Sources of Semantic Similarity

Similarity is the key notion underlying many contemporary theories about the representation of meaning through words or concepts. However, these representations are strongly colored by the kind of information captured by various semantic measures. In this paper we present a systematic comparison of human similarity judgments and calculated similarity coefficients from different sources of semantic similarity based on concept features, word associations, word co-occurrence and expert knowledge. We show that these measures capture our semantic representations to a large extent, but also model different aspects of our semantic knowledge, depending on (a) the semantic domain and (b) the range of similarity comparisons.