Featured ArticleResearch Articles, Behavioral/Cognitive

Lingering Neural Representations of Past Task Features Adversely Affect Future Behavior

Benjamin O. Rangel, Eliot Hazeltine and Jan R. Wessel
Journal of Neuroscience 11 January 2023, 43 (2) 282-292; DOI: https://doi.org/10.1523/JNEUROSCI.0464-22.2022

Abstract

During goal-directed behavior, humans purportedly form and retrieve so-called event files, conjunctive representations that link context-specific information about stimuli, their associated actions, and the expected action outcomes. The automatic formation, and later retrieval, of such conjunctive representations can substantially facilitate efficient action selection. However, recent behavioral work suggests that these event files may also adversely affect future behavior, especially when action requirements have changed between successive instances of the same task context (e.g., during task switching). Here, we directly tested this hypothesis with a recently developed method for measuring the strength of the neural representations of context-specific stimulus–action conjunctions (i.e., event files). Thirty-five male and female adult humans performed a task switching paradigm while undergoing EEG recordings. Replicating previous behavioral work, we found that changes in action requirements between two spaced repetitions of the same task incurred a significant reaction time cost. By combining multivariate pattern analysis and representational similarity analysis of the EEG recordings with linear mixed-effects modeling of trial-to-trial behavior, we then found that the magnitude of this behavioral cost was directly proportional to the strength of the conjunctive representation formed during the most recent previous exposure to the same task, that is, the most recent event file. This confirms that the formation of conjunctive representations of specific task contexts, stimuli, and actions in the brain can indeed adversely affect future behavior. Moreover, these findings demonstrate the potential of neural decoding of complex task set representations toward the prediction of behavior beyond the current trial.

SIGNIFICANCE STATEMENT Understanding how the human brain organizes individual components of complex tasks is paramount for understanding higher-order cognition. During complex tasks, the brain forms conjunctive representations that link individual task features (contexts, stimuli, actions), which aids future performance of the same task. However, this can have adverse effects when the required sequence of actions within a task changes. We decoded conjunctive representations from electroencephalographic recordings during a task that included frequent changes to the rules determining the response. Indeed, stronger initial conjunctive representations predicted significant future response-time costs when task contexts repeated with changed response requirements. Showing that the formation of conjunctive task representations can have negative future effects generates novel insights into complex behavior and cognition, including task switching, planning, and problem solving.

Introduction

According to influential theories of human cognition (Hommel et al., 2001; Frings et al., 2020), humans automatically store and retrieve episodic representations of their interactions with the world to guide future behavior. These representations contain context-specific combinations of stimuli, actions, and outcomes, and are also known as event files (Hommel et al., 2001; Hommel, 2009). For instance, in the example of opening a safe, an event file would contain a conjunction of the specific context (e.g., the room with the safe), a stimulus (the combination lock), an action (entering the combination), and an outcome (the opening of the safe). The formation and retrieval of such conjunctive event files can greatly facilitate behavior by making repeated actions more efficient.

However, it is also observed that these event files can have adverse effects on behavior (Hommel, 2004; Hommel and Colzato, 2004; Frings et al., 2020). For example, when the correct response or its outcome contingencies change between successive exposures to the same context (e.g., if the combination of the safe has been recently changed), preexisting event files may contain outdated representations and prime an inappropriate action (Neill, 1997; Waszak et al., 2003; Hommel et al., 2014). In the laboratory, these effects can be measured in the N-2 Repetition Cost (N2RC) paradigm (Mayr and Keele, 2000; Mayr, 2002; Grange et al., 2013; Kowalczyk and Grange, 2016). In such experiments, triplets of successive trials made in separate task contexts (C-B-A) are compared with triplets of trials that include an N-2 repetition of task A (A-B-A). The N2RC effect is a reaction time cost on the third trial of an ABA compared with a CBA triplet when the response on third trial differs from the response on the first trial (Fig. 1).

Figure 1.
Figure 1.

The event-file hypothesis of N-2 repetition costs. A, Diagram of how individual task features are bound as event files and then retrieved at a later time. B, Generic task diagram for ABA and CBA trial triplets. Binding of the event file in trial N-2 occurs for both tasks A and C, but Retrieval of N-2 event file only occurs in trial N of an ABA triplet, not a CBA triplet.

However, there is ongoing debate over the source of the N2RC and whether it indeed relates to the formation of event files (Grange et al., 2017; Grange, 2018; Kessler, 2018; Scheil and Kleinsorge, 2019, 2021; Kowalczyk and Grange, 2020). For example, an influential alternative account attributes N2RC effects to the inhibition of task set A by the intervening switch to task B. This may impair the return to the previous, now-inhibited task context (Mayr, 2002; Philipp and Koch, 2006; Koch et al., 2010; Schuch and Grange, 2015; Sexton and Cooper, 2017).

In the current study, we address this question from a neuroscientific perspective, using a recently developed method for the quantification of the type of conjunctive representation predicted by the event-file framework (Kikumoto and Mayr, 2020). Single-neuron recordings and fMRIs have already suggested that such conjunctive representations may exist in both nonhuman (Rigotti et al., 2013; Parthasarathy et al., 2017) and human brains (Kühn et al., 2011). Kikumoto and Mayr's (2020) method uses multivariate pattern analysis (MVPA) and representational similarity analysis (RSA) to noninvasively track the emergence of such representations with millisecond precision from human EEG recordings. This method also allows for a direct test of the relationship between the strength of these representations and behavior via linear mixed-effects modeling of trial-to-trial reaction time. Kikumoto and Mayr (2020) demonstrated that conjunctive representations decoded from EEG recordings can be used to predict reaction time on the current trial; stronger conjunctions between task context, stimulus information, and response activity accompany faster reaction times.

Here, we leverage this technique to investigate whether conjunctive representations can also account for N2RC effects, that is, whether the formation of a conjunctive event file on one trial has adverse effects on future trials, as predicted by the event-file account of the N2RC effect. Using an N2RC paradigm, we first aimed to replicate Kikumoto and Mayr's (2020) finding that stronger conjunctions of task context, stimulus, and response accompany faster reaction times on the same trial. We then hypothesized that within ABA triplets, the strength of the conjunctive representation formed on the first instance of task A would be detrimental to reaction time on the second instance of task A if a different response was required on the second instance. This would provide direct evidence for the N2RC effect being partially attributable to previously formed, yet now outdated, event files. More broadly, it would indicate that although event-file formation is typically beneficial to immediate behavior, it can also be detrimental when the behavioral requirements associated with a specific context subsequently change.

