Proceedings of the Annual Meeting of the Cognitive Science Society

The complex judgments we make about the innumerable ob-jects in the world are made on the basis of our representa-tion of those objects. Thus a model of judgment should spec-ify (a) our representation of the many objects in the world,and (b) how we use this knowledge for making judgments.Here we show that word embeddings, vector representationsfor words derived from statistics of word use in corpora, proxythis knowledge, and that accurate models of judgment can betrained by regressing human judgment ratings (e.g., femininityof traits) directly on word embeddings. This method achieveshigher out-of-sample accuracy than a vector similarity-basedbaseline and compares favorably to human inter-rater relia-bility. Word embeddings can also identify the concepts mostassociated with observed judgments, and can thus shed lighton the psychological substrates of judgment. Overall, we pro-vide new methods and insights for predicting and understand-ing high-level human judgment.