Verb Noun Construction MWE Token Supervised Classification

Abstract

We address the problem of classifying multiword expression tokens in running text. We focus our study on Verb-Noun Constructions (VNC) that vary in their idiomaticity depending on context. VNC tokens are classified as either idiomatic or literal. We present a supervised learning approach to the problem. We experiment with different features. Our approach yields the best results to date on MWE classification combining different linguistically motivated features, the overall performance yields an F-measure of 84.58% corresponding to an F-measure of 89.96% for idiomaticity identification and classification and 62.03% for literal identification and classification.