Materials and Methods

Participants

Thirty-five healthy young adults participated in the experiment [age mean (SD), 19.6 (3.0), 1 left handed, 19 females]. This sample size was based on previous work using a similar behavioral paradigm (Grange et al., 2017, their Appendix B). Participants were either paid $15 per hour or received course credit for their participation in the study. All participants had normal or corrected-to-normal vision. The experiment was approved by the ethics committee at the University of Iowa (Institutional Review Board #201511709).

Materials and procedure

Experimental stimuli were presented via an Ubuntu Linux computer, running Psychtoolbox-3 software (Brainard 1997) under MATLAB 2015b (MathWorks). Participants sat upright with their arms and the computer keyboard resting on a supportive platform placed on the arm rests of the chair.

Experimental paradigm

Our paradigm was adapted from Grange et al., (2017; experiment 2; Fig. 2). A large hollow square outline (the frame) remained on the screen across trials. Trials began with the presentation of the task cue (the outline of one of three shapes) inside the center of the frame, followed by a 500 ms cue-target interval. The target (filled black dot) would then appear in one of the four corners of the frame.

Figure 2.
Figure 2.

Experimental paradigm. A, Diagram for a single trial. The shape cue is presented for 500 ms before the target (black dot) onset. The target will be displayed for 2000 ms or until a response is made. Subjects responded using the number pad on a full keyboard. Response is immediately followed by a 200 ms blank inter-trial interval before the next cue. B, The three possible shape cues and their meanings for the task. C, Examples of ABA-S and CBA trial triplets.

Based on the combination of cue shape and target location, participants had 2000 ms to respond using their right index finger. The cue shapes Hexagon, Triangle, and Square instructed one of three different response mappings (tasks), which corresponded with Vertical, Horizontal, and Diagonal movement of the target to one of the unoccupied corners of the frame. For example, should the target appear in the top-left corner of the frame, a triangle cue would imply that the target should move horizontally to the top right, hence cuing the response associated with that location. Participants responded using their right index finger and the 4, 5, 1, and 2 keys on the keyboard number pad, which corresponded to the top-left, top-right, bottom-left, and bottom-right corners of the frame, respectfully. Participants were asked to return their index finger to the center of the response keys after each trial.

Responses were followed by a 200 ms blank period where only the frame remained on screen. If participants failed to respond within 2000 ms, or made an incorrect response, the usual blank period was replaced by a 500 ms presentation of the words Faster or Error in red text. The experiment consisted of 800 trials (eight blocks). Cue identity and target location were pseudorandomized to prevent immediate cue (and task) repetitions and an equal number to task conditions (50% ABA triplets, 50% CBA triplets).

The first two trials of each block were removed from the analysis, as they could not be considered part of a triplet. The behavioral data were then screened for response errors/omissions, RTs lower than 150 ms (three trials overall), and RTs outside 2.5 SDs of the mean (2.33% overall). These trials were removed from further analysis, along with the two subsequent trials. There were three possible task triplet conditions of interest (ABA-S, ABA-R, CBA) with the -S and -R denoting either a switch or repeat in the required response between trials N-2 and N of an ABA triplet. (In principle, CBA-S and CBA-R also exist, but as there is no repetition of task cues within these triplets, the overlap of event files should be zero, regardless of response requirement.) Our main comparison of interest was between ABA-S trials and CBA trials. ABA-R trials were of secondary interest as there was only a limited amount of those triplets per subject (as target locations were drawn from a uniform distribution to not introduce a bias, resulting in only one ABA-R triplet for every three ABA-S triplets). The dependent variables of interest were RT and accuracy on the final trial of ABA-S and CBA triplets, which were compared using paired-samples t tests.

EEG recording

EEG data were recorded using a 64-channel active electrode cap connected to an actiCHamp amplifier (Brain Products) at a rate of 500 Hz (10 s time-constant high-pass and 1000 Hz low-pass hardware filters). Electrodes Pz and Fz served as the reference and ground, respectively.

EEG preprocessing

Raw EEG data were preprocessed using custom MATLAB scripts and EEGLAB (RRID:SCR_007292) toolbox functions (Delorme and Makeig, 2004). Individual participant data were imported, resampled to 250 Hz, and filtered using Hamming windowed sinc FIR filters (pop_eegfiltnew.m; high-pass, 0.01 Hz; low-pass, 50 Hz). The resulting data were then epoched into two datasets, one target locked [−600:800 ms] and the other response locked [−700:200 ms], which were used to investigate the emergence of task-feature representations relative to either event. Epochs containing nonstereotypical artifacts were then removed from further analysis (joint probability and joint kurtosis; 5.5 SD cutoffs; compare Delorme et al., 2007). Data were rereferenced to a common average before independent component analysis decomposition (Infomax; Bell and Sejnowski, 1995) with extension to sub-Gaussian sources (Lee et al., 1999). Components representing eye movement and electrode artifacts were identified using outlier statistics and removed from the data. Unlike Kikumoto and Mayr (2020), EEG data did not undergo time-frequency analysis, and the following analyses were performed on the raw EEG channel data (i.e., the event-related potential).

Linear discriminant analysis

Linear discriminant analysis (LDA) was run on one subject at a time. Each subject had 12 randomly selected trials held out as test data, whereas their remaining trial data were used for training. Training trial data were then sorted into 12 classes, one for each unique combination of cue, target, and response (i.e., the event file or constellation in Kikumoto and Mayr, 2020). In the case of unequal numbers of trials across these classes, trials were randomly excluded to achieve equal numbers. The excluded trials were randomly paired with the remaining trials in their respective classes and averaged. This 1–1 pairing and averaging allowed for all data to be included in each training and testing loop while also ensuring an LDA with balanced classes. LDA was conducted using the MATLAB fitdiscr function. Each training-testing loop applied the LDA to each time point, with all 63 EEG channels included as features. Posterior probabilities for the hold-out data (12 trials) were saved for each time point. This process was repeated with a new set of 12 hold-out trials until each trial had been tested. This entire process was then repeated for 10 iterations, resulting in each trial being included in the testing set 10 times.

RSA

The posterior probabilities for all trials and iterations were smoothed using an overlapping moving median window of 20 ms. As in Kikumoto and Mayr (2020), the posterior probabilities of each trial for each time point were regressed onto model vectors representing the idealized classification probabilities for the true class identity of the trial (Fig. 3). Specifically, four vectors were used as predictors in the regression, representing the cue, target, response, and the conjunction of all three (i.e., the event file). The t values of these predictors were saved as measures of similarity. A fifth vector of z-scored/averaged reaction times for each class was also included to account for variations in RT. Following RSA, the median t value for all predictors at each time point of each trial was taken across all 10 iterations. (The choice to use the t statistic as the measure of similarity rather than the raw betas was made to faithfully replicate the methods of Kikumoto and Mayr (2020). They themselves did not explain their choice between the t statistic and the raw betas.)

Figure 3.
Figure 3.

