Statistical learning of unbalanced exclusive-or temporal sequences in humans

The order of stimuli matters when learning second-order transitional probabilities

The order of stimuli within sequences and the transitional probabilities (TPs) it generates are central information in sequence processing. However, less is known about what type of information and how it is extracted by general learning mechanisms. The present study focused on statistical learning of second-order TPs. Second-order TPs are involved when only the combination of two stimuli predicts the third. In a first experiment, TPs depended crucially on the order of presentation of a pair A - B , which led to different predictions depending on the order of the stimuli (i.e., ABC vs. BAF). Eight visuomotor sequences governed by second-order TPs were used and response times (RTs) were recorded for each transition. The task included a learning phase followed by a switch phase during which the second-order TP were reversed (e.g., the sequences ABC and BAF became respectively ABF and BAC). A decrease of RTs between the second and the third stimulus during the learning phase and an increase of RTs during the switch phase suggested that variations of orders within second-order TPs could be learned. Further analyses, however, indicated that such learning was difficult for most participants. A second experiment showed that the difficulty of learning was not solely due to the difficulty to pick up the effect of order of presentation, but that learning second-order transitional probabilities in addition to order would be the main obstacle. These experiments suggest that statistical learning is capable of learning complex associations, even if this remains a challenge for human cognition.

Statistical word segmentation succeeds given the minimal amount of exposure

One of the first tasks in language acquisition is word segmentation, a process to extract word forms from continuous speech streams. Statistical approaches to word segmentation have been shown to be a powerful mechanism, in which word boundaries are inferred from sequence statistics. This approach requires the learner to represent the frequency of units from syllable sequences, though accounts differ on how much statistical exposure is required. In this study, we examined the computational limit with which words can be extracted from continuous sequences. First, we discussed why two occurrences of a word in a continuous sequence is the computational lower limit for this word to be statistically defined. Next, we created short syllable sequences that contained certain words either two or four times. Learners were presented with these syllable sequences one at a time, immediately followed by a test of the novel words from these sequences. We found that, with the computationally minimal amount of two exposures, words were successfully segmented from continuous sequences. Moreover, longer syllable sequences providing four exposures to words generated more robust learning results. The implications of these results are discussed in terms of how learners segment and store the word candidates from continuous sequences.

A neurophysiological perspective on the integration between incidental learning and cognitive control

Adaptive behaviour requires interaction between neurocognitive systems. Yet, the possibility of concurrent cognitive control and incidental sequence learning remains contentious. We designed an experimental procedure of cognitive conflict monitoring that follows a pre-defined sequence unknown to participants, in which either statistical or rule-based regularities were manipulated. We show that participants learnt the statistical differences in the sequence when stimulus conflict was high. Neurophysiological (EEG) analyses confirmed but also specified the behavioural results: the nature of conflict, the type of sequence learning, and the stage of information processing jointly determine whether cognitive conflict and sequence learning support or compete with each other. Especially statistical learning has the potential to modulate conflict monitoring. Cognitive conflict and incidental sequence learning can engage in cooperative fashion when behavioural adaptation is challenging. Three replication and follow-up experiments provide insights into the generalizability of these results and suggest that the interaction of learning and cognitive control is dependent on the multifactorial aspects of adapting to a dynamic environment. The study indicates that connecting the fields of cognitive control and incidental learning is advantageous to achieve a synergistic view of adaptive behaviour.

Probability, Dependency, and Frequency Are Not All Equally Involved in Statistical Learning

The ability to learn sequences depends on different factors governing sequence structure, such as transitional probability (TP, probability of a stimulus given a previous stimulus), adjacent or nonadjacent dependency, and frequency. Current evidence indicates that adjacent and nonadjacent pairs are not equally learnable; the same applies to second-order and first-order TPs and to the frequency of the sequences. However, the relative importance of these factors and interactive effects on learning remain poorly understood. The first experiment tested the effects of TPs and dependency separately on the learning of nonlinguistic visual sequences, and the second experiment used the factors of the first experiment and added a frequency factor to test their interactive effects with verbal sequences of stimuli (pseudo-words). The results of both experiments showed higher performance during online learning for first-order TPs in adjacent pairs. Moreover, Experiment 2 indicated poorer performance during offline recall for nonadjacent dependencies and low-frequency sequences. We discuss the results that different factors are not used equally in prediction and memorization.

Dynamic branching in a neural network model for probabilistic prediction of sequences

An important function of the brain is to predict which stimulus is likely to occur based on the perceived cues. The present research studied the branching behavior of a computational network model of populations of excitatory and inhibitory neurons, both analytically and through simulations. Results show how synaptic efficacy, retroactive inhibition and short-term synaptic depression determine the dynamics of selection between different branches predicting sequences of stimuli of different probabilities. Further results show that changes in the probability of the different predictions depend on variations of neuronal gain. Such variations allow the network to optimize the probability of its predictions to changing probabilities of the sequences without changing synaptic efficacy.

The Evolution of Chunks in Sequence Learning

Chunking mechanisms are central to several cognitive processes and notably to the acquisition of visuo-motor sequences. Individuals segment sequences into chunks of items to perform visuo-motor tasks more fluidly, rapidly, and accurately. However, the exact dynamics of chunking processes in the case of extended practice remain unclear. Using an operant conditioning device, 18 Guinea baboons (Papio papio) produced a fixed sequence of nine movements during 1000 trials by pointing to a moving target on a touch screen. Response times analyses revealed a specific chunking pattern of the sequence for each baboon. More importantly, we found that these patterns evolved during the course of the experiment, with chunks becoming progressively fewer and longer. We identified two chunk reorganization mechanisms: the recombination of preexisting chunks and the concatenation of two distinct chunks into a single one. These results provide new evidence on chunking mechanisms in sequence learning and challenge current models of associative and statistical learning.