Dependency Parsing

Dependency parsing is a popular approach to natural language parsing. NLTK includes some basic algorithms, but we need more reference implementations and more corpus readers.

Existing functionality

Existing functionality is in the parse package. It includes:

• projective dependency parser (similar to Eisner 1996) and probabilistic projective dependency parser (Eisner 1996 Model C), projectivedependencyparser.py
• non-projective dependency parser and probabilistic non-projective dependency parser (still includes diagnostic print statements?), nonprojectivedependencyparser.py
• interface to the MaltParser, malt.py
• corpus reader for the CoNLL 2007 shared task, and for a 10% sample of the dependency version of the Penn Treebank, dependency.py
• corpus reader for the Universal Dependency Treebank
• Labeled and unlabeled attachment scores
• Nivre's arc-eager and arc-standard algorithms
• Visualization of dependency trees

Planned functionality

• Add interface to Universal Dependencies Dataset
• Adapt Honnibal's shift-reduce dependency parser