Adaptivity in speech-based multilingual e-mail client

Abstract

domain is particularly suitable area for adaptive speech applications since dialogues tend to be more openended than structured. This is because there is no single static task that the user tries to solve. This makes speech based e-mail application very different from, e.g. timetable systems that are commonly modeled as forms. The lack of fixed goal means that the interaction model for this kind of system is user-initiated. However, users must somehow be aware of what can be done with the system. To address this issue we have implemented different adaptive features that help the users in learning how such systems can be used. These features include integrated tutoring, output generation that takes user expertise into account and context adaptive universal commands. ABSTRACTIn speech interfaces users must be aware what can be done with the system -in other words, the system must provide information to help the users to know what to say. We have addressed this challenge by using adaptive techniques that support the learning and use of speech applications. We describe how adaptivity can be supported on architectural level, how user modeling can help to make the interface more adaptive, how integrated tutoring teaches the users to use speech applications and how context adaptive universal commands support cross domain learning. Specific issues concerning e-mail domain are discussed and examples from a working speech-based e-mail application are given.