The Subworld Concept Lexicon and the Lexicon Management System

Natural language processing systems require three different types of lexicons: the concept lexicon that describes the (sub)world ontology and the analysis and generation lexicons for natural languages. We argue that the acquisition of the concept lexicon must precede any lexical work on natural language and that a comprehensive lexicon management system (LMS) is necessary for lexicon acquisition in large-scale applications. We describe the interactive concept lexicon acquisition module of the LMS for TRANSLATOR, a knowledge-based, sublanguage-oriented machine translation project. This project belongs to two of the fastest growing fields of computational linguistics and artificial intelligence in general: the lexicon and knowledge acquisition for AI systems. The work in lexicon has centered on a) studies concerned with the utilization of conventional human-oriented dictionaries, newly available in machine-readable form, for computational tasks (e.g., Amsler 1984a,b; Chodorow et al. 1985; Ahlswede 1985; Marko-witz et al. 1986) improving the ancillary capabilities for lexicon systems, such as, for instance, morphological processors and descriptions (e.g., Nirenburg and Ben Asher 1984; Byrd et al. 1986; Boguraev et al. 1987); c) hand-building of lexicons necessary for natural language systems, often with considerations about extensibility (e.g., Zernik and Dyer 1985; Bessemer and Jacobs 1987). An interesting perspective on the field is given in Miller (1985). Knowledge acquisition is a central topic in AI and expert systems. A number of systems exist for assistance in acquiring specialized knowledge for natural language processing. Thus, TEAM (Grosz et al. 1985), TEL1 (Ballard and Stumberger, 1986), IRACQ (Ayuso et al. 1987) and some others are devoted to facilitating customizations in the framework of database query systems; INKA (Phillips et al. 1986) and CYC (Lenat et al. 1986) are examples of knowledge acquisition systems aimed at expert system design. We would like to address some important method-ological and strategic points in providing the lexicon support in the context of a text processing system, such as the machine translation system TRANSLATOR (Ni-renburg et al. 1987). A natural language lexicon is a necessary part of any natural language processing (NLP) system. Building such a lexicon is an important Rnowledge acquisition task. The format of the lexicon has been the subject of substantial number of research projects, both in linguistics and in artificial intelligence (AI). In linguistic semantics , the efforts have focused on the formal notation for representing semantic knowledge with a special emphasis on the involved categories and rules. The projects have ranged from simple …