Paper: Partial Parsing from Bitext Projections

ACL ID P11-1160
Title Partial Parsing from Bitext Projections
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2011

Recent work has shown how a parallel corpus can be leveraged to build syntac- tic parserfor a target languageby project- ing automaticsourceparseonto the target sentenceusingword alignments.The pro- jectedtarget dependency parsesare not al- waysfullyconnectedto be usefulfortrain- ing traditionaldependency parsers.In this paper, we presenta greedynon-directional parsing algorithm which doesn’t need a fully connectedparse and can learn from partial parses by utilizingavailable struc- tural and syntactic informationin them. Our parser achieved statistically signifi- cant improvementsover a baselinesystem that trains on only fully connectedparses for Bulgarian, Spanishand Hindi. It also gave a significantimprovement over pre- viouslyreportedresultsfor Bulgarian and set a benchmarkfor Hindi.