Abstract
We discuss and solve the task of Khmer name Romanization. Although several standard Romanization systems exist for Khmer, conventional transcription methods are applied prevalently in practice. These are inconsistent and complicated in some cases, due to unstable phonemic, orthographic, and etymological principles. Consequently, statistical approaches are required for the task. We collect and manually align 7, 658 Khmer name Romanization instances. The alignment scheme is designed to reach a precise, consistent, and monotonic correspondence between the two different writing systems on grapheme level, through which various machine learning approaches are facilitated. Experimental results demonstrate that standard approaches of conditional random fields and support vector machine supervised by the manual alignment achieve a precision of .99 on grapheme level, which outperforms a state-of-the-art recurrent neural network approach in a pure sequence-to-sequence manner. The manually aligned data have been released under a license of CC BY-NC-SA for the research community.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
- 2.
Some diacritics always stand for one vowel. There are also diacritics serving as “shifter” to switch the series of certain consonant letters, which affect the vowel sound of other diacritics.
- 3.
Actually, Khmer script has this diacritic, called viriam, but uses obsoletely.
- 4.
The original names are western names in the Thai-to-English task, which may make the grapheme correspondence more varied and inconsistent.
- 5.
Generally, the diacritics in an abugida system are observed as vowel shifters to change the inherent vowel. From the viewpoint of this study, the diacritics are actually treated as vowel notes and the inherent vowel is treated as one with a zero-alternant form, which the insertion processing makes explicit.
- 6.
and 
are diacritics for 
and 
, respectively, and 
and 
are followed by staking operators. So these four consonant letters are not bare. - 7.
As the task is to transform Khmer script to Latin script, the graphemes are not guaranteed to be single letters on the Romanization side, e.g., the
corresponds to CH here. - 8.
The stacking operator is always aligned to a silent placeholder.
- 9.
Spaces are not counted because they are always maintained constant. Stacking operators as well as other silent Khmer letters are counted.
- 10.
Spaces are taken as one character in the BLEU calculation.
- 11.
A Python implementation for the rule-based transcription with different layers of rules is available at http://www2.nict.go.jp/astrec-att/member/mutiyama/software.html.
- 12.
An open-sourced tool is available at https://github.com/lemaoliu/Agtarbidir.
- 13.
- 14.
and 
are diacritics for 
and 
, respectively, and 
and 
are followed by staking operators. So these four consonant letters are not bare.
corresponds to References
Banchs, R.E., Zhang, M., Duan, X., Li, H., Kumaran, A.: Report of NEWS 2015 machine transliteration shared task. In: Proceedings of NEWS, pp. 10–23 (2015)
Costa-jussà, M.R.: Moses-based official baseline for NEWS 2016. In: Proceedings of NEWS, pp. 88–90 (2016)
Ehrman, M.E., Sos, K., Kheang, L.H.: Contemporary Cambodian – grammatical sketch (1974). https://www.livelingua.com/fsi/Fsi-ContemporaryCambodian-GrammaticalSketch.pdf
Finch, A., Liu, L., Wang, X., Sumita, E.: Neural network transduction models in transliteration generation. In: Proceedings of NEWS, pp. 61–66 (2015)
Finch, A., Liu, L., Wang, X., Sumita, E.: Target-bidirectional neural models for machine transliteration. In: Proceedings of NEWS, pp. 78–82 (2016)
Huffman, F.E.: Cambodian system of writing and beginning reader with drills and glossary (1970). http://www.pratyeka.org/csw/hlp-csw.pdf
Kunchukuttan, A., Bhattacharyya, P.: Data representation methods and use of mined corpora for Indian language transliteration. In: Proceedings of NEWS, pp. 78–82 (2015)
Lafferty, J., McCallum, A., Pereira, F.: Conditional random fields: probabilistic models for segmenting and labeling sequence data. In: Proceedings of ICML, pp. 282–289 (2001)
Liu, L., Finch, A., Utiyama, M., Sumita, E.: Agreement on target-bidirectional LSTMs for sequence-to-sequence learning. In: Proceedings of AAAI, pp. 2630–2637 (2016)
Neubig, G., Nakata, Y., Mori, S.: Pointwise prediction for robust, adaptable Japanese morphological analysis. In: Proceedings of ACL-HLT, pp. 529–533 (2011)
Papineni, K., Roukos, S., Ward, T., Zhu, W.J.: BLEU: a method for automatic evaluation of machine translation. In: Proceedings of ACL, pp. 311–318 (2002)
Sha, F., Pereira, F.: Shallow parsing with conditional random fields. In: Proceedings of HLT-NAACL, pp. 134–141 (2003)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer Nature Singapore Pte Ltd.
About this paper
Cite this paper
Ding, C., Chea, V., Utiyama, M., Sumita, E., Sam, S., Seng, S. (2018). Statistical Khmer Name Romanization. In: Hasida, K., Pa, W. (eds) Computational Linguistics. PACLING 2017. Communications in Computer and Information Science, vol 781. Springer, Singapore. https://doi.org/10.1007/978-981-10-8438-6_15
Download citation
DOI: https://doi.org/10.1007/978-981-10-8438-6_15
Published:
Publisher Name: Springer, Singapore
Print ISBN: 978-981-10-8437-9
Online ISBN: 978-981-10-8438-6
eBook Packages: Computer ScienceComputer Science (R0)