THE BERMUDA-TRIANGLE - NATURAL-LANGUAGE SEMANTICS BETWEEN LINGUISTICS, KNOWLEDGE REPRESENTATION, AND KNOWLEDGE PROCESSING

Linguistic parameters alone cannot determine the interpretation of natural language utterances. They can only constrain their interpretation and must leave the rest to other knowledge sources and other processes: language understanding is not just a matter of knowing the language, but also to a considerable degree a matter of logical inference and world knowledge. This is no news as far as the interpretation of referential expressions is concerned. Predicate expressions, however, tend to be treated as if they were functional or relational constants that are directly interpreted with respect to a model. In this paper an attempt is made to treat them too as referential. The only real difference is that the referents of predicate expressions are of a different type: concepts rather than first-order objects. This generalized notion of reference gives us not only a natural way of understanding the interaction of knowledge representation and knowledge processing on the one hand and linguistic processing on the other, but also opens up a perspective for the modularization of Natural Language Understanding (NLU) systems that provides for a very high degree of independence of the more strictly linguistic component from the specific tasks and domains of a particular application. The overall result should be a less language-dependent knowledge representation and less domain-dependent linguistic components, i.e., overall improved portability of the modules.