Yue Li (李悦)

PhD Candidate in Linguistics at Purdue University. I study how language is constructed and represented in both humans and machines.

I am a fifth-year PhD candidate in Linguistics @ Purdue University, affiliated with the Experimental Linguistics (ExLing) Lab and the Computation and Linguistic Meaning (CALM) Lab. Most of my work sits at the intersection of computational modeling and psycholinguistics, asking: How can humans and language models mutually inform our understanding of language?

Research Interests

  • Computational Psycholinguistics — I apply experimental and computational methods to study how humans and language models process and produce language.

  • LLM Interpretability — I probe the internal mechanisms of language models to understand how linguistic knowledge is encoded and used.

News 🎉

  • Apr 2026 Our recent work, LLM evaluation on semantically ambiguous structures (Semantic Capacity in Language Learners and LLMs), accepted at LREC 2026. Talk available soon.

  • Mar 2026 Three first-author poster presentations at HSP 2026 (Human Sentence Processing, MIT) on animacy & mechanistic interpretability, L2 ORC production, and relative clause extraposition.
    • Li, Y., Cong, Y., & Francis, E. J. Illuminating Human Language Production via Mechanistic Interpretability of LM Circuits: Animacy Effects on Sentence Structure Choice. [Poster]
    • Li, Y., Francis, E. J., & Mack, M. Similarity-Based Interference in L2 Production of English Object Relative Clauses. [Poster]
    • Li, Y., Francis, E. J., Khodadadi, G., Mack, M., Ok, S., Bahmanian, N., Fang, S., Michaelis, L. A., Sheu, V., & Weirick, J. Effects of Verb Type and Prior Context on the Production of Relative Clause Extraposition in English. [Poster]
  • Jan 2026 Two presentations at LSA 2026 (Linguistic Society of America) on animacy effects in ORC production/comprehension and quantifier scope in LLMs.

Methods

  • Computational & LLM Methods — mechanistic interpretability (activation patching, attention head ablation, representational decoding), LLM evaluation (BERT, GPT-2, LLaMA, ChatGLM, DistilGPT2), surprisal-based analysis, attention and entropy analysis, persona-based prompting via API, model fine-tuning and training, corpus-based analysis

  • Psycholinguistics & Behavioral Experiments — picture-based elicited production, self-paced reading, structural priming, truth-value judgment, working memory assessment (backward digit span), language proficiency testing (LexTALE)

Beyond Research

I am passionate about building tools that make science more accessible. Check out some of my contributions to Gorilla Open Science, including designing features and tools that (I hope to) make experimental workflows more intuitive and data processing easier for others.

Contact

📧 li4207@purdue.edu · Department of Linguistics, Purdue University, West Lafayette, IN 47907