Models of Sequential Learning

The expression of music, language and physical motor skills share the need to execute well-learned plans of sequential behavior. We can think of each of these as governed by a set of syntactic principles that instantiate organizational rules. From a computational perspective, it has been frequently observed that a fair amount of apparently rule-driven behavior can be captured by simple statistical learning models that identify sequential dependencies within frequently repeated sequences. Across many paradigms, the basal ganglia has been a brain region closely associated with this type of learning, suggesting that a common computational mechanism in this region may be involved in many types of syntax. Computational models that do statistical learning can be based on simple Bayesian statistical principles or simple recurrent connectionist networks. Using set of recent experiments examining skill learning in a novel task (SISL; Serial Interception Sequence Learning), successes and failures of the models to capture key learning, transfer and interference effects will be used to identify candidate mechanisms for neutrally plausible models of sequential learning. These mechanisms may eventually explain key challenges in the learning of syntax in language, music structure or sequential skills.