Xian Li


Home

Research

Teaching

Demo

CV (PDF)



Google Scholar GitHub Email Bluesky

Research

How do meaning and feeling emerge from the experiences we live through?

Lived experience is not a sequence of isolated trials. It unfolds continuously, accumulates over a lifetime, and is shaped by our choices and punctuated by interruptions. I study the emergent properties of naturalistic cognition: the memories, interpretations, and feelings that arise when a mind with a history meets events as they unfold. Memory research increasingly uses stories and films, yet many paradigms still overlook three defining features of lived experience: we actively shape its course, we continually navigate internal and external interruptions, and our feelings reflect both the present moment and our past. My research brings these features together to understand how cognitive and neural processes support everyday life.


The three branches below build on one another. Choosing what happens next personalizes memory, suggesting that memory functions as more than a record: it is a scaffold built around the person. As events unfold, that scaffold must survive the interruptions of everyday life, raising the question of how the brain preserves it against interference. That same scaffold may also shape how new events feel—a question at the heart of my postdoctoral work. I combine naturalistic experiments in which people read, listen to, or choose their way through stories with functional magnetic resonance imaging (fMRI) and large language models (LLMs) to study how experience takes shape over time. Understanding memory as a constructed, personal scaffold also has implications for AI: it helps clarify what systems that read, remember, and personalize should emulate about human cognition, and what they may not capture on their own.

1. Agency: how acting on events shapes what we remember

In real life we are not passive recipients of events; we choose, and our choices change what happens next. This branch asks how that sense of agency changes memory for the events themselves.

Choose your own adventure: when readers steer the story, their memories become their own

Branching story graph of the choose-your-own-adventure paradigm, with two readers' paths highlighted
Two readers' paths (red and blue) through the branching graph of one adventure story.
To study agency in a naturalistic setting we built the choose-your-own-adventure (CYOA) paradigm: a branching story presented a few sentences at a time, in which readers choose what happens next. Comparing readers who had full, partial, or no control over the path of the same story, we found that agency personalizes episodic memory: readers who chose recalled more idiosyncratic combinations of the same events, and their memories were less predictable from generic sentence embeddings, rather than simply better or worse. Choice did not add memory; it moved memory away from the space that a model of everyone's text would predict, and toward the person. That makes memory look less like a record and more like a scaffold built around the individual, and it raises the question of the next branch: how is that scaffold held when experience is interrupted?
paper (Nature Communications, 2026) preprint try the paradigm

2. Continuity: how the mind keeps a thread through interruptions and wandering thoughts

A memory scaffold is built while experience unfolds, yet experience is interrupted all the time, by pauses, by other people, by our own thoughts, and we pick the story back up. This branch asks how the brain keeps prior mental context available across interruptions, and what happens when thought runs freely.

Where was I? How the brain keeps a story alive across interruptions

Three-panel cartoon: a reader absorbed in a book, interrupted, then asking where was I
Mental continuity: an absorbed reader is interrupted, then picks the thread back up.
Seventy-three adults listened to fourteen-minute stories interrupted after every forty seconds by silent pauses or by a competing theory-of-mind task, or heard scrambled controls. Using fMRI, we asked how the brain keeps prior mental context available across interruptions. Two pathways emerged. Hippocampal activity rose at each interruption onset, and the size of that rise predicted later recall, a signature of indexing the epoch just ended. Independently, posterior medial cortex carried the interrupted narrative as a reliable, gradually evolving pattern, an "episodic background" that persisted through the pauses and even through the competing task. The persisting cortical pattern was not a copy of the story pattern that preceded it: it was inverted, and the degree of inversion predicted how well it persisted. If a many-minute mental context is protected by transformation rather than by rehearsal, then continuity is a representational format, not an act of will.
preprint (bioRxiv, 2026) data and code CCN 2025 abstract

Thinking out loud: how memories and plans surface in the free flow of thought

Think-aloud paradigm: speech bubbles segmented into thought units, and five participants' thought timelines colored by category
Ten minutes of spoken thought, segmented into units and labeled by what each thought is about (figure adapted from Su et al., 2025).
Interruptions also come from within. Participants verbalized their thoughts continuously in the scanner. With Dr. Hongmi Lee and colleagues, we characterized the neural dynamics of spontaneous memory recall and future thinking in that continuous flow of thought, and released the annotated dataset. Transitions in spontaneous thought engage the same brain networks that external event boundaries do, which is what lets the interruption work above be extended to interruptions that people generate themselves. Together, the first two branches describe a structured, person-specific representation that is built while events unfold and held across interruption. The third asks how it is reused to construct emotion and social meaning.
paper (Nature Communications, 2025) dataset paper (Scientific Data, 2026)

