the whole complex word, argument inheritance will be blocked since the resulting structure turns out to be uninterpretable. It is the semantics of er which predicts argument inheritance not to occur.19 There appear to be two m ajor alternatives to this approach. Since the Uniformity Principle of Chomsky (1981) requires that each morphological operation is connected to a unique semantic effect (cf. Randall, 1987 for some discussion) we cannot assign two different meaning representations to er. Consequently, one would have to assume two homophonous ermorphemes, the meanings of which apply to one- and two-place properties, respectively, if argument inheritance is to be made interpretable within the limits of the approach developed in section 3. Clearly, this move to two hom ophonous er-affixes is not only completely unmotivated in German grammar, it will also predict th at the fact that the same phonetic sequence applies to all verbs in order to form agent nouns is a curious idiosyncrasy of the Germ an language. In an obvious sense, this prediction is incompatible with the fact th at in most (if not in fact all) languages the very same affix attaches to any kind of verb in the process of agent noun formation. Consequently, this proposal has to be dismissed. 108 G. Fanselow The other alternative would be to claim that the set o f possible semantic operations proposed in section 3 is too much restricted. In fact, there is a mathematical operation that would allow for an interpretation of argument inheritance, functional composition (cf. M oortgat, 1983 and 1985). But while adding a further semantic operation is unproblematic in the ‘traditional* Montague Grammar framework, since its application will be restricted to the syntactic branching it is stipulated to apply to, a semantic system that tries to be compatible with UG-syntax like the one proposed above cannot be enriched arbitrarily. If functional com position were among the set of possible semantic operations, we would predict that argument inheritance is possible for any kind of structure. E.g., it will be predicted to apply to compounds as Such-hubschrauber der Verletzten ‘search helicopter of the injured’ and to all derivations, an incorrect result for obvious reasons. Consequently, the following conclusion seems to be appropriate: the choice of semantic operation available in the hum an conceptual system does make strong predictions about apparent syntactic facts such as argument inheritance. The minimal set of semantic processes proposed here predicts argument inheritance not to occur (in a sense to be made clearer below), and it does not appear to require any major syntactic stipulations in this domain. If it were to be enriched with functional composition, (a move that again would have to be m otivated with semantic arguments), argument inheritance is predicted to occur with any type of structure, and consequently, a complex theory would have to be developed in order to block the greater number of illicit applications. Obviously, if it turns out to be tenable, the first approach is to be preferred. 4.2. Argument inheritance II: vacuous affixes While interpretive facts about German [ [ n V + er] PP/N P] constructions seem to be due to a process of inducing prominent relations rather than to argument inheritance, the reality of the latter type of operation cannot be denied for forms such as English the putting o f men on the moon, the pasting o f stamps onto envelopes, letters given to the child or German den Mann des Mordes anklagend ‘the man the murder accusing, accusing the man of murder’, des Mordes angeklagt ‘the murder accused, accused of murder’, Peters sich mit Maria Vergleichen ‘Peter’s himself with Mary compare, Peter’s comparing himself to Mary*. These forms appear to involve Kiparsky’s and Randall's Level 3 morphemes mostly. The complex words in question are complemented by arguments that are formally and obliquely selected for by the derivational base, a fact that proves the complement specification of the verb to have percolated up to the derivative. In her discussion of these data, Randall (1987) suggests that the ‘Word Syntax ' and Semantic Principles 109 percolation properties to be found with participles and lexically derived ergatives are due to the fact that the item in question has not undergone a shift in category specification, i.e., in general, blocking effects on percolation would show up with affixes changing category only. As for wig-process nominals, which are formed by a category-changing operation, she stipulates th at the operation does not block 0-role assignment. A consideration of interpretive facts, however, suggests yet another generalization. Processive z/tg-nominals, and processive nominal infinitives and G erm an participles and lexically derived ergatives do not appear to involve a (significant) change in meaning. Consequently, it might be claimed that argum ent inheritance is possible if the morphological operation does not affect the m eaning of the underlying predicate. This assum ption seems to be corroborated by action nominals as discussed by R appaport (1983). D ata like the general*s command to the troops to leave or the sale o f missiles to Iran suggest that argument inheritance is possible, but, as R appaport observes, it will be blocked if the nouns are not taken to refer to the action itself, but to its result, i.e., if a shift in meaning has occured. Similar facts appear to hold for German. To sum up, ju st those affixes appear to allow for argument inheritance that are semantically empty. If correct, this generalization is already fully predicted within the semantic system of section 3: argument inheritance will be semantically blocked only if some meaning element a has to be added to the verbal meaning that requires that the argument slots have to be closed in order to make semantic composition possible. If no meaning element has to be added because the affix is meaningless, the verbal ar gument slots may be closed by a semantic process, but do not have to. Consequently, these arguments may still be filled by elements external to the complex word. In a sense, then, the predictions of my approach come close to Booij’s (1987:2) claim that “ the argument structure of a base word is inherited by the derived word unless the semantics of the word form ation process involved gives rise to changes in this argument structure” . Let me conclude this section by noting that not every aspect of a pretheoretical notion of meaning will be relevant for argument inheritance. Germ an process nominalizations are a case in point. Besides the neutral suffixes en and ung, it is possible to form nominalizations with erei and ge-e that bear a pejorative meaning. Thus, Red-en ‘talking’ is a neutral process name, whereas Red-erei and Ge-red-e mean something like ‘endless goddam talking’. Obviously, the difference in question cannot be handled in terms of truth conditions or the nature of the concepts involved, but should be accounted for in a pragmatic theory of emotional meaning and connotation. Consequently, Reden, Gerede, and Rederei have identical semantical values, and the suffixes erei and ge-e should be semantically 110 G. Fanselow vacuous. This will correctly predict argument inheritance to be possible: das Reden über Maria ‘the talking about M ary’, das Gerede über Maria, die Rederei über Maria ‘endless and useless talking about M ary’, Lachen über Maria ‘laughing at Mary’, die Lach-erei / das Ge-lach-e über Maria ‘laughing at Mary endlessly’.20 4.3. Constraints on productivity and argument inheritance III: the case o f Dutch tx-nouns Leaving aside the topic of argument inheritance for a m om ent, let us briefly discuss the second kind of syntactic facts in complex words, constraints on the productivity. A rather simple solution for constraints on prepo sitional compounding can be derived in semantic terms, [ n P + N] ■ compounds are quite frequent in German, cf. words such as Vor-Zug ‘beforetrain - train leaving before sth.*, Zwischen-Bericht ‘between-report intermediate report*. Meanings are assigned in quite a straightforward manner: first, the open object argument place of the preposition is closed, and the result will be conjoined with the meaning of the noun. More often than not, the contextually specified object closing the preposition’s open object argument will be one of the objects referred to by the noun, but not necessarily so, cf., e.g., Zmschen-schlaf ‘between-sleep - a sleep interrupting some activity’. Further kinds of semantic operations are logically applicable, but will be blocked by the constraint requiring that no semantic operation may assign meanings incompatible with categorytype-correlations. Thus, e.g., the meaning of vor might be functionally applied to a quantified version of the interpretation o f Zug, yielding being before some train, but this is a possible PP meaning only, incompatible with the set of meanings available for nouns. The productivity of P+N combinations is in sharp contrast with the non-existence of N+P-compounds. The solution of the problem, however, is quite straightforward, if we consider semantic restrictions that go beyond logical types. Obviously, prepositions may just refer to a restricted class of two-place relations, and any conceivable com bination of nominal and prepositional meanings will go far beyond this restricted set of possible P meanings. Since N4-P combinations have to be prepositions by virtue of the preposition being the compound’s head, again we are faced with a violation of the constraint that no semantic rule may assign improper meanings, and N+P compounds will be blocked in a correct fashion without any syntactic constraints being necessary. Bearing this discussion in mind, let us now turn to [N-fV] combinations which are fairly unproductive, a point Booij (1987) used to argue against Lieber’s and the present account of forms such as truck driver. Some observations suggest, however, that the unproductivity of N + V com pound ‘Word Syntax’ and Semantic Principles 111 ing might not be due to a peculiarity of word formation processes, but rather stem from facts of verbal meanings in general. A closer inspection of the range of possible denotata for verbs (cf., e.g. Carter, 1975; Dowty, 1979) again leads far beyond logical typifications. Verbal meanings appear to be composed of some restricted set of quasi-logical operators such as CAUSE, BE, BEG IN , and so on, possibly applied to a set of verbal core concepts (which might all be stative in nature, a point I would not like to pursue here, cf. Dowty, 1979 for some discussion). Most importantly, the num ber of core concepts present in a given verbal meaning appears to be highly restricted. In most cases, (cf. Carter, 1975), the verb’s meaning will contain one core concept only, but combinations of verbal and nominal core concepts are possible, cf., e.g., electrocute, drown, etc. With the exception of verbs such as bleed, where the nominal meaning appears to be the only core concept present in the verbal meaning combined with an operator such as PRODUCE, the nominal core concepts seldom or never occupy argum ent slots of the verb, but are indicators of the specific m anner the action is performed in. Thus, electrocuting refers to the activity of killing with electricity.21 If these observations are correct, we may conclude that there is a constraint operating on English