SigDial 2013
Department of Computer Science, University of Texas at El Paso
|
Abstract: Some things people say are more important, and some less so. Importance varies from moment to moment in spoken dialog, and contextual prosodic features and patterns signal this. A simple linear regression model over such features gave estimates that correlated well, 0.83, with human importance judgments. |
technical report with additional information
importance annotations (data downloads)