Initial Explorations in Two-phase Turkish Dependency Parsing by Incorporating Constituents

This paper describes a two-phase Turkish dependency parsing which separates dependency and labeling into two similar to (McDonald et al., 2006b). First, in order to solve the long distance dependency attachment problem, the sentences are split into constituents and the dependencies are estimated on shorter sentences. Later, for better estimation of labels, Conditional Random Fields (CRFs) are used with previously learned chunk and several dependency and morphosyntactic features. Finally, a post-processing step is applied to “correct” some of labels, if necessary.