Automatic Segmentation and Part-Of-Speech Tagging For Tibetan: A First Step Towards Machine Translation
Paul G. Hackett · 2000
This paper presents what we believe to be the first reported work on Tibetan machine translation (MT). Of the three conceptually distinct components of a MT system — analysis, transfer, and generation — the first phase, consisting of POS tagging has been successfully completed. The combination POS tagger / word-segmenter was manually constructed as a rule-based multi-tagger relying on the Wilson formulation of Tibetan grammar. Partial parsing was also performed in combination with POS-tag sequence disambiguation. The component was evaluated at the task of document indexing for Information Retrieval (IR). Preliminary analysis indicated slightly better (though statistically comparable) performance to n-gram based approaches at a known-item IR task. Although segmentation is application specific, error analysis placed segmentation accuracy at 99%; the accuracy of the POS tagger is also estimated at 99% based on IR error analysis and random sampling.