Spoken Language Interaction in a Goal-Directed Task

To study the spoken language interface in the context of  a complex
problem-solving task,  a  group  of  users  were asked  to  perform  a
spreadsheet task, alternating voice and keyboard input.  A total of 40
tasks were performed by  each participant, the first thirty in  a group
(over several days),  the  remaining ones a month later.  The voice
spreadsheet program used in this study was extensively instrumented
to provide detailed information about the components of  the inter-
action. These data,  as well  as analysis of  the  participants's  ut-
terances and  recognizer  output, provide a fairly detailed picture of
spoken language interaction.

Although task  completion  by  voice took  longer than  by  keyboard,
analysis shows that users would be  able to perform the spreadsheet
task faster  by  voice,  if  two key  criteria could be  met  recognition
occurs in real-time, and the error rate is sufficiently low.  This initial
experience with a spoken language system also allows us to identify
several metrics,  beyond  those traditionally associated  with  speech
recognition, that can be used to characterize system performance.