Vidal, Y. 1 , Brusini, P. 2 , Bonfieni, M. 4 , Mehler, J. 1 & Bekinschtein, T. 3
1 Language, Cognition and Development Lab, International School for Advanced Studies, Trieste, Italy
2 Center for Neuroscience in Education, University of Cambridge, Downing St, Cambridge CB2 3EB, United Kingdom
3 Consciousness and Cognition Lab, Department of Psychology, University of Cambridge, UK
4 School of Philosophy, Psychology and Language Sciences, The University of Edinburg, UK
Predictive Coding has emerged as a leading theory in neuroscience. This theory states that the brain constructs a hierarchy of predictions of incoming stimuli at multiple levels of processing. Most experimental designs used in the study of Predictive Coding are variations of the OddBall paradigm, using simple tones to establish and violate predictable sequences. Beyond simple-tone paradigms, the use of speech stimuli offers a number of advantages. Within speech, abstract rules are ubiquitous, allowing to test abstract predictions that go beyond physical stimuli features and local transitional probabilities. These properties make speech processing an excellent testbed for the study of Predictive Coding. We propose that during speech perception, phonological prediction are generated even in the absence semantic, syntactic or phonotactic information. To test this hypothesis, we performed 2 electroencephalography (EEG) experiments with an OddBall design. The use of speech stimuli allowed us to test for predictions based on an abstract rule that go beyond local transitional probabilities, and with stimuli far more complex than the tones used in typical Predictive Coding experiments. Our results show that the presentation of deviants, constructed using the same phonemes as standard pseudowords but in an unexpected sequence, elicit an early prediction error signal, which implies that listeners' brains generate predictions about incoming phonemes. Furthermore, the amplitude of prediction error is modulated by the amount of phonemes presented before the point of deviance. This modulation indicates that predictions are not based solely in local transitional probabilities between phonemes. Finally, experiments 1 and 2 differed with respect to the attentional set required to the participants. Despite this, an early prediction error signal was detected in both experiments, implying that predictions are deployed even if the task at hand does not require error detection.