Explaining away the impossible: Effects of plausibility on predictive sentence processing and memory

Abstract This study is concerned with the impact of plausibility on predictive sentence processing and memory for sentence content. Previous research provides inconsistent findings on how implausible input affects processing difficulty and memory performance. In a self-paced reading task and a subsequent word recognition memory task, we tried to disentangle the effects of prediction error magnitude and integrability with the situation model by contrasting unpredictable but plausible and semantically anomalous (implausible) input. To encourage deep processing of differences in plausibility, we incorporated a plausibility judgment in the self-paced reading task. Unlike comparable studies, we found longer reading times and better recognition memory for unpredictable but plausible than for implausible input. Our results suggest that when thorough relational processing is encouraged by the experimental task, unpredictable but plausible input entails slower processing and deeper memory encoding than implausible input. At the same time, knowing that anomalous sentences will occur in the task may have enabled participants to quickly discard the implausible input as uninformative, which led to quicker but also shallower processing. We suggest that prediction error caused by implausible input can be largely explained away top-down by high-level expectations even though it is not possible to predict the more specific semantic and lexical features of this input.