Psycholinguistics · Cognitive Neuroscience · Computational Modeling

Research

One question runs through all of my work: how does the brain construct, maintain, and break down the mental models that make language meaningful?

Research Statement

When we understand language we build a mental model of what is being described: who is involved, where they are, what happened. My research asks how the brain builds and keeps that model, and how it falls apart. I study this in three ways.

What breaks after a stroke

When a stroke damages the language system, how it fails tells us how it was built. Using lesion-symptom mapping and connected speech, I ask whether the kinds of naming errors a person makes (a related word, a similar-sounding word, a non-word, no answer) point to where the lesion is more reliably than the number of errors does. The pattern of a breakdown says more about the system underneath than its severity does.

Breaking a language model the same way

Can a large language model be damaged so that it fails the way a person does? I perturb parts of a model's architecture (attention heads, MLP layers, the residual stream) and look for the changes in discourse that go with aphasia and mild cognitive impairment. I check the model's behavior against speech and cognitive data from the Aging Brain Cohort at the University of South Carolina (ABC@UofSC), a sample of healthy and preclinical older adults, treating cognitive decline as a matter of degree rather than a diagnosis.

III

Where a voice came from

My dissertation asks how listeners keep track of where voices and sounds come from while they follow what is being said, and how that spatial record interacts with reference and discourse. Participants listen to stories over a ring of loudspeakers, and I measure what they remember about location and how it affects their comprehension. The question is whether tying words to places is a general part of how we understand discourse, not just something readers do with text on a page.

A newer line of work takes methods from the study of human conversation to a different question: when two AI agents talk to each other, can people, or other AI systems, keep up?

Current Projects

Late to the Conversation: Can People and AI Monitors Follow AI–AI Conversations?

AI Safety Multi-Agent Communication Psycholinguistics LLM Evaluation

As AI agents increasingly communicate with each other, human oversight depends on whether people can follow what they say. In human conversation, partners quickly develop shorthand that outsiders struggle to understand (Schober & Clark, 1989). This project asks whether AI–AI conversations show the same drift away from legibility, and whether people and AI monitors are equally able to keep up. Pairs of multimodal AI models play a repeated visual reference game (tangram figures), and human participants and AI models then try to identify the figures as “overhearers,” either from the start of the conversation, joining late, or seeing isolated descriptions. A further condition adds communication pressure to test whether efficiency demands make AI conventions harder for people to follow than for AI monitors. The pilot is complete and human data collection is under way.

Spatial Memory of Auditory Sources in Language Comprehension

Spatial Cognition Psycholinguistics Auditory Processing Language Comprehension

This project investigates how people encode and retain information about the spatial locations of auditory sources while comprehending spoken language. Using behavioral and neuroimaging paradigms, I examine the degree to which spatial context — specifically where a voice or sound originates in space — is integrated into the mental models listeners construct during discourse processing.

Results were presented at Human Sentence Processing 2026 (Cambridge, MA), Psychonomics 2025 (Denver, CO), and the Graduate Students in Linguistics Student Research Symposium 2025 (Columbia, SC), in collaboration with Dr. Amit Almor.

Error Profiles as Behavioral Biomarkers of Lesion Location

Aphasia Lesion-Symptom Mapping Neuroimaging Connected Speech

This project examines the cross-task stability of naming error profiles across the Philadelphia Naming Test (PNT) and Western Aphasia Battery-Revised (WAB-R) using lesion-symptom mapping (LSM). The goal is to establish whether error profile distributions — rather than overall accuracy — serve as reliable behavioral biomarkers of lesion location in post-stroke aphasia. Manuscript under review at Brain Communications with co-authors Roger Newman-Norlund, Saeed Ahmadi, Nadra Salman, Yong Yang, Regan Willis, Xiang Guan, and Julius Fridriksson.

LLM Perturbation Modeling of Cognitive Decline

LLM Cognitive Decline MCI Computational Modeling Discourse Analysis

This project develops a computational framework for modeling mild cognitive impairment (MCI) by systematically perturbing the internal architecture of large language models (e.g., attention heads, MLP layers, residual stream). Perturbation-induced changes in discourse output are validated against linguistic and cognitive data from the Aging Brain Cohort at the University of South Carolina (ABC@UofSC) across three discourse genres: picture description, story retelling, and procedural discourse. The project treats MoCA scores as a continuous cognitive dimension rather than a categorical MCI label.

