Using Morphosyntactic Post-processing to Improve POS-tagging Accuracy

Many of the errors produced by up-to-date POS-taggers could be considered as morphologic, syntactic or semantic. Once statistical tagging does not deal with semantic ambiguity, the correction of (morpho)syntactic errors emerges as one of the possibilities to improve the accuracy of this task. This work describes a method that applies a robust parser with correction rules over a POS-tagging output. We outline its preliminary results on a European Portuguese corpus, showing that the most common tagging errors could be corrected up to 50% with very basic and linguistically motivated rules.