Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Strengths:
This manuscript has many strengths, including a clever study design, thoughtful integration of multiple neurocognitive measures, and a set of rigorous and technically sophisticated analyses, which reveal a large set of relationships among the measures and behavior. The findings demonstrating brain/physiology-behavior relationships are particularly important, in that they point to potential functional consequences of MPES.
We thank the reviewer for noting these strengths of the work along with the below encouragement to revise the manuscript to better highlight the key findings and their implications.
Weaknesses:
The technical proficiency and complexity of the study and analysis also present a clear limitation and challenge for interpretation. As a reader, even those who are quite knowledgeable about the methods, constructs, and questions being addressed will often struggle (as this reviewer did) to keep the large set of findings in mind and gain an understanding of how they all fit together.
Indeed, it seems like there are many threads running together in the paper, which makes it challenging to find the through-line of the key findings, or to understand how they might relate to some pre-existing hypotheses, rather than merely interesting patterns detected in the data. In the Introduction and Discussion, it seems as if the key question is to understand the pathways by which MPEs impact cognition, but this is a rather broad topic, so it is not clear exactly what the authors are aiming at with this question and study design.
As an example, authors operationalize frontal theta power as an index of cognitive control demand, and one of the pathways by which MPEs impact cognition. But this point becomes somewhat circular, since it is not clear how or why the Mismatch x Strength interaction in frontal theta reflects that demand. It would have been better to set this pattern up in the Introduction as a theoretically driven hypothesis, since it currently appears more like a post-hoc interpretation. This is mirrored by how the issue is first brought up in the Introduction, where it states somewhat vaguely: "whether MPEs are followed by an increase in frontal theta... warrants closer examination".
Again, we appreciate the reviewer’s thoughtful feedback on where the manuscript can be clearer, especially given the rich set of results it reports. Following the reviewer’s guidance, we restructured and revised the Introduction to further motivate the hypotheses that (a) MPEs increase both attention/arousal (grounded in studies of Event Segmentation Theory) and cognitive control (given findings on reward prediction errors and other types of prediction errors), and (b) there are greater increases in these processes triggered by strong compared to weak MPEs. To better link these hypotheses to resulting statistical tests, we note that hypothesis (a) was tested in our trial-level regression models in Fig. 1 by examining main effects of Strength, whereas hypothesis (b) was tested in our models in the Mismatch x Strength interactions. On point (b) and potentially circularity, we note that previous work indicates that frontal theta scales with negative reward prediction errors; as such, we hypothesized that stronger MPEs would elicit more frontal theta, as evidenced by a robust Mismatch x Strength interaction during the probe period.
Later in the results, there are findings relating frontal theta to pupil dilation, posterior alpha suppression and then subsequent memory. It was hard to understand how all the findings might be linked together functionally or conceptually. Are the authors potentially postulating a mediating or mechanistic pathway, in which the MPE leads to increased cognitive control (frontal theta), which then leads to enhanced subsequent memory of those events? If this is the case, then maybe a formal path analysis would be the best way to test or state this hypothesis. It would also be useful to specify more clearly how the pupil components and alpha suppression factor into this mediating path, since it was not clear.
Relatedly, the authors suggest that internal attention and arousal also play relevant roles in this pathway, but these are also not clear. In some cases, it is stated as if this is a distinct pathway from the cognitive control one, since there is a focus in the results on the independence of frontal theta and posterior alpha, but elsewhere they seem to be treated as two aspects, or distinct steps, within a single pathway. Again, these different threads of the findings were quite challenging for the reader to follow. Pathway analyses, such as with multiple mediation or moderated mediation, could be a useful way to address this question. For example, it seems as if readiness-to-remember is another behavioral outcome (like subsequent memory) that could be used in the search for mediators.
We thank the reviewer for highlighting these ambiguities in the original submission and for the thoughtful encouragement to leverage mediation models to more formally test the hypothesized relationships between MPE magnitude and constructs of control, attention, and arousal. In the revision, we now more clearly hypothesize that the effects of strong MPE-driven increases in attention and arousal might be explained, in part, by cognitive control (as indexed by frontal theta) upregulating attention and arousal. To more explicitly test this model of the relationships between our measures, as recommended by Reviewer #1, we now include multivariate mediation analyses to assess whether, at a trial-level, changes in posterior alpha and the immediate pupil effect PC3 are explained in part by increases in frontal theta. Because changes in posterior alpha following MPEs and PC3 scores did not predict subsequent memory, mediation analyses addressing the hypothesis that our attention/arousal measures mediate the effect of frontal theta on subsequent memory were not conducted. Following insights from a cross-correlation analysis, as recommended by Reviewer #2, we tested an additional model to examine whether the effects of MPE magnitude on frontal theta were explained in part by changes in posterior alpha. Examination of the posterior distributions of the indirect effects did not favor our hypothesized model, nor the alternative model that attention upregulates control. Altogether, these outcomes suggest that increases in control, attention, and arousal following strong MPEs may be elicited independently. Yet, we also note that the current set of experiments may be underpowered for these mediation analyses; future work can further investigate the directionality between these effects. Finally, with respect to the readiness-to-remember findings, they were removed in the interest of space, as recommended by Reviewer #2.
At the minimum, it would be quite helpful to have diagrammatic figures that specify the hypothesized and observed relationships between independent variables (Strength, Mismatch), physiological indices (pupil dilation components, frontal theta, posterior alpha) and key outcome measures (accuracy, RT, next-trial retrieval success, subsequent memory), so that the reader can refer back to them as each component of the analyses is conducted.
To further increase conceptual clarity, we also followed this helpful suggestion, adding diagrammatic figures to illustrate our hypotheses regarding interactions between the effects (Fig. 3a) and to summarize the observed relationships between MPEs, control, attention, and arousal (Fig. 6).
Minor Points:
Many figures had x-axes showing a pupil component or EEG power metric broken down by quartile or quintile. Yet nowhere is it ever explained why this graphical (or analytic?) approach is used and what it reflects, or how it is decided which break down to use (quartile/quintile). If the data are analyzed as a correlation, why is a scatterplot not shown instead?
In the linear mixed effects models, continuous values were used to assess relationships between variables. In the figures, the continuous variables were binned into quartiles or quintiles for ease of visualization. We opted to visualize the data using this approach, rather than with a scatterplot, given the large number of trials. We updated the figure captions to clarify the approach.
It was surprising that, unlike readiness-to-remember, which was analyzed via logistic regression and odds-ratio, subsequent memory was not analyzed in the same fashion (i.e., as a binary outcome variable predicted by frontal theta), rather than in a reverse chronological one (subsequent memory predicting frontal theta). Historically, it was the case that subsequent memory was analyzed in this manner, but that was before the era in which trial-level linear mixed-effect models were in wide usage, as they are implemented in this study. Thus, the choice seems like a wasted opportunity or a step backwards analytically.
We thank the reviewer for this encouragement and agree with the point. In the revision, we note that the readiness-to-remember results were removed in the interest of space and clarity, as was recommended by Reviewer #2. With respect to the subsequent memory analyses, they are now analyzed via logistic regression.
Reviewer #2 (Public review):
Strengths:
The study has a clear behavioral paradigm with multiple measures - behavioral, EEG, and pupillometry that offer an investigation into different aspects of MPE response and memory.
The study is also very comprehensive in looking at multiple phases in processing MPEs: the prediction phase (prior to the violation), the response to MPEs, and subsequent memory of MPEs, all within one study. Specifically, the link between neural mechanisms and subsequent memory is a major advancement, as most prior studies did not include this component. Mechanisms underlying subsequent memory of MPEs are theoretically important, as a primary function of MPEs is to promote learning and memory. As the authors mention, the different neural and pupillary signals are not robustly correlated, suggesting multiple mechanisms underlying MPE detections, which is interesting, offers avenues for future research, and can facilitate a better theory of how MPEs are processed in the brain. Finally, the decomposition of pupil response into different components and their correlation with behavior (RT during match/MPE detection) is interesting.
We thank the reviewer for noting these strengths of the work along with the below encouragement to revise the manuscript to better highlight the key findings and their implications.
Weaknesses:
The methods are rigorous, and the claims are mostly supported by the data, but there are a few weaknesses or places that could be improved:
(1) The authors conduct PCA analysis to identify different components of the pupillary response to MPE and relate them to behavior. Specifically, the authors identify components PC3 and PC4, which they interpret as related to MPE. However, some parts of the interpretation could be clearer or better justified:
(a) The authors refer to PC4 as "post-decision cognitive processing". But, given that RT was between .5-.7s, and PC3 peaked after more than 1s, wouldn't it be cautious to interpret PC3 as postdecision as well?
Thank you for raising this point. Given that pupil is a relatively sluggish response, it is possible that both components reflect post-decision cognitive processing, even if PC3 peaks before PC4. Following the reviewer’s guidance to adopt more cautious language, we replaced “post-decision” with “post-MPE”.
(b) MPEs overall elicit longer RTs in this study, suggesting that long RT is a behavioral marker of MPE. Nonetheless, the authors argue on p. 12: "Altogether, these findings indicate that when stronger mnemonic predictions (as indexed by shorter RTs) were violated." And, PC3 is correlated with shorter RTs for mismatches, meaning that behaviorally, these trials were more similar to matches. Thus, how do the authors interpret shorter versus longer RTs for MPEs, and what processes do these RT reflect?
We thank the reviewer for stressing the need for greater clarity regarding the relationships between RT and the constructs of interest. With respect to RT, we interpret the condition-level difference in RTs between mismatch and match trials as a behavioral marker of an MPE. However, when comparing mismatch trials within a given strength condition to each other (i.e., an analysis at the trial-level), shorter RTs may reflect a stronger prediction, greater certainty that the probe is a mismatch, and therefore the experience of a stronger MPE. Note that while larger PC3 scores were associated with shorter RTs for mismatches (Fig. 2b, right), the mismatch RTs in the largest PC3 quartile were still longer than those on match trials in the corresponding quartile (in other words, there was still a condition-level difference in RTs in the largest PC3 quartile, suggesting that the mismatch trials in this bin are behaviorally still likely to be different from match trials in the corresponding bin).
To clarify these relationships and our interpretation, we modified the referred to text: “(as indexed by shorter strong mismatch RTs).” Moreover, we added text to the Discussion, further delineating our reasoning here and the implications of our findings for understanding the mechanisms giving rise to and triggering by MPEs of varying strengths. This includes adding an explicit summary of the logic and findings that notes that, at the trial-level, shorter RTs may reflect stronger predictions; at the match/mismatch condition-level, longer RTs may reflect the experience of a MPE. For PC3, shorter RTs (trials with a stronger prediction) in the Strong Mismatch condition were associated with a larger pupillary response. The added text notes that “while longer mean RTs for mismatches compared to matches are a behavioral marker of a MPE, within-condition differences in RTs (i.e., between mismatch trials) may reflect more subtle differences in MPE magnitude, with shorter RTs reflecting stronger predictions and thus stronger MPEs. Strong mismatch RTs were used in the mediation models as a proxy measure of MPE magnitude; this estimate may be noisy because RTs in this experiment are likely sensitive to factors independent of mnemonic prediction strength (e.g., preparatory attention (Supplementary Fig. 6) or memory strength of the mismatch probe). This limitation may have additionally reduced sensitivity to detecting indirect effects.”
(2) The brain to pupil relationship (p. 13-14): If I understand correctly, this was done on a trial-by-trial basis, but the high temporal resolution allows doing the analysis in a time-resolved manner - does brain activity at a certain time point preceding/following the pupil response correlate with the pupil response? It might be that cognitive control influences attention mechanisms or vice versa (because there is some overlap in the response). Although not testing causality, this temporally resolved correlation would be an interesting way to start probing how signals might influence each other.
Thank you for this suggestion. We now report a cross-correlation analysis (Fig. 3d) that suggests that cognitive control increases precede attention decreases at retrieval, whereas in response to a strong MPE, cognitive control increases follow attention increases. There were no significant clusters for the temporal relationships between frontal theta and pupil, nor for posterior alpha and pupil.
(3) The relationships the authors find between brain measures and pupil components were largely not specific to mismatches/matches. However, are they specific to this task? I think it would benefit the paper to show that these relationships are potentially specific to making match/mismatch memory decisions, versus, e.g., any stimulus processing. For example, the authors could run the same analyses locked to stimuli in the study phase, anticipating a different pattern, if indeed these findings are specific to the associative memory task.
Many of the associations between our measures indeed did not show an interaction with Mismatch. Due to jitter in the ISI in the study phase, some of the analyses in the retrieval phase cannot be performed in the exact same way for the study phase. We will leave these questions to be addressed in future research. We agree that an important question for future research is to address whether these responses depend on making match/mismatch decisions and now include consideration of this point in the Discussion: “Finally, the magnitude of observed increases in control, attention, and arousal following strong MPEs may be influenced by the decision-making process engaged when making match/mismatch judgments. Not all MPEs necessitate behavioral responses. Whether similar magnitudes in neurocognitive responses and consequences for learning are observed upon detection of an MPE, but in the absence of a decision remains unclear.”
(4) During memory retrieval (i.e., before the probe), the authors find that frontal theta, a marker of cognitive control, was associated on a trial-by-trial basis with more posterior alpha (i.e., less alpha suppression, potentially reflecting less attention), and that this association was stronger for weaker predictions. The authors interpreted this as weaker predictions necessitating more cognitive control, and that more cognitive control was recruited specifically in trials where retrieval included less content (memory reinstatement) to attend to. Generally, cognitive control is recruited to facilitate memory retrieval. If so, one possible interpretation is that this correlation reflects cognitive control effort that has failed to produce enough memory reinstatement. The other possibility is that this correlation reflects more specific retrieval of the correct probe, without retrieval of interfering items (i.e., overall less content). I believe that the former explanation predicts that this correlation would be associated with longer RTs (more difficult decisions), while the latter predicts shorter RTs (easier decisions due to successful retrieval), at least for matches.
Thank you for these insightful comments. Because this analysis is not key to the main questions about MPEs and given both reviewers’ concerns that the manuscript can be overwhelming for the reader given the sheer number of findings reported, we opted to move this point from the main text to the Supplement. However, following the reviewer’s guidance here, we conducted the proposed analyses and tested these alternative accounting by modeling RTs. The results favour the former interpretation:
“Greater control being associated with less attention could reflect failure in controlled retrieval efforts to reinstate sufficient memory evidence of the probe and thus fewer retrieval products to which attention is allocated. Alternatively, greater cognitive control could increase the likelihood of retrieval success, eliciting selective retrieval of the correct probe and inhibition of interfering items. To address these alternatives, we examined how the association between frontal theta and posterior alpha related to the difficulty of a trial, as assayed by RTs. The former failure-of-control account would predict that a stronger positive frontal theta- posterior alpha association would relate to longer RTs, whereas the greater-retrieval-specificity account would predict that a stronger positive association would relate to shorter RTs. In a model predicting RTs as a function of frontal theta, posterior alpha, Strength, Mismatch, and their interactions, we found a two-way frontal theta × posterior alpha interaction (β=0.016, CI=[0.001, 0.031], p=0.033), such that a stronger positive association between frontal theta and posterior alpha predicted longer RTs. This relationship did not differ as a function of Strength (no frontal theta × posterior alpha × Strength interaction: β=-0.015, CI=[-0.033, 0.004], p=0.124), Mismatch (no frontal theta × posterior alpha × Mismatch interaction: β=-0.016, CI=[-0.050, 0.018], p=0.342), or interact with Strength and Mismatch (no frontal theta × posterior alpha × Strength × Mismatch interaction: β=0.024, CI=[-0.014, 0.062], p=0.218). Together, these outcomes support the idea that during memory retrieval, positive coupling between frontal theta and posterior alpha may reflect failure or inefficiency of cognitive control efforts to rapidly accumulate mnemonic evidence to which to attend in support of a memory decision.”
(5) In section 3, the authors found a positive relationship between alpha during memory retrieval and PC3 during MPE. If I understood correctly, this means that less attention during retrieval (less suppression) is correlated with a stronger PC3 response. How do the authors interpret this? Maybe along the same lines as in (5), specifically retrieving the correct information (i.e., less retrieved content to attend to) means a stronger prediction, leading to a stronger MPE, and a stronger MPE response, as reflected by PC3?
We appreciate this comment, as it highlights a need for greater clarity here. The observed relationship was actually negative (Fig. 4f), meaning that more attention during retrieval was associated with a stronger PC3 response. We suspect that the lack of clarity here may be due to the original statement that “there was a positive relationship between posterior alpha suppression during memory retrieval [and PC3 scores]”. To increase clarity, we have modified this statement to “there was a negative relationship between posterior alpha during memory retrieval [and PC3 scores]”. We interpret the negative relationship with posterior alpha (i.e., positive relationship with posterior alpha suppression) to indicate “that greater attentional allocation during memory retrieval, which occurs when memories are stronger and more retrieval products can be reinstated and attended to (Fig. 1e; Fig. 4c), predicts the magnitude of immediate pupil responses to MPEs.”
(6) The results with subsequent memory are important and address a major gap in the field that largely did not relate neural effects of MPE to subsequent memory. However, one major limitation of the study is that the authors did not test memory for matches. I understand the logic of avoiding testing matches. Because matches were repeated more times in the study, it's not a fair comparison, and could change participants' overall criterion for old/new decisions. However, one possibility would have been to test only the weak prediction; this could have given some specificity to the neural subsequent memory findings.
We were indeed concerned about the change in decision criterion and did not include match items for this reason. Nonetheless, this is a useful suggestion that future work could include the weak match items for a better comparison of subsequent recognition memory. We now comment on this in the Discussion: “Whether control-associated enhancements in learning are specific to learning from MPEs can be further tested in future work by including a test of subsequent memory for weak match probes as an additional control condition for comparison.”
(7) The authors nicely characterized the different PC of pupillary MPE response. But, with respect to subsequent memory, they only present pupil size. Unless there is some methodological reason that prevents testing subsequent memory on the PC, I think this will be very informative about the potential mechanisms underlying memory of MPE.
The pupil PCs were not associated with subsequent memory, though there are some interesting trends in the 48-delay condition which could be explored in future work. These findings are now reported in the Supplement (Supplementary Fig. 16).
(8) This paper includes many interesting findings, and I am not sure how they all come together into a cohesive mechanistic understanding of MPE response and subsequent memory. I think the paper would benefit from either a conceptual mechanism figure or, in the Discussion, have a summary of a proposed mechanism integrating the findings together.
We thank the reviewer for stressing this point, which also was raised by Reviewer #1. To better emphasize the novel contributions of the work and to assist the reader’s understanding of the key findings, we: (a) revised the Introduction to more explicitly describe our hypotheses about the relationships among our measures as they relate to MPE responses and subsequent memory; (b) now report mediation models to more directly test these relationships and include diagrams of the hypothesized relationships; and (c) include a schematic figure at the end of the Results to highlight the key mechanistic relationships supported by the data.
(9) Relatedly, the section "Immediate, strength-sensitive neurocognitive impacts of MPEs" does not link the arguments to specific data points, so it's hard to follow which data specifically the authors are interpreting.
The discussion in this paragraph rests on the Strength × Mismatch interactions observed in Figures 1d-f, summarized in the first sentence of the section. To increase clarity, we changed the title of this section to “Neurocognitive impacts of MPEs are strength-sensitive”.
(10) If I understand correctly, the authors did not find improved memory for strong compared to weak MPE. First, I think this behavioral result should be incorporated in the main paper and in the interpretation of the results. Second, given that the neural effects the authors tested either correlated with memory for strong MPE or did not show a relationship with memory, what neural/pupil response could explain memory for weak MPE?
Thank you for raising these points. The behavioral result and the frontal theta trial-level regression model for subsequent memory are now described in the main results and included in the Discussion. As noted in the Discussion, the trial-level regression model indicates pre-probe and post-MPE frontal theta effects may explain memory for weak (and strong) mismatch probes. While the magnitude of probe-period MPE frontal theta is additionally predictive of memory for strong mismatch probes (as indicated by the Subsequent Memory × Strength interaction in the trial-level regression model in Fig. 5c and the logistic regression in Fig. 5b), frontal theta during this period does not additionally enhance memory for strong mismatch probes above that of weak mismatch probes (Fig. 5a). We added more discussion on why memory for strong and weak mismatch probes in the current experiment did not differ. We also discuss potential directions for future research to further probe mechanisms underlying MPE-driven learning.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
It is recommended that the authors determine whether formal path analyses, testing for mediation and moderation, would provide a useful approach from which to better integrate the disparate set of findings and make clear their causal/functional implications.
At a minimum, adding a diagrammatic figure is recommended to visually depict the key components of the study (independent variables, physiological indices, outcome measures) and how they relate to each other both conceptually (ideally in a theoretically hypothesized manner) and in terms of the observed findings. Such a figure will help the reader keep track of the many types of findings and results threads, and with the goal of better organizing the results into a clearer narrative through-line.
Thank you for the suggestions to add mediation analyses and diagrammatic figures. To more formally address our hypothesis that increases in attention/arousal following MPEs are explained in part by increases in cognitive control, we tested two mediation models: one with posterior alpha as the measure of attention and the other with pupil PC3 scores. We now diagram this hypothesis in Fig. 3a. Given the outcomes of the cross-correlation analysis suggested by Reviewer #2, we also tested an additional model where posterior alpha might explain the impact of strong MPEs on frontal theta (now diagrammed in Fig. 3e). We did not find credible evidence for an indirect effect in any of the models, suggesting that strong MPE-driven increases in control, attention, and arousal may be elicited independently. Finally, in a newly added summary diagram (Fig. 6), we highlight the observed effects of strong vs. weak MPEs on RTs, control, attention and arousal; differences in control, attention, and arousal for strong vs. weak predictions preceding the MPE; trial-level mismatch-specific or strong-specific effects on subsequent memory; and mismatch-specific interactions between processes during retrieval and responses to MPEs. We hope these revisions address Reviewer #1’s concerns and that these figures help the reader keep track of the key hypothesized relationships and main findings.
Reviewer #2 (Recommendations for the authors):
(1) The relationship between event segmentation and prediction errors has been reviewed recently in two papers (Nolden et al., 2024, Neuroscience & Biobehavioral Reviews; Rouhani et al., 2024, JOCN for a potentially relevant computational model). I wonder if insights from these papers can inform the Introduction/Discussion of the current manuscript.
Thank you for these suggestions. These papers are now incorporated into the Introduction and the Discussion.
(2) Brod et al. (2022, Psych. Bull. Rev.) have previously reported increased pupil dilations for MPE correlating with subsequent memory, specifically for strong prediction errors. I think it's worth including this paper in the Introduction as the finding is highly relevant. The Brod paper might provide more direct evidence of "MPE-related increases in pupil size" than the evidence the authors provide (p. 3).
Thank you for this suggestion. This paper is now incorporated into the Introduction and Discussion.
(3) I'm confused about the temporal analysis: "Trial-level regression analyses were conducted on the frontal theta, posterior alpha, and pupil time series from the associative retrieval test to identify temporal clusters that were sensitive to the factors of Mismatch (i.e., mismatch vs. match probes) and/or Strength (i.e., strong vs. weak associative pairs). For each participant and each time point, a linear regression model testing main effects of Mismatch and Strength, and a Mismatch × Strength interaction was run using R to compute beta weights for each regressor." What regression exactly was run? A separate model for each participant and time point? Across trials, then? Later, the authors mention that beta weights were averaged across participants and t-tests and permutation tests were conducted. However, if the data were averaged, what t-test was conducted? And how was the permutation test conducted? It's also unclear what the authors mean by "the sign of each participant's beta weights" - what sign?
Thank you for raising this point. A regression was run separately for each participant and each time point, across trials: neurocognitive measure ~ Mismatch + Strength + Mismatch: Strength. Beta weights were averaged only for visualization; t-tests were conducted on each set of beta weights (across participants), separately for each time point. We modified the text of the Methods to describe our procedure more clearly.
(4) The associative memory and recognition accuracy data are presented as d'. In addition, the authors should provide hits and false alarms to facilitate a better interpretation of the results.
We now report these outcomes in Table 1 and refer to them in the main text.
(5) The authors argue regarding the frontal theta that "Qualitatively, the main effect of Strength emerged later than the main effect of Mismatch, suggesting that the increase in frontal theta evoked by the probe was more sustained for weak compared to strong trials (or, as a corollary, that the greater control elicited by strong MPEs enabled more rapid resolution of conflict and ultimate choice selection)." It was unclear to me how that stems from the data.
This statement has been removed altogether.
(6) Especially in Figure 1, I think clarity can be improved if the authors would indicate the specific subsection they are referring to in the text, because even within, e.g., 1d, there are different graphs, so mentioning which graph is relevant for which statement would be helpful to the reader.
Thank you for this suggestion for improving the clarity of the manuscript. Fig. 1d-f includes multiple parts because we wanted to show the raw data (the mean time series on the left) as well as the model coefficients (time series on the right). We are hopeful that, given that each subplot is titled and that the main text refers to both the subplots on the left and on the right, this will be clear to the reader as is. We welcome further guidance if this remains a concern.
(7) This seems highly speculative to me: "Elevated PC4 scores on weak hits, where no error was made, may instead reflect retrieval practice and thus internally oriented attention. On these trials, when cueelicited retrieval may have been weaker, the probe may have provided additional support for pattern completion of the learning episode for that association, and this engagement in memory retrieval may have elicited pupil dilation (c.f., Strength effect in Figure 1f). Overall, PC4 may therefore reflect an attentional orienting response that, depending on the relative success of memory retrieval and probe identity, may direct attention internally or externally. " (p.13). In my opinion, the authors make a lot of assumptions about underlying processes. I'd consider removing.
We removed this text.
(8) In Figure 3c, should the x-axis be PC3 quintile? And (c) is not in the figure caption.
Thank you for this note. We addressed these issues.
(9) The carryover effects the authors report are interesting, but they seem detached, and it is unclear how they fit with the additional findings. In a paper that already includes many findings, I'd recommend either integrating better or removing.
Following this guidance, we removed the readiness-to-remember findings in the interest of space and clarity.
