One strategy for making software more accessible is to develop a way to automatically create semantic parsers that convert domain-specific natural language into code that carries out the desired task. Doing this is challenging, and current methods require massive amounts of data with high-quality annotations by domain experts with programming expertise. I intend to eliminate or greatly reduce the need for such expensive data collection by developing a model that takes better advantage of the shared similarities across domains via a more flexible learned semantic representation space.

I’ve been participating in UROPs for over two years now, and that has prepared me to take on the exciting challenge of a yearlong research project through SuperUROP. I see SuperUROP as a great way to get a taste of and prepare for what graduate school might be like. I’m also very excited about the prospect of making a more substantial contribution to the field of natural language processing than I previously have through smaller UROP projects.