Task-Oriented Modeling for Natural Language Processing Systems
Gudrun Klose · 1993
In this work, I focus on the crucial significance of task orientation as a design criterion for knowledge bases in natural language processing systems, a perspective which I realized with the implementation of the LEU/2 knowledge base for text understanding. The approach outlined in this thesis transcends current contributions in the field of Knowledge Modeling by offering a systematic account of modeling factors. On this basis, I give insights intodesign decisions and the access to knowledge elements during system performance by means of examples from three different natural language processing applications. Starting with a discussion of the key role of task orientation at the intersection of cognitive science, epistemology, and computational linguistics as related disciplines, I continue with the outline of a differentiated task-based description scheme for knowledge base contents and functionalities. For this purpose, I draw on elements of a methodology originally developed within the field of Open Distributed Processing. The notions of aspects and views are tailored to theanalysis of knowledge bases in natural language processing systems. The resulting model of task orientation is validated by applying it to the respective knowledge bases of the LEU/2 system (IBM Scientific Center), the FAST machine translation system (Technical University Berlin), and the PENMAN natural language generation system (Information Sciences Institute, Los Angeles). I conclude by sketching future perspectives for a constructive modeling methodology, and by characterizing ongoing research efforts concerning the investigated knowledge bases in terms of the introduced methodological aspects.