Hi! I'm a postdoc at Princeton working with Adele Goldberg and Tom Griffiths. Previously, I earned my Ph.D. in Psychology from the University of Wisconsin-Madison, advised by Gary Lupyan. I also visited Bill Thompson's lab at UC Berkeley in 2024. My research focuses on understanding how human and artificial minds flexibly draw similarities between seemingly unrelated things and make analogical inferences, using both empirical methods and computational models. Before my PhD in psychology, I earned a Master of Science degree in Statistics and a Bachelor of Arts in English.

Currently, several questions keep me awake at night: 

How do humans and AIs infer hidden connections between concepts or sensory phenomena?

Science is green, philosophy is purple; Bach is blue, and heavy metal is black. When asked to choose a color that best describes an academic subject or a music genre, people and AIs converge on the answers at rates far beyond chance. These cross-domain mappings permeate our lives, from metaphorical language to multisensory experience. What enables them? And are analogy, metaphor, synesthesia, and multisensory matching different faces of one general mechanism?

How is our ability to draw similarities affected by/learnable from language? 

Does language merely reflect pre-existing similarities, enabling us to describe and communicate them? Or does it fundamentally shape the way we perceive, structure, and form these similarities in our minds? To what extent does the similarity we perceive depend on our experience with and use of language? And how far could an intelligent system trained solely on language (like an LLM, or a person born blind learning about color entirely through words) go in drawing human-like similarities, without ever seeing, hearing, or touching the world? 

How much meaning could be recovered from structure?

Take a passage, replace every content word with nonsense, e.g., "The court orders the defendant to..." becomes "The blicket vorks the daxon to...", and large language models can still largely reconstruct the original (!) Scramble the nonce words, and models can still tell you it's a legal document. What makes language so robust to severe degradation? What statistical principles must a structure follow to be recoverable?

How do word meanings evolve? Why some words are more "metaphorical" than others?

Over time, words can acquire new meanings, lose old ones, or develop additional connotations. This evolution is often driven by the need to communicate new concepts or experiences and can result from metaphorical extensions where a word used in one context is applied to another, seemingly unrelated context. But what kind of words tend to be extended in meaning, and why can't similar words be extended in similar ways? For example, why do we only have "small talk" but not "little talk," "big talk," or "large talk"?

How can cross-domain mappings change how we think — from stereotypes to creativity?

Cross-domain mappings don't just describe the existing connections in the world; they can rewire the connections. Having people reason through counter-stereotypical mappings ("if females were an instrument, they'd be drums") measurably shifts social attitudes (as compared to reasoning through "if females were an instrument, they'd be harps"). And having AIs, lay people, and scientists reason through novel analogies (e.g., "design a car that's like an octopus," "explain memory as if it were a pickle") sparks new ideas and hypotheses, just as Darwin's borrowing from economics once sparked the theory of natural selection. Which mappings are we trapped in now, and which ones could set our imagination free?