THE SYNTAX OF ATTRIBUTIVE ADJECTIVES IN ENGLISH AND THE PROBLEMS

THE SYNTAX OF ATTRIBUTIVE ADJECTIVES IN ENGLISH AND THE PROBLEMS

OF INHERITANCE*

Andrew Radford, University of Essex (1989)

In an influential MIT PhD thesis, Abney (1987) has recently proposed an interesting (and largely novel) account of the syntax of attributive adjectivals in English (ibid. chapter 4). The particular aspect of his work which will be the focus of this paper is how Abney deals with the two different uses of premodifying attributive adjectivals illustrated in (1) and (2) below:

(1)(a) It was [a difficult question]

(b) There was [a rather large discrepancy]

(d) I've never seen [a really big spider]

(2)(a) It was [too difficult a question]

(b) I didn't expect [so/as large a discrepancy]

(d) I've never seen [this/that big a spider]

Since the italicised adjectival expression precedes a nominal in the examples in (1), we shall refer to this as the prenominal use of adjectivals; since the adjectival precedes a Determiner in (2), we shall refer to this as the predeterminer use of adjectivals.

Abney proposes an interesting and imaginative analysis of the syntax of these two types of premodifying adjectival. The essence of his analysis is the assumption that in structures such as those bracketed in (3) below (from (1a/2a) above):

(3)(a) a difficult [question] (b) too difficult [a question]

the bracketed nominal expression functions as the complement of the italicised Adjective in both types of structure: the main difference between the two structures is that the (bracketed) complement of the Adjective has the status of an NP (= Noun Phrase) in the first example, and a DP (= Determiner Phrase) in the second. A second difference noted by Abney (1987: 336) is that only in their prenominal use (and not in their predeterminer use) can Adjectives be preceded by a (functional) DEG (= Degree) constituent such as so/as/too/how/this/that – hence the contrast between the grammaticality of (3)(b) above and the ungrammaticality of (4) below:

______________________________________________________________________________

*This is a slightly revised version of a paper entitled 'The Syntax of Attributive Adjectives in English: Abnegating Abney' presented at the Colloquium on Noun Phrase Structure, University of Manchester, 1989. It is to be published in J. Payne (ed.) Empirical Approaches to Language Typology, Mouton, The Hague. I am grateful to Bob Borsley, Andrew Carstairs, Wynn Chao, Annabel Cormack, Dick Hudson, Mike Jones, and Anna Szabolsci for helpful discussion of some of the points raised in this paper.

(4) *a too difficult question

Abney also posits that adjectivals premodified by DEG constituents have the status of APs which function as the complements of DEG, and which combine with DEG to form a DEGP constituent (in much the same way as D combines with NP to form a DP constituent). On the basis of these assumptions, Abney concludes that sequences such as (3)(a) and (b) above have the respective structures (5)(a) and (b) below (Here and elsewhere, we simplify structural representations by omitting single-bar constituents not immediately relevant to the discussion at hand):

(5)(a) [DP [D a] [AP [A difficult] [NP [N question]]]]

(b) [DEGP [DEG too] [AP [A difficult] [DP [D a] [NP [N question]]]]]

Thus, under Abney's analysis the essential differences between the two uses of attributive Adjectives would be that (i) prenominal Adjectives take NP complements, whereas predeterminer Adjectives take DP complements, and (ii) AP has a further projection into DEGP it its predeterminer use, but not in its prenominal use.

Although Abney's analysis provides an interesting account of the dual use of attributive Adjectives, an additional principle is required in order to overcome problems which will otherwise arise under the analysis. One such potential problem (relating to Abney's analysis of prenominal Adjectives) concerns the assumption in (5)(a) that D subcategorises an AP complement. If this is so, then we need to find some way of handling contrasts such as those in (6) below:

(6)(a) He caught [DP a [AP big [NP fish]]]

(b) *He caught [DP a [AP big]]

Given that (on Abney's analysis) the Determiner a takes an AP complement in both structures, we need to explain why (6)(a) is grammatical, but (6)(b) ungrammatical.

Abney's solution to this problem is to posit an Inheritance Principle which we might characterise informally in the following terms (Abney does not attempt any formulation of the principle):

(7) INHERITANCE PRINCIPLE

A matrix phrase inherits the categorial features of the complement of its head, iff the

head f-selects the complement

The key to (7) lies in the distinction between f-selection (= functional selection, a relation between functional modifiers and their complements), and l-selection (= lexical selection, a relation between lexical predicates and their complements). Abney (1987: 54-5) points to two main differences between lexical and functional heads: firstly, functional heads 'select a unique complement' - i.e. they do not permit more than one complement (unlike lexical heads); and secondly, functional heads 'do not describe a distinct object from that described by their complement'. Given these criteria, Determiners like the are functional categories, firstly because the permits only one complement (= NP), and secondly because tomatoes and the tomatoes both describe objects (and so are NPs), whereas eat the tomatoes does not describe an object but rather an action, and so is a VP.

