FUNCTIONAL PROJECTIONS OF THE NOUN*
Andrew Radford
University of Essex
This paper is concerned with the syntax of nominal and pronominal constituents of various kinds, focusing particularly on the structure of nominals such as those below which contain (italicised) adnominal 'modifiers' of various kinds (Determiners, Quantifiers and Adjectives):
(1)(a) all the good students of Linguistics
(b) few good students of Linguistics
Within the classic NP analysis of such nominals (as outlined e.g. in Radford 1988), prenominal Adjectives are analysed as N-bar adjuncts, prenominal Quantifiers like few/many and Determiners like the are analysed as NP-specifiers, and predeterminer Quantifiers like all/both are analysed as NP-adjuncts. Thus, (1)(a) and (b) would be assigned the respective structures indicated in (2)(a) and (b) below:
(2)(a) [NP [QP all] [NP [D the] [N' [AP good] [N' students of Linguistics]]]]
(b) [NP [QP few] [N' [AP good] [N' students of Linguistics]]]
The classic NP analysis has the virtue of providing a fairly straightforward account of a number of aspects of the syntax of nominals. For example, if universal Quantifiers like all/both are NP-adjuncts, then since adjuncts are often positioned to the left or right of the expressions they modify (cf. Radford 1988: 255-6), this would provide a natural account of the dual prenominal and postnominal position of the italicised universal Quantifiers in structures such as:
(3)(a) both you and me/you and me both
(b) tot textul/textul tot (Romanian)
all text+the/text+the all (= 'all the text')
Similarly, if both Determiners and prenominal Quantifiers such as few/many function as NP-specifiers, and if we assume that Maximal Projections are binary-branching, we correctly predict that they are generally mutually exclusive: a structure such as *few the remaining hostages would be ill-formed because the overall NP would be a ternary-branching structure comprising the QP few, the D the and the N-bar remaining hostages. What remains to be accounted for under such an analysis is why Quantifiers like many/few can be positioned after Determiners, e.g. in nominals such as the few remaining hostages. One possible solution might be to posit that many/few in such uses function as Adjectives: this would account for the fact that they can precede or follow other Adjectives, and can be used predicatively: cf.
______
*This is the text of a paper presented at a workshop on functional categories at a meeting of the Linguistics Association of Great Britain in York, 16 September 1991. It was published under the same title in Radford, A. (ed) Functional Categories: Their Nature and Acquisition, Dept. Language and Linguistics Occasional Papers 33: 1-25 (1992), University of Essex.
(4)(a) the remaining few hostages/the few remaining hostages
(b) His friends are few; his enemies are many
It might also account for the fact that few triggers 'Auxiliary Inversion' when used as a Quantifier, but not when used as a (seeming) Adjective: cf.
(5)(a) Few of the remaining hostages will they release unharmed
(b) *The few remaining hostages will they release unharmed
As Adjectives, few/many would serve as adjuncts to N-bar, and hence follow Determiners like the.
If we assume that prenominal Adjectives are adjuncts to N-bar, we provide a straightforward account of the fact that Adjectives follow Determiners (cf. 'the good students'/*'good the students'). More specifically, if we assume that Adjectives are potentially iterable adjuncts to N-bar, as in (6) below:
(6) N' = AP* N'
we then predict that Adjectives can be recursively stacked either linearly or hierarchically: linearly stacked Adjectives would have an appositive interpretation, and hierarchically stacked Adjectives would have a restrictive interpretation. This would enable us to provide a structural account of the ambiguity of a string such my first(,) disastrous marriage, along the lines suggested in Radford 1988: 222. The two interpretations would correspond to the two structures in (7) below, with the Adjectives hierarchically stacked in (7)(a), and linearly stacked in (7)(b):
(7)(a) [NP [D my] [N' [AP first] [N' [AP disastrous] [N' marriage]]]]
(b) [NP [D my] [N' [AP first], [AP disastrous] [N' marriage]]]
If we posit that prenominal APs have scope over any nominal they c-command, it follows that disastrous in both structures will have scope over the N-bar marriage, but that the two structures will differ in that first has scope over the N-bar disastrous marriage in (7)(a) (implying that I had other disastrous marriages), but has scope only over the N-bar marriage in (7)(b). Of course, the structural differences here correlate with phonological differences (relating to stress pattern and intonation contours), and these are reflected in the orthography by the presence or absence of a comma between the stacked Adjectives.
