4.5.7 Mixed initiative

Only the system has initiative.

Only the system has initiative, apart from the initialisation of the interaction. Still that can only be done by one of the pre-specified keywords. If something goes wrong, the system is reset and the dialogue has to start anew, with loss of all information. The system prompts the user for a specific kind of contribution and waits for a relevant utterance in the very limited context of the current state. The user cannot provide more information that could be helpful, or underspecify a parameter required by the system, as only the utterances defined in the recognition grammar can be recognised. Users cannot even choose to answer questions still in the pre-defined way, but in an order that better fits their purpose. In addition, the system cannot reason about anything that is not hard-coded. It cannot make suggestions based on context, for example.

There is no possibility of extending the elevator for mixed initiative other than defining states that allow the system to take further initiative itself. An extension to accept user initiative can also partially be handled in the same way. It is, however, more problematic as there is no upper limit as to what the user can say, or request. This is a general problem with allowing a high degree of user initiative - even sophisticated plan recognition techniques are faced with it.

A fairly easy extension would be for the system to inform the user of the options they have at each point, so that the dialogue does not come to a break. In a way, this is modeled now by the system telling the user to use the keyboard after two misrecognitions in a row.