This paper addresses the rule-based POS tagging method of Brill, and questions the importance of rule interactions to its performance. Adopting two assumptions that serve to exclude rule interactions during tagging and training, we arrive at some variants of Brill's approach that are instances of decision list models. These models allow for both rapid training on large data sets and rapid tagger execution, giving tagging accuracy that is comparable to, or better than the Brill method.
No takes yet. Share an insight, caveat, or question.
Mark Hepple (2000) studied this question.