Towards Context-adaptive Natural Language Processing Systems, in Computational Linguistics.pdfVIP

Towards Context-adaptive Natural Language Processing Systems Robert Porzel and Michael Strube European Media Laboratory GmbH Schloß-Wolfsbrunnenweg 33 69118 Heidelberg, Germany {Robert.Porzel jMi chael.Strub e}@eml.v illa -bosch. de Abstract In this paper we will discuss several issues and requirements for enabling natural language processing systems to become context-adaptive. Given the fact that emerging systems feature speaker indepen- dent continuous speech recognition restricted to individual domains and are equipped with syntactic and semantic parsers for understanding input from these domains, we should begin to envision open multi-domain natural language processing systems. We will address the following issues arising in such open systems: How to divide natural language processing modules in context-variant and context-invariant components and how to enable the ensuing systems to detect contextual changes for processing the context-variant phenomena. 1 Introduction The complete understanding of naturally occurring discourse is still an unsolved task in computational linguistics. Early approaches attempting to solve this task produced only toy systems. Their aim was to cope with special linguistic problems and/or to model particular cognitive capacities of natural language users. Broad coverage of syntactical constructions, lexical information sources and semantic/pragmatic behavior was not the primary concern and also far outside the scope and capabilities of these systems. Today’s linguistic development environments, representations and methodologies have shown that