Tài liệu Constituent Structure - Part 24 doc

the S-syntax. For example, while we want to retain a subject–object hierarchical distinction to explain eVects like binding phenomena in scrambling/non-conWgurational languages, we do not necessarily want these to correspond with the word order. L-syntax, then, can represent the semantic and syntactic relations that refer to constituency. The S-syntax may realize the elements diVerently. Imagine, for example, a sentence with VSO order. The L-syntax would represent the [ S NP subj [ VP VNP Obj ]] constituency consistent with our understanding of argument structure and compositionality. The S-syntax by contrast would represent this as a Xat [ S VNP subj NP obj ] structure, reXecting the actual pronunciation of the sentence. In a conWgurational language like English, the S-syntax and L-syntax are more tightly linked with both, exhibiting the X-bar theoretic projection properties. Around the same time, Dowty (1982)24 reintroduced the distinction between tectogrammatical and phenogrammatical structures in the Montague Grammar version of categorial grammar, based on a pro- posal by Curry (1961). Tectogrammatical structures represent the means by which a sentence is composed semantically from its component parts. It largely ignores the actual order the words appear in. Pheno- grammatical structures by contrast reXect the actual order and mor- phological composition of the sentence. Tectogrammatical structures are typically hierarchical trees; phenogrammatical ones are strings. Take the phenomenon of VSO order. On a semantic level we want the subject to be predicated not only of the verb, but of whole verb phrases. For example, in the sentence John likes apples the denotation of this sentence is true precisely when John is a member of the set of people who like apples, not just the set of people who like. On the other hand, there appears to be no surface VP constituent (see the next chapter for a discussion of this claim). This can be captured by creating a prototypical hierarchical tree to reXect compositionality, but allowing the linear string to be aVected by special ordering functions that apply as each level of structure is created. For VSO languages there are two such functions: (34) (a) F 1 (Æ, â) ¼ the result of inserting â after the Wrst word in Æ. (b) F 2 (Æ, â) ¼ Æ ^ â (the linear concatenation of â after Æ). The second rule is the rule that presumably applies in a language like English which constructs the linear order in strict parallel to the hier- archical structure. Rule(8a) is essentially Bach’s (1979) ‘‘right wrap’’rule, 24 See also the discussion in Huck and Ojeda (1987). 210 controversies which inserts the newly hierarchically attached item inside the already created structure (this amounts to a syntactic inWxation operation). So an Irish sentence like (35a) has the proof tree in (35b). The tree represents the tectogrammatical structure created by the lexical properties of the words involved. The strings are the result of the parallel application of the above-mentioned functions to the phenogrammatical string: (35) (a) D’o ´ l Sea ´ n poitı ´ n. drank John homebrew ‘‘John drank homebrew.’’ (b) [D'ól Seán poitín] F 1 Seán [D'ól poitín] F 2 D'ól poitín The verb and its object are joined to form a VP, and the words are concatenated in the order they appear in the sentence (VO). At the next level up in the tree the subject is predicated of the VP, the F 1 rule inserts the subject between the verb and the object. ‘‘Free’’-word-order languages are dealt with in a similar way; such languages are viewed to have a single word-order function that creates an unordered set (36). Take the Latin sentence in (37) as an example: (36)F 1 (Æ, â) ¼ Æ [ b (37) (a) Marcus Fluvian amat. Marcus Fluvian loves ‘‘Marcus loves Fluvian.’’ (b) Marcus amat Fluvian. (c) Fluvian Marcus amat. (d) Amat Fluvian Marcus. etc. (e) {Marcus, Fluvian, amat} Marcus {Fluvian, amat} amat Fluvian Dowty (1996) presents an updated version of his theory using GPSG- style LP principles and allowing reference to the heads of phrases. Another version of the pheno/tectogrammatical distinction can be found couched within the HPSG framework as described by Kathol multi-structures 211 (2000) (see also Kathol and Levine 1993). Kathol draws on Reape’s (1994) idea that word order in HPSG should be described in terms of a special word-order domain feature (DOM). Word order is determined by a feature structures that constrain the linear order of elements within that feature. Scrambling is accomplished by virtue of a set theoretic operator that creates ‘‘shuZed’’ sets (subject to restrictions based on the original order of elements within the original ordered sets). See also Langendoen (2003), who pursues a version of Dowty’s approach but couched within Minimalist set theoretic Bare Phrase Structure. 10.4.3 Wheel-and-spoke multiplanar approaches In the parallel multiplanar approaches the link between planes of representation is largely abstract and concerns semantic relationships that do not appear to be represented in the surface word order (for example, in scrambling or VSO contexts). The next group of ap- proaches uses the surface string of the sentence itself as the glue that holds together the multiple planes. The motivations here are not usually a mismatch between semantic/syntactic criteria and surface order, but mismatches in the apparent behavior of linearly ordered strings with respect to diVerent constituency tests, such that all the constituency tests reveal the same linear order, but diVerent hierarch- ical properties. The evidence for parallel structures, by contrast, came from situations where the linear order did not reXect some predicted