Effect of Minimal Semantics on Dependency Parsing
Bharat Ram Ambati, Pujitha Gade, Chaitanya Gsk, Samar Husain · Recent Advances in Natural Language Processing · 2009
In many languages general syntactic cues are insufficient to disambiguate crucial relations in the task of Parsing. In such cases semantics is necessary. In this paper we show the effect of minimal semantics on parsing. We did experiments on Hindi, a morphologically rich free word order language to show this effect. We conducted experiments with the two data-driven parsers MSTPaser and MaltParser. We did all the experiments on a part of Hyderabad Dependency Treebank. With the introduction of minimal semantics we achieved an increase of 1.65% and 2.01% in labeled attachment score and labeled accuracy respectively over state-of-the-art data driven dependency parser.