Brain Network Reconfiguration for Narrative and Argumentative Thought
Curation statements for this article:-
Curated by eLife
Summary: The reviewers thought this was a nicely written paper, and were interested in the idea of extending intersubject correlation (ISC) and intersubject functional connectivity (ISFC) work on narratives to arguments. One major concern was that effects reported here may be driven in part by the scrambled conditions. Specifically, the scrambled argument seems to have resulted in stronger and more widespread ISC than the intact argument: this would call into question assumptions about the scrambled version being a control condition. Another, related concern raised is that perhaps argumentative texts are very different from narrative texts: perhaps argumentative texts are less structured, or less interesting (?) and this is why the intact and scrambled versions are so similar. Together with other issues, relating to the interpretation of the findings, it was felt that while of interest, the study's major conclusions could not be justified without additional experiments.
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
Our brain constructs reality through narrative and argumentative thought. Some hypotheses argue that these two modes of cognitive functioning are irreducible, reflecting distinct mental operations underlain by separate neural bases; Others ascribe both to a unitary neural system dedicated to long-timescale information. We addressed this question by employing inter-subject measures to investigate the stimulus-induced neural responses when participants were listening to narrative and argumentative texts during fMRI. We found that following both kinds of texts enhanced functional couplings within the frontoparietal control system. However, while a narrative specifically implicated the default mode system, an argument specifically induced synchronization between the intraparietal sulcus in the frontoparietal control system and multiple perisylvian areas in the language system. Our findings reconcile the two hypotheses by revealing commonalities and differences between the narrative and the argumentative brain networks, showing how diverse mental activities arise from the segregation and integration of the existing brain systems.
Article activity feed
-
Reviewer #2:
The paper titled "Brain Network Reconfiguration for Narrative and Argumentative Thought" sought to uncover the common neural processing sequences (time-locked activations and deactivations; inter-subject correlations and inter-subject functional connectivity) underlying narrative and argumentative thought. In particular, the study aimed to provide evidence that would help adjudicate between two current theories: the Content-Dependent Hypothesis (narrative argumentative) and the Content-Independent Hypothesis (narrative = argumentative). In order to assess these possibilities they tested participants in an fMRI scanner as they listened to validated narrative and argumentative texts. Each text condition was directly compared to resting state and scrambled versions of the texts. Across a range of interesting analyses that …
Reviewer #2:
The paper titled "Brain Network Reconfiguration for Narrative and Argumentative Thought" sought to uncover the common neural processing sequences (time-locked activations and deactivations; inter-subject correlations and inter-subject functional connectivity) underlying narrative and argumentative thought. In particular, the study aimed to provide evidence that would help adjudicate between two current theories: the Content-Dependent Hypothesis (narrative argumentative) and the Content-Independent Hypothesis (narrative = argumentative). In order to assess these possibilities they tested participants in an fMRI scanner as they listened to validated narrative and argumentative texts. Each text condition was directly compared to resting state and scrambled versions of the texts. Across a range of interesting analyses that focus on how each participant's brain synchronized with other participants' brains throughout the same narrative and argumentative texts, they primarily found support for the content-dependent hypothesis with a few differences and commonalities across text conditions. Relative to the scrambled conditions, listening to narrative texts was more associated with default mode activity across participants and listening to argumentative texts only activated a common network of superior fronto-parietal control regions and language regions. Argumentative texts did not differ much from scrambled versions of the same text. These patterns reveal themselves in both ISC and ISFC data. Overall, I feel like this paper is really well written and is a novel approach to distinguishing the neural processes between similar, but different types of thought. At times the manuscript loses touch with its primary brain coordination metrics (ISC and ISFC), describing the findings more like a GLM or functional connectivity study.
Comments:
Introduction:
The introduction is very clearly written and uses a wonderful variety of sentence structure. Well done!
While the writing is beautiful, a few sentences are less easy to comprehend than others. For example the use of outstands in line 36 is a bit difficult to parse on first read. Consider simplifying the language some.
There seems to be an opportunity to discuss this work and its findings in a broad context of narrative or argumentative self-generated internal thought (not based on listening to texts). For instance, I think there could be a few sentences tying this work to studies of autobiographical memory retrieval or mind wandering (for argumentation perhaps studies of the cognitive and neural processes behind complex decision making). This is captured to some extent in the introduction and discussion, but I think it could go further with citations beyond those just associated with listening to various types of text.