Analysis logic, analogous to Kikumoto and Mayr (2020). A, LDA was run on single-subject EEG data, using all 63 channels as features. Each time point was decoded (trained/tested) individually across all trial epochs. B, Posterior probabilities indicating how likely a single sample belonged to each of the 12 classes were saved in a 4D matrix (Class × Time × Trials × Iterations). C, The RSA consisted of taking individual sample points and regressing them onto the corresponding vector from idealized model matrices. D, The t values for each predictor in the regression were saved as a measure of similarity. Once all samples had been regressed, the median t values across the 10 iterations were used for further analysis.

Mixed-model analysis

As in Kikumoto and Mayr (2020), the representational strengths of each of the four factors of interest (cue, target, response, conjunction) in the neural data were related to behavior (reaction time) using a mixed-effects model. The t values from the RSA were baseline corrected (first 100 ms) for each subject. Values outside ±5 * SDs the mean were excluded, and the data were then smoothed using an overlapping 80 ms moving mean window. Subject RTs were then log transformed (ln) and detrended to remove any linear and/or quadratic trends. Subjects' data were first modeled with the RSA values for cue, target, response, and conjunction added as fixed effects predicting the RT on the same trial. Subject was included as the only random effect (Eq. 1). Subject behavioral data and RSA values were then trimmed to include only trials that made up a complete ABA or CBA triplets (uninterrupted by behavioral or EEG rejections). Two models were created to predict trial N RT, one using trial N-2 data predictors (Eq. 2) and the other, N-1 predictors (Eq. 3). In both cases the N-1 or N-2 RSA values for cue, target, response, and conjunction were added as fixed effects, as well as the trial N triplet (ABA-S or CBA). Interactions for all RSA values with the trial N conditions were also included. Subject was included as the only random effect. This analysis was applied to all time points of the trials. All mixed-model analyses were conducted in R (version 4.0.4, packages car, lme4) as follows: ln(rt(N)) ∼ 1 + cue(N) + tar(N) + resp(N) + conj(N) + (1|subject)(1) ln(rtN) ∼ 1 + (cue(N−2) + tar(N−2) + resp(N−2) + con(N−2))*tripletABA−S & CBA + (1|subject)(2) ln(rtN) ∼ 1 + (cue(N−1) + tar(N−1) + resp(N−1) + con(N−1))*tripletABA−S & CBA + (1|subject)(3)

Cross-correlation analyses

To investigate the relationships between the strengths of the neural task representations across the trials of a triplet, the same response-locked data used in the final mixed model (Eqs. 2, 3) were then cross-correlated for each subject. The RSA values were first averaged down to 45 time points (900 ms in 20 ms steps) for the following analysis. The conjunction strength in trial N was predicted using four mixed models, one for each N-2 predictor (cue, target, response, conjunction). These models included both ABA-S and ABA-R triplets, with CBA triplets as the reference group and subject as the random effect (Extended Data Fig. 8-1). These models were run for each time point in trial N and N-2, producing 45 × 45 correlation matrices of β coefficients. A similar analysis was performed using N-1 RSA values to predict trial N conjunction strength. However, these models differed in that they were performed for each triplet separately and included all trial N-1 task features as predictors (Extended Data Fig. 8-4). The purpose of this was to remove the comparisons between triplets, so the true pattern of N-1 to N correlation could be observed.

To obtain single values for each subject, trial triplets were sorted into three condition groups (ABA-S, ABA-R, CBA). The RSA values for all task features at every time point in trials N-2 and N-1 of each group were compared with the corresponding RSA values at every time point in trial N, using Pearson's correlations. This produced a 225 × 225 matrix of Pearson's r values comparing the task features in trials N-2 and N-1 with those in trial N for each condition and subject. These matrices were Fisher's z transformed before plotting and further testing (Fisher, 1915). The difference between the cross-correlation matrices comparing the conjunctions in trial N-2 and N for ABA-S and CBA triplets was tested using a one-sided permutation test for dependent measures (10,000 permutations; Gerber, 2022). A difference matrix (ABA-S – CBA) was produced for each subject and then averaged across both dimensions, postbaseline (first 100 ms), resulting in a single value per subject (Extended Data Fig. 8-2). This was correlated with each subject's N2RC (Pearson's).

Results

Behavior

Subjects failed to respond on 1.09% of trials and committed response errors on 4.05% of trials overall. As predicted, a comparison of ABA-S and CBA trials reveals N-2 Repetition Cost effects on both RT (ABA-S, mean = 748 ms, SE = 26.3; CBA, mean 707.4 ms, SE = 26.1; Fig. 4A) and accuracy (ABA-S, mean = 94.2%, SE = 0.82; CBA, mean 95.1%, SE = 0.74). Paired-samples t tests revealed that these differences were highly significant for both RT (t(34) = 7.48, p < 0.001) and accuracy (t(34) = −3.74, p < 0.001; Fig. 4A).

Figure 4.
Figure 4.

Average RT performance per subject. A, Plots depicting the mean performance of subjects in trial N of ABA-S and CBA triplets. Individual subject performance is represented by gray spheres, whereas the grand mean and SEM are represented by black horizontal and vertical lines, respectively(** p < .001). B, Comparison of trial numbers and median RT between the current study and that of Kikumoto and Mayr (2020). The average trial number for both experiments only exclude trials not responded to accurately. The median RT (box, 25th and 75th percentiles) for each subject is shown in A, B; outliers are not shown.

Decoding conjunctive representations on the current trial

In a first step, we aimed to replicate the results from Kikumoto and Mayr (2020) to validate that we could identify the conjunctive representation of cue, target, and response (i.e., the event file) from the neural data. Following EEG epoch rejections, each subject had an average of 517.4 (±58.0) cue/target-locked trial epochs and 515.7 (±58.2) response-locked trial epochs. From the EEG data of each subject, we successfully produced significant point-by-point estimates of representational strength for all task features (cue, target, response) and their unique conjunction (event file). Single-subject RSA averages (t values from regression of models onto LDA output) were plotted as a grand average, time locked to both cue/target onset (Fig. 5A) and response onset (Fig. 5B). Notably, the representations for the task cue, target, and conjunction were significantly decoded across both time series. The response representation was only significantly decoded when the data were response locked, likely because of variance in RT (Fig. 4B).

Figure 5.
Figure 5.

Decoding current trial representations of cue, target, response, and their conjunction from the neural data. A, Grand average of the cue/target-locked RSA output. B, Grand average of response-locked RSA output.

Furthermore, the mixed-effects model approach produced results similar to those of Kikumoto and Mayr (2020; Fig. 6). Analysis of the response-locked RSA values (Fig. 5B) showed that increased strength of the cue, target, and conjunctive representations predicted faster RTs on the same trial. The response representation did not produce a significant relationship by itself. (However, it did in the stimulus data; Extended Data Fig. 6-1.)

