Determining the Compositionality of Noun-Adjective Pairs with Lexical Variants and Distributional Semantics

In this work we tested whether a series of compositionality indices that compute the distributional similarity between the vector of a given expression and the vectors of its lexical variants can effectively tell apart idiomatic and more compositional expressions in a set of 13 idiomatic and 13 non-idiomatic Italian target noun-adjective constructions. The lexical variants were obtained by replacing the components of the original expressions with semantically related words automatically extracted from Distributional Semantic Models or manually derived from Italian MultiWordnet. Indices based on the Mean or the Centroid cosine similarity between the target and the variant vectors performed comparably or better than the addition-based measure traditionally reported in the distributional literature on compositionality.