Modeling a Dialog Task
Consider a form-filling task:
- Set of states:
String representing status of each slot: { 2, 0, 0, 1, 0}
where 0 = slot empty, 1 = slot filled, 2 = slot verified.
- Set of actions:
{Request_all, Request_i, Verify_all, Verify_i, Quit}
- Rewards/Costs:
- Fixed cost for every prompt/question to the user.
- Final cost for each unfilled or incorrectly filled slot.