Measuring the Compositionality of Arabic Multiword Expressions

Abstract

This paper presents a method for measuring the compositionality score of multiword expressions (MWEs). Based on Wikipedia (WP) as a lexicon resource, the multiword expressions are identified using the title of Wikipedia articles that are made up of more than one word without further process. Through the semantic representation, this method exploits the hierarchical taxonomy in Wikipedia to represent the concept (single word or multiword) as a feature vector containing the WP articles that belong to concept of categories and sub-categories. The literality and the multiplicative function composition scores are used for measuring the compositionality score of an MWE utilizing the semantic similarity. The proposed method is evaluated by comparing the compositionality score against human judgments (dataset) containing 100 Arabic noun-noun compounds. © Springer-Verlag Berlin Heidelberg 2013.