Preparing an accurate Persian POS tagger suitable for MT

Preparing an accurate Persian POS tagger suitable for MT

In this paper an accurate Persian POS tagger suitable for MT is prepared. First a new set of POS tags is defined which is general and more usable for MT rather than detailed ones; Then an accurate tagged corpus is prepared with modifying Bijankhan corpus. Stanford POS tagger is trained on the modified Bijankhan, the resulting tagger gives a 99.36% accuracy which shows significant improvement over previous Persian taggers. Result of utilization of this tagger for statistical machine translation is investigated. Outputs show better performance compared to simple SMT, while using previous tagger in SMT drops the BLEU compared to simple SMT.