Construction-Based MT Lexicons

Abstract

This paper presents a novel view of the boundary between the generalizable and the idiosyncratic in MT lexicons. We argue that the domain of the idiosyncratic should, in fact, be broader than in most current approaches. While at present most MT systems involve phrasal lexicons, these typically contain terminology from a particular field. In order to facilitate naturalness of translation, specifically, to carry the level of “conventionality” of meaning expression across languages, it becomes necessary to use the concept of a grammatical construction, a (possibly, discontiguous) syntactic structure or productive syntactic pattern whose meaning it is often impossible to derive solely based on the meanings of its components. Identification of constructions allows an MT system to select the most appropriate conventional way of expressing a meaning from among the available ways. After discussing the notion of construction, we suggest the format for a construction lexicon for a knowledge-based MT system.