Track:
Contents
Downloads:
Abstract:
A semantic cognitive map of natural language is constructed using dictionaries of synonyms and antonyms with a simple algorithm described by the authors previously that differs from prior work on embedding of words in metric spaces: Isomap, latent semantic analysis, multidimensional scaling. Previous and new results exhibit semantic invariance across languages and extend to psychometric data. Semantics of the two most significant dimensions can be approximately characterized as "good vs. bad" and "calming vs. exciting". Applications of this technique in the present work include (i) quantitative definitions of universal semantic dimensions and their experimental validation, and (ii) computation of semantic biases and preferences expressed in an arbitrary given text segment: "mood sensing from text". The latter, practically important capability can be used in search engines, in human interfaces of intelligent agents, etc. and is illustrated here using a diverse variety of text samples.