Combining multiple information sources for ellipsis

1992

Abstract In human discourse there is much that is communicated without being explicitly stated. The grammar of natural language provides a broad array of mechanisms for such implicit communication. One example of this is verb phrase ellipsis, in which a verb phrase is elided, its position marked only by an auxiliary verb. Such elliptical constructions are generally easily and unambiguously understood. In this proposal I will attempt to explain how this is accomplished.

The area of ellipsis resolution and generation has long been neglected in work on natural language processing, and there are few examples of working systems or computational algorithms. However, the misuse or non-use of ellipsis in highly preferred contexts can make a dialogue difficult to understand similar to the way inappropriate referential expressions can impede comprehension. Furthermore, we think ellipsis has many more functions than are currently recognized, an idea discussed in our own contribution to this volume. We are grateful that the ESSLLI organizers have provided a forum for researchers to present new data and analyses.

This chapter emphasizes the shared communicative motivation of ellipsis constructions that leads to cross-linguistic similarities and certain predictable functional constraints. More specifically, ellipsis is licensed by a system of motivated constructions; i.e., learned pairings of form and function. Constructions capture a range of restrictions on form and function, including those related to semantics, discourse context, register, genre, and dialect. Generalizations across constructions are captured by a network of constructions with partially overlapping representations. Semantic recoverability is facilitated, not by copying and deleting syntactic structure at the ellipsis site, but by a cognitive process that " points " to information that is, typically but not always, available from the memory trace of an antecedent. This mechanism is independently required for fragments, non-elliptical expressions such as ditto and respectively, and the many examples that do not display the " connectivity " effects that are predicted by copy and deletion accounts.

Lecture Notes in Computer Science, 2011

In this paper we set out the preliminaries needed for a formal theory of context, relative to a linguistic framework in which naturallanguage syntax is defined as procedures for context-dependent interpretation. Dynamic Syntax provides a formalism where both representations of content and context are defined dynamically and structurally, with time-linear monotonic growth across sequences of partial trees as the core structure-inducing notion. The primary data involve elliptical fragments, as these provide less familiar evidence of the requisite concept of context than anaphora, but equally central. As part of our sketch of the framework, we show how apparent anomalies for a time-linear basis for interpretation can be straightforwardly characterised once we adopt a new perspective on syntax as the dynamics of transitions between parse-states. We then take this as the basis for providing an integrated account of ellipsis construal. And, as a bonus, we will show how this intrinsically dynamic perspective extends in a seamless way to dialogue exchanges with free shifting of role between speaking and hearing (split-utterances). We shall argue that what is required to explain such dialogue phenomena is for contexts, as representations of content, to include not merely partial structures but also the sequence of actions that led to such structures. 1 Preliminaries Despite extensive research on the context-dependence of natural-language (NL) understanding over the last thirty years, with formal modelling of a wide range of individual phenomena, there has been little attempt to bring everything together in order to seek an overall concept of context-dependence. In particular, ellipsis has been treated wholly differently from anaphora, despite the fact that, like anaphora, ellipsis is a phenomenon which, by definition, exhibits radical context-dependency. An elliptical construction is one "that lacks an element that is, nevertheless, recoverable or inferable from the context" (Wikipedia: ellipsis). This characterisation corresponds to the robust folk intuition that, in

2004

Abstract It is generally assumed that ellipsis requires parallelism between the clause containing the ellipsis and some antecedent clause. We argue that the parallelism requirement generated by ellipsis must be applied in accordance with discourse structure: a matching antecedent clause must be found that locally c‐commands the clause containing the ellipsis in the discourse tree.

Research on Language and Computation, 2007

This paper challenges the tradition of defining grammars and grammaticality independently of the context of utterance. Using dialogue phenomena, in particular elliptical utterances, it argues that the obvious dependence of such utterances on context to recover the intended interpretation should be regarded as an inherent characteristic of natural language grammars and thus applicable to the characterisation of grammaticality for all natural language strings. The paper adopts the framework of Dynamic Syntax which shifts the burden of syntactic explanation away from the definition of de-contextualised syntactic structures defined over strings of words towards the characterisation of syntax as a context-dependent, incremental process whereby interpretations of strings in context are progressively built up as an utterance proceeds. This change in the way syntax is conceived, together with a demonstration that the same processes for building interpretations are used in generation as in parsing, is shown to allow a unitary account of anaphora and a range of elliptical phenomena that is typically precluded in non-dynamic, structure-based theories of syntax. The paper ends by providing formal definitions of well-formedness with respect to context that preserve traditional notions of grammaticality while allowing more fine-grained characterisations of well-formedness to distinguish acceptability from full (un)grammaticality.

2004

AbstractY Since Webber (1978) it has been known that ellipsis interpretation sometimes requires inference. However, the literature on ellipsis has largely Ignored this: prominent accounts such as (Sag (1976); Williams (1977); Dalrymple et al.(1991); Fiengo and May (1994) and Hardt (1999)) provide no mechanism for inference in ellipsis resolution. The proposal in Rooth (1992) is an exception to this: Rooth gives several examples in which inference is required in ellipsis interpretation.

2009

1992

Abstract A computational account of VP ellipsis is described, in which VP's are represented in the discourse model as contextually dependent semantic objects. It is argued that this approach can handle examples that are not allowed by alternative accounts. An implementation is defined in terms of extensions to the Incremental Interpretation System. The treatment of VP ellipsis is analogous to that of pronominal anaphora.