Combining dependency and constituent-based resources for structure disambiguation

Unrestricted text analysis requires an accurate syntactic analysis but structural ambiguity is one of the most difficult problems to resolve. Researchers have tried different approaches to obtain the correct syntactic structure from analyzed sentences but no successful results have been obtained. Two different approaches have traditionally applied to syntactic analysis: constituent grammars and dependency grammars. We propose a model for syntactic analysis and disambiguation combining lexical dependencies and semantic proximity. Lexical dependencies are applied by means of a government pattern dictionary following the dependency approach. The semantic proximity is introduced by means of semantic closeness among constituents. Examples are given to illustrate the method's contributions.