3. Feeling: how memory builds social and emotional experience

If memory is a scaffold rather than a record, it should shape how new events feel. Constructionist and predictive theories of emotion make exactly that claim, yet direct tests are almost absent, because in ordinary stimuli the emotion of a scene cannot be separated from the scene. This is the focus of my postdoctoral work at UCLA with Dr. Maureen Ritchey and Dr. Ajay Satpute, which began in October 2026.

To be continued…

The instrument: humans and LLMs as subjects of the same science

The three branches share a method that did not exist five years ago. LLMs can annotate narrative experience at scale, and they can be studied as a second kind of comprehender. Where models and people agree, naturalistic memory research gains tools that scale by an order of magnitude; where they diverge, the divergence maps what human comprehension adds.

The causal backbone of a story: what holds events together, as judged by people and by language models

Icon of the Causal Rater for Story tool: an open book surrounded by linked nodes
A long story is not a list of events but a web of causes and consequences, and that web may be what memory holds on to. I measure how closely language models agree with people on the causal structure of stories, and whether model-derived causal centrality predicts human recall as well as human-derived centrality does. Treating the model as a rememberer, I also compare the organization of its recall (compression, ordering, causal clustering, temporal contiguity) with the human distribution. Presented at the Conference on Cognitive Computational Neuroscience (CCN) 2026 (Li, Li & Chen); manuscript in preparation.

narRaters app icon narRaters: open-source narrative recall processing, with a web interface at every step

narRaters workflow: story side (transcribe, segment events, causal rating) and recall side (correct text, parse clauses, match to story), with human review at every step
The narRaters workflow. Story side: transcribe, segment into events, rate causal links. Recall side, per participant: correct, parse into clauses, match to story events. Every step can be checked and edited in the browser before export.
Rich narrative data have always been the field's bottleneck: scoring a ten-minute story's free recall means hundreds of clause-to-event judgments per participant, and rating the causal relations among events grows with the square of story length. narRaters (narrative + raters) is an open-source pipeline and web interface that transcribes recall, segments stories into events, matches recall to events, and rates causal relations, using rule-based methods, local models, or cloud models, with a human review screen after every automated step. It is public on GitHub and PyPI with an archived release: python3 -m pip install narraters, then narraters serve.
narRaters project page narRaters on GitHub PyPI archived release (Zenodo)

Related and earlier work

The methods above were built in projects on perception and brain networks that preceded or ran alongside the three branches.

Each viewer's own path: how self-guided exploration of a scene shapes the brain's response to it

A painted diner scene and its segmentation into numbered regions
A painted scene and its segmentation into regions that viewers visit in their own order.
Studies of scene perception often use brief presentations of under a second, yet when people freely explore a complex photograph or painting, each person follows their own "exploration path" through the scene, and the history of that path may shape how they understand the next moment. Using fMRI, we tested whether a viewer's exploration path predicts the brain's response patterns to the scene, both within and across people. Presented at the Society for Neuroscience (SfN) 2021 (talk and poster).

Foundations: faces, attention, and brain networks

Before the Ph.D. I worked on the behavioral and neural signatures of developmental prosopagnosia (face blindness) with Dr. Joseph DeGutis and Dr. Michael Esterman at Harvard Medical School, on reward and visual attention with event-related potentials (ERPs) and fMRI at Capital Normal University, and on functional and structural brain networks in multiple sclerosis at the University at Buffalo. These collaborations are listed under Papers & Preprints.

Software and data releases


Preprints and manuscripts in preparation

Cortical and hippocampal pathways to mental continuity
Li, X., Lee, H., & Honey, C. J. (2026). bioRxiv. Under review.
preprint, data and code


Human and LLM alignment in causal reasoning for narrative events and their structural association with human memory
Li, X., Li, X., Antony, J. W., & Chen, J. (in preparation).


