Grounding Extension 2
Modeling grounding (2nd).
The grounding method we've just seen covers (in a very trivial way) the case that the system doesn't recognize the user's input at all. Yet even if the system does recognize the user's input, grounding is important to make sure that the input has been understood correctly. For this, one solution is to add extra states to every state that expects input. Those states are responsible for asking for explicit or implicit confirmation based on what the proposed recognition is.
Adding states
The resulting dialogues would be as follows:
User: | Elevator. |
System: | Which floor do you want? |
User: | Professor Bill Barry. |
System: | Did you say professor Barry? |
User: | Yes |
System: | [Trip...] |
Or:
User: | Elevator. |
System: | Which floor do you want? |
User: | Professor Bill Barry. |
System: | Did you say professor Barry? |
User: | No |
System: | Which floor do you want? |
It is of course quite aggravating for the user to have to confirm what he has said every time he says something. That would amount to almost double the number of turns that humans need to hold a dialogue with the same outcome. If the number of the original questions that the system needs to ask is more than a couple, which is true for most domains, this becomes a significant problem.