AUTHOR: Cinque, Guglielmo TITLE: The Syntax of Adjectives SUBTITLE: A Comparative Study SERIES TITLE: Linguistic Inquiry Monographs PUBLISHER: MIT Press YEAR: 2010
Denis Bouchard, Département de linguistique, Université du Québec à Montréal
The book is a revised analysis of DP-internal word order in Romance and Germanic, and the interpretation of the adjectives. Cinque's original 1994 analysis was very influential on a way to approach data that has crystallized into the ''Cartography Project'', so it is interesting to see what progress has been accomplished on that front to further our understanding of language.
Chapter 1 presents two empirical problems that have been pointed out for the 1994 analysis based on N Movement: postnominal adjectives in Romance must appear in the mirror order of the corresponding adjectives in prenominal position in Germanic (1) (my examples, DB), and a postnominal ADJ can take scope over a prenominal one (2) (Cinque’s example (9), p. 4).
(1) a. a [round [Chinese [table]]] b. un [[[tavolo] cinese] rotondo]]
(2) E' una giovane promessa sicura. He-is a young promise sure 'He is a sure young promise'
These facts are unexpected in an analysis where adjectives enter the structure of DPs as phrasal specifiers of functional projections that are hierarchically merged in such a way that an AP has scope over the APs that follow it.
Chapter 2 reviews a number of semantic distinctions that fall into two sets (using the terminology of Sproat & Shih 1988):
Indirect modification: stage level, restrictive, implicit RC reading, intersective, relative to a comparison class, comparative (superlative), epistemic, discourse anaphoric.
The positions where ADJs can appear with these readings differ systematically between Germanic and Romance:
(3) Germanic: IndirectModifAP - DirectModifAP - N - IndirectModifAP Romance: DirectModifAP - N - DirectModifAP - IndirectModifAP
Chapter 3 presents the core of the analysis. First, DP-internal adjectives are generated in two different ways: either as direct modifiers in the specifiers of functional heads dedicated to particular semantic classes of adjectives, as in the 1994 analysis, or as predicates of reduced relative clauses (RRC) that are indirect modifiers. RRCs are merged above all the functional projections hosting DirectModifAPs. The second important point is that Cinque retains the assumption that ''there is only one order/structure available for all languages'' (p. 39). The hierarchical positions he attributes to the various FPs hosting the APs derives the following universal basic order of DP-internal APs (for reasons of space, I show only some of the semantic classes of ADJs):
Restrictive RCs > complex AP RRCs > bare AP RRCs > APsize > APcolor > APnationality > NP
This basic order is assumed to be rigid across languages: this is the ''natural'' order for these APs (p. 58). To account for the surface orders that depart from the basic order, Cinque proposes various phrasal movements, which are detailed in chapters 5 and 6.
In Chapter 4, Cinque assumes that DirectModifAPs are functional because they constitute a closed class in many languages. Moreover, despite the fact that they do not take complements when in prenominal position in Germanic and Romance, they are nevertheless phrasal. He gives as evidence examples from Bulgarian and Greek where ADJs are followed by adjuncts in prenominal position (but at least in the Greek examples, the NP is articulated, so this is an IndirectModifAP, not a DirectModifAP). ''Languages simply differ as to whether they allow a complement or adjunct to follow the adjective'' (p. 46). Moreover, we need two sources for ADJs because if we try to derive DirectModifAPs from RCs, ''we are forced to posit more and more complex derivations from sources that differ more and more from one another'' (p. 51).
Chapter 5 gives the analysis of English (Germanic). Germanic manifests the ''natural'' order of DirectModifAPs. Though violations of this order are possible, they are marked: for instance, the normal order is that size ADJs precede color ADJs, but their order can be reversed by introducing the color ADJ as an RRC. Since this source is only available for predicative ADJs, nonpredicative ADJs are always rigidly ordered (*He is an eléctrical old engineer; cf. an old electrical engineer). The class of bare ADJs that can appear postnominally in English is very restricted, even those that have a predicative use. He suggests that there are two kinds of RRCs, complex and bare AP RRCs. The few postnominal bare ADJs are introduced as complex RRCs: these are postnominal because the whole extended NP domain below them obligatorily raises above them (and also above Finite restrictive RCs). ADJs with complements or adjuncts are postnominal because they can only be introduced by this complex RRC source. The bare AP RRCs are part of the extended NP domain that must raise above the other RCs, so they remain prenominal.
