Liz Black
A generative framework for argumentation-based inquiry dialogues
Abstract
My PhD (2003-2006) focuses on argumentation-based communication between agents.
Communication is a key issue in multi-agent systems as it allows agents
to share their knowledge and coordinate their actions in order to
jointly reach their goals. The use of argumentation-based communication
allows a rich flow of data, with agents able to state the reasons behind
their position and to change their position in light of new information.
It also allows agents to deal with the uncertain and often inconsistent
information that is common in the real world and, in particular, common
in the medical domain. I have adopted a standard approach to specifying
argumentation-based agent communication in the form of dialogue games.
I take as a starting point the argumentation system proposed by Garcia
and Simari [1], that was intended for internal reasoning, and adapt this
system to allow it to also deal with inter-agent argumentation. A key
novel feature of this work is a clear delineation of the internal and
the external argumentation that occurs within the framework, that is
made possible by the definition of both internal and external types of
argument tree. I define a general framework for agent dialogues that
allows nesting of dialogues of different types, and give two specific
protocols for two different dialogue types. Both these dialogue types
are forms of inquiry dialogues, a type of dialogue that is particularly
useful for the medical domain. One is at the level of finding a
justification for a single argument from a set of arguments, and the
other is at the level of finding a set of facts (the support) that will
act as a justification for a particular claim (the conclusion). I also
provide strategies for use with these dialogue protocols, meaning this
theory is capable of not only modelling dialogues but also of generating
them.
This work is, as far as I know, the only work that provides a mechanism
for automatically generating inquiry dialogues, by defining strategies
that allow an agent to select exactly one move to make at any point in a
dialogue. This allows further investigation into the outcome of these
dialogues. I propose a benchmark against which I compare my dialogues
and use this to define soundness and completeness properties for inquiry
dialogues. I show that these properties hold for all well-formed inquiry
dialogues in my system.
Reference
[1] Alejandro J. Garcia and Guillermo R. Simari. Defeasible logic programming an argumentative approach. Theory and Practice of Logic Programming, 4(1–2):95–138, 2004.
cancer.org.uk