Statistical Spoken Language Generation of Stylistic Variation

March 13, 2008
Marilyn Walker | Department of Computer Science at the University of Sheffield

Utterance generation for dialogue applications such as spoken dialogue systems and some types of computer games is typically highly handcrafted and domain dependent. This leads to problems of portability and scalability, or what has been called the ‘authoring bottleneck’ in the case of interactive drama systems. In this talk, I will describe a novel statistical spoken language generation framework, and our research investigating the performance of this approach across three successively more challenging application domains. I will focus in particular on our recent work on statistical generation techniques capable of generating a wide range of stylistic variants for an input communicative goal, based on learning the settings for linguistic parameters identified by research on personality psychology.

Speaker Details

Professor Marilyn Walker is the Head of the Cognitive Systems group in the department of Computer Science at the University of Sheffield, and the holder of a Royal Society Wolfson Research Merit Fellowship. Her research focuses on the design and evaluation of interactive multimodal systems, and methods for automatically adapting such systems to individual users or new domains or user groups. She received a Ph.D. in Computer and Information Science from the University of Pennsylvania in 1993. Before joining the faculty at Sheffield, she was a research scientist at Hewlett Packard Labs from 1984 to 1989, at Mitsubishi Electric Research Labs from 1993 to 1996, and at AT&T Bell-Labs from 1996 to 2003.www.dcs.shef.ac.uk/~walker/cv.html