I work on trustworthy NLG. More specifically, I build controlled text generation models and enhance the robustness of NLG systems, including NLG evaluation and fairness. More specifically:

Better Generation Evaluation: Dialect-Robust Evaluation [In Submission]; Evaluate the Helpfulness of Dialogue Responses [In Submission]; Robustness of Bias Evaluation [EMNLP 2022]
Generation Models: Paraphrase Generation [EMNLP 2021]; Controlled Pun Generation [EMNLP 2022]
Data Curation and Collection: Pun with Explanation [EMNLP 2022]; Investigate the Benefits of Free-form Rationales [EMNLP 2022]