Interaction and feedback in a spoken language system: A theoretical framework.

TOWARD A MORE ROBUST SPEECH INTERFACE Traditional approaches to improving the performance of spoken language systems have focused on improving the accuracy of the underlying speech recognition technology. The assumption is that if a system can translate exactly what the user said into text and then map this onto an application command, speech will be a successful input technique. With this approach, improving speech recognition accuracy requires asymptotic effort that ultimately is reflected in the cost of the technology. We argue that perfect performance by a speech recognizer is simply not possible, nor should it be the goal. There are limiting factors that are difficult or impossible to control, such as noise in the environment. Moreover, many words and phrases in English are homophones of other words and phrases, so in some situations, both human and machine listeners find them ambiguous. People frequently have trouble sticking to the grammar and vocabulary that a spoken language system expects. Finally, because people have many other demands on them while they are speaking such as planning what to say next and monitoring their listeners and the environment, they frequently do not produce the kind of fluent speech that a recognizer has been trained to process. Even though human utterances contain "speech errors," most human listeners hardly notice. The intrinsic limits on the wellformedness of utterances and on the accuracy of speech recognition technology suggest that to solve the problem, we must first redefine it. Let us start by considering how people handle these problems in conversation