Abstract Matt Purver 13 October 2014
Ask Not What Semantics Can Do For Dialogue - Ask What Dialogue Can Do For Semantics
Semantic frameworks and analyses are traditionally judged by sentential properties: e.g. truth conditions, compositionality, entailment. A semantics for dialogue must be consistent not only with these intrinsic properties of sentences, but with extrinsic properties: their distribution, appropriateness or update effects in context. The bad news, of course, is that this means our analyses and frameworks have to do more, and fulfilling these requirements has been the aim of a great deal of productive and influential research. But the good news is that it also means that dialogue can act as a "meaning observatory", providing us with observable data on what things mean and how people process that meaning -- data which we can use both to inform our analyses and to learn computational models. This talk will look at a few ways in which we can use aspects of dialogue --- phenomena such as self- and other-repair, situation descriptions, the presence and distribution of appropriate and informative responses --- to help us choose, learn or improve models of meaning representation and processing.
This talk describes joint work with a number of colleagues, but particularly Julian Hough, Arash Eshghi and Jonathan Ginzburg.
This talk was given at SemDial 2014 in September 2014 (here under Programme > Invited talks)
Back to Cognitive Science Seminar Series