| << Prev | - Up - | Next >> |
Grounding in the speaking elevator is implicit.
Grounding in the speaking elevator is implicit. Traversing the edge 4->5, the floor requested is looked up in the system's globl memory and is then mentioned in a standard confirmation utterance: ``I'm taking you to the x-th floor.'', which the user can then object to.
As we've discussed above, by using an extra global memory to store the floor requested we can keep the number of states and edges needed for grounding quite small: There is no need to add explicit states for every different target floor in order to produce the correct confirmation sentence. A nice extension would be to remember the floor that was misunderstood, so that it does not recognise it wrongly twice, that is, if the user has already stated that the original recognition was wrong.
Repair is also possible, although in a very limited fashion in the depicted dialogue model. It is defined as a possible transition from the state after the confirmation utterance has been produced (5->4). The elevator can only recognize correct floor requests over this transition. Everything else activates the Trip edge. That is, you can only correct it by giving alternative (and consistent) floor requests, not by saying for instance ``No!'' or ``Stop!''.
?- Question!
How would the speaking elevator-automaton have to be changed to allow for more natural corrections by the user?
| << Prev | - Up - | Next >> |