Semantic Similarity from Syntactic Relations

This work presents the results of the application of a technique for automatic extraction of semantic relations between words from a corpus. The technique used is the one proposed by Grefenstette in [8]. We proposed contributions for the syntactic context notion adopted in [8], aiming to improve the identification of semantically related words. Such contributions include new information into the contexts, besides those used by Grefenstette, as well as a new sort of context. We carried out some experiments in order to investigate and validate Grefenstette’s technique and the proposed contributions for Portuguese language. The analysis of the experiments results is detailed, and a comparison is given between the lists of similar words generated through the original Grefenstette’s technique and the lists generated through the modified technique proposed.