Research Interests

State-of-the-art AI models have accomplished amazing things via scale: bigger models and bigger training corpora. Humans are able to learn language with orders of magnitude less training data and only a fraction of our brainpower. What lessons can computer scientists take from cognitive psychology, neuroscience, and linguistics to help us build more efficient models? And what can we in turn offer that might help advance those fields as well? As both AI and related fields grow and evolve, I think it is important to regularly come back to these questions, as the answers may change.

Current projects:

• Embodied language learning: Humans perceive and act in the world as we learn language. How can perception (like vision) and action help an artificial agent learn language?
• Remote webcam eye-tracking: Cognitive psychologists use eye tracking to understand human attention. Experiments usually require human participants to come to a lab with special hardware and highly controlled conditions. We are developing datasets and deep learning models to enable remote eye-tracking experiments using participants’ own computers and webcams.