A Unified Sentence Space for Categorical Distributional-Compositional Semantics: Theory and Experiments
Dimitri Kartsaklis, Mehrnoosh Sadrzadeh, Stephen Pulman · 2012
This short paper summarizes a faithful implementation of the categorical framework of Coecke et al. (2010), the aim of which is to provide compositionality in distributional models of lexical semantics. Based on Frobenius Algebras, our method enable us to (1) have a unifying meaning space for phrases and sentences of different structure and word vectors, (2) stay faithful to the linguistic types suggested by the underlying type-logic, and (3) perform the concrete computations in lower dimensions by reducing the space complexity. We experiment with two different parameters of the model and apply the setting to a verb disambiguation and a term/definition classification task with promising results.