Learning dependencies between case frame slots

Abstract

A theoretically sound method for learning dependencies between case frame slots is proposed. In particular, the problem is viewed as that of estimating a probability distribution over the case slots represented by a dependency graph (a dependency forest). Experimental results indicate that the proposed method can bring about a small improvement in disambiguation, but the results are largely consistent with the assumption often made in practice that case slots are mutually independent, at least when the data size is at the level that is currently available.