Jump to ContentJump to Main Navigation
At the Syntax-Pragmatics Interface$

Lutz Marten

Print publication date: 2002

Print ISBN-13: 9780199250639

Published to Oxford Scholarship Online: January 2010

DOI: 10.1093/acprof:oso/9780199250639.001.0001


(p.209) 7 Conclusion
At the Syntax-Pragmatics Interface


Oxford University Press

Abstract and Keywords

This concluding chapter summarizes the arguments presented and draws out the main consequences. It points out the basic idea of the study, namely that verbal subcategorization is underspecified and that verbs allow for a free process of adjunction resulting in verb phrases with varying complementation patterns, built anew on every occasion of use, and interpreted with recourse to pragmatic enrichment. The relation between the conceptual content of the study and the particular implementation is shown, and possible improvements or alternatives are sketched. Further possible empirical developments are mentioned, such as a closer study of other valency changing operations such as passives. The most important result of the study is, however, that the interpretation of verbs and their syntactic environment is thoroughly context-dependent, thus providing evidence against a syntactic level of logical form. Rather, in the process of utterance interpretation, lexical, syntactic, and pragmatic knowledge interact so as to jointly derive an interpretation of the words encountered.

Keywords:   syntax-pragmatics interface, verb phrases, complementation patterns, valency changing, passives, logical form, utterance interpretation

7.1 Summary

In this study I have argued that subcategorization information provided by verbs is underspecified. Verbs may specify how many Ty(e) expressions, including NPs and PPs, they minimally require in order to derive an expression of Ty(t), but they are in general flexible with respect to how many Ty(e) expressions can optionally be introduced into the verb phrase. This view is supported by linguistic evidence which shows that arguments and adjuncts behave alike in a number of respects, where the only difference between the two is the obligatoriness of arguments as opposed to the optionality of adjuncts. Of two possible ways to analyse adjuncts, I have argued that an analysis of VP adjunction which treats the adjunct as a functor taking the verb or the VP as argument does not adequately explain this fact, and that furthermore an analysis involving an Adjunction rule implies the restructuring of trees, which goes against the overall aim of the DS system to model utterance interpretation on-line. In contrast, an analysis which assumes that verbs are lexically underspecified, and thus permits adjuncts to be integrated into the VP under the same structural relation as arguments, provides an incremental parse of adjuncts which accounts for the parallelism of arguments and adjuncts, as well as for the optionality of adjuncts. The formal statement of verbal underspecification involves the underspecified type value e*, which is defined as follows:

  1. (1) Definition of (e* → t)

  •      (e* → t)   = def   {(t) v (e → (e* → t))}

(p.210) The definition states that verbs with such a type can be used with any number of Ty(e) expressions, since the underspecified type can be recursively resolved by introducing additional Ty (e) expressions. The introduction of the notion of underspecification into the type specification of verbs is motivated also by the fact that structural underspecification is already employed in the DS system for values of the tree node and formula predicates. The definition of e* for verbs thus extends the notion of underspecification, which is used in the DS system mainly for the analysis of preposed constituents, to the analysis of predicate–argument structure. The e* analysis entails that PPs are Ty(e) expressions, and that the function of prepositions is mainly to license the building of additional Ty(e) nodes. In English, underspecified verbs are also unfixed, since they are lexically projected onto an unfixed node. However, the combination of these two types of underspecification is not a necessary feature of the analysis, since in verb-final structures, underspecified verbs are projected onto fixed nodes.

The analysis of adjunction as involving verbal underspecification is motivated empirically by the parallelism of arguments and adjuncts in several respects, and from considerations of the overall DS model. However, as shown in Chapter 5, there is no clear analysis of the interpretation of underspecified verbs in the traditional semantic sense. This results from the fact that under-specified verbs are essentially dynamic, and only interpreted in context, with recourse to the utterance situation, while model-theoretic approaches to verb meaning assume that predicates are static, have a fixed arity, and are diectly related to natural language verbs, without the mediation of intermediate representational structure. After considering two possible semantic analyses of underspecified verbs, I have turned to providing a novel, pragmatic approach to verb meaning in context with recourse to the notion of enrichment developed in relevance theory.

