Lexical and Conceptual Structure for Knowledge-Based Machine Translation

The most salient feature of a knowledge-based machine translation system is its reliance on understanding the semantics (and, if possible, pragmatics) of the input text as the prerequisite for successful generation of its target language correlate. Meaning extraction is understood as representation in a specially designed artificial meaning representation language, interlingua. In the knowledge-based machine translation project at Carnegie Mellon University the result of meaning analysis is called interlingua text or ILT. An ILT has a propositional-semantic and a pragmatic component. The former is produced by instantiating tokens of concepts from an underlying ontological and domain model (called concept lexicon in the KBMT-89 machine translation system, an earlier system we developed, see Goodman, 1989). The latter is a set of structures determined through various lexical and syntactic clues in the input, largely independently from the domain model.