Peer-reviewed publications

An fMRI dataset of verbalized spontaneous thought with annotated transcripts and self-report trait measures
Zhang, M., Liu, P. R., Su, H., Zhao, M., Li, X., Born, S., Lee, Y., Honey, C. J., Chen, J., & Lee, H. (2026).
Scientific Data.
paper, preprint


Interactive narratives reveal the personalizing effect of agency on episodic memory
Li, X., Ni, N. K., Born, S. J., Gualano, R. J., Lee, I., Bellana, B., & Chen, J. (2026).
Nature Communications, 17, 7272.
paper, preprint


Left hemisphere abnormalities in face-selective activation and functional connectivity in developmental prosopagnosia
Campbell, A., Li, X., Rothlein, D., Esterman, M., & DeGutis, J. (2025).
Imaging Neuroscience, 3, IMAG.a.971.
paper


Neural dynamics of spontaneous memory recall and future thinking in the continuous flow of thoughts
Su, H., Li, X., Born, S., Honey, C. J., Chen, J., & Lee, H. (2025).
Nature Communications, 16, 6433.
paper


Effects of congruent emotional contexts during encoding on recognition: An ERPs study
Xie, M., Han, M., Liu, Z., Li, X., & Guo, C. (2024).
Psychophysiology, 61, e14516.
paper


Investigating the influence of autism spectrum traits on face processing mechanisms in developmental prosopagnosia
Fry, R., Li, X., Evans, T. C., Esterman, M., Tanaka, J., & DeGutis, J. (2023).
Journal of Autism and Developmental Disorders, 53, 4787–4808.
paper


From encoding to retrieval: Change in level of unitization resolves debate about unitization's effect on associative recognition
Liu, Z., Li, X., Li, X., Yuan, J., & Guo, C. (2023).
Psychophysiology, 60, e14161.
paper


Functional alteration due to structural damage is network dependent: Insight from multiple sclerosis
Bartnik, A., Fuchs, T. A., Ashton, K., Kuceyeski, A., Li, X., Mallory, M., et al. (2023).
Cerebral Cortex, 33, 6090–6102.
paper


Not so fast! Response times in the computerized Benton Face Recognition Test may not reflect face recognition ability
DeGutis, J., Li, X., Yosef, B., & Mishra, M. V. (2022).
Cognitive Neuropsychology, 39, 155–169.
paper


Neural mechanisms of reward-by-cueing interactions: ERP evidence
Li, X.*, Zhang, M.*, Wu, L., Zhang, Q., & Wei, P. (2021).
Frontiers in Human Neuroscience, 15, 608427. (*equal contribution)
paper


Preserved network functional connectivity underlies cognitive reserve in multiple sclerosis
Fuchs, T. A., Benedict, R. H. B., Bartnik, A., Choudhery, S., Li, X., et al. (2019).
Human Brain Mapping, 40, 5231–5241.
paper


Electrophysiological evidence of encoding in self-referential effect
Yang, X., Mao, X., Han, M., Li, X., & Guo, C. (2019).
NeuroReport, 30, 901–907.
paper


White matter tract network disruption explains reduced conscientiousness in multiple sclerosis
Fuchs, T. A., Dwyer, M. G., Kuceyeski, A., Choudhery, S., Carolus, K., Li, X., et al. (2018).
Human Brain Mapping, 39, 3682–3690.
paper


Semantic relationship shared between words: Influence on associative recognition supported by event-related potentials
Lyu, Y., Wang, Y., Mao, X., Li, X., & Guo, C. (2018).
NeuroReport, 29, 71–78.
paper



Human memory area data blitz (organized by Dr. David Clewett), Department of Psychology, University of California, Los Angeles (October 2026).


Tompary Lab (Dr. Alexa Tompary), Drexel University (September 2026).


Naturalistic Cognition Lab (Dr. Coraline Iordan), University of Rochester (August 2026).


Colloquium, Cognitive Neuroscience program, Ohio State University (April 2026).


Voss Lab and Jacobs Lab (Dr. Joel Voss, Dr. Josh Jacobs), University of Chicago (March 2026).


MemoLab (Dr. Maureen Ritchey), University of California, Los Angeles (March 2026).


Stanford Memory Lab (Dr. Anthony Wagner), Stanford University (February 2026).


