who: Shrimai Prabhumoye (CMU))
when: Oct 3 (Thursday) 11:45a – 1:15p
where: Computer Science Building Rm 150
food: Athena’s Pizza
generous sponsor: ORACLE LABS
“Controlling style, content, and structure in natural language generation“
The 21st century is witnessing a major shift in the way people interact with technology and Natural Language Generation (NLG) is playing a central role. The increasing ubiquity of computing technologies has led to situation-aware applications that are required to produce naturalistic (informative, coherent, and appropriate) outputs. But situation-aware NLG is hard. Devices must not only generate natural sentences – already a challenging task – but include ever more complex data as inputs, and sound more natural every year. How can we expect machines to understand this and make the right choice for what to say? Solving a problem like this requires tackling at least three core tasks of NLG. The first is content determination: the information to be conveyed. Next comes discourse planning: the structure that a document will take on when it is articulated, given the preceding context and the rhetorical intent of the speaker. And finally, lexicalization: the particular words and phrases that convey the content of a sentence with a particular style or tone, given the appropriate structure in the discourse context.
Shrimai is a second-year Ph.D. student at Language Technologies, School of Computer Science, Carnegie Mellon University. She is advised by Prof. Alan W Black and Prof. Ruslan Salakhutdinov. She is broadly interested in natural language generation with special focus on style transfer and content transfer. During the course of her Ph.D., she has interned at Facebook AI Research and Microsoft Research.