The paper has described an architecture for plan recognition where problem solving plans play the role of a meta-level which controls the selection and execution of domain and linguistic plans. The paper shows that many dialogues can be analyzed in terms of our tripartite model, that covers examples similar to the ones reported in the literature, but with a better factorization of knowledge in different libraries.
Actually, more special cases can be accounted for by the model, which have not been discussed in depth since in these cases the behavior of the system depends also on the structure of the linguistic plans. For instance, in
A: ``Heuristics are used in problem solving"
B: ``I don't know what `heuristic' means".
the problem solving plans are applied at a lower level of linguistic analysis, on the basis of the knowledge that among the preconditions for interpreting an utterance there is the comprehension of the meaning of words.
So, the adoption of this model guarantees a good level of flexibility, but much work remains to be done on the contents of the problem solving library, especially as concerns the selection of the candidate actions and the comparison among alternative ways to reach a goal.