Unlexicalized dependency parser for variable word order languages based on local contextual pattern

Abstract

We investigate the effect of unlexicalization in a dependency parser for variable word order languages and propose an unlexicalized parser which can utilize some contextual information in order to achieve performance comparable to that of lexicalized parsers. Unlexicalization of an early dependency parser makes performance decrease by 3.6%. However, when we modify the unlexicalized parser into the one which can consider additional contextual information, the parser performs better than some lexicalized dependency parsers, while it requires simpler smoothing processes, less time and space for parsing. © Springer-Verlag 2004.