Lions, Tigers, and Barracudas: What can vector space models tell us about semantic fluency?
Neuropsychological examinations routinely include a semantic fluency task in which a patient has one minute to list members of a particular semantic category. Individuals with certain neurological conditions can show unusual patterns in their responses to this task.
In this talk, I will discuss the promise and the pitfalls of using distributional semantic models to automate the discovery of these patterns. After comparing human-generated pairwise word similarity measures to those derived using several of the most widely used neural word embedding architectures, I will explore the utility of applying these similarity measures to the task of analyzing semantic fluency reponses in two clinical populations.