and German verbs, or verbs in general, ruling out meanings in which one argument slot of a verbal core is filled by a nominal designator. If this is correct, N + V combinations would be syntactically well-formed, but would violate a meaning constraint on simple and complex verbs in general. Indeed, the few existing N-fV combinations in German such as hand-fegen ‘hand-sweep’ or rad-fahren ‘bike-drive, go by bike’ refer to activities where the nominal contribution will always be interpretable as some - mostly instrumental - adverbial modificator. But while there may be no verb apfel-essen in G erm an referring to the activity of eating an apple for semantic reasons, constraints on verbal meanings obviously will not apply for nominalizations such as Apfel-esser ‘apple-eater’ if we make the natural assumption that meaning constraints apply to words only, and not to their parts. An argument in favour of this approach seems to derive from the well-formedness of words such as Pillen-schluck-maschine ‘pill swallow machine, machine swallowing pills’. N +V combinations being interpreted by functional application are always well-formed as non-head parts of com pounds, and this appears to hold for affixation as well, as the data discussed above suggest. Consequently, the second problem Booij (1987) notes with Lieber’s and my analysis o f truck-driver can be overcome as well. Turning to affixation, it is obvious that there are syntactic restrictions on the class o f words a given affix may be attached to. As far as major class differences are concerned, this appears to be unproblematic since no logical problem appears to emerge in the acquistion of these regularities, 112 G. Fanselow if we make the natural assumption that a child will assume that affixes just apply to those lexical classes for which there is positive evidence in his/her input. The situation is different for restrictions within m ajor classes, however, since the child’s input will not tell him /her which of the number of conceivable subclassifications of verbs or nouns is relevant for the constraint in question. In particular, if, for instance, er-derivations would be impossible for ergative verbs (cf. Burzio, 1980 for some suggestions), this would prove notions to apply in word syntax which can be defined in terms of UG only, i.e., it would prove er-affixation to be part of syntax proper. A closer inspection of constraints on German er-affixation suggests a semantic account, however. Firstly, er-nominalizations generally belong to the class of concrete nouns, a fact that might be expressed in a pertinent feature specification of the affix and that will also have to be represented in its meaning structure. Consequently, even if some fact has proven some theorem, this fact cannot be referred to as a Theorem-Beweis~er, ‘theoremprover’, a word that denotes persons or machines proving theorems only. Since quite a number of ergative verbs such as passieren ‘happen* require non-concrete objects in subject-disguise, the meaning of er itself will suffice to block words such as Passier-er ‘happener*. er-nominalizations display temporal and modal differentiations in their interpretation, including perfective, and ‘processive’ aspects and a habitual reading. Thus, Tiinz-er ‘dancer’ may either refer to a person who is actually engaged in the activity of dancing, or to a person who habitually dances under certain circumstances. This diversification can be fully reduced to a corresponding difference in the interpretation of the base verb, as was demonstrated by Oh (1985). In this respect, a constraint on nominal meaning comes into play. In contradistinction to verbs, nouns never refer to punctual properties or predicates in the sense of Vendler’s classification. This alone will suffice to account for the fact that forms such as Aufwach-er ‘wake-up -e r’ or Ankomm-er ‘arrive-er’ are impossible with a non-habitual reading, since the underlying verbs are not only ergative in nature, but punctal as well. Crucially, these words are well-formed in German with a habitual reading, in particular, when they are composed with temporal specifications such as Friih-aufwacher ‘early wake-up-er’ or Spat-ankomm-er ‘late-arrive-er’. Obviously, there can be no restriction blocking er-affixation to ergative verbs at work. Rather, ergative verbs typically denote meanings which are in conflict with the requirements of the category-meaning correlation of nominals. The same point may be made for impossible words such as *Steiger ‘riser*. V+ er seems to require that the objects in its denotation are actively involved in the process expressed by V. On the one hand, this implies ‘Word Syntax9and Semantic Principles 113 that V + er structures are typically bad if the verb does not express an activity but a state, consequently, er does not attach to mssen ‘know’, lieben ‘love’ in German. Those combinations of er with stative verbs which are acceptable (Amerifca-hasser ‘America-hater’) are typically interpreted with respect to specific activities connected with the relevant state, as Oh (1985) has observed. In a similar sense, while German fahren may be understood either as an activity (to drive) or as a non-active process (as in m it dem Zug fahren ‘to go by train ’), Fahrer exclusively refers to people or objects that are actively moving a vehicle. Since no activities are involved in the processes o f growing or rising, logically possible derivations appear to be blocked by the semantic properties of the er- suffix.22 While there are further constraints on er-derivation (cf. Koch, 1976 and Oh, 1985 for detailed discussion) major restrictions appear to follow from semantic specifications, and not from syntactic notions such