Associative adjectives in English. and the lexicon syntax interface 1 HEINZ J. GIEGERICH. School of Philosophy, Psychology and Language Sciences

Associative adjectives in English and the lexicon syntax interface 1 HEINZ J. GIEGERICH School of Philosophy, Psychology and Language Sciences University of Edinburgh (Received 7 September 2004; revised 15 February 2005) Running heads: recto ENGLISH ASSOCIATIVE ADJECTIVES verso HEINZ J. GIEGERICH

2 This article argues that of English adjective noun constructions involving associative adjectives ( associative AdjNs ), some originate in the lexicon and others in the syntax. While in many cases such constructions are unambiguously and for identifiable reasons located on one or the other side of the lexicon syntax divide, variation being possible only among speakers, a range of associative AdjNs is identified which must be simultaneously, and for the same speakers, of both lexical and syntactic provenance. There is therefore no lexicon syntax divide : the two modules overlap.

3 1. INTRODUCTION English adjectives are held typically to have the following characteristics, among others. In syntactic terms, they are typically pre-modifiers ( attributes ) of a nominal, e.g. a beautiful picture, recursively so, as in a beautiful small picture; or they are predicates (e.g. this picture is beautiful). In both functions they are themselves amenable to pre-modification by degree and other adverbs (e.g. very small, exceptionally beautiful), thus demonstrating their property of being heads of adjective phrases. And they are typically gradable, e.g. smaller, more beautiful etc. In semantic terms, typical adjectives are ascriptive: they denote a property which is valid for the entity instantiated by the noun (Ferris 1993: 24), such that beautiful expresses a property of the picture. And finally, typical adjectives are said to be intersective: a beautiful picture is a member of the intersection of the set of pictures and the set of beautiful objects (Siegel 1980). While these properties (among a few others, here irrelevant see e.g. Pullum & Huddleston 2002: 527 ff.; Baker 2003: 190 ff.) serve to identify prototypical adjectives, many members of the category fail to display one or perhaps several of those aspects of typical behaviour. Thus, dead and alive aren t gradable; nor will they readily permit modification. The latter moreover cannot occur in the attributive position (*an alive animal). Small is not intersective: a small elephant is not a member of a definable set of small objects (especially when compared to a large flea).

4 Such failure of certain members of a linguistic category to conform absolutely to prototypical behaviour is hardly surprising. But if an identifiable and substantial subset of the members of the category absolutely fails to conform to the category s defining characteristics, then we have either cause for alarm or an opportunity to learn something fundamental about the grammar of that linguistic category. One such case is that of the associative adjectives in English. Such adjectives, exemplified for the moment by dental in dental decay, express a property which does not apply directly to the denotation of the head nominal, but rather to some entity associated with it (Pullum & Huddleston 2002: 556; similarly Ferris 1993: 24). Thus, dental does not describe the nature of the decay 2 (as slow or unexpected would, for example) but identifies what is decaying. Such adjectives violate by definition not only the standard adjectival characteristic of ascriptiveness as well as that of intersectiveness but also, as I shall show below, all the other criteria of prototypical adjectival behaviour. Indeed, one might legitimately ask whether, in terms of its semantics, dental is an adjective at all, given its synonymy with tooth in tooth decay. I want to demonstrate below that adjective-plus-noun constructions involving associative adjectives (henceforth associative AdjNs ) have an irreconcilably hybrid status in English: their structural characteristics identify them as objects of the lexicon, while at the same time they may behave as though they were syntactic constructions. Thus, in terms of the familiar (but

5 much-debated) distinction between compound and phrase (see e.g. Bauer 1998, Olsen 2000, Giegerich 2004), dental decay is in many ways a compound just as tooth decay is, bovine disease a compound like cattle disease, Medical Faculty one like Law Faculty. But probably unlike many of their noun-plus-noun counterparts, such construction also have characteristics that are strongly consistent with phrasal status. Indeed, I hope to show in this paper that under clearly-identifiable structural and behavioural conditions, a given associative AdjN will be simultaneously both a compound and a phrase. From this follows a rather interesting point regarding the traditional view of syntax and the lexicon, whereby these two modules are held to be formally distinct. This view cannot be upheld: the behaviour of associative AdjNs suggests that the two modules have at least one significant area of overlap. 2. ASSOCIATIVE ADJECTIVES AND THE LEXICON 2.1 On the lexicon syntax distinction Before presenting my analysis of associative AdjNs, I will spell out some of the assumptions made in this study regarding the nature of the lexicon, in relation to the syntax and otherwise. Different linguists mean different things when they talk about this component of the grammar, but one shared assumption is that the lexicon supplies words complete with lexical features such as transitivity for the syntax to concatenate into larger, phrase-level units. Syntactic units are

6 crucially assumed to be semantically transparent, and to be the outcomes of fully productive operations. If there are exceptions to syntactic patterns, they can be traced back to exception features which typically attach to specific lexical items (see e.g. Pustejovsky 1995). Words have a naming function. Sentences are uttered and forgotten but words are coined and then often retained; and once retained in the speech community, both their meanings as well as their forms are prone to change through time. The lexicon therefore has a dual function in that it is both a repository of words, or more generally of listemes (Di Sciullo & Williams 1987), which may or may not have internal structure, and an active component of the grammar, called the morphology, in which words are assembled from the familiar morphological building blocks by means of, in many cases, fully productive operations. The mechanisms involved in ensuring this duality of function continue to be imperfectly understood; and the distinction between list and productive operation is itself almost certainly an oversimplification. Stratified models of the lexicon (Kiparsky 1982, Giegerich 1999, McMahon 2000) are one way of accounting for, among other things, gradient differentials in analyseability and productivity. Given, then, the existence of fully productive and transparent morphological operations, at least one aspect of what happens in the lexicon strongly resembles syntactic processes. In a stratified model of the lexicon, such syntax-like operations would be sited on the final stratum, which interfaces

7 with the syntax. Given moreover that the category Word is recursive, its members capable of being concatenated not only into syntactic objects (phrases) such as white board but also into compound words (and hence morphological objects) such as whiteboard, a grammar in which the lexicon and the syntax constitute distinct modules has potential for ambiguity regarding the sites on which words are concatenated. Like the noun-plus-noun constructions (NNs) discussed in recent years by, for example, Bauer (1998), Olsen (2000) and Giegerich (2004), productive AdjN patterns are a particularly interesting case in point: are they assembled on clearly distinct sites, in the lexicon and in the syntax, and if so, what is the evidence for the distinctness of the two sites? Or is there just a single assembly site which straddles the so-called divide between the lexicon and the syntax? And, if that is the case, where does that leave our assumption of that divide s existence in formal grammar? The distinction between word and phrase is not only subject to distributional criteria. More importantly, as we shall see, under the lexicalist hypothesis (Chomsky 1970) no syntactic process can manipulate linguistic units below the word level. Words are the atoms of the syntax; their constituent morphemes are amenable to manipulation by morphological ( lexical ) processes but not by syntactic processes. If a given AdjN is lexical that is, a compound then, for example, neither the adjective nor the noun contained therein can attract modifiers.

8 Compare a brilliantly white board and *a brilliantly whiteboard, 3 and observe how a white wall-mounted board allows the phrasal interpretation only. While the set of diagnostics for wordhood connected with the lexicalist hypothesis will loom large in the argument that follows, it is also important to observe beforehand that similar firewall effects are said to maintain the integrity of individual strata within a stratified lexicon. The Bracket Erasure Convention, for example, not only makes the morphological structure of a given word invisible to the syntax; it also makes any stratum-1 complexity of a given form invisible to morphological and phonological processes associated with stratum 2 (Kiparsky 1982; Mohanan 1986: 29ff.; Giegerich 1999: 9, 129; McMahon 2000: 35ff.). There is no reason to believe, therefore, that the divide placed between the lexicon and the syntax under the lexicalist hypothesis is necessarily more robust than that between strata within the lexicon. And since, in English, the stratal divide within the lexicon is indeed straddled by a large number of morphological processes (Giegerich 1999: chapter 2), we may have cause to expect there to be concatenative processes which occur in similar form on both sides of the lexicon syntax divide. I want to begin this discussion by arguing in section 2.2 below that in terms of their structural characteristics, associative AdjNs unequivocally have lexical status on both syntactic and semantic grounds. In section 2.3 I shall explore and adopt the opposite position of according phrasal status to such constructions, before concluding this exercise in dialectics in section 3.

9 2.2 The lexical nature of associative AdjNs 2.2.1 Syntactic considerations An instructive case study regarding the structure of associative AdjPs is provided by the subset of collateral adjectives (Pyles & Algeo 1970: 129; Koshiishi 2002) such as those in (1). (1) bovine tuberculosis equine arthritis feline behaviour phocine distemper vernal equinox dental decay Adjectives of this kind always stand in a recurrent semantic relationship, though (crucially) not in a transparent morphological relationship, to a noun, 4 such that the meaning of the adjective is pertaining to, associated with. Bovine, for example, partners the noun cattle in that the adjective means pertaining to cattle. Similarly, dental means pertaining to teeth. The behaviour of such adjectives suggests very clearly that associative AdjNs originate in the lexicon.

10 In their primary, associative sense, such adjectives do not occur in the predicative position (*this tuberculosis is bovine). They are also without exception non-gradable, and they cannot be modified (*more feline, *incredibly equine). Note that some such adjectives (bovine, feline, equine) have a second, metaphorical or figurative sense under which they are ascriptive and intersective, 5 and not subject to such syntactic restrictions, e.g. John s behaviour was rather bovine. Given that the syntactic unit occupying the predicative position is phrasal John is very clever, John is a genius etc. a possible explanation for the various deficiencies in the behaviour of associative adjectives might be that such adjectives cannot form adjective phrases. As an observation, this is clearly correct; but such behaviour is itself in need of explanation while leaving other aspects of behaviour unaccounted for. It is not only the adjective in such constructions that is not amenable to syntactic modification. Just as in whiteboard (section 2.1 above), both elements of associative AdjNs are subject to the same constraint: *bovine contagious tuberculosis, for example, is ill-formed. In contrast, juvenile cardiac arrest and phocine pneumonic diseases, for example, are well-formed. In an account which treats associative AdjNs as lexical constructions (hence as AdjN N ), this is so because cardiac arrest, pneumonic diseases etc. are themselves associative AdjNs (and hence AdjN N constructions), which can then be heads within associative AdjNs. Given the well-attested recursiveness of better-known

