Nanyun (Violet) Peng

Q&A with Professor Violet Peng

Nanyun (Violet) Peng is an assistant professor of computer science at the UCLA Samueli School of Engineering. She was recently highlighted during UCLA Samueli’s first Engineering in Action panel discussion around the concept of equity in artificial intelligence. Her research focuses on the robustness and generalizability of Natural Language Generation models, with applications to creative language generation and low-resource information extraction. Learn more about her work in our exclusive Q&A.

Discretionary funding has enabled me to support undergraduate, graduate and Ph.D. students in conducting research in these fields. Student involvement is critical to helping us with creating the datasets that are necessary to facilitate the research.

Q: What are some of the research projects that you are focusing on for this year?
A:
One of the research projects that I have been focusing on this year is natural language generation (NLG), where we teach artificial intelligence (AI) to generate human language so that it can have natural and engaging communication with humans. I have mainly been working on the generation of storytelling and figurative language, like puns, sarcasms, metaphors, etc.

Another area of research that I have been looking into is biases and fairness in natural language generation systems. This is where AI can learn implicit societal biases from the training data, and even amplify the biases to generate disturbing outputs and create crises in trust. We have done some pioneering work on analyzing, quantifying and reducing such biases in NLG systems.

Q: How do undergraduate and graduate students fit into these research projects?
A:
Natural language generation is an intuitive and fun research area that undergraduate students usually develop an immediate interest in. There are many tasks that undergraduate students can be involved in such as dataset creation and human evaluation of the system output. Undergraduate students can use these tasks to develop their experience and expertise in this research area.

For graduate students, NLG is a very challenging research area, as language is extremely intricate with many subtleties and nuances. Graduate students need to have rigorous training in mathematical foundations, in addition to a machine learning background and linguistic knowledge to do well in this field. Our students are currently helping to develop new models, algorithms and systems that can handle the challenging problems of generating creative natural language content, while avoiding the generation of results that can be offensive or harmful.

Q: How will your research be translated into new technologies?
A:
My research helps computers to have more natural and seamless communication with humans. This has many practical downstream applications such as conversational systems, interactive companions that are versed in things like storytelling and asking/answering questions, and personal assistance systems.

Q: How has discretionary funding through donor gifts enabled you to further your research at UCLA?
A:
Discretionary funding has enabled me to support undergraduate, graduate and Ph.D. students in conducting research in these fields. Student involvement is critical to helping us with creating the datasets that are necessary to facilitate the research. Donor support also allows us to conduct human evaluations of the system outputs, which is essential at the current stage of the research.