Interesting though the classic NP analysis is, there are some aspects of it which seem potentially problematic. One of these is the treatment of prenominal Determiners. In early versions of the 'classic' analysis, these were treated as unprojectable (i.e. zero-level categories which had no phrasal projections into single-bar or double-bar categories). However, such an analysis is objectionable on both theoretical and empirical grounds. From a theoretical point of view, if D is unprojectable, then it is anomalous in that all other categories have phrasal projections (including other functional categories such as C and I, which have phrasal projections into CP and IP within the framework of Chomsky's 1986 Barriers monograph). Moreover, if we assume that Determiners are generated by a PS rule of the form NP = D N', there will be an obvious violation of the constraint proposed by Stowell (1981, p. 70) to the effect that 'Every non-head term in the expansion of a rule must itself be a maximal projection of some category.' Thus, theoretical considerations would lead us to conclude that if Determiners are specifiers of N-bar, they must have the status of DP (Determiner Phrase) constituents.
Empirical considerations lead us to the same conclusion, since we find that
Determiners can be premodified by a variety of expressions which could arguably be analysed as their specifiers: e.g. the italicised constituents in the bracketed NPs below might be argued to function as the specifiers of the bold-printed Determiners:
(8)(a) He made [precisely this point]
(b) Diamonds sell at [several times the price of rubies]
(c) He is [quite the best student] we've ever had
(d) There is [many a slip] twixt cup and lip
(e) [What a fool] I was!
(f) I have never witnessed [so tragic an accident]
(g) He made [rather a mess]
(h) [Not a drop] did I spill
(g) Tu veux [encore une pomme]? (French)
You want [again an apple]? (= Do you want another apple?)
(h) N-a venit [nici o persoana] (Romanian)
Not-has come [not a person] (= 'Not a single person came')
If the italicised constituents do indeed function as the specifiers of the bold-printed Determiners, then it seems clear that Determiners must have phrasal projections (from D into DP), just like other categories. Within the spirit of the NP framework, we might therefore propose (cf. Radford 1988: 263) that Determiner expressions which function as NP-specifiers project into DP, so that nominals such as [several times the price of rubies] or [so tragic an accident] would have the structure (9) below:
(9)(a) NP (b) NP
DP N' DP N'
NP D' N PP AP D' N
several D price of rubies so tragic D accident
times
the an
If we analyse s-phrases (i.e. so/such phrases) as specifiers of D, we can provide a straightforward account of paradigms such as the following:
(10)(a) I have never before witnessed [such a quite so tragic accident]
(b) *I have never before witnessed [such quite so tragic an accident]
(c) *I have never before witnessed [quite so tragic such an accident]
If we assume (as earlier) that maximal projections are binary-branching and that an s-phrase can serve as the specifier of the Determiner a, then it follows that only one of the two s-phrases can occupy the predeterminer DP-specifier position in structures like (10) - in precisely the same way as only only wh-phrase can serve as the specifier of C in English.
