Three-Level Knowledge Representation of Predicate-Argument Mapping for Multilingual Lexicons

Constructing a semantic concept hierarchy covers only part of the problem in building any knowledgebased natural language processing (NLP) system: Language understanding requires mapping from syntactic structures into conceptual (i.e. interlingua) representation (henceforth verb-argument mapping), while language generation requires the inverse mapping. Jacobs [13] proposed a solution in which he augmented his concept hierarchy in the knowledge base with verb-argument mapping information. While his solution works for monolingual understanding/generation, associating each semantic concept of a verb with English specific mapping information in the knowledge base is not ideal for an interlinguabased NLP system, since verb-argument mapping varies greatly among languages. In this paper, we discuss how the lexicon of our interlingua-based NLP system 1 abstracts the language-dependent portion of predicate-argument mapping information from the core meaning of verb senses (i.e. semantic concepts as defined in the knowledge base). We also show how we represent this mapping information in terms of cross-linguistically generalized mapping types called situation types and word sense-specific idiosyncrasies. This three-level representation enables us to keep the languageindependent knowledge base intact and make efficient use of lexical data by inheritance. We take advantage of large multilingual corpora to automatically acquire languageand domain-specific lexical information. In addition, this representation framework easily deals with some of the MT divergences which