Chapter 6 gives the analysis of Italian (Romance): ''the order of Merge is not immediately visible due to the intervention of (various) movements, which in some cases are obligatory and in others optional'' (p. 69). First, since IndirectModifAPs necessarily follow N and any postnominal DirectModifAPs, this implies that the extended NP with the DirectModifAPs raises not only above bare RRCs (as in Germanic) but complex ones as well. Second, internally to the direct modification domain of NP, the NP raises obligatorily above ADJs of nationality, ''but appears to raise above higher adjectives (of color, shape, size, value, etc.) only optionally'' (p. 71). This difference in obligatoriness ''remains to be understood'' (p. 72). Since the order of postnominal DirectModifAPs in Romance is the mirror image of the Germanic prenominal order, the raising of NP is of the ''roll-up'' kind: so for instance in (1), 'tavolo' raises above 'cinese', and then 'tavolo cinese' raises above 'rotondo'. When APs appear in the reverse order of what these movements predict, the ''more special order'' is ''the result of merging the rightmost adjective higher up as a reduced relative clause (which eventually ends up postnominally'' (p. 74). Complements and adjuncts of the NP are stranded at the end of the DP, because (following Kayne 2000) prepositions ''attract their 'complements,' and force (in VO languages) the entire remnant to raise to their left, which makes them final in the DP'' (p. 79).
Chapter 7 briefly presents three differences between Germanic and Romance that relate to the position of DirectModifAPs, concerning referential epithets, idiomatic readings and special cases of agreement with a series of coordinated nouns.
A conclusion summarizes the main points of the analysis, and an Appendix presents additional evidence for the dual source of adnominal adjectives from various languages.
The book presents some data in a new light, which is stimulating. It also takes into account some empirical problems that had been pointed out with the author's previous analysis. Note that the mirror order in Romance in (1) and the scope effects in (2) are not actually captured by the switch from Move N to Move NP, but by adding that pied-piping (roll-up) is obligatory. This could just as well take place with N movement.
More crucially however, Cinque barely addresses the conceptual problems that have been raised for this kind of account. Though he regularly claims the superiority of a movement analysis over base generation, there are so many key elements that are left undefined, in the end there is no analysis in the present state of the proposal. Consider the following points.
(i) The scope of DirectModifAPs with respect to one another. To complete the analysis, we have to know the subcategorization that determines the hierarchical embedding of the functional categories: why the color FP is the complement of the size FP, and so on. Cinque seems to have given up on this hopeless task, and states that the order of DirectModifAPs is part of UG: it reflects their ''natural'' positions, due to some unknown principle (''whatever that turns out to be'' (p. 38)). There is also the problematic assumption that there are FPs that redundantly replicate the classes of ADJs.
(ii) IndirectModifAPs have wider scope than DirectModifAPs. The positions of bare and complex RRCs are also determined by some unidentified UG principle. The only defining property of these RRCs is semantic: they are ADJs interpreted predicatively. So the claim is simply that ADJs used predicatively have wider scope. The fact that their wide scope is directly reflected in their being further from the noun than DirectModifAPs follows in base generation but is an accident for movement: the movements conspire to give that result, but we are not told what triggers the movements, what the empty FP landing sites are, or what determines their position.
(iii) The direct modification use of an ADJ is less marked than its indirect modification use. The ''natural'' order [size > color] can be reversed as in (4) by using a RRC source for 'brown' ''if one wishes to distinguish different groups of individuals of the same size'' (note 3, p. 131):
(4) I've shown you my black small dogs, now, these are my two brown small dogs.
This appears to be a distinction made on the basis of pragmatic markedness, the direct modification source of 'brown' being less marked than the RRC source. So the ''natural'' order is the one that is less marked pragmatically, for some unknown reason. This assumption of markedness is crucial, otherwise a RRC source for 'Chinese' and 'cinese' could produce a reversed ''natural'' order in (1).
(iv) Surface orders often differ from the basic order. These orders are due to obligatory or optional movements, which vary across languages for some unknown reasons.