Figure 6.
Figure 6.

Relationship between behavior and neural representations on the same trial. Plot shows the β coefficients from the mixed-model analysis of response-locked RSA values, fitted at each time point for all subjects and trials. Negative β values indicate that greater representational strength predicts faster RT during the same trial. The results from the same analysis run on stimulus-locked data can be seen in Extended Data Figure 6-1.

Figure 6-1

Plot shows the β coefficients from the mixed-model analysis of Cue/Target locked RSA values, fitted at each time point for all subjects and trials. Positive β values indicate that greater representational strength predicts slower RT during the same trial. Of note, these results differ from those of Kikumoto and Mayr (2020). However, we believe this is again because of the large variability in our subjects' RTs compared with Kikumoto and Mayr's (2020; Fig. 4B). Download Figure 6-1, EPS file.

Main analysis: trial N-2 conjunctions explain RT cost on trial N

We then tested our main hypothesis, that is, that in ABA-S triplets, the neural representation of the conjunction (event file) on trial N-2 would negatively affect the reaction time on the current trial N. A mixed-effects model was used to investigate the relationship between the trial N-2 representational strength of each task factor (cue, target, response, conjunction) and the reaction time of trial N, using the response-locked RSA values. The β coefficients for the interaction terms of each model were plotted at each time point in Figure 7. As hypothesized, the strength of the conjunctive representation in trial N-2 significantly predicted increased trial N RTs. Significant time periods were found from −175 to −83 ms and from −63 to 33 ms, relative to the response, for ABA-S triplets compared with CBA triplets (FDR < 0.05; Fig. 7, left). In other words, the greater the strength of the conjunctive representation on trial N-2, the slower the RT of trial N in an ABA-S triplet. This was only true for the conjunctive representation and not for any individual factor (cue, target, response) by itself. Moreover, the conjunctive representation on the immediately preceding trial N-1 did not predict anything about trial N RT (Fig. 7, right). Beyond these interactions, the mixed-model analysis did not show any simple main effects that survived FDR correction (p < 0.05; Extended Data Fig. 7-1). These analyses were also repeated using the stimulus-locked data but yielded no significant interactions or main effect (Extended Data Fig. 7-2).

Figure 7.
Figure 7.

Plots show the β coefficients of the interaction terms (trial condition × RSA output) from the mixed-model analysis, fitted at each time point, across all subjects and trials. No significant main effects were observed (Extended Data Figure 7-1). Positive β values (left y-axis) indicate that increases in the RSA values (greater representational strength) on trial N-2 correspond with slower trial N RT in ABA-S trials, compared with CBA trials (right y-axis). Left, The output of the mixed model, which only included the RSA values from trial N-2 as predictors for trial N RT. Right, The same but for the RSA values of N-1 trials only. The same mixed-model analysis run on stimulus-locked data did not produce any significant main effects or interactions after FDR correction (Extended Data Fig. 7-2).

Figure 7-1

The main effects from the mixed-model analysis of response-locked RSA data (Eqs. 2, 3). The shaded regions indicate the SEM at each time point. Plots show the β coefficients for the main effects for all task features in N-2 trials (left) and N-1 trials (right), predicting trial N RT. No main effects were significant following FDR correction (p < 0.05). Download Figure 7-1, EPS file.

Figure 7-2

Plots show the β coefficients from the mixed models ran on the stimulus-locked data. The shaded regions indicate the SEM at each time point. A, The main effects from the mixed-model analysis of stimulus-locked RSA data (Eqs. 2, 3). B, The interactions from the mixed-model analysis of stimulus-locked RSA data (Eqs. 2, 3) No main effects or interactions were significant following FDR correction (p < 0.05). Download Figure 7-2, EPS file.

Exploratory analysis: quantifying cross-trial interactions between conjunctive representations

To further explore the cross-trial dynamics that give rise to the N2RC, we calculated cross-correlations between the representational strengths of the conjunctive representations on the first and third trials within ABA and CBA triplets (Fig. 8). The output of the mixed-model analysis shows a significant negative main effect (FDR < 0.05), which is likely driven by CBA triplets. For ABA-S triplets, the interaction shows a notable but not significant positive difference from CBA triplets, peaking after the response in trial N (Fig. 8A). This suggests that in CBA triplets, stronger conjunctions on trial N-2 corresponded to the formation of weaker conjunctions on trial N, indicating possible interference between the different event files corresponding to tasks C and A. This effect is either absent or marginally reversed on ABA-S trials, whose event files partially overlap. These findings are also consistent with the Pearson's correlation matrices (Extended Data Fig. 8-2). For ABA-R triplets, a large and significant positive difference from CBA triplets was observed. As expected, the identical event files of trials N-2 and N in these triplets are positively correlated. To investigate the critical contrast underlying the N2RC, we took the difference between the N2-N Pearson's correlation matrices of ABA-S and CBA triplets and averaged them to obtain a single value per subject (Extended Data Fig. 8-2). These differences in correlation directly related to the size of the N2RC across subjects; that is, subjects with greater differences in the Pearson's cross-correlations between CBA and ABA-S trials showed greater N-2 repetition costs (Fig. 8B). Control analyses revealed that the same was not true for the correlations between nonconjunctive representations (cue, target, response). As would be expected, this effect appears to be driven by an increase in the RTs of ABA-S triplets, as opposed to a decrease in CBA triplet RT (Extended Data Fig. 8-3).

Figure 8.
Figure 8.

Exploration of the representational cross-correlation matrices. A, The three correlation matrices represent the relationship between the conjunction strength of trials N-2 and N at each time point. The main effect and two interactions (left to right) were produced using a mixed model (Extended Data Fig. 8-1). Regions of significance following FDR correction (p < 0.05) are denoted by black lines. B, Scatter plots show the relationship between subjects' averaged difference in ABA-S and CBA Pearson's cross-correlation matrices (Extended Data Fig. 8-2, N-2 and N) and their individual N2RC for all task features. Only the conjunction shows a significant relationship (p = 0.03). This finding remains true even when comparing ABA-S and CBA trials separately (Extended Data Fig. 8-3). C, These example matrices show the distinct patterns of nonoverlapping (N-1 and N) and conjunctive representations for all three triplet conditions. The mixed model also included the individual task features from trial N-1 as predictors, which similarly showed an absent or negative correlation with trial N conjunctive strength (Extended Data Fig. 8-4). For ABA-S triplets, the N-2 conjunction was unable to be decoded on trial N (Extended Data Fig. 8-5).

Figure 8-1

The above correlation matrices show the β coefficients produced by the four models above, run on response-locked RSA values. These models predicted the strength of the conjunction on trial N, using a single N-2 task-feature as the main predictor. Interactions between these task features and the triplet conditions were included (ABA-S/R vs CBA as reference). Each column denotes the main effects and interactions from each model. Black outlines indicate where an effect was significant following FDR correction (<0.05, tiles 4, 11, 12). Download Figure 8-1, TIF file.

