Template-Based English-Filipino Machine Translation System

This paper presents a template-based machine translation system that extracts templates from a given bilingual corpus, then uses these templates to perform bi-directional EnglishFilipino translations. The system extended the similarity template learning algorithm of Cicekli and Guvenir [2] by refining existing templates and deriving templates from previously learned chunks. Chunk alignment and splitting algorithms are integrated into the training process to improve the quality of the extracted templates. Test results verified that a strict chunk alignment scheme used in the training process, including the filtering of commonly occurring words, generated better templates and chunks. Correct extraction of templates and chunks during the learning process led to reduced word and sentence error rates by as much as 50% during translation. Tests also showed that the translation with the highest score selected from a set of candidate translations is consistently the best choice as validated against automatic evaluation methods.