Tutorial questions: argument-based dialogues
1. Consider the following simple dialogue system with two participating agents. Each agent has its own knowledge base (its beliefs), which is a finite set of arguments about beliefs.
Moves:
An agent can ASSERT an argument. An agent can make a CLOSE move.
Protocol:
• The agents take it in turn to make a move.
• The first agent to move ASSERTs an argument, which causes that argument to
become the TOPIC of the dialogue.
• For all other turns:
– An agent can ASSERT an argument as long as it not previously been asserted and it attacks some argument that has been previously asserted during the dialogue. An agent can always make a CLOSE move.
Termination rules:
A dialogue terminates if each agent makes a CLOSE move in succession (i.e. one agent makes a CLOSE move and then the other agent makes a CLOSE move immediately afterwards).
Dialogue outcome:
If the TOPIC is acceptable under the grounded semantics given the abstract argumen- tation framework that is constructed from all of the arguments that have been asserted during the dialogue, then the outcome of the dialogue is Acceptable(TOPIC).
If the TOPIC is not acceptable under the grounded semantics given the abstract argumen- tation framework that is constructed from all of the arguments that have been asserted during the dialogue, then the outcome of the dialogue is NotAcceptable(TOPIC).
Lets consider a few different strategies.
Strategy 1:
If it is permissible to ASSERT an argument from your beliefs, then ASSERT some such an argument;
otherwise make a CLOSE move.
Strategy 2: If it is permissible to ASSERT an argument from your beliefs and adding that argument to the abstract argumentation framework that is constructed from all off the arguments that have been asserted so far during the dialogue causes the label (i.e. IN, OUT, or UNDEC) of the TOPIC to change, then ASSERT such an argument; otherwise make a CLOSE move.
1
Strategy 3:
If it is permissible to ASSERT an argument from your beliefs and adding that argument to the abstract argumentation framework that is constructed from all off the arguments that have been asserted so far during the dialogue causes the label of the TOPIC to change to IN, then ASSERT such an argument;
otherwise make a CLOSE move.
Strategy 4:
If it is permissible to ASSERT an argument from your beliefs and adding that argument to the abstract argumentation framework that is constructed from all off the arguments that have been asserted so far during the dialogue causes the label of the TOPIC to change to OUT, then ASSERT such an argument;
otherwise make a CLOSE move.
(a) Assume both of the agents are using Strategy 1. Can you say what type of dialogue (from Walton and Krabbes dialogue typology) this would produce? If you assume an omniscient perspective where you can see inside the knowledgebase of each agent, can you say what the outcome of the dialogue would be?
Answer:
If both agents are using strategy 1, well end up with each agent asserting all of their arguments that are connected in some way to the topic argument. This leads to an (inefficient) inquiry dialogue, where the outcome of the dialogue is sound and complete in relation to reasoning with the union of the two agents knowledge bases. The outcome will be Acceptable(TOPIC) if and only if the TOPIC argument is acceptable under the grounded semantics applied to the union of the two agents knowledge bases. The outcome will be NotAcceptable(TOPIC) if and only if the TOPIC argument is not acceptable under the grounded semantics applied to the union of the two agents knowledge bases.
(b) Assume both of the agents are using Strategy 2. Can you say what type of dialogue (from Walton and Krabbes dialogue typology) this would produce? If you assume an omniscient perspective where you can see inside the knowledgebase of each agent, can you say what the outcome of the dialogue would be?
Answer:
If both agents are using strategy 2, the dialogue will continue until neither agent has any arguments that they could add to the argumentation framework constructed from the asserted arguments that would cause the status (IN/OUT/UNDECIDED) of the TOPIC argument to change under the grounded semantics. This leads to a (more efficient) inquiry dialogue, where the outcome of the dialogue is sound and complete in relation to reasoning with the union of the two agents knowledge bases. The outcome will be Acceptable(TOPIC) if and only if the TOPIC argument is acceptable under the grounded semantics applied to the union of the two agents knowledge bases. The outcome will be NotAcceptable(TOPIC) if and only if the TOPIC argument is not acceptable under the grounded semantics applied to the union of the two agents knowledge bases.
(c) Assume the agent who starts the dialogue was using Strategy 3 and the other agent is using Strategy 4. Can you say what type of dialogue (from Walton and Krabbes dialogue typology) this would produce?
Answer:
2
The agent using Strategy 3, will only assert arguments that, when added to the argumentation framework that is constructed from the arguments asserted during the dialogue, will cause the status (IN/OUT/UNDECIDED) of the TOPIC argument to change to IN under the grounded semantics. It is therefore trying to persuade the other agent that the TOPIC argument is acceptable.
The agent using Strategy 4, will only assert arguments that, when added to the argumentation framework that is constructed from the arguments asserted during the dialogue, will cause the status (IN/OUT/UNDECIDED) of the TOPIC argument to change to OUT under the grounded semantics. It is therefore trying to persuade the other agent that the TOPIC argument is not acceptable.
3