About

Susan E. Brennan is a cognitive scientist who studies the psychology of language use, particularly how people adapt their speaking and understanding in conversational contexts. Her research includes examining how individuals modify their speech and comprehension strategies to facilitate effective communication, with a focus on interactive spoken dialogue. She is interested in how people adopt their speaking styles and understandings to their conversational partners and the variations that occur in speech. Her recent interests involve the neural circuits that support interactive communication, and she also studies the human use of technology, especially speech and language interfaces to computers. Brennan has developed computational models of caricature and has worked on projects such as the Walking Around Project, the Adaptive Spoken Dialogue Project, and the Shared Gaze Project. Her work emphasizes the importance of eye-tracking in language processing and communication, and she explores how gesture focus groups and other non-verbal cues contribute to effective interaction. Brennan's background includes previous research in psychology, cognitive and empirical approaches to discourse, and the study of language and gesture. She is a professor in the Department of Psychology at Stony Brook University, where she also holds a joint appointment in the Department of Linguistics.

Research topics

• Computer Science
• Psychology
• Political Science
• Public relations
• Linguistics
• Meteorology
• Human–computer interaction
• Geography
• Communication
• Cognitive psychology
• Social psychology

Selected publications

• NSF GRFP Data (publicly available, 2025)

  OSF Preprints (OSF Preprints) · 2026-02-20

  other1st authorCorresponding

  Analyses of publicly available data about the National Science Foundation's Graduate Research Fellowship Program

  Publisher

• Selection Bias for Fields of Study in NSF’s 2025 GRFP Awards

  2026-02-20

  articleOpen access1st authorCorresponding

  This paper identifies striking selection biases based on STEM field of study in the 2025 National Science Foundation Graduate Research Fellowship Program (GRFP) competition. Since 1952, GRFP has recruited “the best and the brightest” early-career individuals into STEM careers by providing support for them to attend graduate programs and to pursue research of their choice. The fellowship was founded to support talented people rather than particular research projects, with awardees going on to win Nobel prizes and lead innovation in U.S. science and engineering. Not only did 2025 see major cuts to the number of GRFP awards, but for the first time there was a finger on the scale: rather than recognizing talent across all STEM fields of basic research, the 2025 GRFP competition prioritized a narrow set of “priority areas.” We curated the publicly available data from 2025’s unprecedented two-step release of results (only 1,000 awards in April rather than the 2,300 projected, followed by another 500 in June) in order to identify and analyze bias in the selection process. The distribution of awards among fields of study appeared to depart significantly from previous years, when numbers of awardees within fields of study were driven primarily by application pressure. We found particularly strong bias in the second release of awards, which excluded the life sciences entirely—even though the life sciences had the highest number of applications. We discuss the risks of redefining NSF’s long-standing GRFP funding policy in order to favor narrow sectors of the STEM research workforce while disadvantaging other fields. Although the favored fields, artificial intelligence and quantum science, are national priorities, it is counterproductive for GRFP to prioritize them, as merit is independent of an early-career applicant’s field of study. We discuss the potential harm of such biases for seeding innovation in science as well as for broadening participation of women and others in basic research and the U.S. STEM workforce.

  PublisherDOI

• Impact of a Workshop with Visualization and Ethics Discussion on Awareness of Flood Risk and Intent to Protect

  Weather Climate and Society · 2025-09-08

  articleSenior author

  Abstract Disasters such as storm surge flooding pose an escalating threat to vulnerable coastal communities. While advances in weather models and forecasts are essential for informing protective actions, improving communication with the public for heightened storm preparedness is equally important. In this report, we provide a quantitative evaluation of lessons learned in an online workshop involving over 150 college students. The workshop employed simulated visuals of flooding and role-playing scenarios about a fictitious college campus. In addition, we used an “ethical matrix” (EM) tool to enable stakeholders to systematically represent, discuss, understand, and weigh trade-offs and perspectives pertaining to potential impacts of anticipated flooding from an impending hurricane. Building on a previous summary of the workshop (Colle et al.), this report presents quantitative and qualitative results from hypotheses about the workshop’s effects on feelings of worry, intent to take protective action, and increased awareness of others’ situations and concerns. These findings provide insights for refining hypotheses and designs for workshops with communities vulnerable to storm surge flooding. Significance Statement Traditionally, flood warnings rely on forecasts, storm path diagrams, and evacuation orders. Additional communication strategies are needed to help people grasp individual and communal impacts. This paper presents quantitative evidence that communication strategies that help people “feel” the likely impact of flooding not only on themselves but also on those around them are associated with intent to protect oneself and others, including vulnerable populations. We also provide a toolkit of supplemental material for replicating some or all of the approaches tested.

