Unsupervised Compositionality Prediction of Nominal Compounds

Abstract

Nominal compounds such as `red wine` and `ivory tower` display a continuum of compositionality, with varying contributions from the components of the compound to its semantics. We propose a framework for compound compositionality prediction using distributional semantic models, evaluating to what extent they capture idiomaticity compared to human judgments. For evaluation, we introduce datasets containing human judgments in three languages: English, French and Portuguese. The results obtained reveal a high agreement between the models and human predictions, suggesting that they are able to incorporate information about idiomaticity. We also present an in-depth evaluation of various factors that can affect prediction, such as model and corpus parameters and compositionality operations. General crosslingual analyses reveal the impact of morphological variation and corpus size in the ability of the model to predict compositionality, and of a uniform combination of the components for best results.