syntactic or semantic property. These two approaches thus emphasize diVerent kinds of data and may not be entirely incompatible. Indeed, one Wnds approaches like RRG, which has aspects of both (mapping from basic predicate and semantic relations to the constituent struc- ture—called the LSC—is done by mapping rules, as is the relationship between the argument structure and the focus structure) but various kinds of constituency relationship are mapped from plane to plane through the surface linear string. In this subsection, we consider a variety of approaches in a variety of frameworks that take a wheel-and-spoke approach to multiplanar structures. Starting in the mid 1980s, this view was adopted by a variety of scholars and at least some versions of it were inXuenced by the theory of autosegmental phonology (Goldsmith 1976). I am not in a position to decipher who proposed a multiplanar model for syntax Wrst, so I will simply describe the arguments of the various main players. As elsewhere in this part of the volume, the seas of shifting assumptions make it nearly impossible to fairly compare the various 212 controversies approaches described here with each other or with non-multiplanar accounts. Instead I hope the reader will be able use my brief descrip- tions as a rudder for negotiating said shifting sea. Haegeman and van Riemsdijk (1986), working within GB theory, suggest that an analysis of pseudopassives and cross-serial dependen- cies motivates multiple constituent structures for a single clause. Con- sider Wrst the case of Zuritu ¨ u ¨ sch and Dutch cross-serial dependencies that we have mentioned several times in this book. Haegeman and van Riemsdijk propose that these have a normal non-line-crossing repre- sentation consistent with the usual understanding of tree structures. However, Dutch (and other languages with crossing dependencies) has a rule of reanalysis that takes the verbal string and reanalyzes it as a single V r (in eVect, unioning all the embedded clauses). This unioned tree is on a separate dimension from the one that is represented basic compositionality. There is then a morphological rule that reorders the verbs (indicated by the arrows in the lower tree). (38) (a) . . . dat hij het probleem probeert te begrijpen. that he the problem tries to understand ‘‘. . . that he tries to understand the problem.’’ (b) SЈ S NP VP S NP VP PRO NP C V 1 V 2 probeert dat hij het probleem te begrijpen C V 1 V 2 NP V r NP VP S SЈ multi-structures 213 They also propose that such trees with reanalysis are the explanation of pseudopassives (39). In such forms the object of a preposition moves to subject position even though they are not the direct object of the verb: (39) a) John talked to Mary. b) Mary was talked to. In eVect the verb þ preposition sequence behaves as if it is were a verb particle construction (e.g. blow up), even though the verb and the preposition never actually form a constituent. Haegeman and van Riemsdijk suggest that this is due to a process of reanalysis which is represented in a second tree structure: ()S S NP VP PP NP V talked V P to P John Mary VNP NP VP The top constituency structure represents our lexical understanding of the verb talk with a full preposition to. The bottom tree represents the reanalyzed form where Mary is the direct object of the complex verb talk to. This bottom tree is the input to the passive operation. In both of these cases, we have a situation where one set of syntactic or semantic information points to one constituent structure, but another set of information points to a diVerent constituency. These kinds of phenomenon we can loosely term ‘‘bracketing paradoxes.’’ A diVerent type of bracketing paradox motivated multiple planes of structure in the Autolexical syntax approach (Sadock 1991):25 25 See Zwicky (1986c) for discussion of how Autolexical syntax and Zwicky’s version of GPSG achieve similar results. 214 controversies mismatches between the syntactic form, the morphological form, and the semantic representation. Sadock lists a number of these, including noun-incorporation structures, various kinds of morphologically complex inXectional structure, and clitics. Let us take the last as an example. As is well known, the possessive ’s in English attaches not to the head of the possessor nominal but to the end of it. However, from a morpholophonological perspective, we see that it must be attached to the Wnal word as it is subject to the typical patterns of assimilation found with other s morphemes in English. There is thus a conXict between the morphological properties of the morpheme, which re- quire it to be bound to the Wnal word in the possessor, but structurally and syntactically it behaves as if it were an independent word that follows the possessor. This can be expressed in terms of multiple planes. (The trees that follows is not identical to those in Sadock, but are consistent with his analysis of possessive ’s.) ()DP DP DЈ D the NP D NP Syntax N man ’s N book DN NMorphology Pesetsky (1996) considers a diVerent kind of bracketing paradox, namely, where the constituency evidence contradicts c-command prominence in the tree. We have conXicting information about the italicized string in (42a). The evidence from binding shows that the DP them must c-command the reciprocal (each other) that it binds. This c-command relationship is reXected in the structure in (43a) which Pesetsky calls a cascade. In the cascade, in order to insure that them c-commands each other, there is no constituent consisting solely of the V and DP to the exclusion of the PP. However, the data in (42b) shows that such a constituent must exist, The ellipsis of visit them excludes the PP. Pesetsky suggests then that the same