Article activity feed

  1. Evaluation Summary:

    Cohen and Baldassano present analyses of a large publicly available set of neuroimaging data from children and adolescents watching an animated movie, and is likely be of interest to neuroscientists interested in methods for analyzing naturalistic neuroimaging data, or those interested in the development of narrative processing in the brain. The results are not embedded in a strong theoretical framework and it is not yet clear which hypotheses are supported and which are refuted. However, the methodological approach developed here is a valuable addition to the repertoire of developmental neuroscience.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1, Reviewer #2 and Reviewer #3 agreed to share their names with the authors.)

  2. Reviewer #1 (Public Review):

    Cohen and Baldassano study the neural response to narrative videos over development in a very large sample (N > 400), especially for an fMRI study. They investigate these neural responses using sophisticated computational analyses, including intersubject correlations and hidden Markov models of event segmentation. These outstanding features of the study are especially impressive given the naturalistic nature of the experimental paradigm, i.e. having participants watch narrative movies (cartoons). Analysing such continuous rather than trial-based data in a meaningful way is challenging, and Cohen and Baldassano's efforts at this are highly impressive.

    Cohen and Baldassano reveal that, with age, cortical event segmentation becomes more consistent over participants. Furthermore, event representations shift earlier in time, possibly reflecting increased anticipation of upcoming events. The relatively unconstrained nature of the stimuli does make it difficult to fully rule out all possible alternative interpretations of the findings, such as differences in eye movements and head motion between age groups.

    Both the insights gleaned from these results, as well as the experimental approach developed here, are likely to have a big impact on the field of developmental neuroscience.

  3. Reviewer #2 (Public Review):

    The authors present an analysis of publicly available data collected from children and adolescents watching a professionally made, animated movie. Previous research in adults suggests that cortical regions represent the content of events in the narrative, while hippocampal activity marks transitions between these events. A large dataset of children and adolescents could therefore be used to test the development timecourse by which these adult mechanisms arise. The paper finds that in this dataset, models fit better in many cortical regions in older adolescents than young children, and (unexpectedly) the hippocampal response at event boundaries is smaller in older adolescents than younger children. We have two major conceptual suggestions for improving the manuscript: (a) clarify which hypotheses were tested, supported, and refuted; and (b) provide stronger tests of data quality in the younger children. In addition we have other technical suggestions to clarify both the measures and the interpretations.


    1. In its current form, it is not clear what is at stake in the present study, or what specific hypotheses the manuscript hopes to test. The strongest claim in the paper is that child development involves a "shifting division of labor" between "episodic encoding processes" and "schematic event representations", but these terms are not clearly defined, and it is not clear which specific findings support or refute this statement. The paper could be significantly improved if the authors provide: i) clear definitions of these various terms, and thus the overall goal of the study (e.g., what is a "schematic event representation"? Is a "schematic event" defined with respect to the narrative arc of the story, or is it really any temporal segmentation that might be found in any region? ); ii) specific neural predictions (i.e.,, can all of cortex really be said to support "schematic event representation"? If not, which regions are most critical?), and iii) alternative hypotheses (e.g., what pattern would disconfirm the hypothesis that schematic event representations are increasing over child development? Will any detected developmental change in any direction and in any cortical region support this hypothesis, or would failure to find developmental change in certain cortical networks refute the hypothesis? And should we expect that the same regions will show developmental change across the various analyses, or is it fine for the overall hypothesis if different sets of regions show significant effects across analyses?).

    To illustrate the problem, consider the relationship between this work and previous work by the same group, which found that certain regions show stronger "schematic event representation" (e.g., mPFC) than others (e.g., auditory cortex) (Baldassano et al., 2018). Despite the claim that schematic event representation is increasing over development, the current study finds weak intersubject correlations (Supplemental Figure 1), and no developmental change in model fit (Figure 4) nor event boundaries (Figure 5) in mPFC, alongside strong effects in auditory cortex for these same analyses. Yet the authors do not emphasize the stability of event representation over development, and instead claim to find evidence of "shifting schematic event representations". In short, a more specific definition of the hypothesis space is required, to determine what pattern could confirm or disconfirm the overall claim, what alternative explanations should be considered, what control analyses ought to be run, or how the current set of results relate to each other or previous results in the literature.

    One possibility is that the study does not test any particular developmental hypotheses, and is completely exploratory or methods-oriented. If so, the authors could improve the manuscript by stating in more detail why such an exploratory analysis might be beneficial (e.g., what hypotheses and theories we might hope to generate from such a data-driven exploration?), and/or reframing the paper in terms of the methodological approach (e.g., as a new way of exploring naturalistic pediatric fMRI data).

    1. The authors should make a more convincing case that results are not explained by lower data quality in younger children than older ones, and that the data from the youngest children on their own are strong and believable. Some results do suggest that the youngest children's data are meaningful, such as the surprising finding that hippocampal responses to event boundaries actually decrease with age. On the whole, however, this issue needs more serious consideration.
  4. Reviewer #3 (Public Review):

    In this manuscript, Cohen and Baldassano sought to determine whether and how neural representations of ongoing experiences change with early-life development by analyzing a large, publicly available dataset consisting of children and adolescents. The participants watched a narrative-driven cartoon while in the scanner, and intersubject correlations (ISC) were used to evaluate both within and between-group neural similarity during viewing. The authors report that ISC increases with age in sensory regions, but surprisingly decreases with age in higher-order regions including the posterior medial cortex (PMC) at event boundaries. The authors further report that event model fits worsened, and neural evidence for anticipation of event transitions decreased with age. The authors attribute this finding to younger children relying more heavily on the PMC because of their limited knowledge for schemas. Additionally, the authors looked at hippocampal responses and found that correlation between the hippocampus (HPC) and event boundaries decreases with age, which was limited to the anterior HPC. The authors suggest that the decreases in HPC responsiveness with age may be related to increases in the reliance of schematic representations, with adolescents relying more on schemas than younger children.

    Overall, the study is interesting and uses a novel measure of response similarity to better understand neural representations of naturalistic stimuli with respect to brain development. The paper is written well, and the design and results are clearly communicated. I think that a major strength of this paper lies in the novel application of these analytical methods to a developmental sample. These analyses are thoughtfully and appropriately executed, and the paper should be of fairly broad interest. I do not have any particularly serious issues with the study, the data, or the manuscript, but one point of concern is that event boundary ratings used to analyze the data were derived from adults, and not individuals from the age range of interest. This could potentially raise interpretational challenges, and so a discussion of this or perhaps additional analytical steps would be helpful. Furthermore, the authors attempt to explain what some might consider to be rather unexpected results (particularly results in PMC), but I think that clearer reasoning and argumentation would benefit the paper in terms of better contextualizing these findings. Finally, there are some lingering concerns about the potential role of signal-to-noise ratios (SNR) in the data, which could affect age differences in the neural findings. I detail these concerns and potential paths forward in the separate recommendations for authors.