Figure 8-2

The above correlation matrices show the grand mean Pearson's correlations between the response-locked RSA values of the trial N-2 conjunction and trial N conjunction, at each time point, for ABA-S triplets (left) and CBA triplets (middle). The difference between subjects' ABA-S and CBA matrices was tested using a one-sided permutation test for dependent measures (10,000). The marginally significant cluster is highlighted in the difference matrix (right). The difference between each subject's ABA-S and CBA matrices was averaged across both dimensions postbaseline (first 100 ms) to produce a single value (Fig. 8B, x-axes). Download Figure 8-2, TIF file.

Figure 8-3

Scatter plots show the relationship between subjects' averaged difference in ABA-S and CBA Pearson's cross-correlation matrices (Extended Data Fig. 8-2, N-2 and N) and their RTs for ABA-S (colored plots) and CBA (gray plots) trials. Only the relationship between the conjunction and ABA-S RT is significant (p = 0.03). Download Figure 8-3, EPS file.

Figure 8-4

The above correlation matrices show the β coefficients produced by the model above, run on response-locked RSA values for each triplet condition. These models predicted the strength of the conjunction on trial N, using all single N-1 task-features as predictors. Thus, each row denotes the main effects from each model. The choice to separate triplets rather than have them compared as in previous models (interactions) was made to preserve the interpretability of the matrices. For example, a positive difference in the correlation of N-1 and N conjunctions between ABA-R and CBA trials would manifest as a positive effect in a correlation matrix, although as a whole, there is a negative relationship between N-1 and N conjunctions in ABA-R triplets. Download Figure 8-4, EPS file.

Figure 8-5

The above plots show the RSA output for trial N of ABA-S triplets for stimulus-locked (left) and response-locked (right) data. Positive t values indicate greater representational strength, and the colored bars below these traces indicate where the RSA values are significant (FDR < 0.001). Download Figure 8-5, EPS file.

To buttress this analysis, we explored whether the correlation patterns of other trial combinations matched the above-mentioned interpretation. Regardless of triplet condition, trial N-1 never fully overlaps with the conjunctions of trial N-2 or N. As such, the mixed-model output matrices for trial N-1 conjunctions predicting trial N conjunction strength all show a notable negative correlation consistent with nonmatching event files (Fig. 8C). This suggests that when there is a mismatch in task features as early as the cue (i.e., a task switch), a stronger conjunctive representation of a past task feature combination will impair the formation of the new conjunction on the current trial.

Together, these correlation patterns suggest that the formation of a neural representation of the task feature conjunction on a past trial (N-2) benefits the formation of the conjunctive neural representation on the present trial (N) if the task sets match and is detrimental to the formation of a different conjunction if they mismatch. Moreover, the cross-subject Pearson's correlations suggest that the strength of these processes is directly proportional to the size of the N2RC effect. In contrast to our main analysis, these exploratory analyses were performed post hoc and therefore must be interpreted with caution. However, they generate further insights into the neural dynamics of how the trial N-2 task set may affect behavior on trial N to affect the N2RC.

Discussion

In the current study, we used a recently established neurophysiological method for tracking conjunctive representations between specific task contexts, stimuli, and responses (i.e., event files; Hommel et al., 2001) to address an ongoing debate about the role of such event files in task switching. Based on the methodological groundwork of Kikumoto and Mayr (2020), we leveraged the combination of RSA and MVPA of EEG data to investigate the trial-to-trial relationship between the strength of conjunctive event-file representations and behavior. However, although Kikumoto and Mayr's (2020) investigation evaluated immediate task repeats/switches, we assessed whether the formation of an event file on the current trial can influence behavior two trials later. Using the N-2 Repetition Cost as a behavioral model, we found that the strength of the conjunctive representation of cue, target, and response on a previous trial indeed adversely influences performance on the current trial, specifically, in cases in which response requirements changed between two instances of the same task. In this way, we provide a direct test of a critical component of the event file hypothesis, that is, that conjunctive representations are encoded and retrieved for the performance of future actions (Hommel et al., 2001; Hommel, 2019; Frings et al., 2020).

Kikumoto and Mayr (2020) demonstrated that partial-repetition costs can be partially explained by the strength of the conjunctive representation formed in the immediately preceding trial. The more strongly encoded the conjunction (event file) is on trial N-1, the more difficult it would be to unbind the constituent task features and reintegrate them into a new conjunction, leading to greater partial-repetition costs. By replicating this effect across trial triplets (ABA and CBA), we were able to show that conjunctive representations formed on trial N-2 can also affect behavior on trial N, despite there being an interleaving task switch (N-1). These findings further support the Theory of Event Coding (Hommel et al., 2001), which proposed that event files are created during each trial of a task and can also be automatically retrieved and reintegrated across extended periods of task performance.

Furthermore, the exploration of correlation patterns between the strength of conjunctive representations on different trials within triplet conditions generated further insight into the dynamics by which this adverse influence of past event files may manifest itself. First, the comparison of nonmatching task contexts (trial N-1 vs trial N comparisons) revealed a stereotypic pattern of decorrelation between the strength of the respective conjunctive representations of successive, different task contexts (Fig. 8C). In other words, the stronger the representational strength of feature conjunctions on the current trial, the weaker the representational strength of feature conjunctions of subsequent nonmatching trials. This suggests that once-formed conjunctive representations may generally interfere with the formation of feature conjunctions on subsequent trials, a phenomenon that may underpin task-switch costs more generally. Moreover, as we observed in the main effect of our mixed-model output comparing the conjunctive strength between trials N-2 and N (Fig. 8A), the comparison of task C and task A conjunctions of CBA triplets revealed that this pattern extended beyond the immediately subsequent trial; that is, stronger representations of feature conjunctions on a given trial still impair the formation of feature conjunctions two trials later. Most interestingly, the comparison of ABA-R triplets to CBA triplets provided a possible explanation on why this takes place, as it showed the opposite pattern between the N-2 and trial N conjunctions (i.e., between two spaced trials with matching event files). In such triplets, stronger conjunctive representations on the first instance of task set A (trial N-2) correlated with stronger conjunctive representations on the second instance (trial N). In line with the event-file hypothesis, this suggests that the adaptive function of this apparent forward propagation of a task set is to benefit behavior if the task set repeats/remains the same. However, if the task set switches, the anticorrelation pattern ensues. Such an automated storage-and-retrieval process makes sense in everyday scenarios the brain has evolved to handle, in which task sets do not typically change with the same regularity as in laboratory experiments such as the current one. ABA-S trials then form a middle ground between the mismatch scenarios provided by CBA triplets, on the one hand, and the full-match scenarios provided by ABA-R trials, on the other; the decorrelation observed on mismatch trials does not take place as the strength of the N-2 conjunction does not adversely influence the formation of the task N conjunction (Fig. 8). However, as the trial unfolds and the target indicates that response requirements have changed compared with the most recent presentation of the cue, conflict ensues and results in the behavioral cost expressed in the N2RC effect. Indeed, as seen in Figure 8B, the difference in correlation between N-2 to N trial conjunctions in ABA-S triplets compared with CBA triplets (i.e., the main contrast of interest in the current study) was directly correlated with the N2RC effect across subjects. This correlation is very much in line with the results of our main analysis, that is, that the strength of N-2 conjunctive representations predicted longer reaction times in trial N of an ABA-S triplet. Indeed, subjects for whom the conjunctive trial N-2 representation was stronger—and therefore, purportedly interfered more strongly with the formation of the conjunctive trial N representation within a CBA triplet (or primed the response contained in the trial N-2 event file to a greater extent on trial N within an ABA-S triplet)—showed a larger N2RC effect.

