Structuring Terminology using Anal- ogy-Based Machine learning

In the field of computational terminology, in addition to work on term extraction, more and more research highlights the importance of structuring terminology, that is, finding and labeling the links between terminological units. Retrieving such relations between terms is usually undertaken using either “external” or “internal” methods (see Daille et al. (2004) for an overview). External methods rely on the (automatic) analysis of corpora to see what kind of words can be associated with a term in context (e.g. Claveau & L'Homme, 2004). Internal methods rely only on the form of the terms to make such associations. Some of this research relies heavily on the use of external knowledge resources (Namer & Zweigenbaum, 2004; Daille, 2003), which implies a lot of human intervention if the technique is defined for another domain or language. Others add little information and make the most of existing data, such as thesauri (Zweigenbaum & Grabar, 2000) or corpora (Zweigenbaum & Grabar, 2003) but aim to identify morphological families without distinguishing the semantic roles of the individual members.