A Large-scale Study of Statistical Machine Translation Methods for Khmer Language

A Large-scale Study of Statistical Machine Translation Methods for Khmer Language Ye Kyaw Thu†, Vichet Chea‡, Andrew Finch†, Masao Utiyama † and Eiichiro Sumita† †Advanced Speech Translation Research and Development Promotion Center, NICT, Kyoto, Japan ‡Research and Development Center, NIPTICT, Cambodia {yekyawthu, andrew.finch, multiyama, eiichiro.sumita}@nict.go.jp vichet.chea@niptict.edu.kh Abstract This paper contributes the first published evaluation of the quality of automatic translation between Khmer (the official language of Cambodia) and twenty other languages, in both directions. The ex- periments were carried out using three different statistical machine translation approaches: phrase-based, hierarchical phrase-based, and the operation sequence model (OSM). In addition two different segmentation schemes for Khmer were studied, these were syllable segmentation and supervised word segmentation. The results show that the highest quality ma- chine translation was attained with word segmentation in all of the experiments. Furthermore, with the exception of very distant language pairs the OSM approach gave the highest quality translations when measured in terms of both the BLEU and RIBES scores. For distant languages, our results showed a hierarchical phrase-based approach to be the most effective. An analysis of the experimental results indi- cated that Kendall’s tau may be directly used as a means of selecting an appropriate machine translation approach for a given language pair. 1 Introduction Natural language processing for the Khmer lan- guage is currently at an early stage and linguis- tic resources for the language are scarce. As far as the authors are aware there has been only one published work on Khmer statistical machine translation (Surabaya Jabin and Sok- phyrum, 2013). In the paper a step-by-step pro- cedure for implementing an English-to-Khmer machine translation system using the Do Moses Yourself (DoMY) Community Edition1 was de- scribed. The system was developed using a small parallel corpus of 5,734 sentence pairs of English-Khmer. The paper mentions that the system obtained good performance compared with Google Translate for in-domain sentences, however, no numerical evaluation of the system was given. The main contribution of this paper, is the first large-scale study of Khmer statistical ma- chine translation. 40 language pairs was used in the experiments, and translation quality was evaluated using both the BLEU and RIBES evaluation metrics. We developed the SMT sys- tems using a parallel corpus of 162,121 sentence pairs for each language pair and studied the ma- chine translation performance using three differ- ent SMT techniques (phrase-based SMT, hierar- chical phrase-based SMT, and the operation se- quence model), using two different segmentation schemes for Khmer. The structure of the paper is as follows. In the next section we briefly introduce the Khmer language, outline the approaches taken so far to Khmer word segmentation, and describe the two approaches we have chosen to examine in this study. These are a simple approach that divides Khmer into its component syllables, and a more sophisticated supervised word segmen- 1http://www.precisiontranslationtools.com/ ?option=com_content&view=article&id=1&Itemid=22 PACLIC 29 259 29th Pacific Asia Conference on Language, Information and Computation pages 259 - 269 Shanghai, China, October 30 - November 1, 2015 Copyright 2015 by Ye …

Content truncated to 500 words