Thematic role assignment through a biologically plausible symbolic-connectionist hybrid system

There are several connectionist systems concerning natural language computational processing applications in artificial intelligence literature. Some try to predict the next word in a sentence context. Others attempt to discover grammatical symbolic rules from the neural network architecture, and others deal with semantic relationships between words. Unfortunately, very few of them employ psycholinguistic theories. Recently, some systems have made use of biologically plausible approaches to connectionist natural language processing, which are computationally as efficient as conventional methods. The system described here, in a way, presents all the properties and characteristics enumerated previously: it is a symbolic-connectionist hybrid system, designed to "predict" thematic (semantic) roles assigned to words in a sentence context with a biologically inspired training algorithm and architecture, and adopting a psycholinguistic view of thematic theory. Its lexicon, which includes lexically ambiguous nouns and "thematically ambiguous" verbs, can be increased, provided that semantic "features" of words are entered. The system has proven to be even more computationally efficient than conventional connectionist systems.