  PublisherDOI

• LVLMs are Bad at Overhearing Human Referential Communication

  ArXiv.org · 2025-09-15

  preprintOpen accessSenior author

  During spontaneous conversations, speakers collaborate on novel referring expressions, which they can then re-use in subsequent conversations. Understanding such referring expressions is an important ability for an embodied agent, so that it can carry out tasks in the real world. This requires integrating and understanding language, vision, and conversational interaction. We study the capabilities of seven state-of-the-art Large Vision Language Models (LVLMs) as overhearers to a corpus of spontaneous conversations between pairs of human discourse participants engaged in a collaborative object-matching task. We find that such a task remains challenging for current LVLMs and they all fail to show a consistent performance improvement as they overhear more conversations from the same discourse participants repeating the same task for multiple rounds. We release our corpus and code for reproducibility and to facilitate future research.

  PublisherOA PDFDOI

• Training LLMs to Recognize Hedges in Dialogues about Roadrunner Cartoons

  2024-01-01 · 1 citations

  articleOpen accessSenior author
  PublisherOA PDFDOI

• Training LLMs to Recognize Hedges in Spontaneous Narratives

  arXiv (Cornell University) · 2024-08-06

  preprintOpen accessSenior author

  Hedges allow speakers to mark utterances as provisional, whether to signal non-prototypicality or "fuzziness", to indicate a lack of commitment to an utterance, to attribute responsibility for a statement to someone else, to invite input from a partner, or to soften critical feedback in the service of face-management needs. Here we focus on hedges in an experimentally parameterized corpus of 63 Roadrunner cartoon narratives spontaneously produced from memory by 21 speakers for co-present addressees, transcribed to text (Galati and Brennan, 2010). We created a gold standard of hedges annotated by human coders (the Roadrunner-Hedge corpus) and compared three LLM-based approaches for hedge detection: fine-tuning BERT, and zero and few-shot prompting with GPT-4o and LLaMA-3. The best-performing approach was a fine-tuned BERT model, followed by few-shot GPT-4o. After an error analysis on the top performing approaches, we used an LLM-in-the-Loop approach to improve the gold standard coding, as well as to highlight cases in which hedges are ambiguous in linguistically interesting ways that will guide future research. This is the first step in our research program to train LLMs to interpret and generate collateral signals appropriately and meaningfully in conversation.

  PublisherOA PDFDOI

• Risk Perception and Preparation for Storm Surge Flooding: A Virtual Workshop with Visualization and Stakeholder Interaction

  Bulletin of the American Meteorological Society · 2023 · 5 citations

  Senior authorCorresponding
  • Computer Science
  • Political Science
  • Psychology

  Abstract Many factors shape public perceptions of extreme weather risk; understanding these factors is important to encourage preparedness. This article describes a novel workshop designed to encourage individual and community decision-making about predicted storm surge flooding. Over 160 U.S. college students participated in this 4-h experience. Distinctive features included 1) two kinds of visualizations, standard weather forecasting graphics versus 3D computer graphics visualization; 2) narrative about a fictitious storm, role-play, and guided discussion of participants’ concerns; and 3) use of an “ethical matrix,” a collective decision-making tool that elicits diverse perspectives based on the lived experiences of diverse stakeholders. Participants experienced a narrative about a hurricane with potential for devastating storm surge flooding on a fictitious coastal college campus. They answered survey questions before, at key points during, and after the narrative, interspersed with forecasts leading to predicted storm landfall. During facilitated breakout groups, participants role-played characters and filled out an ethical matrix. Discussing the matrix encouraged consideration of circumstances impacting evacuation decisions. Participants’ comments suggest several components may have influenced perceptions of personal risk, risks to others, the importance of monitoring weather, and preparing for emergencies. Surprisingly, no differences between the standard forecast graphics versus the immersive, hyperlocal visualizations were detected. Overall, participants’ comments indicate the workshop increased appreciation of others’ evacuation and preparation challenges.