11 compounding constructions (compare e.g. childhood heart disease), the ability of associative adjectives to attach recursively in the lexicon is not really surprising. Note also that in both adjectival and nominal concatenation, the dependents are subject to the same sequencing restrictions: *heart childhood disease, *cardiac juvenile arrest, *pneumonic phocine disease. Another feature of associative AdjNs suggesting lexical origin is the very restricted distribution of some associative adjectives. For many speakers, vernal occurs only with equinox and not, for example, with flowers, cabbage, weather etc.; 6 and while dental collocates with care, decay, treatment, floss etc., it apparently does not go with mug or brush (or indeed fairy). Note moreover that associative adjectives and only those are often interchangeable with nouns (cat behaviour, seal distemper, spring equinox, tooth decay) without of course displaying the morphosyntactic characteristics of nouns (Levi 1978: 37ff.). It will become apparent throughout this paper that associative adjectives are indeed so noun-like semantically that associative AdjNs and certain NN compounds (Giegerich 2004) are virtually identical in many aspects of their behaviour. Suffice it to say for the moment that associative AdjNs are complex nominals in the sense of Levi (1978); they originate in the lexicon rather than in the syntax. If that is the case, then associative adjectives constitute a subclass of adjectives which is available for concatenation in the lexicon only, so as to be dependents of nouns. They are, in that sense, not free forms. 7 This may be

12 due to their hybrid nature, that is, to the category mismatch of adjectival morphology and nominal semantics: it is only in the non-head position that the categorial ambiguity of the adjective does not damage the categorial integrity of the construction. A lexical AdjN is a noun just like a lexical NN is. More importantly, the assumption of lexical provenance accounts at a stroke for such adjectives failure to occur in the predicative position, for their restricted distribution even in the pre-noun position recall *vernal cabbage and for both their own and their heads inability to attract syntactic modifiers, which is then predicted under the lexicalist hypothesis. The observation made above whereby associative adjectives cannot be heads of adjective phrases is merely part of the wider generalisation whereby associative AdjNs are of lexical origin: adjective phrases are not formed in the lexicon but in the syntax. 2.2.2 Semantic considerations I want to argue in this section that in associative AdjNs, the semantic relationship between the adjective and the head noun is essentially unpredictable and in many cases ambiguous, depending crucially on the context as well as speakers encyclopaedic knowledge. In (2) below, containing adjectives which denote persons, at least three relationships are exemplified: object and subject, respectively, of a predicate contained in the noun, in (2a, b), and, in (2c), a more general pertaining to, associated with, similar to the collateral adjectives discussed in section 2.2.1 above. 8

13 (2) (a) papal murder presidential election professorial appointment (b) papal visit presidential lie professorial comment (c) papal emissary presidential plane professorial salary Note, however, that the object vs. subject interpretations imposed on the phrases in (2a, b) are dependent on speakers encyclopaedic knowledge as well as on established usage, such that the pope is unlikely to murder anyone but is known to go visiting. Compare terrorist murder, where the opposite interpretation seems more likely. And presidential election is like parliamentary election an established term following a recurrent pattern. In appropriate contexts, the opposite interpretations are feasible, e.g. the papal murder of the cardinal, the Prime Minister s papal visit, the presidential election of the national security advisor, the parliamentary election of the Speaker. (Presidential lie is unambiguous only because lie is intransitive.) Such phrases raise interesting structural questions: under a lexical analysis of

14 associative AdjNs, the head of the papal murder of the cardinal must be papal murder. But the point is that the interpretations imposed above on (2a, b) are in principle interchangeable; even in (2c), ambiguity is not ruled out. Appropriate contextualisation would allow an alternative interpretation for papal emissary ( the pope s emissary ), namely the emissary who is the pope, as in God s papal emissary. In (2), the range of possible interpretations seems to be constrained by possible syntactic argument structures imposed by deverbal heads (Grimshaw 1990). Similar structures are found in (3a) below, where adjectives again constitute arguments inherited by the head noun from its base verb. When there is no such argument structure, as in (3b), the range of possible interpretations snowballs. (3) (a) electrical supply dental treatment musical criticism (b) electrical engineer electrical shock electrical generator electrical clock musical interlude musical comedy musical composition musical clock dental practitioner dental appointment dental care dental floss medical care medical condition medical doctor medical school

15 As Levi (1978: 52) argues, neither the head noun nor the modifier provide reliable guidance to the meaning of the AdjN as a whole. Electrical enters into a different semantic relationship with engineer than it does with clock, whose semantic relationship with musical is in turn different from that with electrical. The range of possible interpretations of such constructions is determined by what we know about electricity, engineers, music and clocks. It would appear that the morphosyntax provides the basic relationship of associated with in these cases; 9 the specifics for example, that a musical clock makes music while an electrical clock doesn t make electricity but is powered by it constitute encyclopaedic knowledge. Similarly, NNs are known to express highly idiosyncratic semantic relationships, such that the interpretations of olive oil, engine oil, baby oil, hair oil etc. are subject to encyclopaedic knowledge (Giegerich 2004). Such encyclopaedic knowledge is in a sense extra-linguistic, of course; but the fact that it is required for the interpretation of such constructions suggests once again that such constructions originate in the lexicon. I return to the question of semantic relationships between adjectives and head nouns further below. Here I conclude preliminarily that associative AdjNs may express an argument predicate structure inherited from a predicate contained in the noun; or they express the less structured relationship of associated with, to do with (which is then often augmented by encyclopaedic knowledge on the speaker s part, cf. Downing 1988). Both

