Eisenhauer, S. 1 , Gagl, B. 1, 2 & Fiebach, C. J. 1, 2
1 Department of Psychology, Goethe University Frankfurt, Germany
2 Center for Individual Development and Adaptive Education of Children at Risk (IDeA), Frankfurt am Main, Germany
Visual word recognition is facilitated through predictive context, e.g., in sentences or texts, resulting in fast reading times and reduced neuronal activation. Conceptually, this facilitation can be reconciled with predictive coding models, which attribute neuronal processing efficiency and facilitated behaviour to the proactive use of prior information to predict upcoming sensory inputs. As a consequence, predictable stimulus characteristics should be encoded in neuronal activation patterns prior to stimulus presentation, which was found previously for low-level visual percepts. We hypothesized that such predictive coding mechanisms should also be active during visual word recognition. Thus, we assumed that for predictable words, pre-stimulus neuronal activation patterns should encode important information about the expected stimulus, i.e., word frequency and orthographic properties quantified by the orthographic Levenshtein distance 20. We tested this hypothesis using a repetition priming task and simultaneous MEG recording (N=39). High predictability of letter string characteristics was realized by a high probability of identical prime-target pairings (75%). We expected that the information provided by the prime is encoded in the neuronal signal prior to target presentation. Time-resolved multivariate pattern classification analyses (MVPA) supported this hypothesis, by revealing that word frequency, but not orthographic familiarity, can be significantly decoded (p < 0.05) from the pre-target MEG activation. Interestingly, these pre-target neuronal codes seem to differ from representations activated during prime presentation, as there was only generalization within the pre-target delay interval but no temporal generalization from prime presentation to delay. These results indicate lexical-level word characteristics are pre-activated to facilitate word recognition in predictive contexts. Our findings support the hypothesized role of expectation-based predictive mechanisms during visual word recognition.