We described a dialogue system in which domain-specific and domain-independent specifications are separated. We showed, as a prerequisite of a domain-independent dialogue strategy, how to determine the semantic content for clarification questions in a domain-independent way. We showed how the underlying dialogue strategy seeks to obtain information specific enough to select one among a set of possible tasks to fulfil and, subsequently, to obtain the information necessary to actually accomplish the task.
We demonstrated that, as a consequence of such design, it is possible to formulate discourse update and dialogue strategy in a generic way, taking advantage of informational differences in different representations. The resulting dialogue specification template is instantiated with domain models and domain-specific lists of actions in order to fulfil the tasks.
We chose to determine the speakers intention in a rather simple fashion, namely by selecting all possible goals that are compatible with the semantic content of the utterances so far. This comes at the expense of being able to deal with indirect speech acts only insomuch as the intended speech act may be inferred during semantic construction, a characteristic that stands in contrast to plan-based approaches. However, it is our hope that a more sophisticated inference procedure intended to determine the purpose of the utterance may overcome this problem by constructing semantic representations that are less closely related to the verbatim interpretation of the utterance. If and how this problem can be solved in a domain-independent way remains an open question for the time being.