Skip to Main content Skip to Navigation

Traduction assistée par ordinateur et corpus comparables : contributions à la traduction compositionnelle

Estelle Delpech 1
1 équipe TALN
LINA - Laboratoire d'Informatique de Nantes Atlantique
Abstract : Our work deals with the extraction of bilingual lexicons from comparable corpora with an application to specialized translation. We started by evaluating classical methods based on the distributional hypothesis (the more two terms appear in similar contexts, the more likely they are translations of each other) in a user-oriented fashion. This evaluation raised the fact that translators feel very uncomfortable with this kind of lexicon: they feel correct translations are uneasy to spot in the lists of candidate translations and would rather use a smaller lexicon but with higher precision rates. Based on this observation, we turned to another approach for term translation which has been recently and successfully experimented on comparable corpora and produce lexicons that meet the demands of the translators: compositional translation. In this framework, the translation of a term is composed of the translation of its parts. We concentrated on the translation of monolexical terms : the source term is decomposed into morphemes, morphemes are translated into the target language and recomposed as a target term. We investigated three lines of research: generation of fertile translations (cases in which the target term has more lexical words than the source term), independence to morphological structure and candidate translation ranking.
Complete list of metadatas
Contributor : Estelle Delpech <>
Submitted on : Wednesday, March 5, 2014 - 7:44:18 AM
Last modification on : Friday, October 23, 2020 - 4:52:06 PM
Long-term archiving on: : Thursday, June 5, 2014 - 10:36:30 AM


  • HAL Id : tel-00905930, version 1



Estelle Delpech. Traduction assistée par ordinateur et corpus comparables : contributions à la traduction compositionnelle. Informatique et langage [cs.CL]. Université de Nantes, 2013. Français. ⟨tel-00905930⟩



Record views


Files downloads