Research Projects

1. Model Interpretability and Robustness

In addition to prediction accuracy, interpretability and robustness should also be considered as two intrinsic properties. The research in this lab focuses on

• providing rich forms of interpretations (e.g., Chen et al., 2020; Chen et al., 2021);
• improving model interpretability and robustness (e.g., Chen and Ji, 2020).

2. Text Generation and Conversation Modeling

Automatically generating texts is a fundamental challenge in natural language processing and connected to many real-world applications, such as conversation modeling and story generation. To address this challenge, we are working on

• building neural network models to incorporate contextual information (e.g., Clark et al., 2018);
• developing algorithms for effectively training neural network models (e.g., Du and Ji, 2019, Du and Ji, 2021); and
• reconsider the whole evaluation pipeline (e.g., Schoch et al. 2020).

3. Text Dataset Analysis

We are interested in various issues of data-related problems in data-driven NLP research

• learning with noisy labels (e.g., Chen et al. 2020);
• data selection to improve learning (e.g., Schoch et al. 2021)