Proceedings of the Annual Meeting of the Cognitive Science Society

Expectation-based models of sentence processing predict lower processing costs for highly expected words, whereas re- search on long-distance dependencies consistently shows that greater memory demands increase processing difficulty. The Lossy-Context Surprisal (LCS) framework integrates these ap- proaches, postulating that surprisal is determined on the basis of imperfect memory representations, predicting reduced ex- pectancy effects as dependency length increases. In both a self- paced reading and ERP experiment, we investigated whether the distance between predictive elements of the context and a tar- get word modulates expectancy as predicted by LCS. While the SPR findings revealed an interaction of these factors, providing preliminary support for unified accounts like LCS, neurophys- iological responses revealed additive effects of expectancy and memory demands, but not the predicted interaction.