Abney posits that the relation between an attributive Adjective and its nominal (NP or DP) complement is one of f-selection, whereas the relation between a predicative (or postnominal) Adjective and its complement is one of l-selection. One reason for assuming this (as Abney himself notes, 1987: 326) is that it provides a straightforward account of why Adjectives which (when used predicatively) take non-nominal (e.g. prepositional or clausal) complements cannot take such complements in their attributive (prenominal or predeterminer) uses, as illustrated by paradigms such as the following:

(8)(a) She is proud of her son

(b) She is a proud woman

(d) *She is a proud woman of her son

(9)(a) She is so proud of her son

(b) I have never met so proud a woman

(d) *I've never met so proud a woman of her son

In (8)(a), proud is used as a predicate (not as a modifier), and thus l-selects its PP complement of her son: in consequence, the AP proud of her son does not inherit the PP-hood of its complement, but remains an AP. In (8)(b), proud is used as a modifier rather than a predicate, and thus

f-selects its NP complement woman; in consequence, the AP proud woman inherits the NP-hood of its complement woman. Because functional heads 'select a unique complement' (Abney 1987: 54), and because Adjectives always f-select an NP (or DP) complement, it follows that proud when used attributively can only take a nominal complement like woman, and not also a prepositional complement like of her son (hence the ungrammaticality of examples like (8)(c/d)). Data like (9) can be accounted for in an analogous fashion. We might note in passing that Abney also posits that the relation between DEG and its AP complement is likewise one of f-selection: the more general idea underlying Abney's analysis would seem to be that predicates l-select their complements, whereas modifiers f-select their complements.

Having clarified the assumptions which Abney makes in relation to the operation of the Inheritance Principle, we can now turn to see how it would work in the case of structures such as (5). In (5)(a) a difficult question, the Adjective difficult f-selects its NP complement question, with the result that the overall AP difficult question inherits the NP-hood of its complement, so satisfying the subcategorisation requirement for D to have an NP complement. In (5)(b), DEG

f-selects AP, and A likewise f-selects DP: in consequence, the AP difficult a question inherits the DP-hood of its complement a question; in turn, the DEGP too difficult a question inherits the (inherited) DP-hood of its complement difficult a question; the result is that the overall sequence too difficult a question is correctly predicted to have the status (and distribution) of a typical DP.

It should be immediately obvious that Abney's Inheritance Principle provides a solution to the problems posed by data such as (6) above. Given the assumption that prenominal Adjectives

f-select their NP complements, it follows from the INHERITANCE PRINCIPLE (7) that the AP big fish will inherit the NP-hood of its complement fish, so satisfying the subcategorisation requirement for DP to have an NP complement. By contrast, the Adjective big in (6)(b) has no NP complement, so that the AP big cannot become an NP by inheritance, but rather remains an AP – so leading to violation of the requirement for D to subcategorise an NP.

However, intriguing and imaginative though Abney's analysis of the syntax of attributive adjectivals may be, I shall argue in the remainder of this paper that the analysis proves problematic in a number of respects, both from a descriptive viewpoint, and from a theoretical viewpoint. For example, one key descriptive claim which Abney makes is that predeterminer adjectivals have the status of DEGP (and so can be premodified by a Degree expression), whereas prenominal adjectivals have the status of AP (and so cannot be premodified by a Degree expression): this claim is illustrated in (3) and (4) above. However, the claim that prenominal adjectivals cannot take premodifying Degree expressions like too/so/as etc. is clearly falsified by examples such as the following (the bracketed nominal in (10)(g) is ironically taken from Abney 1987: 74):

(10)(a) It was [an all too familiar situation]

(b) It is [a much too important occasion]

(d) I've never had [a quite so unpleasant experience]

(e) There was [an ever so disappointing attendance]

(f) I had to face [a just as difficult situation] myself

(g) The enemy mounted [a nearly as devastating attack]

Given Abney's assumptions, a DP such as that bracketed in (10)(a) would be analysed as having the skeletal structure indicated in (11) below:

(11) [DP [D an] [DEGP all [DEG too] [AP [A familiar] [NP [N situation]]]]]

The crucial point about (11) is that although the Adjective familiar is in prenominal position here, the overall sequence [all too familiar situation] clearly has the status of a DEGP, headed by the DEG too; examples such as (10) thus falsify Abney's claim that prenominal adjectives are always contained within AP, never within DEGP. On the contrary, the obvious conclusion which data such as (10) lead us to is that in both their prenominal and predeterminer uses, Adjectives are projectable into DEGP. In more concrete terms, this would mean that the analysis of prenominal Adjectives in (5)(a) would have to be modified in such a way as to allow for AP to be contained within DEGP. Clearly, if prenominal and predeterminer Adjectives are both analysed as projectable into DEGP, this results in a far more symmetrical analysis of attributive Adjectives. Moreover, it also enables us to posit that there is symmetry between attributive adjectivals on the one hand, and postnominal and predicative adjectivals on the other, since in all of their uses Adjectives will permit premodifying degree expressions, as illustrated in (12) below:

