Neural correlates of internal context transitions in continuous thoughts during free speaking
Poster No:
967
Submission Type:
Abstract Submission
Authors:
Kwon Heo1, DASOM KWON1, WON MOK SHIM1
Institutions:
1CNIR/SKKU, Suwon-si, Gyeonggi-do
First Author:
Co-Author(s):
Introduction:
Continuous experiences of external input can be organized into discrete events, and the boundaries of these events elicit neural responses in the hippocampus [1]. Similar neural responses are also triggered during transitions in mental context when recalling movie narratives, indicating that neural dynamics may capture shifts in both external and internal situational contexts [4]. Analogous to context transitions in movie recall, changing topics during spontaneous thoughts coincides with shifts in mental states [7]. However, the neural mechanism underlying such shifts in internally generated context without external input remains elusive. In this study, we examined the neural dynamics of topic transitions during free speaking by quantifying the degree of context changes in one's speech based on a language model.
Methods:
During fMRI scan, independent groups of participants engaged in three speaking tasks: 1) freely expressing spontaneous thoughts (think-aloud, N = 63), 2) alternately discussing two pre-given topics at their own pace (topic-alternating, N = 53), and 3) recalling movies that they had watched (movie-recall, N = 24). Following the scan, participants received a transcript of their speech and retrospectively identified topic boundaries in think-aloud and topic-alternating tasks. For the movie-recall task, topic boundaries were labeled by the author, corresponding to movie event segmentation reported by independent annotators. To assess the degree of context transition of speech, we calculated next sentence prediction (NSP) scores using a pre-trained language model [2], indicating the likelihood of each spoken sentence given the preceding context. Additionally, we measured pause length (PL), the silent interval between spoken sentences to examine the impact of sensory and motor changes in speaking behaviors associated with topic transitions.
Results:
We computed the linear regression coefficient (β) between the predicted single-event hemodynamic response at the end of each sentence and BOLD signals in 1000 brain parcels [6]. As the NSP scores and pause length were correlated in all speech tasks (think-aloud: ρ = − 0.22, p < 0.01; topic-alternating: ρ = − 0.31, p < 0.01; movie-recall: ρ = − 0.15, p < 0.01), partial rank correlations were used to disentangle their individual effects. Specifically, we examined the relationship between β coefficients and NSP scores when controlling for pause length (ρβ⋅NSP | PL) and vice versa (ρβ⋅PL | NSP). In the control network, significant negative correlations were observed between the coefficients and NSP scores when controlling for pause length, indicating increased responses when the context changed. Conversely, in the same network, positive correlations between the coefficients and pause length were observed (see Fig 1). To examine the impact of context transitions estimated by the language model compared to topic transitions perceived by humans, we computed coefficients after topic boundaries indicated by human annotators in each speaking task. Consistent with the NSP results, the control network exhibited increased responses at the annotated topic boundaries compared to the midst of topics across all tasks. In contrast, the hippocampus showed increased responses at topic boundaries only during movie-recall.
Conclusions:
In summary, our findings highlight the unique functional roles of the control network and the hippocampus in organizing internally generated thoughts. The results suggest that the control network contributes to cognitive processing associated with topic transitions during spontaneous thoughts, possibly by regulating introspective processes [3]. Conversely, the hippocampus is more involved in narrative retrieval, regardless of whether observing existing narratives or generating them internally. The distinct roles of these two brain regions may collectively involve a process of discretizing continuous experiences into meaningful chunks [5].
Higher Cognitive Functions:
Higher Cognitive Functions Other 1
Language:
Language Comprehension and Semantics 2
Keywords:
FUNCTIONAL MRI
Other - Continuous thoughts
1|2Indicates the priority used for review
Provide references using author date format
[2] Devlin, J. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
[3] Dixon, M. L. (2018). Heterogeneity within the frontoparietal control network and its relationship to the default and dorsal attention networks. Proceedings of the National Academy of Sciences, 115(7), E1598-E1607.
Lee, H. (2022). A generalized cortical activity pattern at internally generated mental context boundaries during unguided narrative recall. Elife, 11, e73693.
[4] Radvansky, G. A. (2017). Event boundaries in memory and cognition. Current opinion in behavioral sciences, 17, 133-140.
[5] Schaefer, A. (2018). Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cerebral cortex, 28(9), 3095-3114.
[6] Sripada, C. (2020). Structure in the stream of consciousness: Evidence from a verbalized thought protocol and automated text analytic methods. Consciousness and Cognition, 85, 103007.
[7] Yeo, B. T. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of neurophysiology.