The best way to define a conversational style dialogue is to give you a
description of one of the earliest examples of this kind of interface style.
Schmandt (1987) described an automated telephone service called the
Phone Slave which was an answering machine service that allowed users
to retrieve stored messages. The system worked on the basis of asking the
caller a series of questions like ‘who’s calling please?’ and ‘what’s this
in reference to?’. The Phone Slave has no understanding of the content of
the messages left by any of the callers. The users’ responses were stored
digitally by the service and could be accessed in sequential order by
the system’s owner. For example, the owner could ask the service ‘who
left messages?’ and the system would respond by playing back all the
responses to its own query ‘who’s calling please?’.
Schmandt found that the interface was very effective in eliciting appro-
priate voice message components from callers, attributing the success of
the conversational style to the apparent high quality of the spoken prompts
provided by the system. To take a message requires co-operative behav-
iour and there is no reason to think that callers will not follow conventional
rules. By asking a series of questions, as opposed to a message such as
‘leave your message after the beep’, the system makes it easier for the user
to leave a more complete message as a series of components. In doing
so, the system maintains its ability to control the conversation and protects
the system’s limited ‘intelligence’ from being exposed.
Other examples of this type of approach can be seen with the Philips
TABA train timetable information service (Souvignier et al., 2000) and
the SpeechWorks air travel reservation system (Barnard et al., 1999).
Both these systems rely on system initiated dialogue to keep the speech
recognition accuracy levels high.
However, there are still some problems to be overcome for natural
language conversational interfaces, such as increasing the number of
words that can successfully be recognised as a part of the service’s voca-
bulary, and effective recovery methods when misrecognition occurs. It’s
important that the service gets the user back on track and that the user
has confidence in the system’s ability to understand what they are saying.
Blattner et al. (1989) define earcons as being abstract musical tones that
can be combined to produce sound messages to represent parts of
86 Understanding Mobile Human–Computer Interaction
H6352-Ch05.qxd 7/18/05 12:40 PM Page 86