(12)(a) I have never known a patient make quite so rapid a recovery

(b) I have never known a patient make a quite so rapid recovery

(d) I have never known a patient's recovery to be quite so rapid

Given Abney's assumption that so is a DEG word, it is apparent that that all four italicised sequences would (on his analysis) have the status of DEGP constituents. This would mean that in all their major uses, adjectival expressions are always projectable into DEGP, and would clearly provide a more unitary account of the syntax of adjectivals. What remains to be accounted for under our revised analysis, however, is why examples like (10) become marginal or ungrammatical without a premodifier like all/much/far/quite/ever/just/nearly for the Degree word.

Of course, it is perfectly reasonable to argue that the descriptive modification proposed here (to the effect that all Adjectives in all uses are projectable into DEGP) is simply a generalisation of Abney's analysis, and in no sense undermines his essential claim that attributive Adjectives are heads which f-select a following NP or DP complement (with the overall AP inheriting the categorial status of its complement). However, a much more serious descriptive problem for Abney's analysis is posed by degree-modified adjectivals in which the degree expression follows the adjective - i.e. by structures such as those bracketed in the following examples:

(13)(a) You have [a difficult enough life] as it is

(b) You have [difficult enough a life] as it is

If enough here is treated (like too etc.) as a DEG word which f-selects an AP complement, then I can see no plausible way of analysing the NP life in (13)(a) or the DP a life in (13)(b) as the complement of the head Adjective difficult: any such analysis would seemingly require us to posit that the bracketed nominals in (13)(a) and (b) have the respective (much simplified) structures indicated in (14)(a) and (b) below:

(14)(a) [DP a [DEGP [AP difficult enough DEGP] life AP] DP]

(b) [DEGP [AP difficult enough DEGP] a life AP]

However, (14)(a) and (b) are improper bracketings (or, in tree terms, structures with 'crossing branches'), and thus ruled out as ill-formed by principles of Universal Grammar. We can pose the problem in rather different terms by observing that the sequences [difficult...(a) life] are analysed as AP constituents, and yet are discontinuous (and hence, on standard assumptions, cannot be constituents). Examples such as (13) seem to me to deal a descriptive death-blow to Abney's analysis of attributive adjectivals.

An additional descriptive problem posed by Abney's analysis is that of overgeneration. In particular, I see no obvious mechanism by which Abney can generate nominals like that bracketed in (15)(a) below, while blocking those like that in (15)(b):

(15)(a) It was [too difficult a question]

(b) *It was [difficult a question]

Under Abney's analysis, the bracketed nominals in (15)(a) and (b) would have the respective structures indicated in (16)(a) and (b) below:

(16)(a) [DEGP [DEG too] [AP [A difficult] [DP [D a] [NP [N question]]]]]

(b) [AP [A difficult] [DP [D a] [NP [N question]]]]

What Abney needs to be able to say is that only degree-modified Adjectives select DP complements. However, selection is a purely local relation of sisterhood (in that it is a relation between a head and a sister complement contained within the minimal projection of the head): thus, we cannot say that a head Adjective f-selects a DP complement only if the AP headed by the Adjective is itself the (f-selected) complement of a DEG word like too, since this violates the relevant locality condition on selection (by making reference to constituents contained outside the minimal projection of the head, i.e. constituents which are not sisters of the head). In short, I can see no principled mechanism which will license structures such as (16)(a) while blocking those such as (16)(b) (Of course, unprincipled mechanisms could be devised, e.g. to the effect that Adjectives carry an arbitrary subcategorial feature [±ZONK], such that DEG words select an AP headed by a [+ZONK] Adjective, and only [+ZONK] Adjectives select a DP complement: this kind of mechanism is unprincipled in that it involves the postulation of otherwise unmotivated subcategorial features, and involves implicit violation of the putatively universal locality condition on selection).

From a theoretical point of view, the most interesting aspect of Abney's analysis is undoubtedly the INHERITANCE PRINCIPLE (7). This principle would seem to have a number of interesting descriptive spin-offs. For example, it would seem to provide a natural account of the possibility of stacking attributive Adjectives in front of an NP complement, e.g. in structures such as that bracketed in (17) below:

(17) a tall dark handsome stranger

Thus, if an attributive Adjective like handsome f-selects an NP complement like stranger, then the overall AP handsome stranger will inherit the NP-hood of its complement stranger, so that the sequence handsome stranger will become an NP by inheritance. But the resulting NP can then by f-selected by another head Adjective like dark, so resulting in the AP dark handsome stranger: this AP in turn becomes an NP by inheritance from its NP complement handsome stranger. The resultant NP dark handsome stranger can in turn serve as the f-selected complement of another attributive Adjective like tall, so resulting in the AP tall dark handsome stranger, which likewise becomes and NP by inheritance...and so on and so forth. Of course, for this analysis to work, we have to allow selection to be sensitive to the inherited categorial status of constituents.