as ergativity. This conclusion is in line with the results of Abraham (1985), who demonstrated that the notion o f ergativity is also irrelevant in the formation of adjectival participles and has to be replaced by semantic classifications. Interestingly, quite a number of the constraints we have discussed do not apply to D utch er-formation, as Booij (1987) observed. Thus, words such as lijder ‘sufferer’, daler ‘dropper’, stijger ‘riser’, breker ‘breaker’, that object that is breaking’ are well-formed. In any case, this will require that there are differences in meaning between German er-formation on the one hand, and its Dutch counterpart, on the other. These meaning differences might either be directly attributed to er, or connected to the affixation rule. The most radical approach would be to assume that Dutch er is completely meaningless, with interpretive differences between V and V +er being attributable to the predication-reference distinction connected with the respective syntactic slots (cf. note 21). If this is the case, an interesting prediction follows. In contradistinction to Germ an, D utch should allow for argument inheritance with er-nouns, and the data Booij (1987) discusses demonstrate that this prediction is borne out. Obligatory arguments of the verb may be expressed by PPs external to the words, cf. vormer van subjectsnamen ‘former of subject nouns’, initiatiefnemer ‘taker of initiatives’, verbeteraar van mensen ‘im prover of people’. The only restriction appears to be that those arguments whose form requirements cannot be met in an NP will not be inherited, as would be expected in a m odular framework, cf. *woner in Amsterdam ‘liver in A m sterdam ’ or *houder van kaas ‘lover of cheese’ (cf. Booij, 1987 for a detailed discussion). Thus, differences with respect to argument in heritance between Dutch and G erm an seem to be due to differences in meaning of the respective ^r-morphemes. 114 G. Fanselow 4.4. Restructuring Before we come to a general conclusion, let us briefly consider the process of restructuring, that is sometimes claimed to be necessary in order to derive a correct semantic representation of some complex words. I do not claim that there might be no morphophonological reasons for restruc turing, since similar processes appear to be necessary for syntactic structures as well when they enter phonetic interpretation. R ather, I propose to show that a number of the arguments that are typically presented in favour of restructuring are quite problematic. To begin with a simple example, consider a word such as Schnell-denker ‘fast thinker*, and let us assume that its morphosyntactic structure is [Schnell [[<fcnfc][er]]].23 If this structure is the input of semantic interpretation, a problem appears to arise since, in the last step of interpretation, we appear to have to combine the meanings ‘thinker, somebody engaged in the activity of thinking* with the meaning ‘fast’ by conjunction. The result would be ‘someone who thinks, and who is fast’, and not the desired representation ‘someone who thinks in a fast manner’. Thus, it seems as if we had to restructure Schnelldenker to [[[SchneH][denk]]er], perhaps by some process of affix movement. This line of reasoning, however, is in conflict with further data. In German, there is no verb *computern, computieren as a base for the noun Computer (both phonetically and semantically identical with its English counterpart). A compound like Schnell-computer ‘fast com puter’ never theless just refers to those computers that compute in a very fast manner, not to computers that are fast in some unspecified way. We have to conclude that schnell modifies just part of the total meaning of Computer. The relevant meaning representation cannot be gained by affix movement or some other sort of restructuring for the simple fact that Computer is an unstructured, non-complex word in German (unless, again, we were willing to return to full-fledged Generative Semantics). The same point can be made for quite a number of A+N combinations in German; the semantic process in question again appears to involve the process of inducing a contextually prominent relation from the meaning of one part o f the compound. I will not go into details here but just stress the fact th at the very same problem with the interpretation of schnell arises in both Schnelldenker and Schnellcomputer, and cannot be handled by restructuring, consequently. The same point seems to be valid for the problem of idiomatic readings posed by words like transformational grammarian. Here, the morphosyn tactic structure appears to be [[itransformational] [jgrammar-ian]], but the idiomatic reading of the compound seems to require [[transformational grammar]ian] as input to semantic interpretation (cf. Williams, 1981). Again, it seems unclear to me how a process of affix movement or any other ' Word Syntax* and Semantic Principles 115 type of restructuring would help with transformational linguist, or German forms such as transformationeller Linguist or transformationeile Schriften ‘transform ational writings’, since Linguist again is a non-derived form and the ‘linguistic rule p a rt’ necessary for the idiomatic sense of transformational is even more deeply embedded in the meaning structure of Schriften. Problems with idiom atic compounds abound,24 so I will not offer a solution here since the m ajor point seems already to have been established: the problem is not restricted to morphologically complex forms, and cannot be solved by morphological restructuring. A nother type o f reasoning applies to examples such as unhappier that were discussed by Pesetsky (1985), among others. The argument that was already cited above obviously