Transformation-based Khmer Part-of-Speech tagger

This paper introduces an initiative research on Khmer Part-of-Speech (POS) tagger based on Transformation based approach. Due to a few researches on natural language processing for Khmer, many pre-processing tasks are needed before the automatic tagging can take place. The first Khmer annotated corpus is tagged with 27 tags based on the traditional and modern grammar theories. The learner, based on learning algorithm introduced by Brill [2], is built with 32 transformation templates. After applying the transformation rules with our sophisticated ranking algorithm, the error rate of tagging on trained and untrained data can be reduced around 41% and 18% accordingly over the baseline. The experiments provide very encouraging results; however, some future works are drawn to improve the accuracy and the performance of the tagger to reach the better level.