To Prospective Applicants (Ph.D., Postdoctoral Researcher, and Master)

I'm interested in doing research in natural language processing (NLP). In particular, I use and train language models (LMs) to study tasks on summarization, reasoning, QA, narrative understanding, argument mining, and AI alignment. I'm also interested in applying AI techniques for interdisciplinary subjects, e.g., computational social science and educational technology. Specifically, our group has the following ongoing research projects:

• Factuality of LMs: We tackle the challenge of improving the factuality of LM generations, e.g., on tasks of summarization and QA. We are interested in developing efficient techniques that allow LMs to be better calibrated at prediction confidence and attribute their generations to verifiable sources.
• Reasoning: We aim to train smaller LMs to perform reasoning tasks by identifying their limitations and employing training data that can provide fine-grained and high-quality feedback.
• AI Alignment and Safety: We aim to make LMs more helpful and honest. We are interested in training LMs that can better reflect diverse values. We are also interested in measuring the potential bias content generated in real-world applications where LMs are employed as well as their influence on users.
• Summarization: We want to build summarization systems that can handle long inputs, including lengthy text and documents from multiple sources.
• Narrative Understanding: Narratives are arguably the most effective way to communicate cultures, values, and knowledge. We are interested in media narratives, which play a critical role in echoing, influencing, and reinforcing public opinion. We aim to design a computational framework with a unified narrative representation, grounded in social psychological theories, to facilitate narrative analysis at scale.
• Media Bias Analysis: News media play a vast role not just in supplying information, but in selecting, crafting, and biasing that information to achieve both nonpartisan and partisan goals. We aim to automate media bias detection from news articles, and quantify and further highlight biased content in order to promote the transparency of news production as well as enhance readers’ awareness of media bias.
• Argument Mining: Arguments play an important role in decision-making processes and persuasion. We are interested in understanding how people argue with and influence others, as well as form their own opinions on topics of interest.

If any of these sounds interesting to you, please check out more about my research and papers at this website. If you find your research agenda might be aligned with mine and are interested in working with me, please fill in this external contact form and (for Ph.D. applicants) apply to Computer Science and Engineering at University of Michigan and mention my name in your application. Once you've submitted the form, feel free to send me an email.