Finite-State Morphological Parsing Using Register Vector Grammars

Abstract

Although the RVGMP can parse some data which Kimmo cannot parse, the RVGMP is slower on average than Kimmo. Poor program design is responsible for this slow time. A revised program design, based in part on a new design for RVG, is proposed. Since the redesigned RVGs have doubled their parsing speed, similar time gains are expected with the new implementation of the RVGMP. With these time gains, the RVGMP would parse all the data more quickly than Kimmo.