Refining the Design of a Contracting Finite-State Dependency Parser

This work complements a parallel paper of a new finite-state dependency parser architecture (Yli-Jyrä, 2012) by a proposal for a linguistically elaborated morphology-syntax interface and its finite-state implementation. The proposed interface extends Gaifman’s (1965) classical dependency rule formalism by separating lexical word forms and morphological categories from syntactic categories. The separation lets the linguist take advantage of the morphological features in order to reduce the number of dependency rules and to make them lexically selective. In addition, the relative functional specificity of parse trees gives rise to a measure of parse quality. By filtering worse parses out from the parse forest using finite-state techniques, the best parses are saved. Finally, we present a synthesis of strict grammar parsing and robust text parsing by connecting fragmental parses into trees with additional linear successor links.