string has the ‘‘layered’’ structure in (43b) where the lower V’ is the target for the ellipsis. These two structures hold simultaneously of the VP. multi-structures 215 (42)a.Ivisited them i on each other i ’s birthdays. b. John said that he would [visit the children] and visit them i he did on each other i ’s birthdays. (a) Cascade (b) Layered V visit PP PP DP V visit DP on each other’s birthdays on e.o.’s birthda ys them P DP them () Baltin (2006)oVers an alternative analysis of these facts. He suggests that the underlying form of the VP is like that in (43b) (a layered structure), but that there is a more Wne-grained functional structure (44), such that the object raises out of the VP to get case in an AgrP, creating the c-command structure necessary for binding the reciprocal. (Surface word order results from subsequent movement of the remnant VP.) () AgrP DP j AgrЈ c-command them Agr … VP VЈ PP constituent targeted by ellipsis Vt i on e.o.’s birthdays As mentioned brieXy above, the theory of Role and Reference Gram- mar (RRG) (Van Valin 1993, 2003; Van Valin and LaPolla 1997) uses both linking principles (to link between, for example, semantic representa- tions and syntactic ones) and spoke-and-wheel links between multiple planes of constituent structure (which are typically semantically or pragmatically motivated). One plane of representation is the layered structure of the clause (LSC), which represents the non-endocentric predicate (the nucleus) and its arguments (together with the nucleus forming the core) along with various modiWers (including the ‘‘periph- ery’’ and two positions to the left of the clause for left detached items like sentential adverbials and things like wh-phrases). At the same time, grammatical items such as aspect, modality, negation, tense, and illocu- 216 controversies tionary-force markers are represented on a separate plane for operators. Finally, there is a third dimension that represents focus (which as we will discuss in Ch. 11) which is one way in which VP-like (verb–object) constituencies are represented in RRG. () SENTENCE CLAUSE CORE PERIPHERY NUC ARG PRED ARG ARG LCS NP John V NP PP ADV didnot show the book to Mary yesterday V focus domain NUC CORE NEG CLAUSE Operator Structure TNS CLAUSE IF CLAUSE SENTENCE These three constituent structures are motivated by the semantics or discourse and licensed by mapping principles (including those that map to argument structure and other semantic structures), but all pivot around the linear string of words. In the next chapter, we will address exactly what is meant by categories like ‘‘predicate’’, ‘‘NP’’, ‘‘VP’’ and Tense in some detail. 10.5 Conclusions In this third part of this book, we are looking at a number of alternatives to strict compositional phrase structure. In the last chapter, we looked at dependency based alternatives to constituent representations, in this multi-structures 217 chapter, we surveyed a number of alternatives where restrictions on the form of constituency have been relaxed to allow line crossing, multi- domination, multidimensionality, and multiplanar structures, looking at evidence from discontinuous constituents, scrambling, argument sharing, and cross-modular bracketing paradoxes. In the next and Wnal contentful chapter, we address one further way in which phrase- structure systems might be modiWed, focusing on the content and nature of various major constituent types both from a categorial and semantic perspective. 218 controversies 11 Phrasal Categories and Cartography 11.1 Introduction The previous ten chapters of this book have focused on the mechanisms for forming, describing, or licensing constituent structure(s). Up to now, wherever possible, I have tried to abstract away from the content of those structures. Needless to say, the question of the nature of the content or category of constituents is vital to understanding constituent-structure systems. This much should be clear from the role played by heads in most recent theories of syntax. In this chapter, I survey the diVerent proposals about the categories of various constituent types. I also consider evidence for their arrangement in the clause. In the Chomskyan paradigm, the endeavor of Wnding the correct arrangement of the right categories has come to be known as the ‘‘cartographic approach’’ (largely by the com- munity of Italian linguists headed by Luigi Rizzi, who are investigating this question). I will extend the cartographic label to other approaches to the eVort to understand the contentful structure of phrase structure, seeking out commonalities among these approaches. This is an extremely large topic, and could easily be the topic of several book-length treat- ments. Here, we sketch only the rough outline of the debates. Interested readers should follow up on the citations listed throughout this chapter. This chapter starts with perhaps one of the most interesting discov- eries of twentieth-century linguistics: the layered tripartite nature of the clauses. Here I look at how a number of approaches to syntactic structure have converged on the idea that, cross-linguistically, clauses have at their heart a core predication, which is inside a structure of operators that locate that predication in time and space relative to the speech time, and Wnally are closed by a structure that relates the proposition to the speaker’s perspective. I address the structure of . cross-serial dependen- cies motivates multiple constituent structures for a single clause. Con- sider Wrst the case of Zuritu ¨ u ¨ sch and Dutch cross-serial. component parts. It largely ignores the actual order the words appear in. Pheno- grammatical structures by contrast reXect the actual order and mor- phological