Dynamic Perception and Memory Lab (Dr. Chris Baldassano), Columbia University (September 2025).


Princeton Neuroscience Institute seminar (Dr. Kenneth Norman), Princeton University (March 2025).


Decision Neuroscience Lab (Dr. Tobias Donner), University Medical Center Hamburg-Eppendorf (February 2025).


Episodic Memory Group (Dr. Chris Bird), University of Sussex (July 2022).


Invited talk, Vision Sciences Society Annual Meeting (June 2020).



Cortical and hippocampal pathways to mental continuity
Li, X., Lee, H., & Honey, C. J. (2026). Abstract, Society for Neuroscience (SfN) Annual Meeting.


Human and LLM alignment in causal reasoning for narrative events
Li, X., Li, X., & Chen, J. (2026). Conference on Cognitive Computational Neuroscience (CCN), New York.


Cognitive and neural origins of persistent experiences
Kressin Palacios, G., Li, X., Lai, Y., & Honey, C. J. (2026). CCN, New York. (mentored project)


Automated recall-to-story matching for memory research
Arekar, D., Kressin Palacios, G., Li, X., & Chen, J. (2026). CCN, New York. (mentored project)


Two pathways to mental continuity
Li, X., & Honey, C. J. (2026). Poster, Context and Episodic Memory Symposium (CEMS), Philadelphia.


Agency personalizes episodic memories
Li, X., Ni, N. K., Born, S. J., Gualano, R. J., Lee, I., Bellana, B., & Chen, J. (2025). Poster, SfN, Chicago.


Narrative representations resist interference
Li, X., & Honey, C. J. (2025). Poster, SfN, Chicago.


Neural origins of persistent mental content
Kressin Palacios, G., Li, X., Lai, Y., & Honey, C. J. (2025). Poster, SfN, Chicago. (mentored project)


Neural representations that resist interference
Li, X., & Honey, C. J. (2025). Poster, CCN, Boston.
CCN abstract


Agency enhances individuality in memory for narratives
Li, X., Ni, N. K., Born, S. J., Gualano, R. J., Lee, I., Bellana, B., & Chen, J. (2024). Talk, CEMS, Philadelphia.


Agency personalizes naturalistic event memories
Li, X., Ni, N. K., Born, S., Gualano, R. J., Lee, I., Bellana, B., & Chen, J. (2024). Poster, SfN, Chicago.


The role of agency in memory for narratives
Li, X., Ni, N. K., Born, S., Gualano, R., & Chen, J. (2022). Poster, CCN, San Francisco.


The role of agency in memory for narratives: A choose-your-own-adventure paradigm
Li, X., Ni, N. K., Born, S., Gualano, R., & Chen, J. (2022). Talk and poster, CEMS, Philadelphia.


Exploration path trajectory during self-guided viewing of complex photographs and paintings predicts brain activity patterns in the default network
Li, X., Su, H., Born, S., Chen, J., & Honey, C. J. (2021). Talk and poster, SfN, Chicago.


The role of agency in memory for narratives: A choose-your-own-adventure paradigm
Li, X., Ni, N. K., Born, S., Gualano, R., & Chen, J. (2021). Poster, CEMS, Philadelphia.


Deficient functional MRI selectivity and connectivity in developmental prosopagnosia is specific to face regions
Li, X., Arizpe, J., Rothlein, D., Esterman, M., & DeGutis, J. (2020). Journal of Vision, 20(11), 827 (Vision Sciences Society).


Faces are no object: Normal novel object performance in developmental prosopagnosia
Fry, R., Li, X., Wilmer, J., & DeGutis, J. (2020). Journal of Vision, 20(11), 1231 (Vision Sciences Society).
abstract


Alteration of resting-state functional connectivity in response to structural disruption differs between well-established functional networks in multiple sclerosis
Bartnik, A., Fuchs, T. A., Ashton, K., Li, X., Mallory, M., Oship, D., et al. (2020). Poster, American Academy of Neurology (AAN) Annual Meeting, Toronto.


Preservation of functional connectivity moderates the impact of white matter tract disruption on cognition in multiple sclerosis
Fuchs, T. A., Benedict, R. H. B., Choudhery, S., Li, X., Carolus, K., Mallory, M., et al. (2018). Poster, AAN Annual Meeting, Los Angeles.