Does tagging help parsing?

Abstract

The usefulness of POS taggers for syntactic parsing is a question little addressed in the literature. Because taggers reduce ambiguity from the parser's input, parsing is commonly supposed to become faster, and the result less ambiguous. On the other hand, tagging errors probably reduce the parser's recognition rate, so this drawback may outweigh the possible advantages. This paper empirically investigates these issues using two different rule-based morphological disambiguators as preprocessor of a wide-coverage finite-state parser of English. With these rule-based taggers, the parser's output becomes less ambiguous without a considerable penalty to recognition rate. Parsing speed increases slightly, not decisively.