The ability to learn and adapt to new temporal regularities is pivotal to successfully interact with our environment. Research on rhythm processing has identified which features boost the learning of new rhythmic patterns and facilitate predictive behaviors, such as motor synchronization to them. How do humans learn to synchronize with new temporal patterns when these lack strong rhythmicity and predictability? Here we hypothesized that optimal synchronization with such patterns would require a probabilistic form of associative learning. In our experiment participants tapped a finger in synchrony with an auditory stream in which a familiar and a novel pattern alternated. The two patterns were made of the same intervals but arranged in reversed order, thus having different structures. Within the stream, the two patterns occurred with complementary probabilities, following a gradual transition from familiar to novel pattern across trials. We modelled synchronization performances as the result of a Bayesian inference process parameterized by a Rescorla-Wagner learning rule. Results showed that participants learned to synchronize to the novel pattern only when its predictability clearly exceeded that of the old one, i.e., a few trials after the actual transition in the patterns' probabilities. Parameter estimation makes us hypothesize two learning strategies to be tested in future work. On the one hand, “fast learners” may quickly learn to synchronize to each pattern and adapt to changes between them across trials. On the other hand, “slow learners” may slowly learn to synchronize to a pattern and fail to fully adjust to a novel one. Overall, our results provide preliminary evidence that probabilistic associative learning may offer a parsimonious account for how individuals learn to synchronize with new, variably predictable temporal patterns. This learning could be modulated by subject-specific strategies to adapt to rhythmic changes. • Humans can tap in synchrony to novel, weakly rhythmic temporal patterns by leveraging their predictability. • Bayesian inference and associative learning may support the ability to synchronize to such patterns. • These basic learning mechanisms could potentially be modulated by subject-specific strategies to adapt to rhythmic changes.
Giomo et al. (Thu,) studied this question.