Clustering for semantic purposes

Abstract

This paper presents an innovative approach, within the framework of distributional semantics, for the exploration of semantic similarity in a technical corpus. In complement to a previous quantitative semantic analysis conducted in the same domain of machining terminology, this paper sets out to discover fine-grained semantic distinctions in an attempt to explore the semantic heterogeneity of a number of technical items. Multidimensional scaling analysis (MDS) was carried out in order to cluster first-order co-occurrences of a technical node with respect to shared second-order and third-order co-occurrences. By taking into account the association values between relevant first and second-order co-occurrences, semantic similarities and dissimilarities between first-order co-occurrences could be determined, as well as proximities and distances on a graph. In our discussion of the methodology and results of statistical clustering techniques for semantic purposes, we pay special attention to the linguistic and terminological interpretation.