Concrete Models and Empirical Evaluations for the Categorical Compositional Distributional Model of Meaning

Abstract

Modelling compositional  meaning for sentences using empirical distributional methods has been a challenge for computational linguists. The categorical model of (Clark, Coecke, and Sadrzadeh 2008; Coecke, Sadrzadeh, and Clark 2010) provides a solution by unifying a categorial grammar and a  distributional model of meaning. It takes into account  syntactic relations during semantic vector composition operations. But the setting is abstract: it  has not been evaluated on empirical data and applied to any language tasks. We generate concrete models for this setting by developing   algorithms to construct  tensors and linear maps and instantiate the  abstract parameters using empirical data. We then evaluate our concrete models against several experiments, both existing and new, based on measuring how well models align with human judgements in a paraphrase detection task. Our results  show  the concrete implementation of this general abstract  framework to perform on par with or outperform other leading models in these experiments.