Customizing meaning: building domain-specific semantic representations from a generic lexicon

Language input to practical dialogue systems must be transformed into a semantic representation that is customized for use by the back-end domain reasoners. At the same time, we want to keep front-end system components as domain independent as possible for easy portability across multiple domains. We propose a transparent way to achieve domain specificity from a broad-coverage domain-independent parser. We maintain a domain-independent ontology and define a set of mappings from it into a domain-specific knowledge representation. We use the mappings to customize the semantic representations output by the parser for the reasoners, and to specialize the lexicon to the domain, which improves parsing speed and accuracy. This method facilitates our approach to instances of semantic type coercion common in our domains by combining lexical representations with domain-specific constraints on interpretation.