“This summer, I interned with Deepak Ramachandran and Ellie Pavlick at Google Research on improving Question Answering (QA) systems, and will be continuing to work on this project part-time as a student researcher during the Fall semester.

We’ve been looking at unanswerable questions, which are very common in natural language but pose various challenges to existing QA systems. We are especially interested in questions that contain failed premises, such as ‘How old was Mark Zuckerberg when he was acquiring Google?’, which is not a sensible question because he never was acquiring Google. We have been looking at ways to automatically identify such failed premises in questions and thinking about what would be the best way for a QA system to handle such questions.

This is an exciting project for me because it’s at the intersection of linguistics (in particular semantics/pragmatics) and Natural Language Processing.

I was initially worried about the internship being remote, and I still think many things do remain challenging (e.g., it’s much harder and takes longer to ask quick questions), but my hosts and team have been very supportive and helpful.”

Najoung Kim is a fifth year PhD student working with Dr. Paul Smolensky and Dr. Kyle Rawlins. Her main areas of interest are computational semantics/pragmatics, and generalization in humans and machine learning models. She uses computational and experimental linguistic methodologies to explore these areas.