This book presents a new theory of grammatical categories - the Universal Spine Hypothesis - and reinforces generative notions of Universal Grammar while accommodating insights from linguistic typology.
Date: Fri, 22 Aug 2003 13:17:52 +0200 From: Georges Rebuschi <email@example.com> Subject: Representation Theory
Williams, Edwin (2003) Representation Theory, MIT Press, Current Studies in Linguistics.
Georges Rebuschi, ILPGA, Sorbonne Nouvelle (Université de Paris III)
Both the VP Internal Subject Hypothesis and Larsonian shells, in which a small "v" takes the agentive NP as its specifier and a VP as its structural complement, illustrate the asymmetric c-command of the Theme by the Agent; the functional layers above (whether IP, AgrSP or TP), display the same asymmetry; finally, whether only one Wh-Phrase moves to the left periphery or several do, the "subject" wh- P will also asymmetrically c-command the oblique one. In the author's view, such shape preservation cannot be coincidental, and stipulating "equidistance" without even defining "distance" to evade the fact that intersecting dependencies are created by the movements of the arguments from within VP/vP to IP is at best missing what Williams takes to be the central fact about natural language sentences, namely, that they should not be described as one structure, but as a series of structures which, in the unmarked cases, entertain this shape preservation or isomorphism, dubbed "representation". _Representation Theory_ (RT) is thus a programmatic book that endeavours to develop a new grammatical model, in which each level -- Theta Structure or TS, Case Structure: CS, Surface Structure: SS, Focus Structure :FS, etc.) "represents" the preceding one, distortions between them being governed by one of several factors: - the independent requirements of the representing structure itself (e.g. Hungarian FS requires that the focused phrase precede the finite verb, independently of its theta-role and case); - "blocking" (cf. Williams 1997): any more specific structure blocks or bleeds less specific ones; - another functional factor, which amounts to saying that if a mismatch between levels Ln and Ln+1 renders the next mapping, from Ln+1 to Ln+2, isomorphic, it is tolerated. Note besides that intra-level movement is possible, thereby allowing the word order at Ln-1 to differ from that at Ln+1 without any misrepresentation happening.
Chapter 1, 'Economy as Shape Conservation,' first illustrates the basic tenets of RT and its way of dealing with distortions in the domain of derivational morphology and compounding. Bracketing paradoxes (cf. the ambiguity of (1a), vs. its absence in (1b) below) are functionally determined: "[x[y z]] means [[x y] z] only if [x[y z]] is not generable", which is precisely not the case when a relative clause does the job, as in (1b).
(1) a a beautiful dancer b a person who dances beautifully
Next, the domain of theta/case relations is examined, and Exceptional Case Marking is analyzed as another type of bracketing paradox, since the non-isomorphic Case Structure (CS) [[believe Mary] to be alive] is the "most isomorphic structure that satisfies the strictures of [that] level" wrt. the TS [believe [Mary to be alive]]. Finally, Holmberg's Generalization (the fact that, in Scandinavian languages, object shift must be accompanied by verb movement) is argued to be more satisfactorily accounted for by "a constraint on mapping one representation into another, than as a constraint on the coordinated movements, within a single tree, of the items it pertains to" or by massive remnant movement.
Chapter 2, 'Topic and Focus in RT,' introduces further levels: FS, SS, and Quantification Structure or QS (which also takes care of Topics): SS "represents" QS, and FS "represents" SS. Heavy NP Shift (traditionally viewed as rightward movement of the object NP over a PP) is analyzed as an instance of (short) scrambling, i.e. a mismatch between CS and SS, which is "tolerated because of the SS,FS match". Comparing the following sentences, Williams note that (2c) is only felicitous in a corrective context, a question taken up in more detail in chapter 9.
(2) a John gave to Mary [all the money in the SATCHEL] b John gave [all the money in the SATCHEL] to Mary c *John gave to MARY [all the money in the satchel]
Once QS has been introduced in the model, a fundamental dimension of cross-linguistic variation is provided, because conflicts that arise between the requirements of the various levels may be resolved differently. Thus, Williams proposes that, in German, isomorphism between SS and QS ranks higher than isomorphism between SS and CS, whilst the reverse holds for English, whence the existence of more cases of scopal ambiguities in the latter than in the former. Hungarian and Spanish focusing word order is also tackled, and so is Russian word order with quirky subjects.