We might go further and argue that the INHERITANCE PRINCIPLE also provides a natural account of structures such as the following, in which a possessive nominal is 'sandwiched' between two attributive Adjectives:

(18) She bought two green fisherman's chunky sweaters

(We are concerned here only with the most natural interpretation of this structure, on which green modifies sweaters and not fisherman.) Let us assume (e.g. following Fukui 1986) that 's is a head D constituent which f-selects an NP complement. The expression chunky sweaters is an AP which becomes an NP by inheritance from its complement sweaters, and so satisfies the requirement that 's f-selects an NP complement. In consequence of the INHERITANCE PRINCIPLE (7), the DP fisherman's chunky sweaters will become an NP by inheritance from its complement chunky sweaters. But since the expression fisherman's chunky sweaters is an NP by inheritance, it can serve as the f-selected complement of the Adjective green; the overall AP green fisherman's chunky sweaters becomes an NP by inheritance, and can in turn serve as the

f-selected complement of the numeral two, with the overall numeral phrase two green fisherman's chunky sweaters inheriting the NP-hood of its complement. Thus, the possibility of 'sandwiching' a possessive nominal in between two attributive Adjectives seems to be accounted for in a natural fashion under the analysis suggested here.

One interesting consequence of the inheritance analysis is that simple DPs such as that in (19) below:

(19) [DP [D those] [NP pictures of Mary]]

will become NPs by inheritance: given that Determiners f-select their complements, it follows from (7) that they will inherit the NP-hood of their complements, and thus become NPs by inheritance. It is perhaps somewhat ironic that what is widely known as the 'DP analysis' of nominals is actually (because of inheritance) an NP analysis.

Although the inheritance principle seems at first sight to offer a number of apparent descriptive gains, closer reflection shows that it proves problematic in a number of respects. For one thing, the inheritance analysis results in nominal structures containing prenominal Determiners being assigned a different categorial status from those containing pronominal Determiners, as we can illustrate in terms of the bracketed structures in (20) below (It should be noted that Abney argues that personal pronouns have the status of pronominal Determiners):

(20)(a) He chose [DP [D this] [NP one]]

(b) He chose [DP [D it]]

Thus, the DP this one in (20)(a) will inherit the NP-hood of its f-selected complement one, so that the overall sequence this one will become an NP by inheritance. However, the pronominal DP it in (20)(b) has no NP complement (nor does it seem plausible to posit a null NP complement, given that it is never used prenominally), and thus has the status of DP. The result is that this one has the (inherited) status of NP, whereas it has the status of DP. Thus, nominals headed by a prenominal D are assigned a different status from those headed by a pronominal D; and yet, the two sets of nominals have the same distribution and would accordingly seem to have the same categorial status.

The more general nature of the problem posed by structures like (20) is that under the inheritance analysis, structures which have the same external distribution (and would thus conventionally be assumed to have the same categorial status) are assigned to different categories. Another instance of the same problem arises in relation to the two different gerund constructions italicised in (21) below:

(21)(a) [The enemy's constant bombing of the city] frightened us

(b) [The enemy's constantly bombing the city] frightened us

The gerund bombing appears to be nominal in (21)(a) (in that it is premodified by the Adjective constant, and followed by the Preposition of), but verbal in (21)(b) (in that it is premodified by the Adverb constantly, and followed by a 'direct object' complement). Fassi-Fehri (1988) suggests (in relation to parallel Arabic examples) that gerunds like those in (21) are DPs headed by a genitive D which takes an NP complement in the case of nominal gerunds, and a VP complement in the case of verbal gerunds. If this is so, then the two types of gerund would have simplified structures along the lines of those indicated in (22) below (where we follow Fukui (1986) in assuming that 's is a head genitive Determiner):

(22)(a) [DP the enemy [D 's] [NP [N bombing] of the city]]

(b) [DP the enemy [D 's] [VP [V bombing] the city]]

Given these assumptions, the INHERITANCE PRINCIPLE predicts that gerund DPs will inherit the categorial status of their complements, so that nominal gerunds like (22)(a) will become NPs by inheritance, whereas verbal gerunds like (22)(b) will become VPs by inheritance. The obvious problem posed by this claim is that the two types of gerund (nominal and verbal) have the same external distribution (as Abney 1987 argues in detail), in that both occur in typically nominal (case-marked) positions, e.g. as the complement of by in 'I was disturbed by...'. The conventional assumption would therefore be that (because they have the same external distribution) the two types of gerund have the same categorial status – and yet this is not true under the inheritance analysis.