presupposes a semantic analysis of the comparative that is not necessarily correct. There appears to be a growing conviction among semanticists (cf. Stechow, 1983) that the basic meaning of any adjective is comparative in nature, i.e., big, e.g., refers to the property of being bigger than some contextually given class of reference object. The comparative would just be a syntactic means allowing for an explicit specification of the class of reference objects. Adding the comparative morpheme would not lead to a change in meaning, it is semantically vacuous. An explicit theory o f the meaning of adjectives thus might show that the problem was just an apparent one. Facts about G erm an (and English) un-prefixation might find a similar account. If un would be a simple morpheme of negation that would map the property of being p to the property of not being p, a problem would arise with argum ent inheritance: treu ‘loyal’ is a two-place-relation, cf. dem König treu ‘loyal to the king’ and so is untreu ‘illoyal’, cf. dem König untreu. Un9 however, does not merely negate the adjective it is attached to, it rather m aps the adjective’s meaning to its contrary. Thus, e.g., unintelligent is more than not being intelligent, it refers to the property of being stupid. While it is unclear how this mapping operation might be form ulated, no problem would arise since a property and its reverse may be assumed to share logical type and argument structure. Indeed, there is some evidence that this recourse to the difference between negation and reverse-formation is correct. Adjectives may also be combined with nicht in a process of word formation, which is interpreted as simple negation. Here, argum ent inheritance is impossible as predicted, compare nicht-treu ‘non-loyal’ with *dem König nicht-treu ‘non-loyal to the king’. 5. CONCLUDING REMARKS If the discussion of section 2 and section 4 is correct, we have to conclude that syntactic facts do not play a role in the regularities about argument 116 G. Fanselow inheritance, semantic restructuring and constraints on productivity in the process of German nominal word formation, and we have also seen how differences between languages might be captured in term s of a semantic theory. Processes and constraints of the conceptual system are quite in dependent of facts of word formation, so it is quite hard to see what a separate theory of word formation could account for in the domains we have discussed. In particular, this is true if language acquisition is taken into account. Only those regularities appear to be irreducible to semantics for which it can be safely assumed that children are able to learn them without the aid of innate knowledge. Obviously, the approach emerging from the preceding paragraphs will have to be tested with further data from German and other laguages as well. E.g., while there do not appear to be any processess in German nominal word formation that would fall out of the dom ain of what has already been discussed, there are some types of verbal and adjectival complex words that appear to pose problems. German possesses a restricted rule of P+V compounding in which the resulting verb appears to inherit the preposition’s argument: an-fliegen ‘at-fly, to fly to ’, nach-schwimmen ‘after- swim, to follow sb. swimmingly’ or durchfahren ‘through-drive’. Composition is, however, restricted to directional preposition, the verb in question must itself allow for a directional complement, and the preposition partially specifies this direction. Consequently, we are not faced with a real instance of argument inheritance in semantic terms, but we are so in syntax: case marking of the NP-complement depends on the governing feature of the preposition, as both zu ‘to ’ and zufliegen ‘fly to sb.’ govern the dative on the one hand, and both an and anfliegen govern the accusative. Note, however, that the approach developed here does by no means preclude formal specifications of a non-head to be relevant for the complex constructions; it is neutral to that question. Consequently, a major task appears to consist in formulating a general semantic process appropriate for the compounds we have just discussed. NOTES * This article is a completely revised version of my paper ‘Word formation and the human conceptual system’, which I presented at the morphology workshop of the XIV. CIPL World Congress at Berlin/GDR (August 1987). I would like to thank the workshop’s participants for helpful critical remarks and suggestions. In particular, I would like to thank Geert Booij, Sascha Felix and Jindrich Toman for their comments on earlier versions of the paper. 1. Cf. Bauer (1983), Fanselow (1981), Motsch (1970), Rohrer (1967), (1974), among others, for some discussion. 2. It is by no means a priori clear how reduplication, Ablaut or infixation might be reduced to (1) (Booij, personal communication, but cf. also Lieber, 1980). The morphosyntactic ' Word Syntax*and Semantic Principles 117 questions connected with this issue do not appear to affect the validity of the arguments to be presented below. 3. Similar concatenative operations are in the reach of children at two-word-stage even before the X-bar-scheme becomes operative (cf. Felix, 1987 for some discussion of the preX-bar syntax). They may also be attributed to apes ‘learning languages’. Cf. Fanselow (1985b) for a more detailed discussion of the relation of word syntax, language acquisition and ‘ape language’. 4. The situation would be more complex if the distinction between words, stems and roots has to be accounted for in terms of word internal levels in the way proposed by Selkirk (1982, 1984). Even if this is the case, however, it would still have to be shown that interesting generalizations are captured by treating the word-root distinction on the same explanatory level as the distinction between words and maximal projections. 