About

Jonathan Brennan studies the mental structures and computations used to understand words and sentences, with a focus on how these processes are implemented in the brain. His research employs formal computational models of language comprehension to investigate the neural correlates of basic cognitive computations such as lexical access, syntax, and semantics, utilizing techniques like electroencephalography (EEG), magnetoencephalography (MEG), and functional magnetic resonance imaging (fMRI). He has a particular interest in experimental methods that are as natural as possible, such as having participants read or listen to a story, to focus on sentence processing as it occurs during everyday language use. These naturalistic techniques are especially suitable for studying language comprehension in populations such as children with autism spectrum disorder, for whom standard experimental tasks may not be appropriate. Brennan directs the Computational Neurolinguistics Lab at the University of Michigan.

Research topics

• Artificial Intelligence
• Computer Science
• Natural Language Processing
• Psychology
• Machine Learning
• Speech recognition
• Medicine
• Cognitive science
• Genetics
• Biology
• Neuroscience
• Programming language
• Linguistics
• Psychiatry

Selected publications

• Does prosody mark sarcasm early in an utterance? A production and perception study, including listeners who self-identified as being on the autism spectrum

  Journal of the International Phonetic Association · 2026-01-26

  articleOpen access

  Abstract This study examines the utterance-initial prosodic marking of sarcasm in English and its perception in listeners who did and listeners who did not self-identify as being on the autism spectrum. We ask (i) whether speakers use prosody to mark sarcasm in the early, ‘pre-target’ portion of an utterance (that is, in the portion before a ‘target’ word most closely associated with the sarcastic intent occurs), (ii) whether individuals vary in how they mark sarcasm, (iii) whether listeners reliably recognize sarcasm from pre-target prosody alone, and (iv) whether recognition accuracy varies by speaker or self-identified autistic traits. Eight American English speakers were recorded producing utterances presented in contexts conducive to either sarcasm or sincerity. Pre-target parts were presented in a two-alternative forced-choice experiment to individuals who either did (n=51) or did not (n=44) self-identify as being on the autism spectrum, and were examined for syllable duration and f0-related properties (maximum, minimum, range, and wiggliness). Results show that speakers distinguish sarcasm and sincerity in the pre-target region with duration being the most salient marker. Most listeners recognize sarcasm from pre-target fragments, but there is variation in how well each speaker is perceived. Whether the listener self-identified as being on the autism spectrum or not does not predict sarcasm and sincerity recognition accuracy. The results provide evidence that utterance-initial prosody contributes to sarcasm recognition, with the proviso that speaker and listener variation be taken into account.

  PublisherOA PDFDOI

• Beyond next-word prediction: hierarchical linguistic composition modulates LLM-brain alignment in time

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-05-16

  articleOpen accessSenior author

  Abstract The internal representations of large language models (LLMs) correlate, or “align”, with human neural activity during language comprehension. One view holds that this alignment reflects shared sensitivity to statistical patterns in LLMs and humans, while others hold that it reflects, at least in part, the emergence of shared linguistic representations in these systems. Here, we investigate whether hierarchical linguistic composition, a property believed to be fundamental to human language, modulates LLM-brain alignment. To this end, we manipulated syntax, compositional semantics, and associative semantics in English sentences that were presented to both an LLM and human participants during an electroencephalography (EEG) experiment. We matched linguistically manipulated stimuli in predictability, which allows us to tease apart alignment induced by linguistic structure from statistical factors. By comparing LLM-EEG alignment scores that were derived using a linear encoding model across predictability-matched conditions, we evaluate how linguistic manipulations modulate the alignment between human EEG reading data and contextual embeddings extracted word-by-word from the hidden layers of GPT2-XL. Three key patterns emerge: (1) increased alignment for word sequences with syntactic structure, (2) decreased alignment for sentences with compositional semantics, and (3) associative semantics does not modulate alignment. These observed linguistic modulations of LLM-EEG alignment take place above and beyond predictability. Our results indicate that associative semantics is encoded similarly by LLMs and the brain, as are at least some aspects of syntactic structure, while compositional semantics is more uniquely encoded in the human brain.

  PublisherOA PDFDOI

• XCOMPS: A Multilingual Benchmark of Conceptual Minimal Pairs

  ArXiv.org · 2025-02-27

  preprintOpen access

  We introduce XCOMPS in this work, a multilingual conceptual minimal pair dataset covering 17 languages. Using this dataset, we evaluate LLMs' multilingual conceptual understanding through metalinguistic prompting, direct probability measurement, and neurolinguistic probing. By comparing base, instruction-tuned, and knowledge-distilled models, we find that: 1) LLMs exhibit weaker conceptual understanding for low-resource languages, and accuracy varies across languages despite being tested on the same concept sets. 2) LLMs excel at distinguishing concept-property pairs that are visibly different but exhibit a marked performance drop when negative pairs share subtle semantic similarities. 3) Instruction tuning improves performance in concept understanding but does not enhance internal competence; knowledge distillation can enhance internal competence in conceptual understanding for low-resource languages with limited gains in explicit task performance. 4) More morphologically complex languages yield lower concept understanding scores and require deeper layers for conceptual reasoning.

