Accession Number:



Using Case-Based Reasoning in Natural Language Processing

Descriptive Note:

Rept. for 15 Mar 92-14 Mar 93,

Corporate Author:


Personal Author(s):

Report Date:


Pagination or Media Count:



A variety of problems addressed in natural language processing NLP make this area ripe for hybrid system designs and approaches based on multiple technologies. One particularly promising crossover is the application of case- based reasoning CBRto NLP. There are two important facts that make CBR an obvious candidate for innovative investigations in NLP. First, most decision processes in NLP are characterized by shades of grey and different ways of weighting preferences rather than black and white absolutes or right and wrong answers. This holds true for the lowest levels of lexical ambiguity as well as the highest levels of inference and reasoning. The affinity in language for relative preference rather than hard and fast absolutes is fully consistent with CBR capabilities that can produce multiple solutions and then assess each one in accordance with multiple dimensions for evaluation. Second, many aspects of NLP can be studied with the aid of large online text corpora. Full text databases are now being constructed using optical scanners and commercial OCR technologies at cost levels that make it possible for individual labs to customize their own online corpora. Access to large amounts of real language make it possible for NLP researchers to break out of the traditional linguistic paradigm of contrived examples in order to study language processing from a more seriously empirical perspective with relicable experiments and corpus-driven processing techniques. At the University of Massachusetts, we have demonstrated the utility of NLP research lexical knowledge acquisition and content-based document classification. We will describe each of these research efforts in turn.

Subject Categories:

  • Information Science
  • Linguistics
  • Computer Systems

Distribution Statement: