My research group studies data-driven, probabilistic systems for natural language processing and generation. We are interested in empirically accurate methods that, when possible, are rooted in data and guided by efficient inference. Our recent work has focused on the intersection of deep learning and structured prediction, with an application focus on text generation and document-level understanding.
Current Research Projects
- Interpretable and controllable natural language generation for data-to-text summary
- Deep generative models (e.g. VAEs) for probabilistic text processing and understanding.
- Hardware for speech, translation, and dialogue with the Architecture and Hardware Group.
- Visual understanding of neural language models in collaboration with VCG and IBM.