5. Descriptive approaches to German word formation like Fleischer (1973) sometimes erroneously consider en to be a derivational affix in structures like weissen derived from the adjective weiss ‘white*, en, however, is nothing but the normal infinitival ending of German verbs; the derived verbal stem is homophonous with the adjective. This is obvious if one considers the declensional paradigm: weiss-e, 1 sg weis-st, 2 sg weiss-t, 3 sg weiss-en, 1 pi weiss-t, 2 pi weiss-en 3 pi 6. Throughout the whole text, German examples have been enriched with hyphens to make their morphosyntactic makeup explicit. None of these hyphens is represented in standard orthography, however. 7. Cf. Fanselow (1985b) for a discussion of possible reasons for this state of affairs, and an analysis of apparent counterexamples. 8. Fanselow (1985b) gives a more detailed discussion of the relation between complex words and 0-theory. 9. Roeper (1987), however, has proposed a syntactically based account of composition and derivation which addresses still further issues in syntactic terms. For reasons of space, I cannot discuss his account here in detail, but some of the problems he deals with will be discussed in section 4. 10. It might also be objected that possible world semantics is not the most prominent candidate for psychological reality. My arguments, however, primarily refer to the structure of possible semantic operations. This overall structure of semantic theory was demonstrated to be independent of the nature of semantic objects by Montague (1970). 11. Cf. Fanselow (1981) for a full formalization of rules of this kind in terms of sorted intensional logic. 12. This set of interpretive option seems to characterize the construction of complex concepts as meanings of simple words as well, as I have tried to demonstrate in Fanselow (1987). 13. It might seem to be more appropriate to induce some relation ‘somehow prominently related to ’ in semantics, and specify the exact contents of this relation in pragmatics. 14. Interpretive processes appear to apply blindly to any kind of structural branching in both phrasal syntax and word syntax. Thus, inducing some prominent relation is not restricted to compounds, but operates in the interpretation of genitive NPs in NP as well. Applicative options differ in detail, however, a fact that has been discussed for the relation between compounds and genitives in some detail by Heidolph (1962). A more significant difference between phrasal semantics and the interpretation of complex words lies in the absence of the rule of ‘quantifying in’ (cf. Montague, 1973, responsible for binding of anaphors and pronominals), a fact that appears to have to be stipulated. 15. Non-affixal derivation may be problematic for this approach, as was pointed out by Booij (personal communication). Even if the existence of such processes calls for a direct 118 G. Fanselow connection between semantic and morphological operations, the points to be made in section 4 will still hold, as will be argued below. 16. There are, however, some apparently lexicalized exceptions such as der Träg-er des Balkens ‘lit: the carrier of the rafter, girder of the rafter’ or der Anlass-er des Motors ‘the starter of the engine’ that seem to be restricted to technical terminology, usually. 17. In a recent paper, Reis (1987) presents a number of examples which she claims refute the predictions made by my approach which I would like to discuss briefly. Reis argues that examples such as Verschrott-er ‘scrap-er’ or Ausschlacht-er ‘cut up-er’ demonstrate that obligatory arguments of a verb need not be filled at all in a nominal derivate. But since a rather sharp contrast between the acceptability of (1) and (2) can be observed, her examples are not really convincing. (1) (2) Heute verschrotte nicht ich, heute verschrottest du mal wieder today scrap not I today scrap you once again i t is not me who will be doing the scraping today, it’s your turn’ ?*Heute hemme nicht ich, heute hemmst du mal wieder today inhibit not I today inhibit you once again Reis also cites data such as (3) which apparently prove that external satisfaction of obligatory arguments is possible with certain verbs. (3) a. b. c. d. e. Stell-er von Fangfragen put>er of catch-questions ‘sb. who poses tricky questions’ Erricht-er von Luftschlössern construct-er of air-castles ‘sb. who builds castles in the air’ Vergiess-er von Krokodilstränen shed-er of crocodile-tears ‘sb. who squeezes a tear” Erreg-er der Krankheit exciter of-the illness der größte Verschmutz-er der Umwelt the greatest polluter of-the environment The vast majority of her examples patterns with (3a-c), i.e., involves quasi-idiomatic NP+V collocations. If the phrases in (3a-c) are exchanged for real argument expressions or even other idiom chunks, the resulting structure worsens significantly: (4) a, ?Stell-er von Anträgen put-er of proposals b. ?*Erricht-er von Brücken construct-er of bridges c. ?