  PublisherOA PDFDOI

• Age-related changes in patients with upper limb thalidomide embryopathy in the United Kingdom

  Journal of Hand Surgery (European Volume) · 2023-04-06 · 6 citations

  articleOpen access

  We report the long-term upper limb disability, health-related quality of life (HRQoL), functional impairment, self-perception of appearance and prevalence of neuropathic pain in patients with upper limb thalidomide embryopathy in the United Kingdom. One-hundred and twenty-seven patients responded to our electronic questionnaire. Mean Quick Version of the Disabilities of Arm, Shoulder, and Hand score was 54.3 (SD 22.6). Median EuroQoL 5-Dimension 5-Likert index, Work and Social Adjustment Scale, Derriford Appearance Scale 24 and Neuropathic Pain Scale were 0.6 (IQR 0.4 to 0.7), 15.5 (IQR 8.0 to 23.5), 35.5 (IQR 28.0 to 50.5), and −0.8 (IQR −1.4 to 0.8), respectively. Thirty-three patients (26%) reported neuropathic pain. Finger changes associated with radial longitudinal deficiency were an independent predictor of more severe upper limb disability. Eighty-nine patients (70%) reported deteriorating HRQoL with increasing age. Patients with upper limb thalidomide embryopathy experience age-related worsening of symptoms and function, highlighting the need for ongoing specialist care and support. Level of evidence: IV

  PublisherOA PDFDOI

• What is retained about common ground? Distinct effects of linguistic and visual co-presence

  Cognition · 2021 · 21 citations

  Senior authorCorresponding
  • Computer Science
  • Psychology
  • Linguistics
  PublisherDOI

• What is retained about common ground? Distinct effects of linguistic and visual co-presence

  2020-02-20 · 2 citations

  preprintOpen accessSenior author

  Common ground can be mutually established between conversational partners in several ways. We examined whether the modality (visual or linguistic) with which speakers share information with their conversational partners is encoded in memory in a way that affects subsequent references addressed to a particular partner. In 32 triads, directors arranged a set of tangram cards with one matcher and then with another, but in different modalities, sharing some cards only linguistically (by describing cards the matcher couldn't see), some only visually (by silently showing them), some both linguistically and visually, and others not at all. Then directors arranged the cards again in separate rounds with each matcher. The modality with which they previously established common ground about a particular card with a particular matcher (e.g., linguistically with one partner and visually with the other) affected subsequent referring: References to cards previously shared only visually included more idea units, words, and reconceptualizations than those shared only linguistically, which in turn included more idea units, words, and reconceptualizations than those shared both linguistically and visually. Moreover, speakers were able to tailor references to the same card appropriately to the distinct modality shared with each addressee. Such gradient, partner-specific adaptation during re-referring suggests that memory encodes rich-enough representations of multimodal shared experiences to effectively cue relevant constraints about the perceptual conditions under which speakers and addressees establish common ground.

  PublisherDOI

Recent grants

• ITR: Adaptive Spoken Dialog with Human and Computer Partners

  NSF · $1.5M · 2003–2008

Frequent coauthors

• Jeanne Glidden Prickett

  Holtec International (United States)

  13 shared

• John F. Knutson

  University of Iowa

  13 shared

• Pena Lubrica

  Holtec International (United States)

  13 shared

• Lenore Holte

  University of Iowa

  13 shared

• Claudia L. Knutson

  Holtec International (United States)

  13 shared

• Don C. Van Dyke

  University of Iowa Hospitals and Clinics

  13 shared

• Randall J. Olson

  13 shared

• Justina O. Ohaeri

  State University of New York

  11 shared

Education

• Ph.D., Computer Science

  University of California, Berkeley

  1990

• M.S., Computer Science

  University of California, Berkeley

  1985

• B.S., Computer Science

  University of California, Berkeley

  1983

Awards & honors

• Dean's Award for Excellence in Service to Graduate Education…
• Chancellor's Research Recognition Award, The Research Founda…
• NSF Graduate Research Fellowship, Stanford University (1986-…
• American Can Company Full Scholarship, Cornell University (1…