Regarding our main finding—the inverse relationship between the strength of the conjunctive representation on trial N-2 and RT on trial N of an ABA-S triplet—two observations are key to rule out alternative explanations. First, the conjunctive representation of the intervening trial N-1 in an ABA-S triplet did not predict RT on the current trial. This confirms that the N-2 Repetition Cost is indeed specific to a precise match between task contexts. Second, neither the cue nor target representations of trial N-2 by themselves predicted behavior on trial N. This shows that regardless of how strongly the target or cue was encoded on trial N-2, only the strength of their binding with the task context in a conjunctive representation affected future behavior. As such, our results are consistent with previous work, suggesting that the episodic retrieval event files significantly contribute to the N2RC effect (Neill, 1997; Hommel, 1998; Spapé and Hommel, 2010; Kühn et al., 2011; Grange et al., 2017; Grange, 2018; Frings et al., 2020). Additionally, we are the first to demonstrate that a direct measurement of task representational strength from neural data can be used to predict future RT costs.

As such, we successfully used the decoding of conjunctive representations from neural recordings to not only show that greater task-feature binding can both be beneficial to current behavior and detrimental to future behavior but we also demonstrated how neurophysiological methods can provide crucial insights into behavioral phenomena that have hitherto mainly been investigated using experimental psychology. Moreover, our results provide a powerful demonstration of the event-file framework in explaining behavior and linking it to neural activity. Indeed, conjunctive representations of stimuli, actions, and outcomes appears to be a fundamental code that the human brain uses to guide future behavior. Although early conceptualizations of event or object files focused on the conjunction of visual stimulus representations (Kahneman et al., 1992; Treisman, 1998), the idea that actions and their outcomes can be commonly linked with perceptual information is hence a key insight into human adaptive behavior (Hommel, 1998; Hommel et al., 2001; Spapé and Hommel, 2010). This proposal sheds light on how event files allow us to do more than simply recognize a familiar sensory stimulus but also contextualize that stimulus by its expected effects, our own actions, and the surrounding environment.

One open question that remains is the role of backward inhibition in the N2RC effect. Past studies have shown that ABA triplets can show N-2 Repetition Costs even when the current response matches the past response (Grange et al., 2017; Grange, 2018), which is a counterintuitive finding under the event-file framework. The magnitude of these costs tends to be smaller than those in ABA-S trials, suggesting that they may represent a pure form of backward inhibition that can only be properly examined when the effects of episodic retrieval are controlled for (Grange, 2018; Kowalczyk and Grange, 2020). Although ABA-R trials were not the primary interest of the current study (as the experiment only contained a limited number of ABA-R trials and used a cue-target interval that was too long to expect substantial RT effects on ABA-R trials), the framework presented here could be adapted to specifically test this hypothesis. Indeed, neurophysiological signatures of inhibitory processes are well established in both the motor and cognitive domains (Anderson et al., 2016; Wessel and Aron, 2017), and it has long been proposed that task switch cues activate inhibitory networks (Jamadar et al., 2010; Koch et al., 2010). Some EEG studies examining the N2RC as a function of backward inhibition have produced contrary results, with differences in event-related potentials appearing over centroparietal and parietal sites (Sinai et al., 2007; Zhang et al., 2016) rather than over frontocentral electrodes, as would be expected. However, these studies did not properly account for episodic retrieval effects and were also testing additional task conditions that would obscure the results. The combination of the current approach with direct measurements of inhibition could provide insight into whether there is indeed an additional role for inhibition in the N2RC effect. Moreover, the analytical method developed by Kikumoto and Mayr (2020) could be adapted to identify signatures of the latent conjunctive representations established in trial N-2 before its retrieval on trial N (Rose et al., 2016) and to investigate whether task switching cues on the intervening trial N-1 directly affect the strength of the latent conjunctive representation.

Indeed, to perform an additional test of the retrieval hypothesis suggested by a reviewer, we attempted to decode the conjunctions formed during trial N-2 of ABA-S triplets during trial N as a post hoc analysis. In particular, it was suggested that the representation of the trial N-2 conjunction may be explicitly decodable from the EEG data of trial N and that the strength of this reoccurrence may again influence trial N RT, just like we found for its initial strength on trial N-2. This turned out not to be the case as no significant N-2 conjunction was able to be decoded during trial N (Extended Data Fig. 8-5). There are a few possible explanations for this finding. It is possible that the N-2 conjunction may be held as an activity-silent representation in working memory after being retrieved (Rose et al., 2016). Alternatively, the N-2 conjunction may simply lack the representational strength to be decoded from a sample of our size. Future investigations should focus on this aspect of the event-file hypothesis as it may be able to provide more direct evidence of how and when past event files are retrieved and reintegrated into working memory.

In summary, we here used the recently developed methodological approach for the measurement of conjunctive representations of task context, stimuli, and responses (Kikumoto and Mayr, 2020) to demonstrate that such representations, once formed, can adversely affect future behavior, here, specifically in the form of the N2RC effect. This not only highlights the usefulness of this analytical approach for addressing questions relating to cognitive control processes but it contributes to the extant literature on the role of episodic retrieval in N-2 Repetition Costs. Hence, this work provides fundamentally novel insights into a long-standing debate in the behavioral literature and illustrates the potential of this neuroscientific approach to tackling questions of adaptive and flexible behavior.

Footnotes

  • This work was supported by National Institutes of Health–National Institute of Neurological Disorders and Stroke Grant NS102201 to J.R.W.

  • The authors report no competing financial interests.