*Vergiess-er von Wasser shed-er of water Examples (3a)-(3c) consequently do not involve argument inheritance in a strict sense, but rather fall into the domain of the interaction of word formation with idioms, cf. section 4.4 for some comments, but also note 24. Examples such as (3d) become ungrammatical if Erreger is not used in the specific sense o f‘pathogen’ or with another technical interpretation, cf. (4d): ‘Word Syntax’ and Semantic Principles (4) d. 119 *Erreg-er des Volkes exciter of-the people ‘sb. who infuriates the people* (3d) thus appears to involve a lexicalized nominal item, and appears to be on a par with words such as Verfasser ‘writer’ or Retter ‘saver*. Reis herself does explicitly mention the fact “external argument satisfaction” will have to be accounted for in terms of a semantic approach. Examples such as (3e) appear to resist a direct account, but one might invoke beginning lexicalization here as well. Consider finally examples such as (5) that Reis (1987) cites as counterevidence. (5) Klaras Widmung des Buches an den Ehemann Klara’s dedicating of the book to the husband In these constructions, however, semantically empty affixes are used which are predicted to trigger argument inheritance by my approach, cf. section 4.3 and 4.2. 18. An identification of specific constraints on the interpretation of the N-genitive or NPP relation is consequently necessary. Quite a number of 0-relations are strictly connected to specific formal representations that are inaccessible within NPs (e.g., oblique relations), a fact that will imply part of the 0-hierarchy proposed by Randall (1987). E.g., it is impossible to express directional relations in agentive NPs, a fact that both will block *der Pilot nach New York ‘the pilot to New York* and *der Fahr-er nach New York ‘the driver to New York*. On the other hand, for reasons yet unclear to me, instrumental NPs do accept directional complements, a fact exemplified by both non-complex nouns such as Flugzeug nach New York ‘plane to New York’, Taxi zum Bahnhof ‘taxi to the station’ and derived nominals such as der Flieger nach New York ‘the fly-er to New York, the plane to New York’ or der Transporter nach Frankfurt ‘the transporter to Frankfurt’. Again, the close parallelism between complex and simple nouns implies that no account in terms of restrictions on argument inheritance can be derived here. Considerations of this kind might also be relevant for examples like *der Lieferer des Hotels ‘the deliverer of the hotel, sb. delivering something to the hotel’, cited by Reis (1987). The genitive NP may not be interpreted as the goal argument of delivering that is expressed as a dative in verbal constructions. One might argue that this fact would demonstrate that there is a syntactic correlation between accusative verbal arguments and genitive Ncomplements, but consider der Bote der Armee ‘the messenger of the army’. Here, a goal 0-role is impossible for the genitive complement as well. In general, thus, N+NP does not appear to license a goal relation, a fact that will also block the incorrect reading of *der Lieferer des Hotels. This conclusion might appear to be problematic in the light of examples such as Belieferer des Hotels ‘fumish-er of-the hotel’: beliefern is derived from liefern by a productive process of verbal prefixation which appears to map dative object slots onto accusative object positions. But certainly, the process in question also triggers a semantic change which might be described by converting the goal 0-role to some goal-theme complex. If this is correct, no problem will arise with these examples. 19. The same conclusion can be derived if, as Geert Booij has suggested in personal communication, er is intrinsically connected to some meaning operation applied to the verb, but without an interpretation of its own. The meaning operation has to be well-defined itself, i.e., it must be specified to which type of meaning it applies. The grammaticality of Schläf-er forces this domain of application to be one-place-properties. 20. In a similar sense, we have to assume that the difference between predicative and referential use that characterizes a pair such as to drive - the driving should not be directly encoded in the meaning of nominals and verbs, but should rather be derived from the respective positions phrases occupy in sentential structure. 120 G. Fanselow 21. Let us briefly discuss some apparent or real exceptions to the claim made in the text. At first glance, the difference between German essen ‘eat’ and trinken ‘drink* might imply a specification of the aggregate state of the ingested object, but since both die Suppe essen and die Suppen trinken (eat and drink the soup) are well-formed, the meaning difference appears to reduce to the manner of ingestion. There are a handful of complex words such as rasen-mahen ‘lawn-mow’ or schach-spielen ‘chess-play* that also seem to violate the envisaged restriction on verbal meanings. It is not possible to form these verbs productively, wiesen-mahen ‘meadow-mow*, e.g., is not wellformed. In a sense, again, the relation between noun and verb does not have to be considered to be of the predicate-argument type in the well-formed examples. 22. Again, there are some apparently lexicalized exceptions such as Friih-blUh-er ‘earlyblossom-er’, but no productive pattern appears to be involved, cf. *Langsam-wachs-er ‘slowgrower’ or *Schnell-fall-er ‘fast-faller*. 23. No problems would arise if the word is structured as [[Schnell-denk]-er]]. 24. The most puzzling fact appears to be that composition is possible into completely unmotivated idiomatic expressions. Thus, it is possible to transform the idiomatic expressions of the (a) sentences into structures with nominal compounding internal to the idiom. The noun is compounded to a nominal part of the idiom that does not appear to have any significant contribution to the idiom’s meaning. (1) a. b. (2) a. b. Du bist auf dem falschen Dampfer you are on the wrong steamer ‘you are wrong’ Du bist auf dem falschen Erwartungsdampfer you are on the wrong expectation steamer ‘you have got incorrect expectations’ Es ist höchste Eisenbahn it is highest railway ‘it is high time’ Es ist höchste Abrüstungs-eisenbahn it is highest disarmament-railway ‘it is high time for disarming’ Compounds like these seem to pose the greatest problem for a compositional theory of semantic interpretation. REFERENCES Abraham, W. 1985. “Transitivitätskorrelate” . Groninger Arbeiten zur Germanistischen Lin guistik 26, 1-60. Bauer, L. 1983. English Word Formation. Cambridge: Cambridge University Press. Boase-Beier, J. and J. To man. 1986. “On Ö-role Assignment in German Compounds” . Folia Linguistica 20, 319-339. Booij, G.E. 1986. “Form and Meaning in Morphology: The Case of Dutch ‘Agent Nouns’”. Linguistics 24, 503-518. Booij, G.E. 1987. “The Relation between Inheritance and Argument Linking: Deverbal Nouns in Dutch” . Vrije Universiteit Working Papers in Linguistics 26, Amsterdam. Burzio, L. 1980. Intransitive verbs and Italian Auxiliaries. PhD.