In Chapter 3, 'Embedding' is dealt with. According to the general architecture of RT, the Level Embedding Conjecture says that each clause type is embedded at the very level at which it is defined. Thus, clause union at TS results in serial verb structures, clause union at CS, in infinitive complementation; indirect questions governed by bridge-verbs are only embedded at SS, the level at which wh-movement is hypothesized to take place; finally, since non-bridge-verbs like 'exclaim' resist wh-extraction from their complement, they must only be embedded at FS. (In German, V2 sentences are also defined at FS, whereas V-final ones are defined at SS). The author also argues that his model offers a natural derivation of a generalized version of the Ban on Improper Movement, and introduces a further level, Predication Structure/PS, intermediary between CS and SS, thereby simultaneously accounting for ("structural") nominative assignment/checking in English (at CS) vs. quirky/inherent subject-marking in Icelandic (at TS) and for the distinct levels at which control and predication subjects must be defined in Russian.
Chapter 4, 'Anaphora,' develops a typology of anaphoric elements by assigning different anaphors to different RT- structures or levels: tight co-arguments are defined at TS, and long-distance anaphors, only at SS. Well-known cases, such as the opposition between Dutch 'zich' and 'zichzelf', or Reinhart and Reuland's findings concerning the "logophoric" use of reflexives in English are dealt with in this spirit: just as embedding takes place at the level at which the embedded structure is defined, so are anaphors submitted to various locality conditions, depending on the level for which they are defined. Locality and the type of antecedent needed (theta, A, or A') are thus closely correlated.
Chapter 5, entitled 'A/A'/A"/A''',' associates locality and (type of) target as examined in the preceding one with reconstruction. It is shown that the A/A' distinction must be relativized or generalized both with respect to movement proper (wh-movement, defined at one level) and with respect to scrambling (a case of misrepresentation between two levels); moreover, reconstruction effects are shown to follow from the architecture of the model.
In chapter 6, 'Superiority and Movement,' what is standardly analyzed as multiple Wh-movement is argued to be a case of "real" movement for the first Wh-Phrase, which has wide scope, but sheer scrambling for the following one(s) -- this scrambling being due to the strong D-linking flavour of multiple Wh- sentences: a distinction must consequently be established between wh-dependencies, which are established at PS, and Wh-movement proper, which occurs at SS).
Chapter 7, 'X-bar Theory and Clause Structure,' proposes a series of axioms that fix the number of juncture-types of X- bar theory (C-adjunction is allowed). Cinque's (1998) functional heads' hierarchy is adopted, but, crucially, those heads need not project: if they are realized by affixes/features, they simply percolate down to the next (lexical) head, with which they thereby lexicalize a subsequence of that functional structure. Besides, a process of "Reassociation" is defined, which allows a sort of morphological restructuring such that the two sequences [[X>Y]>T] and [X>[Y>T]] (where the caret denotes the head- complement relation) are stipulated to be equivalent. Williams then endeavours to demonstrate that the Head Movement Constraint and Relativized Minimality are sheer effects of his specific approach to X-bar structure.
Chapter 8, 'Inflectional Morphology,' further develops the foregoing ideas, in particular adding to Reassociation the rule "Flip" which allows two items [A>B] to appear in the reverse order [B<A] (and vice versa). Taken together, these two rules allow quite some freedom in the linear order morphemes can exhibit across languages, but crucially not too much -- possibly exactly what is empirically required. They also account for departures from Baker's Mirror Principle, which is itself argued to derive from the architecture of RT and the specific X'-theory developed in the book. This "morphological" approach is also applied to Verb Raising (or Clustering) in Dutch and in Hungarian: once again, massive remnant movement is shown to be unnecessarily complex and unmotivated, whereas the difficulties raised by more ordinary feature-checking theories are avoided.
