Towards Automatic Semantic Classification for a Natural Language Understanding System

The knowledge acquisition component of the natural language understanding (NLU) system VIE-LANG with self-adjusting capabitilies is described. The system VIE-LANG is heavily data-driven and uses a morphological lexicon coupled with a syntactic/semantic lexicon to model language specific and real-world knowledge. The acquisition paradigm adopted here uses the knowledge thus represented, together with flexibly structured heuristics, to aid in enlarging the lexica. Methods for the automatic modification of the acquisition heuristics are investigated and examined for their usefulness and workability. This is done experimentally on a ‘skeleton-system’, implemented in INTERLISP. The heuristic rules are implemented declaratively and thus are decoupled from the program to aid easy modification. Concepts developed in linguistics regarding word formation play an extensive role in the classification heuristics used.