Detection of Multiword Expressions With Word Vector Representations

Abstract

Multiword expressions (MWE) are word combinations where multiple words leave their own meaning to build a new one. These word combinations are important in text summarization, automatic machine translation and language generation fields. In this study, MWEs are detected employing the word vector representations of word combinations and their composing words. In our work, vector representations of words and Word combinations are built employing five different methods. Giving vector representations as inputs to a group of classifiers, the classification performances are examined relatively. Experimental work is performed on Turkish data sets. The classification performance is measured by F1 value employing 5-fold cross validation. Experimental results showed that vector representations can be employed in MWE detection.