My current research focuses on understanding the capabilities and limitations of large language models, making them more explainable through techniques like concept bottlenecks and faithful chain-of-thought reasoning, enabling symbolic reasoning by having models generate executable code, and building applications that use LLMs. I aim to gain insight into these powerful models while also developing methods to make them more interpretable, trustworthy, and beneficial for society. You can find my full research statement on my website.
My publication home is the Association for Computational Linguistics. To date, I have 33 ACL publications, 23 EMNLP publications, 22 NAACL papers, 6 TACL, 6 EACL and 4 Computational Linguistics articles. My publications have been cited more than 25,000 times. I have an h-index of 61.
My Artificial Intelligence course is one of the most popular courses at the University of Pennsylvania. This semester it has an enrollment of 600 students, and my course reviews are excellent despite the enormous size.
I designed a fun set of hands-on extra credit assignments that use programmable toy R2D2s to demonstrate the key ideas in my AI course. I bought 250 Sphero R2D2s - initially hoping to have one robot per student (the course outgrew it). We added a Rasberry Pis plus a custom built senor pack to the R2D2s, and have an API that lets the students to program the robots in Python.
I offer several courses through Penn Engineering’s online master’s degree program: AI, and NLP. I redesigned my NLP course during my sabbatical to cover LLMs and to record high-quality lecture videos for our online degree program.
I have also taught specalized courses on Crowdsourcing, Interactive Fiction, plus one-off seminars on Dialogue Systems, LLMs+programing languages, and an experimental Research Practicum course for masters and undergrad students that resulted in 4 ACL publications.
I won two student-nominated teaching awards during the pandemic, for the efforts that I put into keeping students engaged and the accomodations that I made for remote learning.
Placement of PhD students and postdocs
I have graduated 9 PhD students and supervised 7 postdocs. All of them have gone on to excellent positions. Several have gone on to faculty positions including
Daphne Ippolito – now an assistant professor at CMU LTI
Ellie Pavlick – now an assistant professor in computer science at Brown
Lara Martin – now an assistant professor in computer science at UMBC
Derry Wijaya - now an assistant professor in computer science at Boston University
Wei Xu - now an assistant professor in computer science at Georgia Tech
Anietie Andy - now an assistant professor in computer science at Howard University
Service to ACL
I have extensive service to the ACL community:
General Chair for ACL 2017
Program Co-Chair for EMNLP 2015
Chair of the NAACL Executive Board 2012-2013
Secretary-Treasurer for SIGDAT (EMNLP) 2016-2018
Sponsorship Director for ACL, 2020-present
Recently, I have begun making media appearances and engaging with policymakers about AI. Here are some highlights: