Making term extraction tools usable

Abstract

The paper reviews the extraction of terminology from corpora. It identifies three possible applications, terminology, translation, and retrieval. They differ in the requirements for relevancy of terms to be extracted. Standard evaluation methods based on recall and precision critically depend on the notion of relevancy of a term, which is questionable and possibly outweighed in favour of criteria of usability in practical applications. Two tools, TermExtract and BiExtract, are presented as examples of the integration of extraction tools into different workflows.