The assumption that a DP of the form [D+NP] becomes an NP by inheritance also proves problematic for any attempt to provide a purely syntactic account of contrasts such as the following:

(23)(a) I like [the blue dress]

(b) *I like [blue the dress]

If (following Abney) we take the as a head Determiner here, then (23)(a) would have the simplified structure (24) below:

(24) [DP [D the] [AP [A blue] [NP [N dress]]]]

The AP blue dress would inherit the NP-hood of its complement dress, so satisfying the requirement for the Determiner the to have an NP complement. Thus, we might say that the inheritance analysis accounts for the well-formedness of nominals like that bracketed in (23)(a). What it does not account for, however, is why the bracketed nominal in (23)(b) is ungrammatical: given familiar assumptions, this would be taken to have the structure (25) below:

(25) [AP [A blue] [DP [D the] [NP [N dress]]]]

Under the inheritance analysis, we should expect DP to inherit the NP-hood of its complement dress, and AP to inherit the (inherited) NP-hood of its complement the dress: thus, there would be every reason to expect that (25) should be a well formed NP; the fact that it is not casts serious doubt on the viability of the inheritance analysis.

Moreover, the INHERITANCE PRINCIPLE (7) makes the (false) prediction that Determiners can be recursively stacked. For, given this principle, a DP headed by a D with an NP complement would inherit the categorial status of its complement, and thus become an NP: but this in turn would make the false prediction that such a DP (by virtue of becoming an NP by inheritance) could function as the complement of another D constituent, so allowing recursive stacking of Determiners; however, stacked Determiner structures are generally ungrammatical in English, as the ill-formedness of structures such as the following illustrates:

(26)(a) *a/the/this/that my car

(b) *his every/this/which hat

The ill-formedness of such structures does not appear to be semantic in nature, since the counterpart of (a) is grammatical in Italian, of (b) in Hungarian, and of (c) in Kikongo. Under a conventional analysis, a structure such as a my car is ungrammatical in English because my car is a DP, and a subcategorises an NP (not a DP) complement. However, under the inheritance analysis, the DP my car would become an NP by inheritance from its NP complement car, and thus be eligible to occur as the complement of a Determiner such as a/the/this/that, etc. We see, then, that the inheritance analysis wrongly predicts that Determiners can freely be 'stacked' in English. It should be apparent that (for analogous reasons) it also wrongly predicts that DEG words can be recursively stacked in front of APs. More generally still, it predicts that all modifiers (i.e. heads which f-select their complements) can be recursively stacked.

As a consequence of this, the inheritance analysis further predicts that not only prenominal Adjectives but also predeterminer Adjectives can be recursively stacked. We can illustrate this in terms of our earlier structure (5)(b) above, repeated in skeletal form as (27) below:

(27) [DEGP too [AP difficult [DP a [NP question]]]]

The essence of Abney's analysis of predeterminer Adjectives is that they f-select a DP complement; recall that in order to account for the fact that the overall DEGP too difficult a question in (27) has the distribution of a DP, Abney has to posit that the AP difficult a question inherits the DP-hood of its complement a question, and that the DEGP too difficult a question likewise inherits the (inherited) DP-hood of its complement difficult a question. But this means that we should expect that the resulting DP too difficult a question will then itself be able to serve as the complement of another predeterminer Adjective: and yet, it is not generally possible to have more than one expression in predeterminer position in such structures, as examples such as the following illustrate:

(28)(a) I have never heard [such a quite so ridiculous allegation]

(b) *I have never heard [quite so ridiculous such an allegation]

Abney's analysis would wrongly predict that examples such as (28)(b) and (c) should be grammatical; clearly, their ill-formedness cannot be ascribed to semantic constraints, given the well-formedness of (28)(a). More generally, Abney's analysis is unable to capture the crucial generalisation that only one adjectival premodifier (of the relevant type) is permitted in predeterminer position in English.

It might of course be objected that the problems posed by structures such as (23-28) are artefacts of the assumption that the selection properties of heads can be satisfied by the inherited categorial status of complements. If we were to reject this assumption, and argue instead that selection is sensitive to the intrinsic (and not inherited) categorial features of complements, we can resolve all the problems posed by such structures in a straightforward fashion. Thus, we might argue that an expression like my car has the intrinsic categorial status of a DP (though the inherited status of an NP), and it is its intrinsic DP status which determines that it cannot be used as the complement of a Determiner like a/the which f-selects an NP complement. However, the problem with this 'solution' is that there are many cases in which it is crucial for Abney that selection be sensitive to the inherited categorial status of complements, and not their intrinsic categorial properties. For example, this assumption is crucial to accounting for data such as (6), (17) and (18) above. We thus reach the paradoxical conclusion that in some structures selection is sensitive to the intrinsic categorial features of complements, whereas in others it is sensitive to their inherited categorial features.

