Vector Space Models in Detection of Semantically Non-Compositional Word Combinations in Turkish

Abstract

The semantic compositionality presents the relation between the meanings of word combinations and their components. Simply, in non-compositional expressions, the words combine to generate a different meaning. This is why, identification of non-compositional expressions (e.g. idioms) become important in natural language processing tasks such as machine translation and word sense disambiguation. In this study, we explored the performance of vector space models in detection of non-compositional expressions in Turkish. A data set of 2229 uninterrupted two-word combinations that is built from six different Turkish corpora is utilized. Three sets of five different vector space models are employed in the experiments. The evaluation of models is performed using well-known accuracy and F-measures. The experimental results showed that the model that measures the similarity between the vectors of word combination and the second composing word produced higher average F-scores for all testing corpora. © Springer Nature Switzerland AG 2018.