Choosing the Word Most Typical in Context Using a Lexical Co-occurrence Network

Choosing the Word Most Typical in Context Using a Lexical Co-occurrence Network Philip Edmonds Department of Computer Science, University of Toronto Toronto, Canada, M5S 3G4 pedmonds©cs, toronto, edu Abstract This paper presents a partial solution to a com- ponent of the problem of lexical choice: choos- ing the synonym most typical, or expected, in context. We apply a new statistical approach to representing the context of a word through lexical co-occurrence networks. The imple- mentation was trained and evaluated on a large corpus, and results show that the inclusion of second-order co-occurrence relations improves the performance of our implemented lexical choice program. 1 Introduction Recent work views lexical choice as the process of map- ping fi'om a set of concepts (in some representation of knowledge) to a word or phrase (Elhadad, 1992; Stede, 1996). When the same concept admits more than one lexicalization, it is often difficult to choose which of these 'synonyms' is the most appropriate for achieving the desired pragmatic goals: but this is necessary for high- quality machine translation and natural language genera- tion. Knowledge-based approaches to representing the po- tentially subtle differences between synonyms have suf- fered from a serious lexical acquisition bottleneck (Di- Marco, Hirst, and Stede, 1993; Hirst, 1995). Statistical approaches, which have sought to explicitly represent differences between pairs of synonyms with respect to their occurrence with other specific words (Church et al., 1994), are inefficient in time and space. This paper presents a new statistical approach to mod- eling context that provides a preliminary solution to an important sub-problem, that of determining the near- synonym that is most typical, or expected, if any, in a given context. Although weaker than full lexical choice, because it doesn't choose the 'best' word, we believe that it is a necessary first step, because it would allow one to determine the effects of choosing a non-typical word in place of the typical word. The approach relies on a generalization of lexical co-occurrence that allows for an implicit representation of the differences between two (or more) words with respect to any actual context. For example, our implemented lexical choice program selects mistake as most typical for the 'gap' in sen- tence (1), and error in (2). (1) However, such a move also would run the risk of cutting deeply into U.S. economic growth, which is why some economists think it would be a big {error I mistake [ oversight}. (2) The {error I mistake t oversight} was magnified when the Army failed to charge the standard percentage rate for packing and handling. 2 Generalizing Lexical Co-occurrence 2.1 Evidence-based Models of Context Evidence-based models represent context as a set of fea- tures, say words, that are observed to co-occur with, and thereby predict, a word (Yarowsky, 1992; Golding and Schabes, 1996; Karow and Edelman, 1996; Ng and Lee, 1996). But, if we use just the context surrounding a word, we might not be able to build up a representation satisfac- tory to uncover the subtle differences between synonyms, …

Content truncated to 500 words