Category and semantic distance modulate the impact of prediction on memory

Abstract: Humans often make predictions based on current contexts to guide subsequent behavior. However, there is ongoing debate about whether predictable or unpredictable items lead to better memory performance. Some researchers have found better memory performance for predictable items, while others have found better memory performance for unpredictable items. The present study used EEG to examine the relationship between encoding sub-processes and memory performance across different levels of predictability. This provides a new perspective on resolving the discrepancy in memory performance between predictable and unpredictable items.
We manipulated predictability through category and semantic distance to examine memory performance under different levels of predictability (Experiment 1: item recognition; Experiment 2: associative recognition). Specifically, we set up 3 conditions: C+S+ (e.g. furniture: sofa, within-category and near semantic distance), C-S+ (e.g. furniture: decoration, out-of-category and near semantic distance), and C-S- (e.g. furniture: phase, out-of-category and far semantic distance). In the study phase, prime words (e.g., furniture) and target words (e.g., sofa) were presented sequentially. Participants were asked to judge which condition each word pair belonged to by pressing the corresponding key when the target word appeared. In the test phase, 28 participants completed an item recognition test and 26 participants completed an associative recognition test.
In the study phase, both experiments consistently showed higher accuracy and faster reaction times in the C+S+ and C-S- conditions than in the C−S+ condition. The N400 amplitudes became progressively more negative across the C+S+, C-S+, and C-S- conditions. In Experiment 2, the P600 amplitude was more positive in the C+S+ and C-S+ conditions compared to the C-S- condition. In both experiments, linear discriminant analysis (LDA) successfully distinguished neural representations across conditions during the N400 and P600 time windows. In the test phase, item recognition (Experiment 1) and associative recognition (Experiment 2) both showed a graded decline in memory performance across the C+S+, C-S+, and C-S- conditions. In addition, both experiments consistently showed that N400 amplitudes during the study phase significantly predicted memory performance in the test phase.
In conclusion, these results suggest that category and semantic distance modulate the influence of prediction on memory through different mechanisms: category may influence memory by modulating encoding burden, while semantic distance may influence memory performance through semantic integration. These findings not only support the view that predictable items have better memory performance, but also suggest that the divergence between predictable and unpredictable memory performance may arise from the relative contributions of two independent factors: encoding difficulty and encoding effort.