In short, the proposal is that an unknown principle accounts for a universal basic order of APs, that movements change this order diversely in each language due to unknown triggers, that the movements are sometimes obligatory, sometimes optional for unknown reasons, and the moved elements end up in landing sites of unknown category (''FP'') with unknown principles accounting for the positions of these FPs. Given the tools that are used, it is possible to match any order of ADJs with any scope. This is similar to the use of 'predicate raising' in Generative Semantics, which for example can make an appropriate constituent in (brother of [(mother or father) of Bill]) for 'uncle' to be inserted in its place. As Chomsky (1972: 79) comments, ''Such a device will always be available, so that the hypothesis that Q is a constituent has little empirical content.''
Compare this syntactic analysis with accounts based on semantic, cognitive, pragmatic principles. Cinque, without discussing any, says that ''none seems entirely convincing'' (p. 122). Let's look at one and let the readers make their own decision. Bouchard (2005), following up on Bouchard (2002) and furthering ideas from Ziff (1960), Vendler (1968), Sproat & Shih (1988), Krifka (1995), proposes the following principle:
(5) General principle of seriation of adjectives: The more the property expressed by an adjective makes it likely to form with the noun a relevant and usual Concept, the more this adjective tends to be close to the noun, i.e., to modify the noun more directly than another adjective [my translation from French].
The notion of 'Concept' is used in the sense of Krifka (1995). Concepts are similar to Kinds, but contrary to natural kinds, Concepts need not be well established, and are instead construed from scratch. If the context is appropriate, numerous classes can be considered sufficiently stable for the speaker to present them as forming a Concept. So for instance, ADJs of Provenance tend to be closer to the noun than ADJs of size:
(6) a. enormous Chinese vase; #Chinese enormous vase b. vase chinois énorme; #vase énorme chinois
This is expected according to (5): Provenance is a property that is ''taxonomic'' (Sproat & Shih 1988), i.e., often useful for identifying elements. The strength of the tendencies depends on cognitive and contextual factors. So despite the fact that these two ADJs can have both a direct modification and a RRC source, their order is not free, because they differ strongly in taxonomicity. Cinque himself appeals to pragmatic taxonomicity: ''In its classificatory usage, a color adjective will appear closer to the N than a nationality/provenance adjective, thus apparently contradicting what was taken [...] to be the canonical order color > nationality > N. See la mele verde canadese (è la mela più buona in assoluto)'' (note 2, p. 140).
Cinque objects to a preferential approach because the order and scope of the adjectives appears to be rigid in some cases, such as 'an old electrical engineer'. But in the context of engineering, 'electrical' is highly taxonomic: there are university departments and professional orders that are based on it. It is therefore no surprise that the order/scope is quite rigid (though not absolute: the reverse order could be used jokingly). The reverse order has a meaning, but it does not make sense in the taxonomic system that is commonly shared. Moreover, if we look at ADJs that are clearly direct modifiers (because they are prenominal) but less taxonomic, their ordering is quite free:
(7) a. un supposé nouveau miracle = new miracle which is alleged to have taken place b. nouveau supposé miracle = new type of alleged miracle c. bon futur président = good as a candidate for the presidency d. futur bon président = someone who will be a good president
This is unexpected under Cinque's (yet unknown) principle (though he could say that these ADJs ''belong to two (or more) of the classes of adjectives that can occur prenominally'' (p.118), a doublet strategy that nullifies the claim of rigid ordering; cf. Siegel 1980).
Principle (4) also accounts for the wide scope and fairly rigid placement of evaluative ADJs (bon/good, beau/nice) because the properties they express are quite subjective, hence variable, so it is unnatural to use them as the basis of a taxonomy, which requires some stability.
Cinque feels ''that a system that minimizes base generation (to one structure/order for all languages), deriving all other structures/orders via independently needed types of movement, is more interesting'' (note 20, p.126), as in Kayne's (1994) LCA approach. But a system with a simple head parameter also has only one structure for examples as in (1): merge has applied in the same way in the two languages, but the linearization follows different patterns. The linear order of ADJs directly reflects their order of merge. This dual possibility of ordering follows naturally from design properties of our SM apparatus: two constituents A and B with vocal content can be linearized as AB or BA, and as in other arbitrary choices deriving from design properties, languages must choose and conventionalize this choice (Saussure 1916). Since by design there are two possible orderings of any structure obtained by merge, the single order system must add a costly condition that excludes one, which is exactly what the LCA does by stipulating that there is an abstract terminal 'a' that precedes all the other terminals: so in this model a crucial element for linear ordering at the SM interface has no SM properties (see Bouchard 2002: 378ff for other problems with a universal basic order. Celtic languages appear to raise empirical problems for base generation; however, the facts are far from straightforward; cf. Bouchard 2002: 191-193).