The relevance-theoretic notions of pragmatic enrichment and concept formation help to show that natural language expressions quite generally do not address fully specified concepts, i.e. expressions of the language of thought, but rather provide an instruction for the hearer to construct an occasion-specific ad hoc concept which serves to derive particular occasion-specific inferential effects. Concepts addressed by verbs, in particular, are interpreted with the aid of their complements in the verb phrase, and the propositional structure expressed by an utterance can only be established when the concept addressed by the verb has been fixed in a way appropriate to the utterance situation. Part of this process of fixing the conceptual contribution of the verb is to determine its occasion specific arity, which may vary from one use to another. The syntactic reflex of the indirect way in which lexical verbs address the concepts they eventually stand for is their underspecified subcategorization information. (p.211) Verbal underspecification can be seen as the overt encoding of the need for conceptual enrichment of verb denotations. As further shown in Chapter 5, the resolution of verbal underspecification is thoroughly pragmatic, and involves both the strengthening of the concept addressed and the selection of the hypotheses entertained as a result from the activation of concepts. The analysis thus shows how instructions from words, structural syntactic processes, and pragmatic inference interact in the process of utterance interpretation.

In Chapter 6, I have developed a novel analysis of applied verbs in Swahili as encoding an instruction for concept formation. Applied verbs have often been analysed as involving a process of valency changing, in that an additional Ty(e) expression is introduced. The argument developed in this study is that applied verbs instruct the hearer to build a stronger concept than a possible concept built from the base verb, so that additional inferential effects can be derived. This instruction for concept formation may, but does not necessarily, result in the introduction of an additional Ty(e) expression. From this perspective, the syntactic facts follow from the underlying meaning of applied verbs, which is essentially pragmatic. I have presented evidence for this view, and have argued that it provides a better explanation of the facts than syntactic analyses, which view applied verbs as only encoding a change in valency.

7.2 Concluding Remarks

This study has aimed at providing a comprehensive theory of verb phrase adjunction within a hearer-based, formal model of syntax, which takes account of the facts that adjunction processes are free and general, and that arguments and adjuncts often behave alike, and which at the same time provides a model of adjunction in which structures are built incrementally, on-line. In order to make the overall perspective explicit, the argument developed is embedded in a model of utterance interpretation according to which utterance interpretation involves different aspects of cognitive activity: the ability to relate a physical signal to lexical units, the ability to build structured representations from lexical instructions, and the ability to construct mental concepts from the information provided by words and contextual assumptions. One of the central conclusions of this study is that these different abilities interact freely, and that for the analysis of verb phrase interpretation, all aspects have to be taken into account. More specifically, the analysis of verb phrase adjunction and the interpretation of underspecified verbs shows that there is no well-defined level of an interface between syntactic and pragmatic processes. The two processes can be distinguished, since the former are either computational rules or lexical instructions, which interact to result in increasingly complex structures (p.212) representing the semantic content of the utterance they are modelling, while the latter involve non-demonstrative inference and access to world and contextual knowledge. However, the eventual tree representation of the proposition communicated by the utterance does not result from syntactic knowledge alone. In order to derive a well-formed output tree with all requirements fulfilled and all underspecification resolved, i.e. in order to complete the syntactic part of the parse, the hearer has to engage in relevance-based reasoning about the appropriate construction of the concept meant to be communicated by the speaker, which, as a process of enrichment, involves pragmatic knowledge. The interpretation of underspecified verbs thus shows exactly how these two types of knowledge interact, and in that sense provide a sketch of the syntax-pragmatics interface.

