| Title | : | Style in Authoring – Understanding, Modeling & Generating text like a target author |
| Speaker | : | Dr. Balajivasan Srinivasan (Adobe Research Labs) |
| Details | : | Mon, 24 Feb, 2020 2:00 PM @ AM Turing Hall |
| Abstract: | : | Every enterprise has a unique content style that gets expressed across their communication across a nuanced spectrum of style. A copywriter authoring content for the Enterprise needs to understand these stylistic expressions – both the established and the subtle aspects of the enterprise’s content across different levels and reflect it appropriately in their authored text to maintain a consistent communication language. Not just an enterprise, this occurs in several places. In this talk, I will present our work in understanding different elements of style and utilizing it towards stylistically rewriting an input text to a target style. I will first cover our approach to holistically model style in text at the surface, lexical, syntactic and semantic levels. We provide methods from existing literature that quantize these elements in the text and extend them to find causally significant attributes that contribute towards a target style and objective. Finally, we leverage the power of recent transformer language models to automatically rewrite a text aligned to the target author profile. We achieve rewriting with the author's corpus only (without parallel corpora) and show comparable performances to parallel architecture. Bio: Balaji is a principal scientist at Adobe Research, India. His research interests span the areas of natural language generation and multimodal content modelling. He completed his PhD in computer science at the University of Maryland in September 2011. His thesis was on Scalable Learning Methods for Speaker Recognition and Geostatistics. Balaji completed his M.S. in electrical engineering from the University of Maryland in 2008 and B.E. in electrical engineering from Anna University (India) in 2006. His work experience includes research internships at National Institutes of Health, Bethesda, MD (May – Aug 2007) and Xerox Research Center, Webster, NY (May – Aug 2011). |
