Recent advances in data-driven approaches have demonstrated appealing results in generating natural languages in applications like machine translation and summarization. However, when the generation tasks are open-ended and the content is under-specified, existing techniques struggle to generate coherent and creative sentences. This happens because the generation models are trained to capture the surface form (i.e. sequences of words), rather than the underlying semantics and discourse structures. Moreover, composing creative pieces such as puns, poems, and stories require deviating from the norm, whereas existing generation approaches seek to mimic the norm and thus are unlikely to lead to truly novel, creative composition. In this talk, I will present several of our recent works related to creative story and pun generation, emphasizing the importance of understanding and control for creative generation.
Nanyun Peng is a Research Assistant Professor of Computer Science at the University of Southern California, and a Research Lead at the Information Sciences Institute. She received a Ph.D. in Computer Science from Johns Hopkins University. Her research focuses on creative language generation, and the robustness and generalizability of natural language understanding, with works being featured in major tech media such as Wired and The Register. Nanyun received a Google Anita Borg Scholarship, a Fred Jelinek Fellowship, and multiple DARPA, IARPA, and NIH grants. She has backgrounds in Linguistics and Economics and held BAs in both.
Faculty Host: Yulia Tsvetkov
Language Technologies Institute