16 variants of the semantic relationship, again, strongly support the lexical origin of such constructions. 2.2.3 Associative AdjNs and the pro-form one I established in section 2.2.1 that associative AdjNs satisfy the predictions made by the lexicalist hypothesis as far as their recurrent structure is concerned: the components of such AdjNs are unavailable to those syntactic operations that we would expect to be applicable to phrasal constructions. Hence, the adjective is not gradable; and neither the adjective nor the noun is amenable to modification in the syntax. The facts that associative adjectives occur only in AdjNs never in predicative positions and that in many cases they are even confined to specific heads (vernal equinox, *vernal flowers) gave us additional cause to treat associative AdjNs as lexical constructions. Turning from the recurrent structural characteristics of associative AdjNs to the more dynamic aspects of their behaviour, let us attempt to replace (countable) head nouns by the pro-form one. This is one of the tests commonly invoked to verify the lexical status of a given construction under the lexicalist hypothesis (Bauer 1998, Huddleston & Pullum 2002: 554ff.; for detailed discussion see Stirling & Huddleston 2002: 1511ff.). Clear-cut grammaticality judgments are hard to come by, but informal questioning of some ten native speakers of English reveals that even if associative adjectives mostly lack the ability to modify one, as Huddleston & Pullum (2002: 557) observe, there is a

17 substantial number that can readily occur in that position as long as suitable head nouns are chosen. (4) (a) Is this the bovine strain of the disease or the feline one? Do you have a medical appointment or a dental one? Is he a rural policeman or an urban one? Is this a cold-water fish or a tropical one? (b)?do you mean the presidential murder or the papal one??do you mean the parliamentary election or the presidential one??have they made a junior appointment or a professorial one??do you need a back massage or a cardiac one??is this a subject review or an institutional one? (c) *Do you mean the autumnal equinox or the vernal one? *Is this the Home Office or the Foreign one? *Is he a constitutional lawyer or a criminal one? *Is he a theatrical critic or a musical one? *Is he an electrical engineer or an electronic one? *Is this a mental disorder or a nervous one? *Is he a financial advisor or a legal one?

18 Individual readers will disagree with some of these judgments, as indeed informants did among one another: in variance with the majority judgement as indicated in (4), some regarded all the data in (4) as ungrammatical while others accepted all of (4a, b) and found (4c) distinctly odd. But if this division of the data into three subsets grammatical, questionable, ungrammatical is at least roughly accurate then it raises serious questions regarding my analysis of associative AdjNs as lexical entities. Under this analysis, each of the one examples in (4) would constitute an anaphoric island violation, rendering all of (4) ungrammatical. We may take it, then, that there are speakers (namely those who rule out all of (4)) for whom all associative AdjNs are lexical, in line with the analysis presented above. And there are others for whom at least some associative AdjNs are produced in the syntax and hence eligible for pro-one. If the judgements documented in (4) above are reliable then a cardiac one (= massage) is less likely to be acceptable than a tropical one (= fish), for example. I suggest that because cardiac massage has an argument predicate structure which tropical fish lacks, the former is lexical while the latter may not be: tropical fish is a truly attributive construction (albeit one involving an associative adjective). There is another parallel here with NNs: those with an argument predicate structure are lexical (watch-maker, basket-weaving) while attributive constructions such as steel bridge may be syntactic or lexical (Giegerich 2004). If this reasoning is correct and if therefore all argument

19 predicate constructions are lexical, then why might anyone regard the data in (4c) as acceptable? The answer is, I suggest, that all these examples are also interpretable in the more general attributive sense. A cardiac massage is a massage to do with the heart, a papal murder is a murder pertaining in some way to the pope. Under that less specific semantic interpretation and compare a similarly fuzzy analysis of Tory leader in Giegerich (2004: 19f.) the examples in (4b) may be products of the syntax. The examples in (4c), too, are not random. *The vernal one is ruled out because surely vernal equinox, with what for many speakers is probably the sole occurrence of vernal in English, must be lexical. Similarly, Foreign Office must be lexical, but for different and more interesting reasons: foreign has a more common ascriptive interpretation (cf. foreign car) and a highly restricted associative one: Foreign Office, Foreign Secretary, Foreign Service, foreign affairs. In fact, foreign office is ambiguous. Hence, the foreign one is not ungrammatical in isolation (cf. Did you buy the British car or the foreign one?); but it is odd in parallel to an associative construction. Speakers who accept (4a, b) are unlikely to find (4c) worse than just odd. All other examples in (4c) are subject to the same explanation based on the ambiguity of the adjectives in question. For speakers who don t impose a blanket ban on syntactic associative AdjNs, the reason for the oddness of foreign one, criminal one etc. is, then, that the antecedent giving rise to the pro-one construction requires the adjective to

