Partial repetition costs index a mixture of binding and signaling

Repetition costs in sequence chunking

We examined how flexibly we plan sequences of actions when we switch between multiple action sequences. Mastering a sequential skill is assumed to involve integrating successive actions into groups known as chunks that can be efficiently planned and smoothly executed. Chunking is suggested by gains in planning efficiency for long compared to short action sequences following practice and learning associations between actions and perceptual outcomes. Less is understood about how efficiently we plan sequential chunks when we switch between multiple action sequences. Do we plan learned chunks less efficiently when we switch to a different action sequence? We examined this question by comparing the initiation and execution latencies of long versus short action sequences, performed from memory, when sequences switched or repeated across trials. Additionally, each action within the sequences generated predictable perceptual outcomes that were either spatially compatible or spatially incompatible with the action sequences. Results suggested repetition costs (instead of benefits) when performing long sequences. Repetition, as opposed to switching, prolonged initiation and increased the error rate of long compared to short sequences. We attribute these results to the flexible coordination of chunk planning and execution. Repetition may prolong advanced planning of long sequences in order to resolve conflict between multiple chunks, and switching may allow the planning of later chunks to be postponed until execution. We propose that the chunking of action sequences can both facilitate and interfere with action-switching performance.

Repetition costs in task switching are not equal to cue switching costs: evidence from a cue-independent context

Recent task-switching studies highlighted the presence of feature binding processes. These studies documented that even a task-irrelevant feature (the context, henceforth) may be bound with the task and the response in each trial. When the context repeated in the following trial, it supposedly retrieved the bound features, causing benefits when the task and the response repeated and costs otherwise (i.e. full repetition benefits). In the present study, we aim to rule out an alternative explanation for such full repetition benefits in task switching. These benefits were observed in studies that used a cue-related context so that full repetition conditions always implied a cue repetition. Therefore, these full repetition benefits may be ascribed to the priming of cue encoding, instead of the binding of the context. In the present study, we implemented a similar context manipulation but used univalent target stimuli and did not present any cue. Hence, the varying context was never cue-related. We still found full repetition benefits but only when the context appeared before the target and not when they appeared simultaneously. Thus, full repetition benefits can be observed in the absence of priming of cue encoding. However, the context must occupy a prominent position (i.e. at the beginning of the trial). These results, therefore, reinforce the hypothesis that full repetition benefits stem from binding processes that take place on a trial-by-trial basis and involve both task-relevant (the task and the response) and task-irrelevant features (the context).

Task-relevance and change detection in action-effect binding

Features of actions are bound to coincidentally occurring stimuli so that re-encountering a stimulus retrieves a previous action episode. One hallmark of the purported mechanism in binding/retrieval tasks is a reliable reaction time advantage for repeating a previous response if tone stimuli repeat rather than alternate across trials. Other measures than reaction times yielded surprisingly mixed results, however. This is particularly true for continuous response features like force or response duration. We therefore conducted two experiments to resolve this disconnect between different measures. Experiment 1 tested for a potentially inflated effect in reaction time data, whereas Experiment 2 took the converse approach of studying conditions that would elicit similarly strong effects on alternative measures. Our results show that confounds in terms of auditory change detection do not inflate reaction time differences, reinforcing an interpretation of these effects as reflecting binding and retrieval. Moreover, strong effects on alternative measures appeared if these features were rendered task-relevant and came with sufficient variability. These observations provide critical evidence for binding and retrieval accounts, especially by showing that these accounts extend from binary decisions to continuous features of an actual motor response.

Response-repetition costs in task switching do not index a simple response-switch bias: Evidence from manipulating the number of response alternatives

Response repetitions aid performance when a task repeats but impair performance when a task switches. Although this interaction is robust, theoretical accounts remain controversial. Here, we used an un-cued, predictable task-switching paradigm with univalent targets to explore whether a simple bias to switch the response when the task switches can explain the interaction. In Experiment 1A (n = 40), we replicated the basic interaction in a two-choice task. In Experiment 1B (n = 60), we observed the same interaction in a three-choice task, wherein a bias to switch the response when the task switches cannot prime a specific alternative response because both remaining response alternatives are equally likely. Exploratory comparisons revealed a larger interaction between task repetition and response repetition in the three-choice task than in the two-choice task for mean response time (RT) and the opposite pattern for mean error rate (ER). Critically, in the three-choice task, response-repetition costs in task switches were significant in both RT and ER. Since a bias to switch the response cannot prime a specific response alternative in a three-choice task, we conclude that such a bias cannot account for response-repetition costs in task-switch trials.

What is behind partial repetition costs? Event-files do not fully occupy bound feature codes

Feature binding accounts state that features of perceived and produced events are bound into event-files. Performance while responding to an event is impaired when some, as opposed to all or none, of this event's features already belong to a previous event-file. While these partial repetition costs are generally considered to be indicators of feature binding, their cause is still unclear. Possibly, features are fully occupied when bound in an event-file and must be unbound in a time-consuming process before they can enter a novel event-file. In this study, we tested this code occupation account. Participants responded to the font color (target) of a word (distractor) by pressing one of three keys (response) while ignoring the word meaning. We measured partial repetition costs from prime to probe while introducing an intermediate trial. We compared sequences in which this intermediate trial did not repeat any prime features and sequences in which it repeated either the prime response or distractor. Partial repetition costs occurred in the probe, even when one (vs. none) of the prime features repeated in the intermediate trial, although significantly reduced. Thus, single bindings do not fully occupy feature codes. By ruling out a possible mechanism behind partial repetition costs, the present study contributes to the further specification of feature binding accounts.