David R. Traum and Carl F. Andersen Representations of Dialogue State for Domain and Task Independent Meta-Dialogue

[Full Text] [send contribution] [debate procedure] [copyright]

C1. Ingrid Zukerman (22.2.00):

I have the following questions about this paper.

One question is about these structures: how is the outcome of actions accounted for? i.e., there doesn't seem to be an explicit distinction between the user changing his/her mind about an action, and the outcome of the action not being satisfactory, thereby prompting a repair utterance.

The concept "dialog obligations" is being used without any explanation. While readers can get a sense of what this means (and they can obtain the original 1994 paper), this concept should be explained for the paper to be more accessible.

You mention that a simple dialogue management algorithm is being used. How does this algorithm differ from more complex alternatives? How adequate is it? What is gained by using this simple algorithm (compared to more complex options)?

I would also like to see which parts of this dialogue manager generalize to other types of interactions. Like many of the papers in this issue, this paper deals with short, rather simple interactions. This is a good start, as even those interactions uncover interesting phenomena that should be handled. However, eventually, more complex dialogues will be entered. It would be good to see some extrapolation from these techniques to such dialogues. In particular, I am coming from the argumentation angle, where there is no execution at present, so certain problems are obviated, but other aspects are more complicated, e.g., the comprehension of the user's intent, the changes in topic and endowing the user with the capability to counter-argue.