20 be associative, while the pro-one construction itself is ambiguous but amenable to an ascriptive default interpretation. For those speakers, Foreign Office, criminal lawyer etc. must have lexical status if it is correct that non-default interpretations of a given construction must be lexical while the syntax only provides default interpretations (I return to this issue below). If this is so, then clearly syntactic constructions like the foreign one, a criminal one etc. must have the ascriptive default interpretation, which makes them incompatible with their respective antecedents (the Home Office, a constitutional lawyer etc.). 10 Thus, as long as the examples in (4c) are ruled out, the data in (4) require our lexicalist theory of associative AdjNs to be weakened in just one respect. Rather than insisting that they arise in the lexicon, we must allow for the possibility that associative AdjNs containing attributive adjectives (involving the associated with X interpretation) arise in the syntax. We already know, of course, that the default interpretation of adjectives preceding head nouns is attributive. The problem we face is to do with the structural constraints on associative AdjNs discussed in section 2.2.1 above. If these are not accounted for (under the lexicalist hypothesis) by the lexical status of the constructions involved as they aren t if the construction is in fact shown to be of syntactic provenance then how do we explain them? 2.3 The phrasal nature of associative AdjNs: an alternative account

21 I want to show in this section that of the properties of associative AdjNs which I connected with the lexicalist hypothesis in section 2.2.1 above, none is actually unique to the lexicon. Every one of those properties is also available to constructions originating in the syntax. Therefore, if a construction of the kind discussed here meets the criteria for lexical status then that does not enforce the conclusion that the construction actually has lexical status: it is merely eligible for being lexical. I will argue that therefore, associative AdjNs might have an alternative, syntactic origin and constitute phrases. Further below I will reject the term alternative and replace it with something like also or even simultaneously ; but for the moment let us pursue the rather simpler either-or mode of argumentation. 2.3.1 Restriction to non-predicative positions It is a necessary condition for the lexical status of a given AdjN that the adjective involved should be barred from the predicative position, except perhaps when it has a different sense in that position. Hence bovine tuberculosis qualifies for lexical status (*this tuberculosis is bovine) while bovine people does not (these people are rather bovine). This restriction is not a sufficient condition for lexical status, however: there are plenty of AdjNs which do not have predicative equivalents but which not only fail to display other diagnostics of lexical status but actually bear the characteristics of syntactic origin.

22 An adjective s ability to occur in the predicative position is determined by its intersectiveness: this X N is Y Adj is true only for entities which are both X and Y. In some cases, non-intersectiveness has no syntactic effect: for small elephant, for example, the predicative construction is possible because small is assumed to mean small for an elephant, hence this elephant is rather small is grammatical. But in the class of the process-oriented attributives (Pullum & Huddleston 2002: 557; see also Bolinger 1967; Siegel 1980), nonintersectiveness has the expected syntactic effect. Some examples are given in (5a, b); (5c) gives modal attributives (Pullum & Huddleston 2002: 557), which equally are non-intersective and non-predicative. (5) (a) beautiful dancer big eater heavy smoker sound sleeper (b) fast runner slow learner good singer (c) likely benefit certain winner self-styled genius

23 Adjectives such as those in (5a, c) are barred from the alternative predicative construction. In those instances where the predicative construction does occur, it takes on the intersective interpretation, e.g. this dancer is beautiful. The situation is slightly different in (5b) in that the intersective and non-intersective interpretations are not entirely distinct. While a beautiful dancer can be an ugly person, a fast runner is also a fast person hence the acceptability of this runner is fast. And in (5c), *this genius is self-styled is impossible because the person isn t actually a genius. Despite the absence of parallel predicative constructions, the examples in (5) clearly cannot be regarded as lexical: the adjective is gradable and both the adjective and the noun are capable of syntactic modification, e.g. she is a very beautiful dancer, a beautiful Chinese dancer, 11 the biggest eater of them all, highly likely benefits. It is therefore possible for AdjNs that have no predicative equivalent to originate in the syntax. 2.3.2 Gradability and modifiability Some adjectives are not gradable and of those, many are also resistant to modification. This is true not only for associative adjectives. Some ascriptive examples (mostly of absolute adjectives, cf. Pullum & Huddleston 2002: 531) are given in (6). (6) a complete set

24 an essential ingredient a dead animal the correct answer While in these cases gradation is impossible and modification at least rare (e.g. an almost complete set), these adjectives have no restrictions concerning synonymous predicative usage, e.g. this set is complete, this animal is dead etc. It is therefore perfectly possible for AdjNs containing non-gradable adjectives to originate in the syntax. Recall also that in section 2.2.1 above I observed that within associative AdjNs, the head noun can be separately modified only by another associative adjective, e.g. phocine pneumonic diseases vs. *phocine severe diseases. Pneumonic diseases can be equally well embedded within a phrasal construction, as we would expect, e.g. severe pneumonic diseases. And sequencing restrictions such as that ruling out *phocine severe diseases are well-known among entirely syntactic pre-modified AdjNs, e.g. a wealthy German relative vs. *a German wealthy relative (see e.g. Quirk et al. 1985: 437f., Payne & Huddleston 2002: 452 for details). It is simply unusual for a gradable adjective to be preceded by a non-gradable adjective. So, while the sequencing restrictions observed in section 2.2.1 are fully consistent with a lexical analysis of associative AdjNs, they are also known to apply in similar form within unequivocally syntactic constructions.

