Constraining Lexical Selection Across Languages Using TAGs

Abstract

Lexical selection in Machine Translation consists of several related components. Two that have received a lot of attention are lexical mapping from an underlying concept or lexical item, and choosing the correct subcategorization frame based on argument structure. Because most MT applications are small or relatively domain specific, a third component of lexical selection is generally overlooked - distinguishing between lexical items that are closely related conceptually. While some MT systems have proposed using a 'world knowledge' module to decide which word is more appropriate based on various pragmatic or stylistic constraints, we are interested in seeing how much we can accomplish using a combination of syntax and lexical semantics. By using separate ontologies for each language implemented in FB-LTAGs, we are able to elegantly model the more specific and language dependent syntactic and semantic distinctions necessary to further filter the choice of the lexical item.