Schisma

Abstract: The analysis of natural language in the context of keyboard-driven dialogue systems is the central issue addressed in this paper. A module that corrects typing errors and performs domain-specifc morphological analysis has been developed. A parser for typed unification grammars is designed and implemented in C++; for description of the lexicon and the grammer a specialised specification language has been developed. It is argued that typed unification grammars and especially the newly developed specification language are convenient formalisms for describing natural language use in dialogue systems. Research on these issues is carried out in the context of the Schisma project, a research project of the Parlevink group in linguistic engineering; participants in Schisma are KPN Research and the University of Twente. The aims of the Schisma project are twofold: both the accumulation of knowledge in the field of computational linguistics and the development of a natural language interfaced theatre information and booking system is envisaged. The Schisma project serves as a testbed for the development of the various language analysis modules necessary for dialogue systems.

Abstract: This paper addresses the application of dialogue management techniques in a buyer-seller situation. In particular, we will describe the techniques used in the SCHISMA system, a system that provides the caller with information on theatre performances, and optionally "sells" the caller one or more tickets for a given performance.

Abstract: SCHISMA is a research project that is concerned with the development of a natural language accessible theatre information and booking system. In this project two research approaches can be distinguished. One approach is devoted to theoretical research in the areas syntax, semantics and pragmatics.Research on syntax has been conducted on a unifying parsing approach [7] , left and head corner grammars [6] , on stochastic context-free grammars [1] and on unification grammar parsing [2] [4] . Research on semantics and pragmatics has been conducted on dialogue act classification [3] and from a logical point of view [5] . The other approach is more practically oriented in that it is more focused on the realisation of a fully functional prototype of the SCHISMA system. To this purpose, in the past, the interface development system Natural LanguageTM has been used for implementing a SCHISMA prototype, a Wizard of Oz environment has been developed (with which a corpus of dialogues has been collected), an attempt has been made to develop a SCHISMA prototype for the WWW, and a prototype has been developed for education purposes1 . Our most recent achievement is a fully functional SCHISMA system (implemented in Java) based on a context-sensitive string rewrite mechanism. Although the system is primitive in nature and not necessarily built on linguistic principles (rather on intuitions), we believe its performance is such that the majority of users will be satisfied. Users will adapt to the system rather than reject it because of poor performance (when compared to human­ human dialogues). In the development of a spoken telephone service for the Dutch Public Transport Service a similar position is taken. In this short paper we will give an overview of the two approaches distinguished here as far as parsing is concerned. Section 2 presents work on classical unification-based parsing, section 3 discusses the rewrite and understand approach in some detail. Section 4 considers future developments and concludes the paper.