Reference Processing Modulation During HD-tDCS

HD-tDCS Reference Processing Cognitive Neuroscience Brain Stimulation

In collaboration with Sarah Wilson, Samantha Langley, Dirk-Bart den Ouden, and Amit Almor, this project examined how high-definition transcranial direct current stimulation (HD-tDCS) of the superior parietal cortex modulates linguistic reference processing during reading comprehension (Wilson, Langley, Warren, den Ouden, & Almor, in prep). Findings were presented at the Society for the Neurobiology of Language (2023, Marseille).

StatTeacher Toolkit

Statistics Education Python Pedagogy Open-Source Tool

Drawing on my experience teaching introductory and methods courses in psychology as Instructor of Record, the StatTeacher Toolkit is an interactive application designed to help statistics instructors and students engage with core statistical concepts through live visualization and simulation.

Explore the Toolkit →

Publications

Published and In Press

2026

Fridriksson, J., Newman-Norlund, R. D., Ahmadi, S., Willis, R., Salman, N., Warren, K., … & Rorden, C. (2026). Stroke lesions as a Rosetta Stone for language model interpretability. arXiv preprint arXiv:2602.04074. https://arxiv.org/abs/2602.04074

2025

Warren, K., Nichols, C., Petersen, D., Shalin, V. L., & Almor, A. (2025). Stylistic language drives perceived moral superiority of LLMs. Scientific Reports, 15(1), 39168. https://doi.org/10.1038/s41598-025-25046-9

Under Review

Under review

Nemati, S., Newman-Norlund, R. D., Ahmadi, S., Guan, X., Warren, K., Yang, Y., Nelakuditi, S., Rorden, C., Bonilha, L., & Fridriksson, J. (under review). Head-level lesion-symptom mapping of picture naming in vision-language models. Neuroscience. Submitted September 2026.

Under review

Warren, K., Newman-Norlund, R. D., Ahmadi, S., Salman, N., Yang, Y., Willis, R., Guan, X., & Fridriksson, J. (under review). Cross-task lesion-symptom mapping in chronic aphasia. Brain Communications. Submitted May 2026.

Under review

Ahmadi, S. S., Salman, N., Newman-Norlund, R. D., Warren, K., Willis, R., Yang, Y., Guan, X., Kristinsson, S., Bonilha, L., & Fridriksson, J. (under review). The elicitation effect in aphasic naming. Journal of Speech, Language, and Hearing Research. Submitted April 2026.

Under review

Visikamalla, H., Riccardi, N., Rangus, I., Warren, K., Nichols, C., Horacek, K., Newman-Norlund, S., Fridriksson, J., & Bonilha, L. (under review). Tip-of-the-tongue word-finding phenomenon: The significance of the white matter language network. Neurobiology of Language. Submitted March 2026.

Under review

Warren, K. (under review). StatTeacher Toolkit: A browser-based application for generating and practicing psychological statistics problems. Journal of Open Source Education. Submitted March 2026.

Manuscripts in Preparation

In prep

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Salman, N., Guan, X., Yang, Y., Nelakuditi, S., Bonilha, L., & Fridriksson, J. (in prep). Patient-specific digital twins of aphasia from lesioned and retrained language models.

In prep

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Salman, N., Guan, X., Yang, Y., Nelakuditi, S., Bonilha, L., & Fridriksson, J. (in prep). Perturbation profiles in language models resemble human aphasia.

In prep

Guan, X., Newman-Norlund, R. D., Yang, Y., Ahmadi, S., Willis, R., Salman, N., Warren, K., Nelakuditi, S., Desai, R., Rorden, C., Bonilha, L., & Fridriksson, J. (in prep). PRISM: A causal framework for layer-resolved subtraction analysis in perturbed language models.

In prep

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Willis, R., Salman, N., Guan, X., Yang, Y., Nelakuditi, S., Desai, R., Bonilha, L., Rorden, C., & Fridriksson, J. (in prep). Inverse symptom-lesion mapping in human aphasia: Predicting left-hemisphere damage location from behavioral phenotype.

