Evaluation of Domain Adaptation Approaches to Improve the Translation Quality


Yildirim E. , Tantuğ A. C.

6th International Conference on Computational Collective Intelligence (ICCCI), Seoul, Güney Kore, 23 - 26 Eylül 2014, cilt.572, ss.15-26 identifier identifier

  • Cilt numarası: 572
  • Doi Numarası: 10.1007/978-3-319-10774-5_2
  • Basıldığı Şehir: Seoul
  • Basıldığı Ülke: Güney Kore
  • Sayfa Sayıları: ss.15-26

Özet

This paper focuses on the usage of different domain adaptation methods to build a general purpose translation system for the languages with limited parallel training data. Several domain adaptation approaches are evaluated on four different domains in the English-Turkish SMT task. Our comparative experiments show that the language model adaptation gives the best performance and increases the translation success with a relative 9.25% improvement yielding 29.89 BLEU points on multi-domain test data.