Using lexical and contextual constraints in language understanding

Following the principles of compositionality and locality in language understanding, we propose a representation formalism which manages to model the lexical and contextual similarity constraints. The formalism follows the symbolic-connectionist approach, paraphrases and aggregates every lexical concepts as well as sentential meaning to produce the contextual constraints. The formalism has the ability to tolerate ambiguities and remains evenhanded with respect to the various interpretations until evidences from the preceding contextual constraints are involved. The system works fully automatically and performs particularly well in anaphora resolution. Due to these constraints for semantic interpretation, no particular dominant linguistic knowledge source will be required.