User response to prompt wording styles in an automated banking service with a Wizard of Oz simulation of word-spotting

This paper describes a new Wizard of Oz simulation for spoken dialogue systems, which combines acoustic pattern matching and input from a human operator (the "wizard") to simulate the performance of a word-spotting recogniser at any desired level of accuracy. An experiment using this simulation to compare users' responses to three styles of prompt wording in a telephone banking service is described and the main findings are summarised. Implications for automated service design and for usability research are discussed.