Evaluation of SMT in localization to under-resourced inflected language

Evaluation of SMT in localization to under-resourced inflected language Raivis Skadi��, ������������, Inguna Skadi�����������������jevs Tilde SIA Vienibas gatve 75A, Riga Latvia, LV-1004 {raivis.skadins,maris.purins,inguna.skadina,andrejs}@tilde.lv �Abstract Although Machine Translation is very popular for personal tasks, its use in lo- calization and other business applications is still very limited. The paper presents an experiment on the evaluation of an English-Latvian SMT system integrated into SDL Trados which has been used in an actual localization assignment by a professional localization company. We show that such an integrated localization environment can increase the productivi- ty of localization by 32.9% without a critical reduction in quality. 1 Introduction The rapid advancement of machine translation technologies, especially statistical machine trans- lation (SMT), draws attention to questions of its usability in practical, business-oriented applica- tions. One promising area where industry could ben- efit from SMT is localization. Translation memo- ries (TM) have been in use in localization for more than 10 years to increase productivity. Translation memories can significantly improve the efficiency of localization if the new text is similar to the previously translated material. However, if the text is in a different domain than the TM or in the same domain from a different customer using different terminology, support from the TM is minimal. For this reason the lo- calization industry is increasingly interested in combining translation memories with machine ��2011 European Association for Machine Translation. translation solutions adapted for the particular domain or customer requirements. For the development of MT in the localization and translation industry, huge pools of parallel texts in a variety of industry formats have been accumulated, but the use of this data alone does not fully utilize the benefits of modern MT tech- nology. At the same time, this industry experi- ences a growing pressure on efficiency and per- formance, especially due to the fact that volumes of texts that need to be translated are growing at a greater rate than the availability of human translation, and translation results are expected in real-time. Increasing the efficiency of the translation process without a degradation of quality is the most important goal for a localization service provider. In this paper we present an experiment on the application of an English-Latvian SMT in locali- zation through the integration of MT into the SDL Trados 2009 translation environment. Simi- larly to Plitt and Masselot (2010) we measure performance of a translator translating with and without MT. In addition a quality assessment for texts was performed according to the standard internal quality assessment procedure. 2 Related work Although the idea to use MT in the localization process is not new, it has not been explored widely. Different aspects of post-editing and ma- chine translatability have been researched since the 90-ies. A comprehensive overview of re- search on machine translatability and post- editing ���� ����� ��������� ��� �������� ������. She also analyses the potential of a keyboard- monitoring program and Choice Network Analy- sis for measuring the effort involved in post- editing MT output. However this work mainly …

Content truncated to 500 words