Using Singular-value Decomposition on Local Word Contexts to Derive a Measure of Constructional Similarity

摘要: This paper presents a novel method of generating word similarity scores, using term by n-gram context matrix which is compressed Singular Value Decomposition, statistical data analysis that extracts the most significant components variation from large matrix, and has previously been used in methods like Latent Semantic Analysis to identify latent semantic variables text. We present results applying these scores standard synonym benchmark tests, argue on basis our metric represents an aspect usage largely orthogonal addressed other methods, such as Analysis. In particular, it appears this captures with respect participation words grammatical constructions, at level generalization corresponding broad syntacticosemantic classes body part terms, kin terms like. Aside assessing similarity, promising applications language modeling automatic lexical acquisition.