An Empirical Study of a Deliberation Dialogue System

We present an empirical simulation-based study of the use of value-based argumentation in two-party deliberation dialogues, investigating the impact that argumentation can have on the quality of the outcome reached. Our simulation allows us to vary the number of values, actions and arguments that appear in the system; we investigate how the behaviour of the system changes as these parameters vary. This parameter sensitivity analysis tells us whether a value-based deliberation dialogue system may be useful for a particular real-world application. We measure the quality of the dialogue outcome (i.e. the action that the agents agree to) against a global view of whether that action would be agreeable to each agent if all of the agents' knowledge were taken into account. We compare the deliberation outcome with a simple consensus forming procedure (where no arguments are exchanged). Our results show that the deliberation dialogue system we present outperforms consensus forming.