Latent Semantic Analysis Model as a Representation of Free-Association Word Norms

Abstract

The current work aims to validate, by means of a computational model, an empirical database of free word association norms of Mexican Spanish. Specifically, this work has two main goals: (1) to detect the associated weight of word word pairs, and (2) to provide an understanding of a lexical network formed beyond an input-output word pair, similar to the mediated priming effect reported experimentally. We used the Term Frequency-Inverse Document Frequency weighting (TFIDF) to obtain the associated weight between an input output word pair and to calculate the TFIDF-Matrix which is used as an input in the Latent Semantic Analysis (LSA) Model. The LSA model is a semantic representation at the lexical level that allows us to understand semantic relationships beyond input-output word pairs. Our computational model replicates and further explains previous experimental work on lexical networks.