Corrective modeling for non-projective dependency parsing

Abstract

We present a corrective model for recovering non-projective dependency structures from trees generated by state-of-the-art constituency-based parsers. The continuity constraint of these constituency-based parsers makes it impossible for them to posit non-projective dependency trees. Analysis of the types of dependency errors made by these parsers on a Czech corpus show that the correct governor is likely to be found within a local neighborhood of the governor proposed by the parser. Our model, based on a MaxEnt classifier, improves overall dependency accuracy by .7% (a 4.5% reduction in error) with over 50% accuracy for non-projective structures.