A Myanmar (Burmese)-English Named Entity Transliteration Dictionary

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2980–2983 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2980 A Myanmar (Burmese)-English Named Entity Transliteration Dictionary Aye Myat Mon†‡, Chenchen Ding†*, Hour Kaing†, Khin Mar Soe‡, Masao Utiyama†, Eiichiro Sumita† † Advanced Translation Technology Laboratory, ASTREC, NICT 3-5 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0289, Japan ‡ Natural Language Processing Lab., University of Computer Studies, Yangon, Myanmar † {ayemyatmon, chenchen.ding, hour_kaing, mutiyama, eiichiro.sumita}@nict.go.jp ‡ {ayemyatmon.ptn, khinmarsoe}@ucsy.edu.mm Abstract Transliteration is generally a phonetically based transcription across different writing systems. It is a crucial task for various down- stream natural language processing applications. For the Myanmar (Burmese) language, robust automatic transliteration for borrowed English words is a challenging task because of the complex Myanmar writing system and the lack of data. In this study, we construct- ed a Myanmar-English named entity dictionary containing more than eighty thousand transliteration instances. The data have been released under a CC BY-NC-SA license. We evaluated the automatic transliteration performance using statistical and neural network- based approaches based on the prepared data. The neural network model outperformed the statistical model significantly in terms of the BLEU score on the character level. Different units used in the Myanmar script for processing were also compared and discussed. Keywords: Myanmar (Burmese), named entity, transliteration, machine translation, neural network 1. Introduction Transliteration is the task of transcribing words from a source script to a target script. Generally, the transcription is phonetically based. Transliteration processing is im- portant for many downstream natural language processing (NLP) tasks, such as machine translation and information retrieval. In recent years, general sequence-to-sequence processing techniques for NLP tasks have been developed significantly. However, a lack of resources is still a prob- lematic issue for many understudied languages. In this study, we focus on transliteration between Myan- mar (Burmese) and English. To facilitate the application of data-driven approaches, we manually collected a dic- tionary containing more than eighty thousand Myanmar- English transliteration instances. The data have been re- leased under a CC BY-NC-SA license for research pur- poses. 1 Based on the dictionary, we conducted experi- ments on automatic transliteration between Myanmar and English. Specifically, we conducted experiments using two neural network (NN)-based approaches: the Trans- former model using the OpenNMT system2 (Vaswani et al., 2017; Klein et al., 2017) and a joint agreement bidi- rectional long short-term memory (LSTM)-based recur- rent NN (RNN) using the JANUS3 tool (Liu et al., 2016). A traditional phrase-based statistical machine translation (PBSMT) system using the Moses4 toolkit (Koehn et al., 2007) was set as a baseline. The experimental results were evaluated using the BLEU score (Papineni et al. 2002) on the character level. The experimental approaches per- formed well on transliteration tasks. The NN-based ap- proaches outperformed the traditional PBSMT by large gains. The effect of using units at different granularities in the Myanmar script was also investigated. To the best of our knowledge, this study is the first systematic work on the topic …

Content truncated to 500 words