25 I showed in section 2.2.1 above that the structural characteristics of associative AdjNs are fully consistent (in the sense of the lexicalist hypothesis) with an analysis which treats such constructions as lexical, that is, as compound nouns rather than noun phrases. In the present section I have shown that none of these structural characteristics is in fact unique to lexical constructions, and that indeed every one of these characteristics is also attested among constructions which unequivocally originate in the syntax. This means that at least by those syntactic criteria, an analysis which treats associative AdjNs as (noun) phrases is equally justified. I will show in the next section that semantic criteria for phrasal status are somewhat more restrictive. 2.3.3 Semantic restrictions on phrasal interpretation We saw in section 2.2.3 above that in terms of their behaviour regarding the pro-one construction, associative AdjNs display some variation. Those AdjNs whose semantic relationship is simply one of associated with (e.g. bovine disease, tropical fish) allow pro-one for some speakers; but others resist proone much more strongly. These are, firstly, those where the adjective occurs only with a restricted set of heads recall the extreme case of vernal equinox. Secondly, associative AdjNs where the adjective expresses an argument (for example an object) of a predicate contained in the head noun (e.g. papal murder, cardiac massage) resist pro-one. Finally, associative AdjNs which

26 have an alternative, ascriptive reading (e.g. criminal lawyer, musical critic) also resist pro-one. Let us assume, in line with what was noted in section 1, that in prototypical AdjNs of the kind beautiful picture, the adjective is ascriptive and serves a modifying ( attributive ) function. While an adjective s ascriptiveness or associativeness is a matter of its lexical semantics, its attributiveness arises from the configuration AdjN. In the syntax, ascriptive adjectives in this context are modifiers rather than complements they do not, for example, enter into the argument predicate relationship that can occur in associative AdjNs (e.g. papal murder, cardiac massage). If that is correct then it is not unreasonable to claim that associative adjectives whose semantics is simply of the associated with kind, such as tropical fish and bovine disease, are attributes. Such AdjNs meet the criteria for both lexical and syntactic origin, just as NNs of the type metal bridge, mountain peak do (Giegerich 2004). This would then explain why those forms allow proone for some speakers. Note that even if bovine disease is of syntactic provenance, its distinctness from bovine people is expressed through the associative ascriptive distinction in the adjective s lexical semantics. If that, too, is correct then there is a simple Occam s Razor argument whereby argument predicate AdjNs such as papal murder, cardiac massage must be lexical: the argument predicate relationship is already well established in the lexicon for certain NNs for example, for secondary compounds such as

27 watch-maker, basket-weaving etc. (Giegerich 2004) while establishing it in the syntax for this kind of construction would necessitate the introduction of new mechanisms there. 12 This argument is of course supported by the fact that these forms are resistant to the pro-one construction unless, as I argued above, they are interpreted in a simpler attributive, associated with sense instead. In cases such as professorial appointment, this interpretation may well be the preferred one. This leaves us with the associative AdjNs which have an alternative, ascriptive reading, e.g. criminal lawyer, musical critic; if their resistance to proone is, as I have been assuming, a reliable indicator of lexical status then they too must be lexical. The latter example is of course already accounted for as it exemplifies not only ambiguity but also an argument predicate structure; but criminal lawyer does not have that structure. If it is the case that ascriptiveness is the default interpretation for AdjNs and that associativeness is a more specific and hence non-default interpretation, 13 then the associative AdjN must be generated before the ascriptive alternative is; and if it arises, it will block the latter interpretation. This ordering is predicted by the Elsewhere Condition (Kiparsky 1982, Giegerich 2001). Placing the associative version of criminal lawyer in the lexicon (as is supported by its failure to undergo pro-one) expresses this ordering. 14 Our findings so far may be summarised as follows. Constraints on the internal structure of associative AdjNs suggest that such constructions are

28 lexical and in line with the lexicalist hypothesis. However, their behaviour in the pro-one operation suggests that associative AdjNs whose adjectives are attributive may be of syntactic provenance. Indeed, exactly those syntactic constraints which jointly amount to satisfying the lexicalist hypothesis are also attested individually among clearly phrasal (syntactic) constructions. Thus, AdjNs of the type tropical fish, bovine disease, where the associative adjective has an attributive function, can be analysed equally well as phrases as they can be as compounds. They have this in common with NNs such as metal bridge (Giegerich 2004). What does that mean, though? Are these forms either phrases or compounds, depending on the speaker s individual grammar, or are they both phrase and compound at the same time? Addressing this question is one of two purposes of the next section. The other purpose is to demonstrate that the analysis of associative AdjNs presented here is consistent with an analysis of similar constructions of the type NN presented elsewhere (Giegerich 2004), and especially that the stress behaviour identified for such constructions supports this analysis. 3. ADJNS, NNS AND STRESS The analysis of NNs proposed by Giegerich (2004) draws a basic distinction between two prototype NNs: the attributive construction (e.g. metal bridge), which is typically a product of the syntax but can be lexicalised, and the

