Annotating Concept Abstractness by Common-Sense Knowledge

Dealing with semantic representations of concepts involves collecting information on many aspects that collectively contribute to (lexical, semantic and ultimately) linguistic competence. In the last few years mounting experimental evidences have been gathered in the fields of Neuroscience and Cognitive Science on conceptual access and retrieval dynamics that posit novel issues, such as the imageability associated to terms and concepts, or abstractness features as a correlate of figurative uses of language. However, this body of research has not yet penetrated Computational Linguistics: specifically, as regards as Lexical Semantics, in the last few years the field has been dominated by distributional models and vectorial representations. We recently proposed COVER, that relies on a partly different approach. Conceptual descriptions herein are aimed at putting together the lexicographic precision of BabelNet and the common-sense available in ConceptNet. We now propose Abs-COVER, that extends the existing lexical resource by associating an abstractness score to the concepts contained therein. We introduce the detailed algorithms and report about an extensive evaluation on the renewed resource, where we obtained correlations with human judgements in line or higher compared to state of the art approaches.