Slide Slam L12
Language experience affects predictive coding during auditory rhythm perception
Piermatteo Morucci1, Clara Martin1,2, Nicola Molinaro1,2; 1Basque Center on Cognition Brain and Language, 2Ikerbasque, Basque Foundation for Science
According to Predictive Coding models of the auditory cortex, the auditory system constantly extrapolates the rules governing recent inputs and uses such models to generate prediction of incoming sensory events. This operation is central to tracking meaningful structures during auditory rhythm perception. In this study, we investigate whether the auditory system generates predictions based on lifelong exposure to linguistic regularities, using rules that extend beyond those acquired in the recent past. We compare magnetoencephalography (MEG) data from native speakers of Basque and Spanish who performed a rhythmic version of the alternation paradigm with omission responses. Basque and Spanish are two languages that differ in their syntactic/prosodic structure, thus providing an ideal model to study the effect of linguistic experience on auditory predictive processing. Spanish is a functor-initial language, in which short events (i.e., function words; e.g., la, the) usually combine with long ones (i.e., content words; e.g., casa, house), thus forming “short-long” higher-level chunks. On the other hand, Basque is a functor-final language, in which long events (i.e., content words; e.g., etxera, house) usually form phrasal chunks with short events (i.e., function words; e.g., bat, the), resulting in “long-short” grouping units. We hypothesize that the auditory system extrapolates abstract schemes underlying the phrasal structure of language, and use such knowledge to generate long-term predictions about incoming sounds. To test this hypothesis, we present subjects with 30s rhythmic sequences of two tones alternating in duration (short tone = 0.250s; long tone = 0.435s) at fixed intervals (0.020s). In each sequence, two to six tone omissions occur pseudo-randomly. MEG responses to omitted sounds are recorded. A hierarchical predictive coding model predicts that the omission of a short tone represents the violation of two predictions in Spanish, but not in Basque: a local prediction, based on the transitional probabilities of previous stimuli, and a long-term, language-induced prediction based on the regularities of Spanish syntax/prosody. The opposite pattern is expected in the Basque group. Results show that unexpected omissions elicited a sharp “Mismatch Negativity” (MMN) – a neural response putatively associated to cortical prediction error. Importantly, the amplitude of the MMN varied orthogonally depending on the individual’s linguistic background. The omission of a short tone elicited a larger MMN in Spanish compared to Basque native speakers. On the other hand, the omission of a long tone elicited a larger MMN response in the Basque compared to the Spanish group. This prediction error signals occurred around 0.100s from deviant onset, and had their locus in auditory regions. This finding indicates that the auditory system recycles coding schemes employed to parse linguistic information to implement predictive models of non-linguistic sound sequences. These results provide support for the proposal that shared computational resources underlie speech, sound, and music processing.