The 33rd IPP Symposium

Collecting, Err, Correcting Speech Errors

Mark Johnson, Professor of Cognitive and Linguistic Sciences and Computer Science, Brown University

Speech errors are pervasive in fluent spoken speech, and any speech understanding system will have to find some way of dealing with them. This talk shows how a Bayesian approach based on the "noisy channel model" can be used to identify and correct speech errors. One interesting apect of this approach is that a single sentence is analysed in two different ways by two different components (the source model and the channel model), and the structures they recover correspond to very different properties of the sentence.

Joint work with Eugene Charniak and Matt Lease (Brown CS)