Recent developments in connectionist architectures for symbolic computation have made it possible to investigate parsing in a connectionist network while still taking advantage of the large body of work on parsing in symbolic frameworks. The work discussed here investigates syntactic parsing in the temporal synchrony variable binding model of symbolic computation in a connectionist network. This computational architecture solves the basic problem with previous connectionist architectures, while keeping their advantages. However, the architecture does have some limitations, which impose constraints on parsing in this architecture. Despite these constraints, the architecture is computationally adequate for syntactic parsing. In addition, the constraints make some significant linguistic predictions. These arguments are made using a specific parsing model. The extensive use of partial descriptions of phrase structure trees is crucial to the ability of this model to recover the syntactic st...

Abstract not found

Current research in natural language generation is situated in a computational linguistics tradition that was founded several decades ago. We critically analyse some of the architectural assumptions underlying existing systems and point out some problems in the domains of text planning and lexicalization. Guided by the identification of major generation challenges viewed from the angles of knowledge-based systems and cognitive psychology, we sketch some new directions for future research.

Centering theory (CT) has been mostly discussed from the point of view of interpretation rather than generation, and research has tended to concentrate on problems of anaphora resolution. This paper examines how centering could fit into the generation task, separating out components of the theory which are concerned with planning and lexical choice. We argue that it is a mistake to define a total ordering on the transitions CONTINUE, RETAIN, SHIFT and that they are in fact epiphenomenal; a partial ordering emerges from the interaction between cohesion (maintaining the same center) and salience (realising the center as Subject). CT has generally been neglected by NLG practitioners, possibly because it appears to assume that the center is determined according to feedback from the surface grammar to text planning, but we argue that this is an artefactual problem which can be eliminated on an appropriate interpretation of the CT rules. 1 What is Centering? Centering theory (CT)...

This paper presents a system architecture for producing spoken utterances while collaborating with the human dialogue partner in carrying on a spoken dialogue. The system incrementally produces utterances while managing the responses and interruptions from the human dialogue partner that occur in the middle of the system's utterances. The collaborative principles that the system must use in order to collaborate on a spoken dialogue with the human partner are described. These principles are based on the results of an analysis of the nature and interdependence of utterance units, responses from dialogue partners, and discourse strategies for handling responses in a corpus of spoken dialogues. In particular, it is shown that there are cases where the system has to suspend utterances planned in advance and plan additional utterances to maintain the mutual understanding of exchanges even when the partner does not make any responses. The system has been implemented and t...

This thesis addresses the cognitive basis of syntactic adaptation, which biases speakers to repeat their own syntactic constructions and those of their conversational partners. I address two types of syntactic adaptation: short-term priming and longterm adaptation. I develop two metrics for syntactic adaptation within a speaker and between speakers in dialogue: one for short-term priming effects that decay quickly, and one for long-term adaptation over the course of a dialogue. Both methods estimate adaptation in large datasets consisting of transcribed human-human dialogue annotated with syntactic information. Two such corpora in English are used: Switchboard, a collection of spontaneous phone conversation, and HCRC Map Task, a set of task-oriented dialogues in which participants describe routes on a map to one another. I find both priming and long-term adaptation in both corpora, confirming well-known experimental results (e.g., Bock, 1986b). I extend prior work by showing that syntactic priming effects not only apply to selected syntactic constructions that are alternative realizations of the same semantics, but still hold when a broad

We develop a linearization model that captures a broad range of constituent order phenomena in clauses of Dutch, English and German (clause union, cross-serial dependencies, scrambling, verb clusters, wh-fronting, extraction, extraposition, etc.). The model is part of the psycholinguistically motivated formalism of Performance Grammar, which has separate components for assembling the hierarchical and the linear structure of sentences. We demonstrate that a few narrowly localized, relatively minor variations of the model suffice to account for a great deal of the --- sometimes widely diverging --- word order patterns in the three target languages, and of the constraints that apply to them. This is the first reason for calling our approach `uniform'. The other reason is that the system allows the various linear order phenomena to be viewed as manifestations of the same basic mechanism, which we have dubbed topology sharing. Here, a topology is a one-dimensional array containing a limited number of left-to-right positions (`slots') for clausal constituents (cf. the `topological fields' of traditional German grammar). Topologies at adjacent levels of the syntactic hierarchy sometimes share left- and/or rightperipheral slots, which gives rise to upward `movement' of the slot fillers. We also explore the generative capacity of the linearization system and suggest that it approaches the theoretical minimum of mild context-sensitivity.

In this paper, a new discourse structure model is proposed, and based on the model, we report the results of an analysis on Japanese-language dialogues over the telephone. As a result, a method for describing and analyzing the structure of spontaneous spoken dialogue is provided, and some characteristics of spontaneous spoken dialogue over the telephone were clarified.

is a one-dimensional array called topology, which specifies a fixed number of positions (or slots, landing sites) where segments (branches, constituents) of lexical frames can be stored in left-to-right order. The segments stored in a topology may include constituents originating from a lexical frame lower in the hierarchy. As we will see below, this happens without affecting the frame hierarchy. In this Section, we discuss the hierarchical and linear grammar components in turn. For the empirical psycholinguistic arguments in support of the separation between hierarchical and linear structure we refer to Kempen & Harbusch (2002). 1.1 Hierarchical structures in Performance Grammar PG's hierarchical component generates unordered trees by combining 3-tiered `mobiles ' called lexical frames. Figure 1 shows the eight lexical frames expressing how each of the words in Dutch example (1) is used . For predecessors of PG, see Kempen & Hoenkamp (1987) and De Smedt & Kempen (1990a, b). Ot

Abstract not found