Contextual Meta-Knowledge Acquisition from Corpora
Nigel Collier · 1996
This paper looks at the area of automatic acquisition of metaknowledge for the structuring of very large knowledge bases - (VLKB). It is argued that we will rediscover the need in Natural Language Processing (NLP) for such large knowledge bases and that one possible method for structuring them efficiently lies in associationbased statistics gathered from corpora. The discussion sets out the aims and objectives of a knowledge representation strategy. Key words: Knowledge representation, NLP, statistics. 1 Introduction The large monolithic lexical knowledge base is unfashionable. Toy systems, sublanguage systems, low level-analysis and non-linguistic systems are now at the forefront of Natural Language Processing (NLP) research. The reason for this is that language engineers have found it difficult to empower a computer with the same competence as a human for language analysis and production. The result has often been to avoid the issue rather than to solve it. This has led to NLP sys...