Appreciate the thorough discussion of hypotheses and background.
It is not necessary, but it might be interesting to show some basic functional connectivity analyses of the individual participant activations in supplemental analyses (no ISC or ISFC).
Methods:
- Please clarify how the ISFC analysis can be directional in any way? Does unidirectional mean that you're just taking one value for each pairwise connection Cij?
Results:
To what extent is there a concern that participants would still try to stitch together the scrambled narratives even if they are less coherent? Was this even possible given the nature of the stimuli?
In line 125 and throughout the authors should consistently remind the reader that 'engagement' in this case means that there were consistent and correlated increases in the bold response across participants. This differs in some ways to task engagement in event-related GLM studies.
The language throughout should reflect consistent involvement across participants at particular time points in each of the narratives vs the argumentative.
It seems like argumentative is more similar to the scrambled in many ways. Might it be that argumentative texts are just less coherent and structured than narrative texts?
It seems clear that the neural processing of argumentative texts (64 distinct edges) were very different from the narrative texts (2348 distinct edges), but that the current contrasts did not clearly and consistently distinguish argumentative thought from the scrambled argument conditions. A discussion of the analyses that might be necessary to better elucidate the dynamics of processing for argumentative thought would be helpful.
Discussion:
Were there any neural differences between the narrative vs argument scrambled-texts? This might reveal any differences in the processing of the scrambled texts for each condition and might help shine light on features of the scrambled argument condition that contributed to the overall lack of distinction relative to the narrative vs scrambled narrative conditions.
Throughout the results from ISC and ISFC findings are convolved with the findings from univariate or GLM results from prior studies. Please compare and contrast how ISC and ISFC findings might relate to univariate or GLM findings early in the discussion.
Related to point 2 in the introduction, please also cite studies from autobiographical memory retrieval studies that also show the frontoparietal control system working as information is iteratively accumulated and updated over long temporal windows (St. Jacques et al., 2011; Inman et al., 2018; Daselaar et al., 2008).
Please reconsider how the ISC findings are discussed as 'activation'. While the BOLD activity of these areas are certainly coordinated across participants at similar points in the text, I feel like the term activation fits best with studies that convolve the brain activity with an HRF. In particular, from what I understand ISC, a common decrease in BOLD activity across participants at the same time in a read text would also lead to activity or 'activation' of that area in an ISC analysis. This seems counterintuitive. The 2nd paragraph of the discussion describes ISC and ISFC well in terms of what it shows across a sample (synchronization of fluctuations in BOLD activity across participants for the same stimuli). "Activity" may capture this, but please consider some more nuanced ways to refer to these ISC and ISFC findings.
Figures:
Please double check the box plots in figure 1a for Scene Construction. Another method of displaying this likert rating data might be helpful. While appreciating the attempt to display the individual data points, the simple main points get somewhat obscured by all of the information in the graph.
Overall, I appreciate the attention to detail in all of the figures and the completeness of the data visualization with several useful supplemental figures.
-
Reviewer #1:
Xu and colleagues compared the intersubject correlation (ISC) and intersubject functional connectivity (ISFC) of participants listening to narrative and argumentative texts while undergoing fMRI. Replicating earlier findings, they show that ISC in the DMN was greater when participants listened to an intact narrative than when they listened to a sentence-scrambled version of the same narrative. Listening to a sentence-scrambled argument elicited ISC in language and control regions of the brain, though interestingly, there was no region in the brain where ISC was greater when participants listened to an intact version of the argument. Instead, there was greater ISFC between the IPS and language areas of the brain when participants listened to the intact argument than when they listened to the scrambled argument. The authors …
Reviewer #1:
Xu and colleagues compared the intersubject correlation (ISC) and intersubject functional connectivity (ISFC) of participants listening to narrative and argumentative texts while undergoing fMRI. Replicating earlier findings, they show that ISC in the DMN was greater when participants listened to an intact narrative than when they listened to a sentence-scrambled version of the same narrative. Listening to a sentence-scrambled argument elicited ISC in language and control regions of the brain, though interestingly, there was no region in the brain where ISC was greater when participants listened to an intact version of the argument. Instead, there was greater ISFC between the IPS and language areas of the brain when participants listened to the intact argument than when they listened to the scrambled argument. The authors interpret their results as suggesting that listening to the intact argument did not recruit additional brain systems, but instead promoted the cooperation between regions that were already involved in processing the argument.
