A BAYESIAN-NETWORK APPROACH TO LEXICAL DISAMBIGUATION

Lexical ambiguity can be syntactic if it involves more than one gramma tical category for a single word, or semantic if more than one meaning can be associated with a word. In this article we discuss the applica tion of a Bayesian-network model in the resolution of lexical ambiguit ies of both types. The network we propose comprises a parsing subnetwo rk, which can be constructed automatically for any context-free gramma r, and a subnetwork for semantic analysis, which, in the spirit of Fil lmore's (1968) case grammars, seeks to fulfill the required cases of a ll candidates for verb of the sentence. Solving for the highest joint probability of the variables conditioned upon the evidences to the net work yields the most likely candidate with its meaning, along with its cases and respective meanings. This is achieved by fixing the values of all evidence nodes concurrently, and then performing a stochastic s imulation in which the remaining nodes are updated probabilistically w ith a high degree of parallelism. The process of disambiguation is dir ected neither by the syntax nor the semantics, but rather by the inter relation between the two subnetworks. The use of a Bayesian-network mo del allows us to express this interrelation between the two subnetwork s and among their constituents in a rather direct and rigorous way tha t, in connection with the convergence properties of the stochastic sim ulation, reveals a very robust model.