Improving Dialogs with EMMA

The Extensible MultiModal Annotation specification (EMMA) is a standard of the World Wide Web Consortium for representing user input. EMMA is most commonly used for representing the interpretation of spoken input in dialogs. However, EMMA provides a great deal of additional information that can also be taken advantage of in speech applications. This presentation will discuss various ways in which this additional information can be used. As one example, EMMA includes information about the duration of an utterance. Given the number of words in an utterance and its duration, the speech rate can be determined. A slow speech rate might be an indication of a novice user, which in turn could allow the system to provide novice prompts or to slow its own speech rate. EMMA information can also be useful in log analysis. For example, if the average lag between the end of a specific system prompt and the beginning of the user’s response tends to be relatively long, this could be an indication that the prompt is complex or confusing. A number of other ways to exploit EMMA data will also be discussed.