Evaluation of the system for extraction of multi-word expressions and prediction of their translation

Abstract

During PARSEME meeting in Athens, we proposed an unsupervised learning system intended to extract all the candidate multiword expressions from sentence aligned parallel corpora and to predict their translations. The system was created using the parallel corpora of Orwell’s 1984, which is a part of Multext-East project. In this paper we evaluate the efficiency of the system and try to determine the major drawbacks leading to wrong expressions and inaccurate translation. They will be illustrated with the examples of a bilingual translation of Orwell’s 1984.