-diss., MIT. Carter, R.R. 1975. “Some Constraints on Possible Words” . Semantikos 1,2. ' Word Syntax* and Semantic Principles 121 Chomsky, N. 1970. “ Remarks on Nominalization” . In: R. Jacobs & P. Rosenbaum (eds.), Readings in English Transformational Grammar. Waltham, Mass.: Ginn, 184-221. Chomsky, N. 1980. Rules and Representations. New York.*Columbia. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1986a. Knowledge o f Language. London: Praeger. Chomsky, N. 1986b. Barriers, Cambridge, Mass.: MIT Press. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Fabb, N. 1984. Syntactic Affixation. PhD-diss., MIT. Fanselow, G. 1981. Zur Syntax und Semantik der Nominalkomposition. Tübingen: Niemeyer. Fanselow, G. 1985a. “ W hat Is a Possible Complex Word?*’. In: J. Toman (ed.), Studies in German Grammar. Dordrecht: Foris, 289-318. Fanselow, G. 1985b. “ Zur Stellung der Wortbildung im System kognitiver Module*’. Linguistische Berichte 96, 91-126. Fanselow, G. 1987. “Gemeinsame Prinzipien von Wort- und Phrasensemantik**. In: B. AsbachSchnitker & J. Roggenhofer (eds.), Neuere Forschungen zur Wortbildung und Historiographie der Linguistik. Tübingen: Narr, 177-194. Felix, S.W. 1987. Cognition and Language Growth. Dordrecht: Foris. Fleischer, W. 1973. Wortbildung der Deutschen Gegenwartssprache. Tübingen: Niemeyer3. Handwerker, B. 1985. “ Review of Fanselow (1981)” . Beiträge zur Geschichte der Deutschen Sprache und Literatur 107, 114-117. Heidolph, K.E. 1962. Beziehungen zwischen Kompositum und attributiven Substantiv konstruktionen in der Deutschen Gegenwartssprache. Diss., Humboldt-Univ., Berlin/GDR. Hoeksema, J. 1985. Categorial Morphology. New York: Garland. Hoeksema, J. 1987. “Relating word structure and LF” . Linguistic Inquiry 18, 119-126. Höhle, T. 1982. “ Über Komposition und Derivation: Zur Konstituentenstruktur von Wort bildungsprodukten im Deutschen” . Zeitschrift für Sprachwissenschaft 1, 76-112. Höhle, T. 1985. “ On Composition and Derivation: The Constituent Structure of Secondary Words in German. In: J. Toman (ed.), Studies in German Grammar. Dordrecht: Foris, 119-176. Hornstein, N. 1984. Logic as Grammar. Cambridge, Mass.: MIT Press. Jackendoff, R. 1983. Semantics and Cognition. Cambridge, Mass.: MIT Press. Klein, E. & I. Sag. 1985. “Type-driven Translation” . Linguistics & Philosophy 8, 163-201. Koch, S. 1976. “ Bemerkungen zu er-Nominalisierungen”. Leuvense Bijdragen 65, 69-77. Lees, R. 1960. The Grammar o f English Nominalizations. The Hague: Mouton. Lieber, R. 1980. On the Organization o f the Lexicon. Bloomington, Ind.: IULC. Lieber, R. 1983. “ Argument Linking and Compounds in English” . Linguistic Inquiry 14, 251-286. May, R. 1985. Logical Form: Its Structure and Derivation. Cambridge, Mass.: MIT Press. Montague, R. 1970. “ Universal Grammar” . Theoria 36, 373-398. Montague, R. 1973. “The Proper Treatment of Quantification in Ordinary English” . In: J. Hintikka & J. Moravcsik Sl P. Suppes (eds.), Approaches to Natural Language. Dordrecht: Reidel, 221-242. Moortgat, M. 1983. Synthetic Compounds and Interpretation., Ms., Leyden. Moortgat, M. 1985. “ Functional Composition and Complement Inheritance” . In: Hoppenbrouwers, G. et al. (eds.), Meaning and the Lexicon. Dordrecht: Foris, 39-48. Motsch, W. 1970. “ Analyse von Komposita mit zwei nominalen Elementen. In: M. Bierwisch & K. E. Heidolph (eds.), Progress in linguistics. The Hague: Mouton, 208-223. Pesetsky, D. 1985. “ Morphology and Logical Form” . Linguistic Inquiry 16,193-246. Oh, Y.-O. 1985. Wortsyntax und Semantik der Nominalisierungen im Gegenwartsdeutsch. Konstanz, Gorre. Olsen, S. 1986. Wortbildung im Deutschen. Stuttgart, Kroner. Randall, J. 1987. “Inheritance” . In: W. Wilkins (ed.), to appear, Thematic Relations. New York: Academic Press. 122 G. Fanselow Rappaport, M. 1983. “On the Nature of Derived Nominals” . In: L. Levin, M. Rappaport & A. Zaenen (eds.), Papers in Lexical Functional Grammar. Bloomington, Ind.: IULC, 113-141. Reinhart, T. 1983. Anaphora and Semantic Interpretation. London, Croom Helm. Reis, M. 1987. “ Word Structure and Argument Inheritance: How much is Semantics?” . Ms., Tübingen. To appear in: Linguistische Studien. Roeper, T. 1987. The Syntax of Compound Reference. Ms., Amherst. Rohrer, C. 1967. “Review of Lees (I960)” . Indogermanische Forschungen 71, 161-170. Rohrer, C. 1974. “Some Problems of Word Formation” . In: C. Rohrer & N. Ruwet (eds.), Actes du Colloque Franco-Allemand de Grammaire transformationeile II. Tübingen: Niemeyer, 113-123. Selkirk, E. 1982. The Syntax of Words. Cambridge, Mass.: MIT Press. Selkirk, E. 1984. Phonology and Syntax. Cambridge, Mass.: MIT Press. Stechow, A. v. 1983. “Comparing Semantic Theories of Comparison” . Papiere des SFB 99, Universität Konstanz. Toman, J. 1983. Wortsyntax. Tübingen: Niemeyer. Toman, J. 1987. Issues in the Theory o f Inheritance. Ms., Regensburg. Williams, E. 1981. “On the Notions ‘Lexically Related* and ‘Head of a Word*” . Linguistic Inquiry 12, 245-274.