Discourse and Meaning

Wolfgang Teubert
Institute für deutsche Sprache
Mannheim, Germany
e-mail: wolfgang.teubert@ids-mannheim.de

The traditional AI and MT approach deals with meaning in terms of rule-based commutation and permutation of uninterpreted symbols, i.e. concepts. But this approach reduces semantics to syntax and is desirable only for formal (and perhaps "controlled") languages. Natural languages are symbolic, and symbols cannot be processed in an algorithmic way; neither can they be mapped directly onto extralinguistic entities (classes or individuals). Instead they have to be interpreted. Interpretation, however, presupposes understanding, and understanding presupposes intentionality, a property computers do not have. If meaning in natural language cannot be reduced to concepts and their conceptual relationships in a "language-independent" ontology, we have to find it elsewhere. Texts contain interpretations of symbols, and corpus linguistics is developing operations to extract the meaning of text elements from texts. The traditional analytic approach of breaking down a sentence into atomistic categories (most of which are text-external) is thus complemented by a probabilistic analysis based on lists and statistics.

Together, these two approaches are able to cope with the inherent fuzziness of natural language.

See previous, next abstract.