Applications of Spoken Language Technology and Systems Mike Phillips
Challenge of Dialog Systems
Usability of Dialog Systems Usability Flexibility Like a Human Structured Dialog Add NL/Dialog ??
User Expectations User expectations of what they can say OK for fixed choices (“you can say…”) OK for constrained input (“please say your account number”) But, if we expand it… How do users know what they can say? Will the system understand them? What did it understand anyway? User expectations of dialog OK for structured dialog OK for single turn But, if we expand it… How do users know what they can do? Does it really know what they want? There seems to be an error, now what?
Avoiding the Abyss What the user can say 1.Keep it simple 2.Allow everything Dialog strategy 1.Keep it simple 2.Be as good as a human
Simple Dialog
Telephone applications Depends on application Personal Computers Who needs it? Mobile Devices Mobile devices and networks becoming more capable Other input modalities not so great Personal device Interface can learn about user User can learn about Interface
What About Telephone Applications? Phone network switch IVR Systems Agents
Virtual Agent Telephony Applications User experiences what appears to be fully automated experience But, there are no constraints on what they can say Actual work is mix of automation and agents Usage data can be used to train SLMs, automated dialog managers, etc. Increase % automation over time with same user experience
Predictions Call centers will continue adoption of speech-based systems Incremental improvement of user experience Increased used of more flexible dialog strategies Better mix of automation and humans Mobile devices will be area of fastest growth Data-based services just taking off Broad agreement that there is a user interface problem All players scrambling for ownership of customer experience