A two-stage constraint based dependency parser for free word order languages

Abstract

The paper proposes a broad coverage two-stage constraint based dependency parser for free word order languages. For evaluating the parser and to ascertain its coverage we show its performance on Hindi which is a free word order language. We compare our results with that of two data-driven parsers which were trained on a subpart of a Hindi Treebank. The final results are good with a maximum attachment and labeled attachment accuracy of 90.1% and 75% respectively. The results point towards the phenomena which the data driven parsers find easier to learn vs. those for which the grammar driven approach can prove more effective.