The medical domain is characterized by a rich terminology.  Capturing this terminology, with its variants and relations, provides a key asset for medical language and information processing.  While English medical terms are extensively represented and linked in the UMLS Metathesaurus®, the terminological resources available to researchers or to applications in other languages are much more limited.  A similar situation holds for lexical resources.  This situation motivated work in our group on acquiring lexical and terminological resources for French medical language.

I will describe acquisition methods and experiments:

• to learn lexical relations:
  • within a given language, morphological relations between words based on existing thesauri or based on text corpora;
  • between two languages, identifying translational equivalents through transducer induction, morphological segmentation, or based on parallel or comparable corpora;
• to identify term variants:
  • within a given language, through morphological and syntactic relations;
  • between two languages, based on parallel corpora.

Some of this work was performed within projects UMLF (lexicon) and VUMeF (terminology), while I was at Assistance Publique - Hôpitaux de Paris and at Inserm U729.