Proceedings of the Annual Meeting of the Cognitive Science Society

Human working memory resources are strategically allocated in sentence processing. Recent studies have argued that less predictable linguistic units are prioritized for memory resources to be encoded with higher precision in sentence processing, as a principled solution to minimize the overall memory error given noisy encoding and resource limitations. The investigation of this claim has been operationalized from the perspective of memory robustness, predicting that representations of unexpected linguistic units are more robust against memory interference since they are more precisely encoded. In this study, using a controlled reading experiment in the A-Maze paradigm, we examine this prediction through the lens of the dependency locality effect in comprehension, a classic sentence-processing effect arising from memory decay and interference. Focusing on the subject--verb dependency in English, we find that the locality effect only occurs when the retrieval target (i.e., left codependent) is of high predictability, suggesting that high-predictability linguistic units are less prioritized for memory resources, thus encoded with less robust representation against the interference from the intervening material between codependents. Interestingly, we also observe a mild anti-locality pattern for unexpected retrieval target, pointing to a potential competition between memory- and expectation-based sentence processing mechanisms under the impact of strategic allocation of memory resources.