Lexical Acquisition

This paper describes several aspects of the process of lexical acquisition for one of the most comprehensive applications in natural language processing knowledge-based machine translation. Specifically, the paper concentrates on those components of the lexicon that centrally relate to the task of processing meaning. The usability of the comprehensive ontological semantic lexicon exceeds NLP and extends to theoretical descriptive linguistics and its other applications. Acquisition of lexical knowledge is a crucial component of building natural language processing applications. The requirements for lexical knowledge differ across different applications. Some of them require only a small amount of information. For example, a lexicon supporting a spelling checker must, at a minimum, only list all the possible word forms in a language. Some other applications require vast quantities of diverse kinds of data. For example, a comprehensive text analysis system may require information about word boundary determination (useful for compounding languages, such as Swedish, where the lexical entries would often match not complete words but parts of compound words); information about inflectional and derivational morphology, syntax, semantics and pragmatics of a lexical unit as well as possible connections among knowledge elements at these various levels. In this document, we describe lexical acquisition for a semantic analyzer viewed as a component of a knowledge-based machine translation system. The lexicon whose construction we illustrate has been and still is under development at NMSU CRL. Work on this lexicon has been an integral part of the development of ontological semantics , a semantic theory for computational applications and a practical approach to automatic treatment of meaning in texts. Over the years, our work involved, on the one hand, the development of general ideas about lexicon structure and content and, on the other, development of actual semantic lexicons for English, Spanish and Chinese. The utility of these lexicons extends beyond natural language processing to the needs of language description proper. The lexicons are flexible as to the grain size of description, thus absorbing and actually implementing the idea of variable-depth semantics (Nirenburg and Raskin 1986, 2001). 1. The Lexicon in the Scheme of KBMT Things In any natural language processing system, the lexicon supports the processes of analysis and generation of text or spoken language at all levels tokenization (that is, roughly, lexical segmentation), part-of-speech tagging and morphological analysis, proper-name recognition, syntactic, 1. Also of the Interdepartmental Program in Linguistics and its Natural Language Processing Laboratory, Purdue University, West Lafayette, IN 47907-1356.