29 argument predicate construction (e.g. watch-maker), which can be generated in the lexicon only. Also lexical are primary compounds such as mosquito net, butterfly net, hair net, for whose interpretation the speaker needs to draw on encyclopaedic knowledge (in this instance, regarding the function of the net in question). This analysis is exactly paralleled by that of associative AdjNs presented here. There is a prototypically phrasal variant of this construction, namely one where the associative adjective is attributive (e.g. bovine disease, tropical fish), and a prototypically lexical variant which displays an argument predicate relationship between its elements (e.g. papal murder). In addition there is a large number of AdjNs for whose interpretation encyclopaedic knowledge is required and which therefore must be lexical for reasons of their own, e.g. musical criticism, musical clock, electrical clock. One diagnostic often invoked in the distinction between compound word and phrase is stress. Compounds are said to have fore-stress, phrases are said to have end-stress (see, for example, Bloomfield 1933, Lees 1963, Marchand 1969, Liberman & Sproat 1992). This diagnostic works well in many well-known cases, such as blackbird vs. black bird, whiteboard vs. white board, sweet-corn vs. sweet corn (for more examples see Bauer 2004: 9). However, Giegerich (2004) argues that at least where NNs are concerned the situation is rather more complicated: while phrasal constructions always have end-stress (metal bridge), lexical constructions may have fore-stress or end-

30 stress (London Road vs. London Street, olive oil vs. engine oil). Variation in the stress pattern is common especially among attributive NNs, which can be phrasal or lexical (e.g. olive oil). Again there are noteworthy parallels in the stress behaviour between associative AdjNs and NNs. If we assume, as Giegerich (2004) does, that endstress is available to both lexical constructions and syntactic constructions but (uncontroversially) that fore-stress is available in the lexicon only, then we may expect both end-stress and fore-stress to occur among associative AdjNs. This expectation is met. While it does seem to be the case that fore-stress is somewhat less common among associative AdjNs than it is among NNs (Liberman & Sproat 1992), it is by no means exceptional for associative AdjNs to have fore-stress. I give some examples in (7), mostly from Olsen (2000: 66). (7) polar bear solar system solar panel tidal wave medical profession medical building Medical Faculty electrical worker mental institution mental hospital mental disease pharmaceutical company dental care dental treatment dental appointment circulatory system

31 legal work postal service athletic department financial advisor There will be variation in stress, just as there is among attributive NNs such as olive oil and Bloomfield s (1933: 228) famous example of ice cream; but the point here is that any examples of associative AdjNs with fore-stress, such as those in (7), must be lexical under the accounts presented by a long line of researchers from Marchand (1969) to Liberman & Sproat (1992), Olsen (2000) and Giegerich (2004): the fore-stress pattern simply does not permit otherwise. As we saw above, this is a reasonable prediction by the model for the specific data under discussion here, given especially the idiosyncratic semantic relationships displayed by some of the examples. Now if the grammar is organised in such a way that the lexicon and the syntax constitute distinct modules, with a clear divide between them, then the results obtained so far will prompt us to place the mechanisms which generate associative AdjNs in both modules, such that associative AdjNs which resist the pro-one construction (e.g. papal murder) are lexical while syntactic associative AdjNs show no such resistance (e.g. bovine disease). The assumption of such a divide would moreover serve to predict point-blank that associative AdjNs which are eligible for the pro-one construction, and which are hence indisputably of syntactic origin, cannot have the (equally indisputably) lexical feature of fore-stress. Interestingly, this prediction is wrong.

32 In the sentences below, the attributive AdjNs from which the relevant pro-one forms are derived (dental building, mental hospital, Medical Faculty, financial advisor, dental appointment) are fore-stressed. 15 This requires them to be of lexical origin. Nevertheless, there are native speakers for whom some or all of the sentences in (8) involving the pro-one construction are perfectly acceptable. (8) Is this the medical building or the dental one? Do you have a medical appointment or a dental one? Is this the general hospital or the mental one? Is this the Arts Faculty or the Medical one? Is he a legal advisor or a financial one? This means not only that associative AdjNs can originate variously in the lexicon and in the syntax but also that there are actually individual associative AdjNs (dental building, mental hospital etc.) which are simultaneously lexical entities ( compounds ) in some respects and syntactic entities ( phrases ) in other respects. It follows that the lexicon and the syntax are not separate, distinct modules in the grammar. They overlap. 4. OUTCOMES AND INTERPRETATIONS

33 In this section, I present some comments intended to place both the outcomes and the argumentation of this paper in a broader theoretical perspective. This paper has made three main points. The first of these is that in English, associative AdjNs are in structural terms candidates for lexical ( compound noun ) status on the grounds that they comply with the standard diagnostics connected with the lexicalist hypothesis. If associativeness is taken to be a property of adjectives available to lexical constructions only and I have presented a number of arguments supporting this view then the various aspects of atypical grammatical behaviour of such adjectives are amenable to a single, striking explanation. But I then, secondly, observed that a subset of associative AdjNs cannot be lexical as they are available for the pro-one construction. The pro-one construction is unambiguously the outcome of a syntactic operation: the antecedents of the pro-form one cannot be lexical. 16 I have observed that in this specific subset (e.g. tropical fish), the associative adjective has the straightforward attributive function commonly found in the syntax while other associative AdjNs (e.g. papal murder) display an argument structure strongly suggestive of lexical provenance. Given that the criteria for lexical status invoked earlier indicate mere availability, rather than necessity, of lexical status, I have concluded that associative AdjNs of the tropical fish type are in fact phrasal. Thus, associative AdjNs straddle the lexicon syntax divide. Some are lexical, some are phrasal unsurprisingly so because the very similar NN