In prep

Yang, Y., Newman-Norlund, R., Guan, X., Ahmadi, S., Willis, R., Salman, N., Warren, K., Arheix-Parras, S., Nelakuditi, S., Desai, R. H., Bonilha, L., Rorden, C., & Fridriksson, J. (in prep). Recovering internal perturbations from behavioral error profiles in large language models.

In prep

Warren, K., Ahmadi, S. S., Guan, X., Salman, N., Yang, Y., Willis, R., Newman-Norlund, R. D., Nelakuditi, S., Desai, R., Bonilha, L., & Fridriksson, J. (in prep). Modeling aphasic taxonomic and thematic errors in a lesioned LLM.

In prep

Wilson, S. C., Langley, S., Warren, K., den Ouden, D. B., & Almor, A. (in prep). Transcranial direct current stimulation of the superior parietal cortex modulates reference processing during reading comprehension.

In prep

Warren, K. & Almor, A. (in prep). Hearing where they were: Spatial indexing as a modality-general discourse mechanism.

Conference Presentations

2026

Warren, K., Xie, A., Fahey, D., & Tasseva-Kurktchieva, M. (2026, September–October). Orthography outweighs phonology in bilingual cognate similarity judgments. Poster, Society for the Neurobiology of Language 2026, Geneva, Switzerland.

2026

Warren, K., Newman-Norlund, R. D., Ahmadi, S., Salman, N., Yang, Y., Willis, R., Guan, X., Bonilha, L., & Fridriksson, J. (2026, September–October). Neural–behavioral dissociations in cross-task lesion–symptom mapping of naming errors in chronic aphasia. Poster, Society for the Neurobiology of Language 2026, Geneva, Switzerland.

2026

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Salman, N., Yang, Y., Willis, R., Guan, X., Bonilha, L., & Fridriksson, J. (2026, September–October). Inverse symptom–lesion mapping in human aphasia: Predicting left-hemisphere damage location from behavioral phenotype. Poster, Society for the Neurobiology of Language 2026, Geneva, Switzerland.

2026

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Salman, N., Yang, Y., Willis, R., Guan, X., Bonilha, L., & Fridriksson, J. (2026, September–October). Severity scores blur, error types reveal: Behavioral choice shapes lesion–language mapping. Poster, Society for the Neurobiology of Language 2026, Geneva, Switzerland.

2026

Newman-Norlund, R. D., Warren, K., Ahmadi, S., Salman, N., Guan, X., Yang, Y., Willis, R., Bonilha, L., & Fridriksson, J. (2026, September–October). A universal bidirectional model of aphasic behavior and left-hemisphere lesion in 296 patients. Poster, Society for the Neurobiology of Language 2026, Geneva, Switzerland.

2026

Warren, K. & Almor, A. (2026, March). Spatial memory of auditory sources in language comprehension. Poster, Human Sentence Processing (HSP), Cambridge, MA.

2025

Warren, K. & Almor, A. (2025, November). Spatial memory of auditory sources in language comprehension. Poster, Psychonomic Society Annual Meeting, Denver, CO.

2025

Warren, K. & Almor, A. (2025, November). Spatial memory of auditory sources in language comprehension. Talk, Graduate Students in Linguistics Student Research Symposium, Columbia, SC.

2023

Wilson, S. C., Langley, S., Warren, K., den Ouden, D., & Almor, A. (2023, October). Reference processing modulation during HD-transcranial direct current stimulation. Poster, Society for the Neurobiology of Language, Marseille, France.

2023

Wilson, S. C., Langley, S., Warren, K., den Ouden, D., & Almor, A. (2023, April). Reference processing modulation during HD-transcranial direct current stimulation. Poster, Discover USC, Columbia, SC.

2022

Warren, K., Bury, H., & Weist, M. D. (2022). Comparing clinicians' and graduate students' multicultural knowledge and awareness in school psychology and clinical-community programs. Poster, Southeastern School Behavioral Health Conference, Myrtle Beach, SC.

2020

Che, B., & Warren, K. (2020). Air cleaning drone feasibility: Designing a mobile filtration unit to reduce indoor particulate matter. Poster, South Carolina Junior Academy of Science. scholarexchange.furman.edu →