In this situation, it might seem that the obvious proposal to make is that selection can be sensitive to either intrinsic or inherited properties of complements. More specifically, we might posit that if a given head category f-selects a complement of type XP, then this condition is satisfied either if the complement is an XP intrinsically, or if it is an XP by inheritance. However, this solution would be unworkable, since it would lead to massive overgeneration. For example, if an Adjective like difficult subcategorises an NP complement, then the 'dual status' solution would predict that difficult can have as its complement either an intrinsic NP like problems, or an expression which becomes an NP by inheritance - e.g. an NP such as too complex a problem: however this latter prediction is false, as we see from the ungrammaticality of *difficult too complex a problem. Thus, the overall situation is that in some contexts it is the intrinsic not the inherited categorial status of a complement which 'counts' for selection purposes, while in other contexts it is the inherited and not the intrinsic status which 'counts'. Unless we have a principled way of determining in which contexts intrinsic properties 'count' and in which contexts inherited properties 'count', the whole analysis will massively overgenerate.

Thus far, we have concentrated on the descriptive problems which the INHERITANCE PRINCIPLE poses. However, it might be argued that the principle also runs into a number of theoretical problems, in that it falls foul of rather more established principles of Universal Grammar. One such principle which the Inheritance Principle would appear to violate is the Endocentricity Principle, a (weak) version of which is outlined informally in (29) below:

(29) ENDOCENTRICITY PRINCIPLE

All Phrases are endocentric (i.e. properly headed) constructions.

To see how the inheritance analysis leads to violation of the endocentricity requirement, consider Abney's analysis of a structure such as that bracketed in (30)(a) below (to which he would assign the structure (30)(b)):

(30)(a) The black cars show the dirt more than [the white cars]

(b) [DP [D the] [AP [A white] [NP [N cars]]]]

Given the Inheritance Principle (7), the AP white cars will inherit the NP-hood of its complement cars. However, given the Endocentricity Principle, we should expect this to entail that the head white will consequently take on the status of a Noun (since the Endocentricity Principle requires an NP to be headed by a nominal constituent). However, although white can function as a Noun in some uses (and then behaves morphologically like a Noun in taking the Noun plural inflection +s) – cf. e.g.

(31) The blacks mistrust the whites

it clearly has the morphosyntactic status of an Adjective and not a Noun in structures such as (30)(b) – as we see from the fact that it can take the adjectival comparative inflection +er, but not the Noun plural inflection +s: cf.

(32)(a) whiter cars

(b) *whites cars

Thus, morphological facts require us to posit that white has the status of an Adjective in its prenominal use; but given Abney's claim that the overall structure white cars has the status of an NP, the resulting structure will be as in (33) below:

(33) [DP [D the] [NP [A white] [NP [N cars]]]]

However, a structure such as (33) leads to an obvious violation of the Endocentricity Principle if we posit that the sequence white cars is an NP headed by the A white. Thus, the Inheritance Principle is in obvious conflict with the Endocentricity Principle, and hence the adequacy of Abney's analysis must be called into question on theory-internal grounds. Of course, precisely parallel problems will arise if the DP node in (33) becomes an NP by inheritance, since the resulting NP will then be headed by a D (so violating the endocentricity requirement).

There is, however, one way in which we might seek to circumvent the problems posed by structures like (33): namely, we might argue that the NP white cars is indeed an endocentric structure, and that its head is the NP cars. However, this solution would run into problems with a different principle, which we shall refer to as the Projectability Principle, viz.

(34) PROJECTABILITY PRINCIPLE

All word-level categories are projectable into corresponding phrasal categories (e.g. N

into NP, A into AP, D into DP, C into CP, I into IP, etc.)

The fact that the Adjective white in (33) has no phrasal projection into AP leads to an apparent violation of the Projectability Principle (34). The same will be true of the head D in (33), if its containing DP becomes an NP by inheritance.

A further principle of Universal Grammar which would seem to be violated by the Inheritance Principle is the Projection Principle, which is given the following informal characterisation by Chomsky:

(35) PROJECTION PRINCIPLE

'The lexical properties of each lexical item must be preserved at every level of

representation' (Chomsky Language and Problems of Knowledge, 1988a: 75)

We can illustrate this violation in terms of our earlier structure (5)(b) above, [too difficult a question]. Since too is a DEG word, its lexical entry (under Abney's analysis) specifies that it subcategorises an AP complement; and in (5)(b) above, we see that too has as its complement the AP [difficult a question]. However, by operation of the Inheritance Principle, this AP inherits the status of its DP complement a question, and thus becomes a DP. But this means that too no longer takes an AP complement, but rather has a DP complement. Consequently, the lexical requirement for DEG to have an AP complement is no longer met, so leading to a seeming violation of the Projection Principle.

