Efficient HPSG Parsing Algorithm with Array Unification

This paper presents a method for improving parsing performance of parsers for HPSG. The method was obtained by extending Torisawa’s parsing method for HPSG. His parsing method utilizes a CFG compiled from a given HPSG-based grammar, and the parser predicts the possible parse trees with the CFG. Since the amount of unification is reduced because of this prediction, parsing performance is improved. However, we observed that there is a limit of speedup by the method. This is due to the fact that a compiled CFG cannot capture all the constraints in an original HPSG. We add an extra mechanism called array unification to a parser for Torisawa’s parsing method and overcome this limitation. The mechanism performs part of type unification and improves precision of parse tree prediction with a small overhead. We observed improvement of parsing speed by a factor of 2.8 with an existing Japanese grammar and 1.7 with an existing English grammar.