A grammar of movements: Acquisition and retrieval of biased probability distributions in a finite-state grammar-based serial reaction time task

Motor sequence learning research has been primarily concerned with simple frequentist patterns, thereby neglecting the motor system's ability to handle uncertainty and non-deterministic environments. In this study, we combine methods from motor neuroscience and neurolinguistics to study how probabilistic systems are acquired during motor learning. Two groups of young adults practiced a probabilistic serial reaction time task (pSRTT) for three consecutive days. The pSRTT was generated from a finite-state grammar (FSG). One group trained a probability-biased version of the grammar with an 80 %-20 % probability split between transition cues (HiLo), while the other group trained an equal version with 50 % and 50 % transition probability (Eq). On the final day, the pSRTT was followed by a generation task in which participants were asked to finish grammatical (inclusion) and ungrammatical (exclusion) sequences. Both groups displayed distinct learning characteristics depending on the probability distribution of transitions. The Eq group decreased reaction times equally for all allowed transitions, while the HiLo group had slower reactions for less likely transitions. In the generation task, only the HiLo group was able to generate more grammatical sequences in the inclusion task than in the exclusion task. Furthermore, correctly generated sequences had the same probability distribution as the learned FSG for both groups, indicating that participants developed a representation of the entire probabilistic system and used this representation to adapt their behavior in the generation task. Our study shows that participants can form representations of probabilistic frequency distributions during motor learning.