The only apparent way of overcoming these problems of categorial identity would seem to be to allow constituents have a dual categorial status, so that complements would retain their intrinsic categorial specification in addition to their inherited specification. Under this proposal, an expression such as tall girls would have the dual status of an AP{NP}, i.e. an AP which becomes an NP by inheritance. This would mean that the conventional single-valued category specifications of Government and Binding Theory/GB would have to be replaced by two-valued specifications comprising both intrinsic and inherited categorial features. It goes without saying that this would be a radical departure from current GB practice. Nor would this solution resolve the problem (alluded to earlier) of how we 'know' when it is intrinsic categorial properties which are crucial for selection, and when it is inherited properties.

At this point, having pinpointed a number of descriptive and theoretical problems posed by Abney's inheritance analysis of attributive Adjectives, we shall outline an alternative account of the syntax of adnominal adjectives which provides a unitary account of their use in predeterminer, prenominal, and postnominal positions. We shall attempt to formulate the analysis within the general functional framework which Abney assumes, and to preserve as many of the assumptions underlying Abney's analysis as possible (hence e.g. we shall continue to assume that adjectival expressions have the status of DEGP constituents). The analysis we shall propose here is one in which DEGPs are base-generated as NP-adjuncts (and thus serve to expand NP into NP). We shall assume that they can be base-generated to the left or right of the NPs which they modify – in much the same way as a VP-adjunct like completely can be generated to the left or right of VP, as we see from examples such as the following:

(36)(a) The bombs may completely [VP destroy the city]

(b) The bombs may [VP destroy the city] completely

Completely is generally classed as a VP adverbial (cf. e.g. Jackendoff 1972); within the more recent work of Pollock (1988) and Chomsky (1988b), VP adverbials are analysed as adjuncts to VP. Since Adverbs are a subclass of Adjectives (for the reasons given in Radford 1988), then if Adverbs are VP-Adjuncts, it would seem reasonable to posit that adnominal Adjectives are NP-Adjuncts (so leading to parallelism in structure between a VP like completely destroy the city, and a NP such as complete destruction of the city).

The assumption that adnominal Adjectives are NP-Adjuncts provides a straightforward account of alternations such as the following:

(37)(a) I have never seen [a car quite so impressive]

(b) I have never seen [a quite so impressive car]

Under the adjunct analysis proposed here, the bracketed DPs in (37)(a) and (b) would be assigned the respective (skeletal) structures indicated in (38)(a) and (b) below:

(38)(a) [DP a [NP [NP car] [DEGP quite so impressive]]]

(b) [DP a [NP [DEGP quite so impressive] [NP car]]]

Given that adjunction is a recursive operation, the adjunct analysis would obviously account for the fact that both prenominal and postnominal Adjectives can be recursively stacked, as we see from structures like (17) above and (39) below:

(39)(a) We must choose the best person [available] [suitable for the post]

(b) He is the one person [present] [capable of doing it]

What remains to be accounted for under the adjunct analysis is how we derive structures involving predeterminer Adjectives, such as that bracketed in (40) below:

(40) I have never seen [quite so impressive a car]

What we shall suggest here is that structures like (40) are derived by moving the italicised DEGP out of its underlying position as an adjunct to NP into a superficial position as the specifier of a DP headed by the Determiner a. The kind of movement operation involved can be represented in schematic terms as in (41) below:

(41) [DP [D' [D a] [NP [DEGP quite so impressive] [NP car]]]]

-----------------------------------------

Since the operation affects (inter alia) phrases containing words like such/so, we might refer to it as S-MOVEMENT. Such an S-MOVEMENT operation would show numerous parallels with the operation of WH MOVEMENT. For one thing, the two movement operations affect a similar range of constituents, including wh-phrases and degree-phrases: cf. e.g.

(42)(a) [CP How remarkable [C did] [IP the story seem ---]]?

(b) [DP How remarkable [D a] [NP --- story]]!

(43)(a) [CP So remarkable] [C did] [IP the story seem ---]] that it defied credulity

(b) It was [DP so remarkable [D a] [NP --- story]] that it defied credulity

(The elongated dash indicates the position in which the phrase in question is assumed to originate prior to movement into the italicised position.) In both cases, the movement applies only to APs which are premodified by an appropriate operator (e.g. a wh-operator such as how, or a degree-operator such as so). Furthermore, there are obvious parallels in the landing-sites for the moved constituents: in both cases, movement is into the specifier position of a functional category; in both cases, movement is only licensed when the head functional category position is filled by a specific type of item – viz. by a (preposed) Auxiliary like did in the head C position of CP in the case of direct questions (or by a null C in the case of indirect questions), and by the indefinite article a in the case of DP; and in both cases, only a single operator phrase can be preposed (given that DP and CP have only a single specifier position). The apparent parallels between the two different types of movement lend empirical support to the suggestion made in Szabolsci (1987) and Horrocks (1988) that DP and CP are systematically interrelated.

