Using Weighted Abduction to Align Term Variant Translations in Bilingual Texts

In this paper we describe a method for detecting terminological variants and their translations in bilingual texts. Our approach is based on abductive reasoning and combines various monolingual and bilingual resources. A small scale experiment shows that precision and recall increase when using more resources and when the resources interfere in a less restricted way. In order to tune our system, we develop a weighing strategy based on the precision of term translation alignments in a reference text. We feed back these weights into the linguistic resources and repeat the experiment. The results show that precision values are considerably higher when weighing term alignments.