I’m a third-year PhD candidate at Mila and McGill University, advised by Jackie Cheung in the McGill NLP group.
I study how language models take part in the social processes by which people make meaning. Speakers choose among several valid ways of conceptualizing the same thing (e.g. a riot vs. a protest). In humans this choice is negotiated between partners over repeated interaction and can stabilize into conventions across a community. How does this process change when LLMs are conversation partners?
I am pursuing two directions:
-
Lexical choice in LLMs. What drives how models select a conceptualization in LLM-LLM and LLM-human interaction? Conceptualization affects task success — my Semantic Bandits paper showed that how an environment is labeled changes how LLMs explore it. Outside of toy settings there is rarely one objectively correct labeling.
-
The role of LLMs in social meaning-making. LLMs offer a new way to interface with cultural knowledge, potentially influencing how we articulate and make sense of our own experience. How can lessons from earlier shifts in cultural technology (writing, print, search) inform the role we envision for LLMs within the larger process of meaning-making? This continues work I pursued at Building 21 at McGill.
Feel free to shoot me an email if these topics align with your interests! I am always happy to chat.
Before Mila, I worked with Scott Sanner and Chris McIntosh at the University of Toronto and the Vector Institute on conversational recommendation and multimodal medical models. I did my undergrad in Computer Science and Statistics at the University of Waterloo, completing research internships with Bo Wang and Doug Lee at the University of Toronto.
Outside of my research, I play guitar and bass, read, and play volleyball. I’m currently very into folk rock like Fairport Convention and Gillian Welch / David Rawlings.
News
- Oct 2026Presenting Semantic Bandits at COLM ’26.
- Sep 2026Passed my comprehensive exam. I’m now officially a PhD candidate!
- May 2026Attending ACM FAccT ’26.
- Oct 2025Presenting (RSA)2 as a workshop paper at COLM ’25.
- Aug 2025Presenting (RSA)2, a rhetorical-strategy-aware Rational Speech Act framework, with Cesare Spinoso-Di Piano at ACL ’25.
Publications
* equal contribution. Also on Google Scholar.
- In-Context Exploration-Exploitation is Biased by Semantic Priors
- (RSA)2: A Rhetorical-Strategy-Aware Rational Speech Act Framework for Figurative Language Understanding
- Bayesian Optimization with LLM-Based Acquisition Functions for Natural Language Preference Elicitation
- MEDBind: Unifying Language and Multimodal Medical Data Embeddings
Other
Earlier work in clinical machine learning and statistics.
- The Performance of Different Propensity Score Methods for Estimating the Effects of Multiple Treatments or Exposures: A Neutral Comparison Study
- Comparison of Machine Learning and Conventional Statistical Modeling for Predicting Readmission Following Acute Heart Failure Hospitalization
- Comparison of Machine Learning and the Regression-Based EHMRG Model for Predicting Early Mortality in Acute Heart Failure