  PublisherOA PDFDOI

• Decoding of lexical items and grammatical features in EEG: A cross-linguistic study

  Neuropsychologia · 2025-04-23

  articleSenior author
  PublisherDOI

• Decoding the Neural Dynamics of Headed Syntactic Structure Building

  Journal of Neuroscience · 2025-03-06 · 3 citations

  articleOpen accessSenior author

  The brain builds hierarchical phrases during language comprehension; however, the details and dynamics of the phrase-building process remain underspecified. This study directly probes whether the neural code of verb phrases involves reactivating the syntactic property of a key subcomponent (the "head" verb). To this end, we train a part-of-speech sliding-window verb/adverb decoder on EEG signals recorded while 30 participants read sentences in a controlled experiment. The decoder reaches above-chance performance that is spatiotemporally consistent and generalizes to unseen data across sentence positions. Applying the decoder to held-out data yields predicted activation levels for the verbal "head" of a verb phrase at a distant nonhead word (adverb); the critical adverb appeared either at the end of a verb phrase or at a sequentially and lexically matched position with no verb phrase boundary. There is stronger verb activation beginning at ∼600 milliseconds at the critical adverb when it appears at a verb phrase boundary; this effect is not modulated by the strength of conceptual association nor does it reflect word predictability. Time-locked analyses additionally reveal a negativity waveform component and increased beta-delta inter-trial phase coherence, both previously linked to linguistic composition, in a similar time window. With a novel application of neural decoding, our findings delineate the dynamics by which the brain encodes phrasal representations by, in part, reactivating the representation of key subcomponents. We thus establish a link between cognitive accounts of phrasal representations and electrophysiological dynamics.

  PublisherOA PDFDOI

• Prefix priming within and across languages in early and late bilinguals

  Bilingualism Language and Cognition · 2025-02-12

  articleOpen accessSenior author

  Abstract In contrast to ample evidence for cross-linguistic priming of monomorphemic words, cross-linguistic representation of affixes is not well understood. The current study examines cross-linguistic prefix priming among early and late English-Spanish bilinguals, focusing on prefixes that have the same form and meaning in the two languages. We first confirm robust prefix priming among English monolingual speakers (Experiment 1). We also observe prefix priming from first-language English to second-language Spanish but only for early bilinguals (Experiment 2). On the other hand, both early and late bilinguals do not show reliable prefix priming effects that are dissociated from orthographic or semantic priming from Spanish to English (Experiment 3) or from Spanish to Spanish (Experiment 4). The results suggest that for early bilinguals, the tested prefixes in their L1 and L2 have shared representations. Less reliable results for late bilinguals may reflect their weaker sensitivity to morphological structure in a second language.

  PublisherOA PDFDOI

• XCOMPS: A Multilingual Benchmark of Conceptual Minimal Pairs

  2025-01-01

  articleOpen access

  Linyang He, Ercong Nie, Sukru Samet Dindar, Arsalan Firoozi, Van Nguyen, Corentin Puffay, Riki Shimizu, Haotian Ye, Jonathan Brennan, Helmut Schmid, Hinrich Schuetze, Nima Mesgarani. Proceedings of the 7th Workshop on Research in Computational Linguistic Typology and Multilingual NLP. 2025.

  PublisherOA PDFDOI

• Neural processing of children’s theory of mind in a naturalistic story-listening paradigm

  Social Cognitive and Affective Neuroscience · 2025-01-01 · 1 citations

  articleOpen access

  Theory of mind (ToM) refers to our understanding of people's mental states. This ability develops in childhood and influences later social life. However, neuroimaging of ToM in young children often faces challenges in ecological validity and quality data collection. We developed and implemented an innovative naturalistic story-listening paradigm, which is child-friendly, engaging, and ecologically valid, to shed light on ToM neural mechanisms in childhood. Children (N = 51; age range = 6-12 years) listened to a chapter of Alice's Adventures in Wonderland during functional near-infrared spectroscopy neuroimaging. Methodologically, we showed the feasibility and utility of our paradigm, which successfully captured the neural mechanisms of ToM in young children. Substantively, our findings confirm and extend previous results by revealing the same ToM brain regions found in the adult and adolescent literature, including, specifically, the activations of the right temporoparietal junction. We further confirm that ToM processing has its own specialized neural profile, different from the left frontal and temporal activations found during language processing, with the language being independent of, but potentially supportive, of ToM deployment and development.

  PublisherOA PDFDOI

• Neural tracking of surprisal in Spanish-English bilingual children during naturalistic heritage language listening

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• Large Language Models as Neurolinguistic Subjects: Discrepancy between Performance and Competence

  2025-01-01

  articleOpen accessSenior author

  This study investigates the linguistic understanding of Large Language Models (LLMs) regarding signifier (form) and signified (meaning) by distinguishing two LLM assessment paradigms: psycholinguistic and neurolinguistic.Traditional psycholinguistic evaluations often reflect statistical rules that may not accurately represent LLMs' true linguistic competence.We introduce a neurolinguistic approach, utilizing a novel method that combines minimal pairs and diagnostic probing to analyze activation patterns across model layers.This method allows for a detailed examination of how LLMs represent form and meaning, and whether these representations are consistent across languages.We found: (1) Psycholinguistic and neurolinguistic methods reveal that language performance and competence are distinct; (2) Direct probability measurement may not accurately assess linguistic competence; (3) Instruction tuning won't change much competence but improve performance; (4) LLMs exhibit higher competence and performance in form compared to meaning.Additionally, we introduce new conceptual minimal pair datasets for Chinese (COMPS-ZH) and German (COMPS-DE), complementing existing English datasets. 1

  PublisherOA PDFDOI

Recent grants

• US-French Collaboration: Collaborative Research: Neuro-Computational Models of Natural Language

  NSF · $292k · 2016–2022

Frequent coauthors

• John Hale

  33 shared

• Susan M. Bowyer

  Michigan United

  17 shared

• Ioulia Kovelman

  University of Michigan–Ann Arbor

  16 shared

• Renee Lajiness‐O’Neill

  University of Michigan–Ann Arbor

  16 shared

• Neelima Wagley

  14 shared

• Shohini Bhattasali

  University of Toronto

  13 shared

• Christophe Pallier

  Cognitive Neuroimaging Lab

  11 shared

• Annette Richard

  11 shared

Education

• Ph.D., Linguistics

  New York University

  2010