Statistical machine translation system and computational domain adaptation

Dunđer, Ivan. (2015). Statistical machine translation system and computational domain adaptation. PhD Thesis. Filozofski fakultet u Zagrebu, Department of Information Science
Department of Information Science > Chair of social-humanistic informatics.
(Poslijediplomski doktorski studij informacijskih i komunikacijskih znanosti) [mentor Seljan, Sanja].

Preview

PDF (Croatian) Download (2MB) | Preview

Abstract

Phrase-based statistical machine translation is one of possible automatic machine translation approaches. This work proposes methods for increasing the quality of machine translation by adapting certain parameters in the statistical machine translation model. The idea was to build phrase-based statistical machine translation systems for Croatian and English language. The systems were be trained for two directions, on two domains, on parallel corpora of different sizes and characteristics for Croatian-English and English-Croatian language pair, after which the tuning procedure was conducted. Afterwards, hybrid systems which combine features of both domains were investigated. Thereby the direct impact of domain adaptation on the quality of automatic machine translation of Croatian language was explored, whereas new findings can be utilised for building new systems. Automatic and human evaluation of machine translations were carried out, while obtained results were compared with results obtained from applying existing statistical machine translation web services.

Actions (login required)

View Item