A Three Stage Hybrid Parser for Hindi

The present paper describes a three stage technique to parse Hindi sentences. In the first stage we create a model with the features of head words of each chunk and their dependency relations. Here, the dependency relations are inter-chunk dependency relations. We have experimentally fixed a feature set for learning this model. In the second stage, we extract the intra-chunk dependency relations using a set of rules. The first stage is combined with the second stage to build a two-stage word level Hindi dependency parser. In the third stage, we formulate some rules based on features and used them to post-process the output given by the two-stage parser.