  • Correspondence should be addressed to Benjamin O. Rangel at benjamin-rangel{at}uiowa.edu or Jan R. Wessel at jan-wessel{at}uiowa.edu

References

  1. ↵
    2. Anderson MC,
    3. Bunce JG,
    4. Barbas H
    (2016) Prefrontal–hippocampal pathways underlying inhibitory control over memory. Neurobiol Learn Mem 134:145–161. doi:10.1016/j.nlm.2015.11.008
    OpenUrlCrossRefPubMed
  2. ↵
    2. Bell AJ,
    3. Sejnowski TJ
    (1995) An information-maximization approach to blind separation and blind deconvolution. Neural Comput 7:1129–1159. doi:10.1162/neco.1995.7.6.1129 pmid:7584893
    OpenUrlCrossRefPubMed
  3. ↵
    2. Brainard DH
    (1997) The psychophysics toolbox. Spatial Vision 10:433–436.
    OpenUrlCrossRefPubMed
  4. ↵
    2. Delorme A,
    3. Makeig S
    (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Meth 134:9–21. doi:10.1016/j.jneumeth.2003.10.009
    OpenUrlCrossRefPubMed
  5. ↵
    2. Delorme A,
    3. Sejnowski T,
    4. Makeig S
    (2007) Enhanced detection of artifacts in EEG data using higher-order statistics and independent component analysis. Neuroimage 34:1443–1449. doi:10.1016/j.neuroimage.2006.11.004 pmid:17188898
    OpenUrlCrossRefPubMed
  6. ↵
    2. Fisher RA
    (1915) Frequency distribution of the values of the correlation coefficient in samples from an indefinitely large population. Biometrika 10:507–521. doi:10.2307/2331838
    OpenUrlCrossRef
  7. ↵
    2. Frings C,
    3. Hommel B,
    4. Koch I,
    5. Rothermund K,
    6. Dignath D,
    7. Giesen C,
    8. Kiesel A,
    9. Kunde W,
    10. Mayr S,
    11. Moeller B,
    12. Möller M,
    13. Pfister R,
    14. Philipp A
    (2020) Binding and retrieval in action control (BRAC). Trends Cogn Sci 24:375–387. doi:10.1016/j.tics.2020.02.004 pmid:32298623
    OpenUrlCrossRefPubMed
  8. ↵
    2. Gerber EM
    (2022) permutest. MATLAB Central File Exchange. Available at https://www.mathworks.com/matlabcentral/fileexchange/71737-permutest.
  9. ↵
    2. Grange JA
    (2018) Does task activation in task switching influence inhibition or episodic interference? Exp Psychol 65:393–404. doi:10.1027/1618-3169/a000423 pmid:30638164
    OpenUrlCrossRefPubMed
  10. ↵
    2. Grange JA,
    3. Juvina I,
    4. Houghton G
    (2013) On costs and benefits of n-2 repetitions in task switching: towards a behavioural marker of cognitive inhibition. Psychol Res 77:211–222. doi:10.1007/s00426-012-0421-4 pmid:22327120
    OpenUrlCrossRefPubMed
  11. ↵
    2. Grange JA,
    3. Kowalczyk AW,
    4. O'Loughlin R
    (2017) The effect of episodic retrieval on inhibition in task switching. J Exp Psychol Hum Percept Perform 43:1568–1583. doi:10.1037/xhp0000411 pmid:28383961
    OpenUrlCrossRefPubMed
  12. ↵
    2. Hommel B
    (1998) Event files: evidence for automatic integration of stimulus–response episodes. Vis Cogn 5:183–216. doi:10.1080/713756773
    OpenUrlCrossRef
  13. ↵
    2. Hommel B
    (2004) Event files: feature binding in and across perception and action. Trends Cogn Sci 8:494–500. doi:10.1016/j.tics.2004.08.007 pmid:15491903
    OpenUrlCrossRefPubMed
  14. ↵
    2. Hommel B
    (2009) Action control according to TEC (theory of event coding). Psychol Res 73:512–526. doi:10.1007/s00426-009-0234-2 pmid:19337749
    OpenUrlCrossRefPubMed
  15. ↵
    2. Hommel B
    (2019) Theory of Event Coding (TEC) V2.0: representing and controlling perception and action. Atten Percept Psychophys 81:2139–2154. doi:10.3758/s13414-019-01779-4 pmid:31168699
    OpenUrlCrossRefPubMed
  16. ↵
    2. Hommel B,
    3. Colzato L
    (2004) Visual attention and the temporal dynamics of feature integration. Vis Cogn 11:483–521. doi:10.1080/13506280344000400
    OpenUrlCrossRef
  17. ↵
    2. Hommel B,
    3. Müsseler J,
    4. Aschersleben G,
    5. Prinz W
    (2001) The Theory of Event Coding (TEC): a framework for perception and action planning. Behav Brain Sci 24:849–878. doi:10.1017/s0140525x01000103 pmid:12239891
    OpenUrlCrossRefPubMed
  18. ↵
    2. Hommel B,
    3. Memelink J,
    4. Zmigrod S,
    5. Colzato LS
    (2014) Attentional control of the creation and retrieval of stimulus–response bindings. Psychol Res 78:520–538. doi:10.1007/s00426-013-0503-y pmid:23884516
    OpenUrlCrossRefPubMed
  19. ↵
    2. Jamadar S,
    3. Hughes M,
    4. Fulham WR,
    5. Michie PT,
    6. Karayanidis F
    (2010) The spatial and temporal dynamics of anticipatory preparation and response inhibition in task-switching. Neuroimage 51:432–449. doi:10.1016/j.neuroimage.2010.01.090 pmid:20123028
    OpenUrlCrossRefPubMed
  20. ↵
    2. Kahneman D,
    3. Treisman A,
    4. Gibbs BJ
    (1992) The reviewing of object files: object-specific integration of information. Cognitive Psychol 24:175–219. doi:10.1016/0010-0285(92)90007-O
    OpenUrlCrossRefPubMed
  21. ↵
    2. Kessler Y
    (2018) N–2 repetition leads to a cost within working memory and a benefit outside it. Ann N Y Acad Sci 1424:268–277. doi:10.1111/nyas.13607 pmid:29542817
    OpenUrlCrossRefPubMed
  22. ↵
    2. Kikumoto A,
    3. Mayr U
    (2020) Conjunctive representations that integrate stimuli, responses, and rules are critical for action selection. Proc Natl Acad Sci U S A 117:10603–10608. doi:10.1073/pnas.1922166117 pmid:32341161
    OpenUrlAbstract/FREE Full Text
  23. ↵
    2. Koch I,
    3. Gade M,
    4. Schuch S,
    5. Philipp AM
    (2010) The role of inhibition in task switching: a review. Psychon B Rev 17:1–14. doi:10.3758/PBR.17.1.1
    OpenUrlCrossRefPubMed