A number of comments are in order regarding the implementation of the basic ideas presented here. The most obvious way, maybe, to think about verbal underspecification is that the verb is lexically underspecified and that Ty(e) expressions are just added if and when pragmatically licensed. However, the approach taken here is more cautious, putting more weight of the explanation on the syntax, in that prepositions are lexically defined as building Ty(e) nodes and no general rules for introduction of Ty(e) expressions are available. The more daring alternative would be, of course, to allow a general introduction rule and invoke pragmatic infelicity for examples such as (2):

  1. (2) John met Bill Sally Donovan.

Under the current proposal, the string in (2) is ungrammatical because the NPs Sally and Donovan cannot be introduced into the parse. While this alternative is less radical, it has the advantage that it gives a structural reason for the ungrammatically of strings like (2), and it also provides an analysis of prepositions—and, by extension, case marking—as licensing VP adjuncts. On the other hand, both NPs and PPs are analysed uniformly as Ty(e) expressions, and thus their parallel behaviour finds its formal reflex.

Another aspect of this implementation is that optional arguments have to be analysed as lexically licensed by the verb, as have the various complementation patterns for applied verbs in Swahili. This is, on the one hand, not problematic for a strongly lexically based framework as DS, but, on the other hand, it remains to be seen whether or not a more general formulation of optional arguments can be found. The case of applied verbs points to another area for further research, namely other processes of valency-changing operations. While for applied verbs, it turned out that their valency changing aspects could be related to a more general explanation, a similar argument has still to be developed for verb forms such as passives and causatives.

(p.213) There is more room for enlarging the empirical coverage of the analysis presented. For example, little has been said about the representation of verbs taking a sentential or adjectival complement. Outside the domain of verbs, there is scope for inquiry as to whether other word forms incorporate a similar aspect of underspecification in their lexical information, in particular, of course, those words which have been argued to have argument structure, such as nouns. All these aspects remain for the time being challenges for future research.

However, the ideas developed in this study are relevant for work outside of DS, or logic-based frameworks. Head-Driven Phrase Structure Grammar (HPSG), as discussed in Chapter 5, has mechanisms for introducing adjuncts as syntactic arguments, and the idea has been discussed more generally, as pointed out in Chapter 1. Abstracting away from matters of implementation, the key issue raised in this work is the role and function of verb phrase adjuncts, and the structure of the verb phrase beyond semantically or thematically licensed nominal expressions. A full understanding of this matter, I have argued, takes us beyond the domain of syntax, and leads to questions of how language can be used to convey an unlimited set of meanings, by employing the same structural resources on different occasions. The two complementary notions of underspecification on the one hand and contextual enrichment on the other, independent of their particular detailed characterization, provide powerful theoretical tools which make the interplay between structure and context more explicit. Both notions have been explored independently of each other, and independently of verb phrase interpretation, and one result of this study is that their combination may shed new light on seemingly unrelated problems. It remains to be seen if by continuing research along these lines, more phenomena can be characterized as resulting from the particular interface explored here.

Notwithstanding issues which have to be developed further, the main points and findings of the study can then be summarized as follows. Verbs are lexically underspecified with respect to their subcategorization information. This means that Ty(e) expressions can be introduced into the verb phrase either as obligatorily required arguments or as optional adjuncts. However, every Ty(e) expression is assigned to a position with the same structural relation to the verb, so that once an expression is introduced, there is no longer any difference between adjuncts and arguments. In addition, this analysis entails that verbs are found with different complementation patterns and varying arity on different occasions of their use. The interpretation of verb phrases, which includes the resolution of their type underspecification, is subject to pragmatic enrichment, employing background and contextual knowledge. The main consequence of this analysis is that, since the eventual syntactic structure of the verb phrase can only be established with recourse to pragmatic knowledge, verb (p.214) phrase underspecification provides a strong argument for the free interaction between syntax and pragmatics in the process of utterance interpretation. From this perspective, there is no syntactically defined output level which feeds into a general reasoning module; rather, hearers employ their relevance-based inferential abilities not only for the interpretation but also for the establishment of logico-syntactic structure in context. The syntactic structure of human language thus reflects in a fundamental way our cognitive propensity for communication.