Automatic Syllabification for Manipuri language

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 349–357, Osaka, Japan, December 11-17 2016. Automatic Syllabification for Manipuri language Loitongbam Gyanendro Singh, Lenin Laitonjam, Sanasam Ranbir Singh Computer Science and Engineering Department Indian Institute of Technology Guwahati, Assam, India Email: {gyanendrol9,lenin.lai,ranbir}@iitg.ernet.in Abstract Development of hand crafted rule for syllabifying words of a language is an expensive task. This paper proposes several data-driven methods for automatic syllabification of words written in Manipuri language. Manipuri is one of the scheduled Indian languages. First, we propose a language-independent rule-based approach formulated using entropy based phonotactic seg- mentation. Second, we project the syllabification problem as a sequence labeling problem and investigate its effect using various sequence labeling approaches. Third, we combine the effect of sequence labeling and rule-based method and investigate the performance of the hybrid approach. From various experimental observations, it is evident that the proposed methods outperform the baseline rule-based method. The entropy based phonotactic segmentation provides a word ac- curacy of 96%, CRF (sequence labeling approach) provides 97% and hybrid approach provides 98% word accuracy. 1 Introduction Manipuri language, one of the scheduled Indian languages, belonging to a Tibeto-Burman lan- guages family(Chelliah, 1990) is syllabic in nature. In general, a syllable follows onset-nucleus-coda (consonant-vowel-consonant) structure, where nucleus is the core component defined by a vowel. The preceding and following consonants defined by the onset and coda respectively may or may not be present in a syllable. However every syllable must have a nucleus. Formation of an onset, nucleus and coda, while producing an uninterrupted syllabic sound, greatly affects the pronunciation of a language. For various applications such as Text-to-Speech Synthesis (TTS) (Kishore and Black, 2003; Bellur et al., 2011), Automatic Speech Recognition (ASR) (Wu et al., 1998) etc., proper syllabification of a word is one of the important core issues, especially for syllabic Indian languages. Though there have been sev- eral studies in this direction for the rich resource Indian languages such as Hindi, Bengali, Tamil, Telegu, Malayalam, Marathi (Bellur et al., 2011; Narendra et al., 2011; Kurian et al., 2011), only very few stud- ies have been reported for low resource Manipuri language (Abbi and Awadhesh, 1985; Chelliah, 1990). To the best of our knowledge, no corpus is available for Manipuri language in UTF-8 to study various aspects of corpus-based linguistic understanding. This has motivated us to generate Manipuri corpus in UTF-8 and analyse various characteristics such as onset and coda distributions, syllable formulation and syllabification rules etc. Manipuri language is written generally using two scripts; Bengali scripts and Meitei Mayek 1. At present, majority of the Manipuri documents are written using Bengali scripts and, hence the investiga- tion in this paper focuses on Manipuri words written using Bengali script. Further, processing of Bengali scripts poses more challenges than processing Meitei Mayek because of the phoneme imbalance. Bengali script has 55 symbols to represent 38 phonemes in Manipuri language (Singh et al., 2007). One of the major challenges in syllabifying Indian languages such as …

Content truncated to 500 words