  24. ↵
    2. Kowalczyk AW,
    3. Grange JA
    (2016) Inhibition in task switching: the reliability of the n - 2 repetition cost. Q J Exp Psychol (Hove) 70:2419–2433.
    OpenUrl
  25. ↵
    2. Kowalczyk AW,
    3. Grange JA
    (2020) The effect of episodic retrieval on inhibition in task switching: a diffusion model analysis. Psychol Res 84:1965–1999. doi:10.1007/s00426-019-01206-1 pmid:31177315
    OpenUrlCrossRefPubMed
  26. ↵
    2. Kühn S,
    3. Keizer AW,
    4. Colzato LS,
    5. Rombouts SARB,
    6. Hommel B
    (2011) The neural underpinnings of event-file management: evidence for stimulus-induced activation of and competition among stimulus–response bindings. J Cognitive Neurosci 23:896–904. doi:10.1162/jocn.2010.21485
    OpenUrlCrossRefPubMed
  27. ↵
    2. Lee T-W,
    3. Girolami M,
    4. Sejnowski TJ
    (1999) Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources. Neural Comput 11:417–441. doi:10.1162/089976699300016719 pmid:9950738
    OpenUrlCrossRefPubMed
  28. ↵
    2. Mayr U
    (2002) Inhibition of action rules. Psychon Bull Rev 9:93–99. doi:10.3758/bf03196261 pmid:12026957
    OpenUrlCrossRefPubMed
  29. ↵
    2. Mayr U,
    3. Keele SW
    (2000) Changing internal constraints on action: the role of backward inhibition. J Exp Psychol Gen 129:4–26. doi:10.1037/0096-3445.129.1.4 pmid:10756484
    OpenUrlCrossRefPubMed
  30. ↵
    2. Neill WT
    (1997) Episodic retrieval in negative priming and repetition priming. J Exp Psychology Learn Mem Cognition 23:1291–3105. doi:10.1037/0278-7393.23.6.1291
    OpenUrlCrossRef
  31. ↵
    2. Parthasarathy A,
    3. Herikstad R,
    4. Bong JH,
    5. Medina FS,
    6. Libedinsky C,
    7. Yen S-C
    (2017) Mixed selectivity morphs population codes in prefrontal cortex. Nat Neurosci 20:1770–1779. doi:10.1038/s41593-017-0003-2 pmid:29184197
    OpenUrlCrossRefPubMed
  32. ↵
    2. Philipp AM,
    3. Koch I
    (2006) Task inhibition and task repetition in task switching. Eur J Cogn Psychol 18:624–639. doi:10.1080/09541440500423269
    OpenUrlCrossRef
  33. ↵
    2. Rigotti M,
    3. Barak O,
    4. Warden MR,
    5. Wang X-J,
    6. Daw ND,
    7. Miller EK,
    8. Fusi S
    (2013) The importance of mixed selectivity in complex cognitive tasks. Nature 497:585–590. doi:10.1038/nature12160 pmid:23685452
    OpenUrlCrossRefPubMed
  34. ↵
    2. Rose NS,
    3. LaRocque JJ,
    4. Riggall AC,
    5. Gosseries O,
    6. Starrett MJ,
    7. Meyering EE,
    8. Postle BR
    (2016) Reactivation of latent working memories with transcranial magnetic stimulation. Science 354:1136–1139. doi:10.1126/science.aah7011 pmid:27934762
    OpenUrlAbstract/FREE Full Text
  35. ↵
    2. Scheil J,
    3. Kleinsorge T
    (2019) Effects of global and local task repetition proportion on n − 2 repetition costs. Q J Exp Psychol (Hove) 72:579–588. doi:10.1177/1747021818762087 pmid:29431022
    OpenUrlCrossRefPubMed
  36. ↵
    2. Scheil J,
    3. Kleinsorge T
    (2021) No-go trials in task switching: effects on the task-set and task-space level. Psychol Res 86:1097–1107.
    OpenUrl
  37. ↵
    2. Schuch S,
    3. Grange JA
    (2015) The effect of N-3 on N-2 repetition costs in task switching. J Exp Psychol Learn Mem Cogn 41:760–767. doi:10.1037/xlm0000059 pmid:25197760
    OpenUrlCrossRefPubMed
  38. ↵
    2. Sexton NJ,
    3. Cooper RP
    (2017) Task inhibition, conflict, and the n-2 repetition cost: A combined computational and empirical approach. Cognitive Psychol 94:1–25. doi:10.1016/j.cogpsych.2017.01.003
    OpenUrlCrossRef
  39. ↵
    2. Sinai M,
    3. Goffaux P,
    4. Phillips NA
    (2007) Cue- versus response-locked processes in backward inhibition: evidence from ERPs. Psychophysiology 44:596–609. doi:10.1111/j.1469-8986.2007.00527.x pmid:17451492
    OpenUrlCrossRefPubMed
  40. ↵
    2. Spapé MM,
    3. Hommel B
    (2010) Actions travel with their objects: evidence for dynamic event files. Psychol Res 74:50–58. doi:10.1007/s00426-008-0219-6
    OpenUrlCrossRefPubMed
  41. ↵
    2. Treisman A
    (1998) The perception of features and objects. In: Visual attention (Wright RD, ed), pp 26–56. New York: Oxford UP.
  42. ↵
    2. Waszak F,
    3. Hommel B,
    4. Allport A
    (2003) Task-switching and long-term priming: role of episodic stimulus–task bindings in task-shift costs. Cogn Psychol 46:361–413. doi:10.1016/s0010-0285(02)00520-0 pmid:12809680
    OpenUrlCrossRefPubMed
  43. ↵
    2. Wessel JR,
    3. Aron AR
    (2017) On the globality of motor suppression: unexpected events and their influence on behavior and cognition. Neuron 93:259–280. doi:10.1016/j.neuron.2016.12.013 pmid:28103476
    OpenUrlCrossRefPubMed
  44. ↵
    2. Zhang R,
    3. Stock A-K,
    4. Beste C
    (2016) The neurophysiological basis of reward effects on backward inhibition processes. Neuroimage 142:163–171. doi:10.1016/j.neuroimage.2016.05.080 pmid:27262242
    OpenUrlCrossRefPubMed
Back to top

In this issue

Email
Print
View Full Page PDF
Citation Tools
Respond to this article
Lingering Neural Representations of Past Task Features Adversely Affect Future Behavior
Benjamin O. Rangel, Eliot Hazeltine, Jan R. Wessel
Journal of Neuroscience 11 January 2023, 43 (2) 282-292; DOI: 10.1523/JNEUROSCI.0464-22.2022
del.icio.us logo Digg logo Reddit logo Twitter logo Facebook logo Google logo Mendeley logo

Jump to section

Keywords

Responses to this article

Respond to this article

Jump to comment:

No eLetters have been published for this article.