Automatic identification of non-compositional multi-word expressions using latent semantic analysis

Abstract

Making use of latent semantic analysis, we explore the hypothesis that local linguistic context can serve to identify multi-word expressions that have noncompositional meanings. We propose that vector-similarity between distribution vectors associated with an MWE as a whole and those associated with its constitutent parts can serve as a good measure of the degree to which the MWE is compositional. We present experiments that show that low (cosine) similarity does, in fact, correlate with non-compositionality.

Unsupervised recognition of literal and non-literal use of idiomatic expressions

“…The few token-based approaches include a study by Katz and Giesbrecht (2006), who devise a supervised method in which they compute the meaning vectors for the literal and non-literal usages of a given expression in the training data. An unseen test instance of the same expression is then labelled by performing a nearest neighbour classification.…”

Detecting Metaphorical Phrases in the Polish Language

“…Baldwin et al (2003) showed that LSA-based similarity between the multiword expression and each of its components is indicative for compositionality. Katz and Giesbrecht (2006) compared the actual phrase vector to the estimated compositional meaning vector calculated as a sum of the meaning vectors of the 772 parts. The hypothesis was that the similarity between these two vectors should be larger in case of phrases which are not used non-compositionally.…”

Bilingual Multiword Expression Alignment by Constituent-Based Similarity Score

“…Several researchers have used latent semantic analysis (LSA) to distinguish between compositional and non-compositional preferences of expressions [17,18]. They show that compositional MWEs are generally more likely similar to their constituents than other non-compositional MWEs.…”