Automatic Generation of Term Translations
Estelle Maryline Delpech · 2014
This chapter presents translation generation methods. Contrary to the approaches which try to align predefined terms on the base of distributional similarity, generation consists of generating a translation using a knowledge base of translation equivalences and of possible realizations in the target language. One possible approach is based on compositional semantics. The unit to be translated is decomposed in meaning-bearing sub-units, and these sub-units are translated and recomposed so as to create a term in the target language. A second type of approach, called data driven approach, is to rely on empirical knowledge. These approaches consider the unit to be translated more as a sequence of characters to be rewritten than as a linguistic unit. The translation is carried out to translation models learned from examples of translation pairs. The chapter lists strengths and weaknesses of the distributional, compositional and data-driven approaches.