34 construction in English is also well-known for exactly that behaviour. Note that this second observation did not damage the novel analysis of associative AdjN behaviour presented initially; it simply prompted a modification of that analysis only for a clearly defined subset of associative AdjNs. Thirdly, I have investigated the stress behaviour of associative AdjNs, treating stress on the first element (e.g. déntal care) as a robust diagnostic for lexical status. This link between initial stress and lexical ( compound ) status is uncontroversial and has been well-known since Marchand s (1969) seminal study of compounding in English. (It is the purported converse link whereby final stress is said unambiguously to signal phrasal status that is controversial; see e.g. Liberman & Sproat 1992 vs. Olsen 2000, Giegerich 2004.) It is in this third part of the argument that a result appears which is of theoretical interest beyond the study of a certain construction type in a particular language. It is clear that in English, certain associative AdjNs are simultaneously phrases in terms of one robust criterion (viz. the pro-one construction) and compounds in terms of another, equally robust criterion (stress), while of course their other behaviour is consistent with either status, as we have seen. This must mean that the lexicon and the syntax or, if this is preferred, the morphology and the syntax have a significant area of overlap. A strictly modular view of the lexicon and the syntax, insisting on the presence of a clear partition between the two components, is on the evidence of this particular construction in English simply not tenable. While there is of course

35 no formal reason why anyone should assume such strict modularity in the first place, we now know that such an assumption is downright wrong at least for a language such as English, where compound and phrase are no longer inflectionally distinct. Moreover, similar overlap has been observed where the purported divide between the strata of the English lexicon is concerned (Giegerich 1999: chapter 3). Many processes of the derivational morphology straddle this divide and indeed display characteristics of both strata. The emerging picture of the organisation of the grammar is then one of overlapping rather than discreet modules, and one where the distinction between the lexicon and the syntax is not necessarily more significant in status than is the stratal distinction within the lexicon. The analytical difficulties and ambiguities presented by the various phenomena connected with the lexicalist hypothesis are not all that different from those connected with the various affix ordering generalisations in Lexical Morphology (Szpyra 1989, Giegerich 1999: chapter 2).

36 REFERENCES Baker, M. C. (2003). Lexical categories: verbs, nouns, and adjectives (Cambridge Studies in Linguistics 102). Cambridge: Cambridge University Press. Bauer, L. (1998). When is a sequence of two nouns a compound in English? English Language and Linguistics 2. 65 86. Bauer, L. (2003). Introducing linguistic morphology (2nd edn.). Edinburgh: Edinburgh University Press. Bauer, L. (2004). Adjectives, compounds and words. Nordic Journal of English Studies 3.1 (special issue, Worlds of words: a tribute to Arne Zettersten). 7 22. Bloomfield, L. (1933). Language. Chicago, IL: Holt, Rinehart and Winston. Bolinger, D. (1967). Adjectives in English: attribution and predication. Lingua 18. 1 34. Booij, G. (1985). Coordination reduction in complex words: the case for prosodic phonology. In van der Hulst, H. & Smith, N. (eds.), Advances in nonlinear phonology. Dordrecht: Foris. 143 160. Carstairs-McCarthy, A. (2002). An introduction to English morphology: words and their structure. Edinburgh: Edinburgh University Press. Chomsky, N. (1970). Remarks on nominalization. In Jacobs, R. A. & Rosenbaum, P. S. (eds.), Readings in English transformational grammar. Waltham, MA: Ginn & Co. 184 221.

37 Di Sciullo, A.-M. & Williams, E. (1987). On the definition of word. Cambridge, MA & London: MIT Press. Downing, P. A. (1988). On the creation and use of English compound nouns. Language 53. 810 842. Fanselow, G. (1981). Zur Syntax und Semantik der Nominalkomposition: ein Versuch praktischer Anwendung der Montague-Grammatik auf die Wortbildung im Deutschen. Tübingen: Max Niemeyer. Ferris, C. (1993). The meaning of syntax: a study of adjectives in English. London & New York: Longman. Fries, P. H. (1998). Post nominal modifiers in the English noun phrase. In Collins, P. & Lee, D. (eds.), The clause in English: in honor of Rodney Huddleston. Amsterdam & Philadelphia: John Benjamins. 93 110. Giegerich, H. J. (1999). Lexical strata in English: morphological causes, phonological effects (Cambridge Studies in Linguistics 89). Cambridge: Cambridge University Press. Giegerich, H. J. (2001). Synonymy blocking and the Elsewhere Condition: lexical morphology and the speaker. Transactions of the Philological Society 99. 65 98. Giegerich, H. J. (2004). Compound or phrase? English noun-plus-noun constructions and the stress criterion. English Language and Linguistics 8. 1 24.