Compositional Distributional Semantics with Syntactic Dependencies and Selectional Preferences

This article describes a compositional model based on syntactic dependencies which has been designed to build contextualized word vectors, by following linguistic principles related the concept of selectional preferences. The strategy proposed in current work evaluated syntactically controlled and multilingual dataset, compared with Transformer BERT-like models, such as Sentence BERT, state-of-the-art sentence similarity. For this purpose, we created two new test datasets for Portuguese Spanish basis that defined English language, containing expressions noun-verb-noun transitive constructions. results have obtained show linguistic-based approach turns out be competitive models.