Finally, Cinque claims that in examples like (8), the direct modification reading of the prenominal ADJ is exactly the same reading postnominally, and that this is a problem for Bouchard (2002).
(8) a. le verdi colline della Toscana b. le colline verdi della Toscana
He attributes this to the optional movement of the NP across the ADJ (this optionality ''remains to be understood'' p. 72). He notes that (8a) is slightly different and has a 'poetic' flavor, but gives no indication why. As pointed out by Bouchard (2002: 95ff) and Sadler & Arnold (1994), these examples may have the same meaning in terms of truth conditions, but truth-conditional semantics is not subtle enough to capture the distinction here. In short, the combination N+ADJ defines two sets, each with their property, and a fortuitous, non-necessary intersection of these properties, whereas the combination ADJ+N involves two properties presented as forming a single complex property. The difference between the two interpretations is brought out by three tests: in a context of comparison, in D- and non-D-linked questions, and in a context of negation. The prenominal ADJ has a 'poetic' flavor because ''Something which is not usually considered as a natural kind in the shared knowledge of a community of speakers is introduced in a grammatical construction which says it is'' (Bouchard 2002: 105). It is disheartening that the author missed that point.
Recall the original motivation for transformational grammar: it provided simpler analyses for what seemed to require growing complications in a Phrase Structure Grammar. We seem to be in a somewhat reverse situation here, the movement analysis being forced to posit more and more complex derivations with crucial points that ''remain to be understood'' at every step. Cinque's book demonstrates indirectly that this kind of movement analysis is no match for a well-defined analysis of ADJs built on base-generation. The problem becomes critical if the analysis relies on ''natural positions'': ''as long as they're going to their natural places, there's no science'' (Chomsky 2010).
Bouchard, Denis (2002). Adjectives, Number and Interfaces: Why Languages Vary. Oxford: Elsevier Science.
Bouchard, Denis (2005). Sériation des adjectifs dans le SN et formation de Concepts, Recherches linguistiques de Vincennes 34: 125-142.
Chomsky, Noam (1972). Studies on Semantics in Generative Grammar. Mouton, The Hague.
Chomsky, Noam (2010). Poverty of Stimulus: Unfinished Business. Conference paper presented in Paris, May 29th, 2010.
Cinque, Guglielmo (1994). On the Evidence for Partial N Movement in the Romance DP. In: Paths Towards Universal Grammar (Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi and Raffaella Zanuttini, eds.), 85-110. Georgetown University Press, Georgetown.
Kayne, Richard (1994). The Antisymmetry of Syntax. MIT Press, Cambridge.
Kayne, Richard (2000). A note on prepositions, complementizers, and word order universals. In R. Kayne, Parameters and Universals, 314-326. Oxford University Press, Oxford.
Krifka, Manfred (1995). Common Nouns: a contrastive analysis of Chinese and English. In Carlson, G. & Pelletier, F.J. (eds.), The Generic Book: 398-411. Chicago: The University of Chicago Press.
Sadler, Louisa and Douglas Arnold (1994). Prenominal adjectives and the phrasal/lexical distinction. Journal of Linguistics, 30, 187-226.
Saussure, Ferdinand de (1916). Cours de Linguistique Générale. Payot, Paris.
Sproat, Richard and Chilin Shih (1988). Prenominal Adjectival Ordering in English and Mandarin. Proceedings of NELS, 18, 465-489.
Vendler, Zeno (1968). Adjectives and Nominalizations. Mouton, The Hague.
Ziff, Paul. 1960. Semantic analysis. Cornell University Press, Ithaca, N.Y.
ABOUT THE REVIEWER
ABOUT THE REVIEWER:
Denis Bouchard is professor in the département de linguistique at the
Université du Québec à Montréal. His interests include linguistic theory
with regards to principled explanation, comparative syntax, and the
syntax/semantics interface. He is currently working on a book about the
origins of language.