Unificational Combinatory Categorial Grammar. Combining Information Structure and Discourse Representations

Unificational Combinatory Categorial Grammar: Combining Information Structure and Discourse Representations Maarika Traat The University of Edinburgh 2 Buccleuch Place Edinburgh EH8 9LW, United Kingdom, M.Traat@ed.ac.uk Johan Bos The University of Edinburgh 2 Buccleuch Place Edinburgh EH8 9LW, United Kingdom, jbos@inf.ed.ac.uk Abstract In this paper we present a grammar formalism that combines the insights from Combinatory Categorial Grammar with feature structure unification. We show how information structure can be incorporated with syntactic and semantic representations in a principled way. We focus on the way theme, rheme, and focus are integrated in the compositional se- mantics, using Discourse Representation Theory as first-order semantic theory. UCCG can be used for parsing and generating prosodically annotated text, and therefore has the potential to advance spoken dialogue systems. 1 Introduction The integration of information structure (the way information is “packaged” in a sentence) in practical formalisms in computational lin- guistics has long been ignored. There are two main reasons for this: (1) formalisations of in- formation structure often use variants of higher- order logic to characterise its semantic impact (Krifka, 1993; Kruijff-Korbayova, 1998; Steed- man, 2000), which limits the use of inference in practice (Blackburn and Bos, 2003); and (2) the effect of information structure on the com- positional semantics of an utterance is rarely worked out in enough detail useful for compu- tational implementation. On the other hand, exploring information structure in spoken dia- logue systems is becoming realistic now because of the recent advances made in text-to-speech synthesisers and automated speech recognisers — hence there is a growing need for computa- tional implementations of information structure in grammar formalisms. In this paper we present Unificational Com- binatory Categorial Grammar (UCCG), which integrates aspects of Combinatory Categorial Grammar (Steedman, 2000), Unification Cat- egorial Grammar (Zeevat, 1988; Calder et al., 1988), and Discourse Representation Theory (Kamp and Reyle, 1993). It offers a compo- sitional analysis of information structure, a se- mantics compatible with first-order logic, and a computational implementation for a fragment of English, using unification in combining gram- matical categories. As we will show, this makes UCCG easy to implement, and allows us to in- tegrate prosodic information in the semantics in a transparent and systematic way. Although based on first-order logic, we claim that UCCG has enough expressive power to model informa- tion structure such that it has the potential to improve speech generation with context appro- priate intonation in spoken dialogue systems. 2 Background Categorial Grammars (CG) (Wood, 2000) are lexicalised theories of grammar. The notion of “category” refers to the functional type that is associated with each entry in the lexicon which determines the ability of a lexical item to com- bine with other lexical items. CGs also have a set of rules defining the syntactico-semantic operations that can be performed on the cate- gories. Combinatory Categorial Grammar (CCG) is a generalisation of CG (Steedman, 2000). While the pure CG only involved functional applica- tion rules for combining categories, CCG intro- duces several additional combinatory rules for both syntactic and semantic …

Content truncated to 500 words