Dependency Parsing With DMV

We built a dependency parser based upon the dependency model with valence (DMV) created by Klein and Manning. We used this model to parse part-of-speech (POS) tag dependencies in the Wall Street Journal corpus without punctuation of sentences with less than or equal to ten words: WSJ10. We trained the dependency grammars using the inside outside algorithm for probabilistic context free grammars (PCFGs). To make use of this algorithm we first convert our dependency grammar into a split-head CFG before parsing, and then after parsing we convert the split-head CFG parse tree back into a dependency grammar.

Python code used for this project is available.