Proceedings of the Annual Meeting of the Cognitive Science Society

Distributional models of semantics assume that the meaningof a given word is a function of the contexts in which itoccurs. In line with this, prior research suggests that a word’ssemantic representation can be manipulated – pushed towarda target meaning, for example – by situating that word indistributional contexts frequented by the target. Left open toquestion is the role that order plays in the distributionalconstruction of meaning. Learning occurs in time, and it canproduce asymmetric outcomes depending on the order inwhich information is presented. Discriminative learningmodels predict that systematically manipulating a word’spreceding context should more strongly influence its meaningthan should varying what follows. We find support for thishypothesis in three experiments in which we manipulatedsubjects’ contextual experience with novel and marginallyfamiliar words, while varying the locus of manipulation.