An in-depth look into the co-occurrence distribution of semantic associates

Semantic associations have served as a tool in cognitive science research for decades, and in recent years have also been of importance to computational linguists interested in a corpus-based induction of semantic relatedness. Within both areas, the co-occurrence hypothesis, which holds that free associations reflect co-occurrence in text, has underpinned much of the research that uses associations. However, few investigations have studied the properties of the associates in depth. In this paper, we scrutinise the co-occurrence hypothesis by exploring the distributional characteristics of a large set of stimulus-response pairs in a text corpus. By addressing the question from multiple perspectives we aim to extract more precise information about the generalisability of prior evidence in support of the co-occurrence hypothesis and the validity of many of the assumptions that grow out of the hypothesis.