Semantic Vector Spaces as a Tool for Psychological Investigation

The dramatic increase in the availability of language data through the internet over the past two decades, and the corresponding increase in computational power, provide a rich source of information for psychological research. However, analyzing this information and using it to test psychological theories is not always easy or straightforward. In this presentation I describe a method that uses semantic vector spaces (such as those generated by Latent Semantic Analysis and Topic Models) to quantify patterns of word co-occurrence and test hypotheses. Using this method, I explore how differences in the representation of nouns and verbs affect the stability of their meaning, compare the levels and types of moral rhetoric associated with various concepts and debates, and track the convergence of language use as a measure of reaching agreement in a negotiation.