4.2.1 Processing dialogue with finite state systems

At the heart of the speaking elevator's dialogue processing capabilities, there's a FSA much like the ones we've looked at in the last lectures.

At the heart of the speaking elevator's dialogue processing capabilities, there's a FSA much like the ones we've looked at in the last lectures. We will now have a closer look at how finite state automata can be adapted for dialogue processing. When modeling a dialogue with an automaton, the key idea is to think of the states of that automaton as standing for different states of the dialogue, and of its edges as corresponding to things that happen in the dialogue.

So states are always defined with certain expectations as to what the system can have had as input, and what else can have happened at certain stages of the dialogue. For instance, the initial state of a dialogue automaton is naturally the beginning of the dialogue, and final states will normally correspond to possible end-points of a dialogue.

Edges of a dialogue automaton may for instance be defined to take the user input into account in order to decide what the new system state is, and to activate certain procedures, for instance output or movements. That way whenever a particular input is recognised by the system, a predefined behaviour and the way to realise it can be produced next.

So one adaption that has to be made to use FSA for dialogue processing is to allow various kinds of actions to be connected to edges, other than just consuming given input. Although this step is formally quite far-reaching, it is conceptually simple. For the practical purposes we're going to look at, the striking simplicity of FSAs remains untouched.