The paper has presented an approach for coping with indirect speech acts in an interpreter of natural language. A plan-based representation of speech acts has been adopted. A major advantage consists in the strict integration of the processes of recognition of speech acts and domain plans. In fact, the same representation underlies both processes; however, the speech act analysis is affected also by the presence of some linguistic (syntactic and semantic) features which have been discussed in the paper. These features are related to the politeness of the request. The next step of our work will be an assessment of the evaluation of the politeness level on the basis of the features detected in the sentences and of the intended impact of this level on the receiver of the message.
The speech-acts recognition algorithm is embedded in a plan-recognition system for information-seeking dialogues in a subset of the University domain. The system is implemented in Common Lisp and runs on workstations. The speech-acts analysis exploits the basic action-identification and upward-expansion procedures written for the recognition of the domain plans of the user of the system.