Conversational collaboration in user-initiated interruption and cancellation requests

User-initiated interruption and cancellation (UC) requests are two important means by which a user regains control of the dialogue flow in a human-computer dialogue, but, their definition and implementation are left to ad-hoc methods. This dissertation uses a conversational view of interaction to define how these requests should be designed and implemented. A negotiated process consisting of two parts: the structure of the dialogue and the communication of system state, is used to accept these requests. This dissertation describes a dialogue structure model and a feedback model. The dialogue structure model is a hierarchical model based on conversational principles. The feedback model identifies different system states of feedback each with different user communication expectations. The dialogue structure model is validated by the design and implementation of a runtime software architecture that allows the building of dialogues by hierarchical composition of modular dialogue units (MDU). Composition of the MDUs at runtime forms a tree that defines the context for handling I/C requests. This allows MDUs to be defined locally and still make use of the current context at invocation. This architecture supports concurrent, asynchronous dialogues. The implementation shows that dialogue structure can be defined independently of interaction techniques without compromising modularity, while encouraging the reuse of portions of the dialogue structure. This architecture allows explicit representation of control so that many different negotiations for dialogue control can be expressed. To validate the feedback model, an experiment tested user's expectations when faced with different feedback conditions. The results show that users have different communication expectations based on the signals used for feedback states. Communicating state information to the user for all the states produced behavior that was consistent with each of the states described in the model. Not communicating the states individually produced a higher number of incomplete actions. The results also found sufficient evidence to support the collaborative nature of feedback by showing that when appropriate feedback is not provided, users tend to spend more effort trying to ascertain the state of the system. In general, the research shows the applicability of conversational theories to the study of human-computer dialogues.