Towards an integrated account of incidental learning and cognitive control

Published in Neuroscience
Towards an integrated account of incidental learning and cognitive control
Like

Share this post

Choose a social network to share with, or copy the shortened URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

The problem

Consider the scenario where you and your partner learn a new dance. As you become more skilled, your movements become more automatic and require less conscious effort. However, you must also be able to predict when your partner wants to change direction and adjust your moves. If you are too focused on making these adjustments, the dance will not flow smoothly. Conversely, if you fail to detect these changes at the appropriate moment, one of you may suffer from a painful foot injury and have to leave the dance floor. This example shows that learning and control over our actions are intertwined. However, the relationship between these functions appears to be complex as many factors modulate whether it will be cooperative or competitive simultaneous acting.

We investigated how different forms of incidental learning (i.e., without the intention to learn) influence cognitive control. Participants performed a classical laboratory paradigm, the Stroop task, with a twist. They were asked to respond to the colour of the fonts rather than the actual words. If these two were in conflict (e.g., “BLUE” written in yellow), the likelihood of errors increased and it took longer to respond. The twist was, that the relationship between font colour and word meaning followed a sequence. In this sequence, regularities or patterns can be learnt via the processes of statistical learning and rule-based learning. Statistical learning enables us to build predictions on probable continuations of the presented events. That is, we can learn that A-C-B is more likely than A-C-D. Rule-based learning uses predictions on a higher level of the sequence structure, which was an alternating rule in our case: every second element was 100% predictable, while every other one was random. Both learning types were documented before with physical features, such as learning a sequence of stimulus tones, colours, positions, orientations, etc. Here we asked whether it is possible to learn a sequence of relationship types between font colour and word meaning. If participants were capable to learn such abstract associations, that would pave the way to build predictions on the upcoming conflict.

The most challenging part of the study was to create a novel task that (1) combines the Stroop task with an incidental learning paradigm, (2) has a complex enough sequence structure to study both statistical and rule-based learning, (3) and presents challenging enough conditions for the participants. The task is pictured in Figure 1.

Participants saw a word or coloured ‘XXXX’ characters in the middle of the screen. a The stimulus presentation followed an eight-element sequence, in which pattern and random (R) elements alternate. Numbers refer to the four possible conflict scenarios (1 – congruent: colour and meaning are the same, 2 – incongruent: colour and meaning are different, 3 – word naming: colour is irrelevant, 4 – colour naming: meaning is irrelevant). The timing of the task is presented on the left side of the figure. b Some series of consecutive elements (triplets) were more probable in the task than others. High-probability triplets could either end with a pattern or with random elements, while low-probability triplets always end with a random element. Two types of sequence learning performance can be calculated in the task. Statistical learning is the difference between high- and low-probability random elements. Rule-based learning is the difference between pattern elements (presented according to the serial order) and random elements (not determined by the sequence).

Figure 1. The alternating sequential Stroop task. Participants saw a word or coloured ‘XXXX’ characters in the middle of the screen. a The stimulus presentation followed an eight-element sequence, in which pattern and random (R) elements alternate. Numbers refer to the four possible conflict scenarios (1 – congruent: colour and meaning are the same, 2 – incongruent: colour and meaning are different, 3 – word naming: colour is irrelevant, 4 – colour naming: meaning is irrelevant). The timing of the task is presented on the left side of the figure. b Some series of consecutive elements (triplets) were more probable in the task than others. High-probability triplets could either end with a pattern or with random elements, while low-probability triplets always end with a random element. Two types of sequence learning performance can be calculated in the task. Statistical learning is the difference between high- and low-probability random elements. Rule-based learning is the difference between pattern elements (presented according to the serial order) and random elements (not determined by the sequence).

The findings

Participants’ behaviour reflected the classical Stroop effect: conflict between colour and meaning prolonged reaction times relative to non-conflict events (i.e., red written in green vs red written in red). Importantly, in the conflict events, responses differed between low-probability and high-probability stimulus continuations. Thus, statistical learning interacted with cognitive control. This was corroborated by the neurophysiological analysis. The N450 is a negative deflection in the EEG signal, which in Stroop tasks is typically larger in conflict than in non-conflict events. In our study, the conflict effect on the N450 was larger and conflict detection more pronounced when events could be predicted by statistical learning (Figure 2). In other words, learning statistical patterns was effective in enhancing conflict detection for conflict events and reducing response costs.