So, it would seem that a fairly trivial modification of the traditional NP analysis (viz. treating Determiner expressions as phrasal DP constituents which allow specifiers of their own) will accommodate the fact that Determiners can be 'premodified' by a range of predeterminer expressions. Such an analysis also has obvious theoretical advantages, in that it is no longer necessary to posit that Determiners are unprojectable heads: under the revised analysis, just as N projects into NP, V into VP, I into IP, and C into CP, so too D projects into a DP constituent which functions as the specifier of NP. However, there are both theoretical and empirical problems posed by analysing DPs as specifiers of NPs. From a theoretical point of view, the essential problem is that Determiners are anomalous in that they have no complements of their own - as we see from the fact that D is the sole constituent of D-bar in structures like (9). Thus, DPs fail to conform to the generalised X-bar schema (11) below:
(11) [XP specifier [X' [X head] complement/s]]
What (11) says is that any head category X can combine with a following complement to form an X-bar constituent (which in turn can be combined with a preceding specifier to form an XP). Under the analysis of DPs as NP-specifiers, Determiners are anomalous in that no Determiner ever permits a following complement of any kind - unlike other functional categories (e.g. C takes an IP complement, and I takes a VP complement).
One way in which we might try and overcome this problem would be to suggest that the 'defectiveness' of Determiners in not permitting complements is not restricted to Determiners alone, and that other categories share this property. For example, it is suggested by Jackendoff (1977: 78) that Adverbs are similarly heads which permit specifiers but not complements, and in this respect contrast with Adjectives, which permit both complements and specifiers: cf. e.g.
(12)(a) She smiled, [very confident of success]
(b) She smiled [very confidently]
(c) *She smiled [very confidently of success]
Thus, whereas an Adjective like confident permits both a following complement like of success and a preceding specifier like very, the corresponding Adverb confidently permits a preceding specifier, but no following complement. We might therefore suggest that Determiners are no more anomalous than Adverbs in respect of not permitting complements.
However, the generalisation that Adverbs never permit complements is falsified by examples such as the following:
(13)(a) You must make up your mind independently of me
(b) Are you going to do it differently from me?
(c) She dresses similarly to me
(d) She goes swimming more than me
(e) I'll let you know immediately she arrives
Thus, it is not a general structural property of Adverbial Phrases that they do not license complements, but rather a lexical property of individual Adverbs. By contrast, under the analysis of DPs as NP-specifiers, we are forced to posit either that it is a general structural property of all DPs that they never permit a complement, or that it is an idiosyncratic#lexical property of every individual Determiner in English that (coincidentally) none of them permits a complement. Now, it seems unlikely to be either a general structural property of DPs, or an accidental lexical property of (all) Determiners that they never permit complements, since we do find other uses of certain Determiners in which they permit complements (such as those bold-printed below):
(14)(a) Her behaviour was that of a four-year-old
(b) Those of you who wish to do so may leave
Thus, if DPs are to be analysed as NP-specifiers, we have no principled explanation of the fact that only when used prenominally do Determiners not allow complements. Given that the essential motivation of X-bar syntax is to establish cross-categorial symmetry in the ways in which heads project into phrases, the resulting asymmetry is clearly problematic.
Moreover, the specifier analysis is also questionable on the grounds of its putative descriptive inadequacy, in that the constituent structure which it assigns to DPs is counterintuitive. Thus, it seems counterintuitive to claim that in structures such as (10) above, the strings several times the and so tragic an are constituents. On the contrary, it seems more plausible at an intuitive level to claim that Determiners are closely linked to (and form a constituent with) the expressions which follow them, so that the strings the price of rubies and an accident are constituents of some kind. This kind of 'close relation' between Determiners and the expressions following them seems plausible on phonological grounds, in that the phonological form of a Determiner is often dictated by the initial segment of the first word of the postdeterminer string: for example, a takes the form an and the is homophonous with thee when the initial segment of the postdeterminer expression is a vowel. In Italian, the masculine singular definite article takes the form l' before a vowel (e.g. l'albero 'the tree'), lo before an s+consonant cluster (e.g. lo specchio 'the mirror'), and il before other consonants groups (e.g. il mondo 'the world'). In many languages, this 'close relation' between Determiners and postdeterminer expressions leads to cliticisation of Determiners to the initial word of the postdeterminer expression: e.g. in Yorkshire dialects of English, the definite article procliticises to the following word (cf. 'On Ilkley Moor bar t'hat'), while in Romanian the definite article encliticises to the following word (cf. 'textul', literally 'text+the', i.e. 'the text').