Combining shallow and deep processing for a robust, fast, deep-linguistic dependency parser

Abstract

This paper describes Pro3Gres, a fast, robust, broad-coverage parser that delivers deep-linguistic grammatical relation structures as output, which are closer to predicate-argument structures and more informative than pure constituency structures. The parser stays as shallow as is possible for each task, combining shallow and deep-linguistic methods by integrating chunking and by expressing the majority of long-distance dependencies in a context-free way. It combines statistical and rule-based approaches, different linguistic grammar theories and different linguistic resources.
Preliminary evaluations indicate that the parsers performance is state-of-the-art.