Although we lack the space to mount a detailed defense of our alternative analysis of attributive Adjectives as NP-adjuncts which can undergo movement into the DP-specifier position, suffice it to remark that the proposed analysis provides a straightforward account for data such as (1), (2), (3), (6), (10), (12), (13), (15), and (17) above. Moreover, by eliminating the INHERITANCE PRINCIPLE, our analysis overcomes the problems which beset the principle (discussed in relation to (20-35) above). In addition, the proposed analysis would seem to offer other potential descriptive advantages. There are a number of facts which seem to suggest that adnominal DEG+Adjective sequences form a constituent independent of the following nominal (as would be the case under our analysis, but not under Abney's). Among these are data such as the following:

(44)(a) I have never met a [quite so obnoxious or quite so arrogant] individual

(b) I have never met [quite so obnoxious or quite so arrogant] an individual

(45)(a) He showed [a larger than average capacity for ineptitude]

(b) Linguists are showing [a greater than ever fascination for functional fantasies]

(46)(a) ?So delicate was it a situation that the Prime Minister had to resign

(b) ?How delicate was it a situation?

We shall not discuss the significance of these facts here (since this should be largely self-evident): suffice it to remark that the italicised sequences would be constituents under our proposed analysis, but not under Abney's; in other words, our analysis predicts that examples like (44-46) are grammatical, whereas Abney's predicts that they are ungrammatical.

It would of course be fatuous to claim that our analysis of attributives adjectives as

NP-adjuncts/DP-specifiers solves all the problems associated with adnominal adjectivals. Residual problems which remain under our analysis include the problem of accounting for why structures like (4) are ungrammatical when those like (10) are grammatical; since this is problematic under any analysis (including Abney's), I shall have no more to say about it here. A second problem is accounting for why structures like (8/9)(c/d) are ungrammatical; the answer here may lie in some version of Williams' (1982) HEAD FINAL FILTER. A third problem posed by our analysis is accounting for how it is possible (in structures like (18) above) for a 'descriptive' possessive phrase to be positioned after an attributive Adjective. The 'standard' GB analysis of possessive structures as DPs headed by a genitive Determiner which takes a DP as its specifier and an NP as its complement seems totally inappropriate here. Firstly, the status of 'descriptive' [possessive+nominal] structures seems to be that of NP (not DP), so that an expression such as fisherman's sweaters is an NP which can be premodified by an attributive Adjective, and/or a genitive possessive, as in the president's chunky fisherman's sweaters. Secondly, the supposed 'specifier' in the case of descriptive possessive 's seems to be a Determiner-less NP rather than a DP, since determinate expressions are not permitted - cf. *a tactless the mayor's remark. Thus, 'descriptive' possessive 's seems to be contained within an NP, and to take an NP 'complement' and an NP 'specifier' (whereas 'genitive' possessive 's is contained within a DP, and takes a DP specifier and an NP complement). The exact status of 'descriptive’ 's is something of a mystery to me; it does not seem to be an adjectivalising suffix, since it does not permit the kind of adverbial premodifiers which adjectives formed from nominals typically allow (cf. 'a characteristically childlike expression', but not *'a characteristically child's expression'); it does not seem to be an N, in that it cannot be used pronominally (cf. *'He's wearing a chunky fisherman's'). In the sacred academic tradition of perpetuating problems, I shall leave this question as a 'topic for future research'!

References

Abney, S.P. (1987) The English Noun Phrase in Its Sentential Aspect, unpublished PhD diss.,

MIT

Chomsky, N. (1988a) Language and Problems of Knowledge, MIT press

Chomsky, N. (1988b) 'Some Notes on Economy of Derivation and Representation', ms., MIT

Fassi-Fehri, A. (1988) 'Generalised IP Structure, Case, and VS Word Order', in Fassi-Fehri et al.,

189-221

Fassi-Fehri, A. et al. (1988) (eds) Proceedings of the First International Conference of the

Linguistic Society of Morocco, Editions OKAD, Rabat

Fukui, N. (1986) A Theory of Category Projection and its Applications, unpublished PhD diss,

MIT

Horrocks, G. (1988) 'Movement Rules Within "Noun Phrases"', in Fassi-Fehri et al., 97-105

Jackendoff, R.S. (1972) Semantic Interpretation in Generative Grammar, MIT press

Pollock, J.-Y. (1988) 'Verb Movement, UG, and the Structure of IP', ms., Université de Haute

Bretagne, Rennes

Radford, A. (1988) Transformational Grammar, CUP

Scabolsci, A. (1987) 'Functional Categories in the Noun Phrase', in I. Kenesei (ed.) Approaches

to Hungarian (vol. 2), JATE, Szeged, pp. 167-189

Williams, E. (1982) 'Another Argument that Passive is Transformational', Linguistic Inquiry

13: 160-163