In Chapter 9, 'Semantics in Representation Theory,' Williams suggests that each RT level possibly contributes to the meaning of expressions (up to utterances), noting however that existential closure of implicit arguments may well already take place at TS, leaving other quantificational aspects of interpretation to QS. An important feature of this chapter is the dual theory of focus it offers. Whereas Logical Focus (LFocus) partitions a sentence between what is logically presupposed and what is not, Information Focus (IFocus) manifests itself linearly by the heavy stress that must be carried by the word 'blue' in 'John compared the red hat to the BLUE hat'. Here, there is no logical presupposition, but mere "I(nformation) Presupposition"; the relevant level thus cannot be FS, but yet another, AS (for Accent Structure). A consequence of this approach is that, in corrective utterances, which typically copy the preceding sequence of words, IPresupposition (bracketed below) may include LFocus, as in the following dialogue:
(3) A 'It was JOHN that Bill heard.' B 'No, [it was John that Bill] SAW'
Crucially, however, the reverse never holds: AS (and the Information Structure it feeds) simply appears after FS in any derivation. Various forms of ellipsis (VP-ellipsis, [counter-]Focus- ellipsis, Gapping) are also dealt with in this last chapter, in the spirit and with the techniques anaphora was in chapter 4.
As noted earlier, _Representation Theory_ is very much a programmatic book, whence the presence of many gaps and loose ends in the demonstrations and the evaluation of empirical and/or theoretical consequences, generally acknowledged by the author. In this context, the wide coverage of cross-linguistic data (examples from Germanic, Romance and Slavic languages, but also, outside of the Indo-European family, Japanese, Hungarian, Swahili and a variety of lesser-known ones, are examined) must certainly be welcome. Williams also notes himself that the increasing number of levels or structures he must posit is a methodological difficulty. However, he writes (p. 58): "The limiting case is an RT with exactly the same number of levels as there are functional elements in the structure of a clause in the corresponding Checking Theory [...] Even if this limiting case turned out to be correct, RT would NOT thereby become a notational variant of Checking Theory, because the architecture is different, and the architecture makes predictions that Checking Theory is intrinsically incapable of." Leaving aside data already summarized above, one more such prediction is made on p. 193-194, where the tools and architecture of RT naturally explain why Gapping is possible when two TPs are coordinated, but not when two CPs are ("--" indicates a gap):
(4) a John saw Mary and Bill -- Pete b *I think that John saw Mary and that Bill -- Pete
Clearly, this phenomenon is all the more difficult to account for if one adopts the recent Minimalist suggestion that C is sensitive to T. But in RT, the data fall in naturally because (a) control of the "--" element under T is defined at the level at which T is introduced (CS or PS), and (b) CP embedding takes place much later (at SS); there is thus just no way to independently derive any well-formed CP with a (non-controlled) gap under T (as "that Bill -- Pete"), and coordinating that CP with a well-formed one will not help. Given that FS and AS are distinguished, Williams also predicts in chapter 9 that the two facts that (a) in some languages, such as English, neutral focus is on the right, whilst it is on the left in others, such as Hungarian, and that (b) the focused constituent itself is right accented or left accented (again, as in English and Hungarian) should NOT be correlated, but confesses he cannot illustrate this lack of connection. I am therefore happy to contribute the references to Hualde et al. (1994) and Elordieta (2002), because this disassociation is exactly exemplified by the Northern Biscayan variety of Basque, in which focused XPs must immediately precede the lexical participle+inflected auxiliary cluster, but in which the most prominent syllable in any phrase, non-focused OR focused, is final.
Although it is conceptually fascinating, the book also has its shortcomings, some of which are probably due to the editorial policy of the book series. Thus, I have noted the absence of a badly needed list of abbreviations, or the date "1982" as the year of publication of Chomsky's _Barriers_, both in the text and in the references. But the author himself should have been careful enough to replace, for instance, his numerous references to Richards's (1997) dissertation, now that a (revised) commercial version of it exists, namely Richards (2001). Moreover, some passages are fairly difficult to understand. To take but one example, in chapter 8, the author derives the Swahili order AgrS-T-AgrO-V of inflected transitive verbs on the basis of T taking AgrS as its functional complement (an issue he acknowledges is not without problems), supplemented by the stipulation that T is lexically encoded as an prefix, and AgrS as a stem; this is a bit confusing, as one would expect T to be a suffix, thereby automatically triggering Flip. But there is more to it: the main tense morphemes in Swahili are in fact stems (past tense -li- is etymologically connected with one of the copula stems, present tense -na- with the possessive use of the preposition na, 'with', and future -ta- with the stem -taka 'want'); thus, if, as is usually assumed, AgrS is a prefix, [T>AgrS] will naturally undergo Flip, yielding [AgrS<T], without any further ado. At a more general level, Williams could also have noted potential problems for his theory. We have seen that the Level Embedding Conjecture axiomatically guarantees that certain complex objets (such as CPs) do not appear in the derivation before some processes have been completed (recall the account of the ungrammaticality of (4b)). On such a theory, I keep wondering how Kayne's (1980) discovery that the French Wh-Phrase below was licensed by its higher trace t' receiving Case in COMP (today's Spec,C) from the matrix verb, as in (5), can be explained away (Hungarian displays the same phenomenon, see Chomsky (1981: 174, citing Horvath), and some hypercorrective idiolects of English, as noted by Radford (1988)).
