Investigating Low-resource Machine Translation for English-to-Tamil
Akshai Ramesh, Venkatesh Balavadhani parthasa, Rejwanul Haque, Andy Way · 2020
Statistical machine translation (SMT) which was the dominant paradigm in machine translation (MT) research for nearly three decades has re cently been superseded by the endtoend deep learning approaches to MT.Although deep neu ral models produce stateoftheart results in many translation tasks, they are found to under perform on resourcepoor scenarios.Despite some success, none of the presentday bench marks that have tried to overcome this prob lem can be regarded as a universal solution to the problem of translation of many lowresource languages.In this work, we investigate the performance of phrasebased SMT (PBSMT) and neural MT (NMT) on a rarelytested low resource languagepair, EnglishtoTamil, tak ing a specialised data domain (software localisa tion) into consideration.In particular, we pro duce rankings of our MT systems via a social media platformbased human evaluation scheme, and demonstrate our findings in the lowresource domainspecific text translation task.