Morphology Driven Manipuri POS Tagger
Thoudam Doren Singh, Sivaji Bandyopadhyay · 2008
A good POS tagger is a critical component of a machine translation system and other related NLP applications where an appropriate POS tag will be assigned to individual words in a collection of texts. There is not enough POS tagged corpus available in Manipuri language ruling out machine learning approaches for a POS tagger in the language. A morphology driven Manipuri POS tagger that uses three dictionaries containing root words, prefixes and suffixes has been designed and implemented using the affix information irrespective of the context of the words. We have tested the current POS tagger on 3784 sentences containing 10917 unique words. The POS tagger demonstrated an accuracy of 69%. Among the incorrectly tagged 31% words, 23 % were unknown words (includes 9 % named entities) and 8 % known words were wrongly tagged. 1