(5) Qui crois-tu [t' [t être le meilleur]]? who think you to-be the best
The difficulty is this: if Exceptional Case Marking and infinitival clause embedding both take place earlier than CP embedding, it should be just impossible for 'qui' to get case-marked (cf. '*Je crois Jean être le meilleur') and the subordinate IP string should consequently be ruled out. Consider an alternative: the embedded clause is a wh- clause; therefore, it should not be embedded before SS, but SS is, again, a level at which Case marking cannot apply. I may well have missed something, but I can think of no way out.
Returning to the methodological problem set by the fairly large number of levels or representations Williams defines, I would like to underline the fact that it is precisely this large number that enables him to address in detail questions that have been left aside in the Minimalist Program, as acknowledged by Chomsky himself at the beginning of his famous Chapter 4, where he wrote: "Notice that I am sweeping under the rug questions of considerable significance, notably, questions about what in the earlier Extended Standard Theory framework were called 'surface effects' on interpretation. These are manifold, involving topic-focus and theme-rheme structures, figure- ground properties, effects of adjacency and linearity, and many others. Prima facie, they seem to involve some additional levels or levels internal to the phonological component, postmorphology but prephonetic, accessed at the interface along with PF (Phonetic Form) and LF (Logical Form). If that turns out to be correct, then the abstraction I am now pursuing may require qualification." (Chomsky 1995: 220). In a sense, it is then tempting to regard Williams's work as attempt at rationalizing the 'surface effects' mentioned by Chomsky (cf. the SS, FS and AS RT-levels), and to assimilate his QS with that part of LF that actually deals with scope. However, since he also has defined other levels before SS/SpellOut, and since he considers that the RT-levels are not related by movement, but by (mis)matching or (mis)representation (whereas movement is level-internal), the book definitely represents a real alternative to Minimalist derivations -- the more so as it predicts that Wh-movement determines nesting dependencies, whereas (his account of) scrambling (which, recall, includes secondary Wh-P movement) creates "(the appearance of) intersecting dependencies", and as reconstruction may only be defined for the relations created by movement or by those -- earlier -- relations "misrepresented" in the scrambling cases.
Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht, Foris.
Chomsky, N. 1995. The Minimalist Program. Cambridge (Mass.), MIT Press.
Cinque, G. 1998. Adverbs and Functional Heads. Oxford: Oxford University Press.
Elordieta, A. 2002. 'On the (im)possibility of prosodic focus marking in embedded contexts in Northern Bizkaian Basque.' In X. Artiagoitia, P. Goenaga and J. A. Lakarra (eds.), Erramu boneta: Festschrift for Rudolf P. G. de Rijk (Bilbao, Universidad del Pais Vasco), 153-177.
Hualde, J. I., G. Elordieta, and A. Elordieta. 1994. The Basque Dialect of Lekeitio. San Sebastian: ASJU Supplements, #34.
Kayne, R. 1980. 'Extensions of Binding and Case-Marking.' Linguistic Inquiry 11, 75-96.
Radford, A. 1988. Transformational Grammar. Cambridge: Cambridge University press.
Richards, N. 2001. Movement in Language. Interactions and Architectures. Oxford: Oxford University Press.
Williams, E. 1997. 'Blocking and Anaphora.' Linguistic Inquiry 28, 577-628.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER Georges Rebuschi is professor of general linguistics at the Sorbonne Nouvelle. His main interests are syntactic parametrization, and the syntax/semantics interface. He published a collection devoted to Basque linguistics in 1997, and co-edited a book on the Grammar of focus in 1999. He is currently working on the syntactic typology, and correlated variable semantics, of left-dislocated (or left- hanging), free relative clauses.