Comparative Evaluation of Tagalog Part-of-Speech Taggers
Dalos D. Miguel, Rachel Edita, O. Roxas · 2007
We have conducted empirical tests on part of speech taggers for Tagalog such as the MBPOST, PTPOST4.1 without stemming, TagAlog and TPOST, which implement the rule-based and the probabilistic tagging. Corpora from different domains with 120,000 words were tagged using the revised tagset of Rabo [8] and used for the evaluation. When the test data is not part of the training data, the PTPOST4.1 which is a HMM tagger yielded the highest average accuracy (78.3%). On the other hand, when the test data is taken from the training data, MBPOST which is a memory-based POS tagger yielded the highest average accuracy (85.0%). Based on the tests, the results are statistically significant (at 0.05 level) across taggers and across domains.