The N450 is a frontocentral negative deflection that is typically observed in Stroop tasks with a peak of approximately 450 ms after stimulus presentation. This event-related potential’s (ERP) amplitude is larger in incongruent than in congruent trials, therefore, it was selected to study conflict detection at the neurophysiological level. Time point zero represents the stimulus presentation. The analysed time window (380-480 ms) is marked with a shaded area. The N450 is organised into four conditions: colour naming, congruent, incongruent, and word naming. The data is presented as a function of triplet types: high-probability pattern (purple), high-probability random (blue), and low-probability random (green). The scalp topography plots show the distribution of the mean activity of each presented condition.

Figure 2. Main neurophysiological results: the 450 component on electrode FCz. The N450 is a frontocentral negative deflection that is typically observed in Stroop tasks with a peak of approximately 450 ms after stimulus presentation. This event-related potential’s (ERP) amplitude is larger in incongruent than in congruent trials, therefore, it was selected to study conflict detection at the neurophysiological level. Time point zero represents the stimulus presentation. The analysed time window (380-480 ms) is marked with a shaded area. The N450 is organised into four conditions: colour naming, congruent, incongruent, and word naming. The data is presented as a function of triplet types: high-probability pattern (purple), high-probability random (blue), and low-probability random (green). The scalp topography plots show the distribution of the mean activity of each presented condition.

It was suggested that by binding goal representations (e.g., responding to colour whilst ignoring word meaning) and contextual information (e.g., the sequence of events) together, these associations modulate adaptive behaviour and have a role in control functions when less demanding processes are not sufficient to direct response selection. The current findings support this theory. Moreover, we showed that statistical learning but not rule-based learning played a role in this adaptation, and the role of statistical predictions decreased as participants became more familiar with the task. That is, the interaction between incidental learning and cognitive control might be process-specific and depends on the familiarity and effort level of the task. Overall, the study highlights the multifactorial nature of real-life adaptations and advocates for a synergistic view of adaptive behaviour by connecting the fields of cognitive control and incidental learning.

Significance

The study expands the understanding of how statistical information modulates attention, inhibitory control, or linguistic processes at the neurophysiological level. It also sheds light on the complexity of everyday interactions between learning and cognitive control. The dance floor example above can be applied to any kind of skill in which we need to carefully balance between automatic/procedural and controlled sequences of actions. Furthermore, the relationship between incidentally learnt habits and cognitive control is considered a crucial aspect of addiction. Individuals with addiction may be able to use cognitive strategies, such as mindfulness or cognitive-behavioural therapy, to help them become more aware of their habits and to develop new habits that are more adaptive and healthier. To facilitate these mechanisms, we need to understand when and how can we promote the flexible switch between incidental learning and cognitive control processes. Our study brought us closer to understanding the process-specificity of this switch both at the behavioural and neurophysiological levels.

Future directions

We conducted follow-up experiments on independent groups of participants to investigate the boundaries of interaction between incidental learning and cognitive control. We successfully replicated the original findings in the second experiment. However, when we decreased the effort level in the third experiment, the interaction was not significant anymore. In the final experiment, when we changed the sequence configuration, we did not obtain a significant interaction but individually significant main effects of incidental learning and cognitive control. These follow-up experiments paved the way for further studies. It will be important to (1) understand the nature of the interaction between incidental learning and cognitive control on a longer time scale; (2) how the dynamics change with different types of conflict processes; (3) and how can we enhance the strength of the interaction by behavioural interventions or non-invasive brain stimulation. These are the necessary steps to connect the results presented in Communications Biology to potential applications.

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Neuroscience
Life Sciences > Biological Sciences > Neuroscience

Related Collections

With collections, you can get published faster and increase your visibility.

Biology of reproduction

For this Collection, we encourage submissions that push forward our understanding of reproduction and its impact on offspring in both model organisms and human studies.

Publishing Model: Open Access

Deadline: Jul 10, 2024