Enforcing Subcategorization Constraints in a Parser Using Sub-parses Recombining

Treebanks are not large enough to adequately model subcategorization frames of predicative lexemes, which is an important source of lexico-syntactic constraints for parsing. As a consequence, parsers trained on such treebanks usually make mistakes when selecting the arguments of predicative lexemes. In this paper, we propose an original way to correct subcategorization errors by combining subparses of a sentence S that appear in the list of the n-best parses of S. The subcategorization information comes from three different resources, the first one is extracted from a treebank, the second one is computed on a large corpora and the third one is an existing syntactic lexicon. Experiments on the French Treebank showed a 15.24% reduction of erroneous subcategorization frames (SF) selections for verbs as well as a relative decrease of the error rate of 4% Labeled Accuracy Score on the state of the art parser on this treebank.