Automatic acquisition of domain-specific morphological resources from thesauri
Natalia Grabar, Pierre Zweigenbaum · 2000
These is a growing body of evidence that morphological knowledge, beyond stemming methods, can help Information Retrieval tasks. Automated acquisition methods have been devised to learn morphological knowledge from corpora and thesauri, and show how corpus- or domain-specific morphological knowledge can enhance information retrieval results. In medicine, compounding and derivation are commonly used to form new words (e.g., "neo-classical compounds"). However, although there exist many large medical thesauri, medical morphological resources other than for inflection remain scarce in most languages. We therefore designed a method that learns morphological knowledge from an input thesaurus and identifies morphological families of word forms. It can work with no a priori linguistic resources, and has been applied to several languages (French, English and Russian). Applied to a medical thesaurus in these three languages, it produces morphological families with a precision ranging ...