Most prior work using "naturalistic stimuli" has examined the neural responses to narratives. This manuscript extends this work in an important way by examining how the brain responds to arguments, which comprise a non-trivial proportion of the linguistic content people are exposed to on a daily basis. The ISFC results (Fig. 7) are particularly noteworthy and novel. My main concerns have to do with the possibility that ISC for the scrambled argument seems to be stronger and more extensive than that for the intact argument, and how this might affect the authors' interpretation of their results. Below are some suggestions and comments which I think the paper could benefit from considering further:
I think it would be helpful to run the Scrambled Argument > Intact Argument ISC contrast. Visual inspection of Figure 2 suggests that ISC for the scrambled argument might be stronger than that for the intact argument, especially in control regions. If this is truly the case, I think the authors should discuss what this might imply about what is happening during the scrambled condition and if this affects thinking of the scrambled condition as a control for low-level linguistic features. In particular, the 2.97 out of 5 comprehensibility rating of the scrambled arguments suggests that participants might have understood the scrambled arguments. If participants are actively trying to make sense of the scrambled argument text, it seems like this could then drive observed differences in ISFC between the intact and scrambled arguments as well (e.g., decreased connectivity between control and language regions when trying to make sense of scrambled text, rather than increased connectivity between control and language regions when processing an intact argument).
More broadly, I think the authors need to make sure their effects aren't driven by the scrambled conditions. For example, for Figure 2 - figure supplement 2, the (Intact Narrative - Scrambled Narrative) > (Intact Argument - Scrambled Argument) contrast can be driven by high ISC in the Scrambled Argument condition, which would suggest a different interpretation of the results. My suggestion would be to run the contrast as (Intact Narrative - Scrambled Narrative) > max((Intact Argument - Scrambled Argument),0) to make sure that the contrast isn't driven by a negative value on the right hand side of the inequality.
Point 2 also applies to Figures 6 and 7. Relatedly, the rightmost panel of Figure 6C suggests that the analysis is indeed capturing some edges where the SES of the Scrambled Argument is greater than that of the Intact Argument.
How well do the vertexes identified in Figure 7D overlap with the Intact Argument > Resting map? Given the authors interpretation that the ISFC results suggest cooperation between areas involved in processing the intact stimulus, I think this should be properly assessed.
Both ISC and ISFC capture only signal that is shared across participants. Most narratives are crafted such that all listeners have a similar interpretation. This is unlike arguments, where different listeners might agree with an argument to a different extent. If listeners had differing interpretations of the argument, ISC/ISFC would miss brain activity/connectivity involved in processing an argument. I think this possibility should be considered and discussed, especially given the null DMN finding for the argumentative texts.
For the t-tests on the behavioral ratings , it looks like the authors collapsed over the two texts within a category. This doesn't seem right, given that the ratings for each text are dependent. A mixed model approach would be more appropriate. I doubt this will change the results, but I think it would be good to follow best practices when possible.
-
Summary: The reviewers thought this was a nicely written paper, and were interested in the idea of extending intersubject correlation (ISC) and intersubject functional connectivity (ISFC) work on narratives to arguments. One major concern was that effects reported here may be driven in part by the scrambled conditions. Specifically, the scrambled argument seems to have resulted in stronger and more widespread ISC than the intact argument: this would call into question assumptions about the scrambled version being a control condition. Another, related concern raised is that perhaps argumentative texts are very different from narrative texts: perhaps argumentative texts are less structured, or less interesting (?) and this is why the intact and scrambled versions are so similar. Together with other issues, relating to the interpretation …
Summary: The reviewers thought this was a nicely written paper, and were interested in the idea of extending intersubject correlation (ISC) and intersubject functional connectivity (ISFC) work on narratives to arguments. One major concern was that effects reported here may be driven in part by the scrambled conditions. Specifically, the scrambled argument seems to have resulted in stronger and more widespread ISC than the intact argument: this would call into question assumptions about the scrambled version being a control condition. Another, related concern raised is that perhaps argumentative texts are very different from narrative texts: perhaps argumentative texts are less structured, or less interesting (?) and this is why the intact and scrambled versions are so similar. Together with other issues, relating to the interpretation of the findings, it was felt that while of interest, the study's major conclusions could not be justified without additional experiments.
-
