issues raised and comments about AIF from a presentation at Imperial.

[Jarred]

Hello AIF-ateers!

As some of you may or may not know, I am currently working on the ARGUGRID project and one of other participants of this project is Imperial. So wearing my official AIF badge and secret decoder ring, I presented the work of the AIF as presented in the recent journal paper. Luckily Sanjay Modgil and Matt South were also in attendance to ensure I towed the party line. :)

Below are some comments we received from the non-AIF people that were present at the talk. I say talk but happily it changed to a good long discussion with slides.

Here is the format of these issues and comments

Issues raised:
    "comments made by either myself, Sanjay or Matt"

The need for composition or aggregation of I-nodes:
      "F. Toni et.al.'s requirements for assumption based approach to argumentation can be met by the AIF. Another instance of a rule application node that might be required is one that "collects" arguments into a set of arguments which collectively (but not individually) conflict with another argument. Potential problem is:
A and B collectively, but not individually conflict with C
Two solutions:
1) I-node A and I-node B linked via a conflict application node CA-node1 to I-node C, where CA-node1 specifies that the content of all incoming I-nodes collectively
conflict with the supported I-node
2)  I-node A and I-node B linked via a rule application node RA-node1 to I-node with content A&B that is in turn linked via a conflict application node CA-node1 to I-node C
I prefer the second solution"

User Guide and notes:
    " It was pointed out that we really need to demonstrate that the existing approach allows interchange between different argumentation tools and an examination of the information loss involved. It was suggested that we need some guidelines for constructing these argument networks - e.g. allow duplicate arguments, anticipated use of s-nodes.  Currently we only do this with examples.  We need an application that will consume AIF that can then be used as a basis for these guidelines."
   
    "It was suggested that a user guide or a step by step process exemplify the translation process would be helpful. I think maybe it is too early for this as we first need more users to be able to guide. I think that instead we should get input from people such as you guys and Iyad who have had reification experience. In particular comments similar to Matt's earlier email about dangling S-nodes."

Interoperability (to what degree):

    "Is the semantics rich enough to enable a shared understanding between two agents with
different reps. of arguments. E.g., one agent who uses Toulmin schema, the other uses
first order tree structured arguments. The semantics of the support relations will provide for
a shared understanding, but to what extent. Are the current semantics sufficient?"

    "Ever have an argument with someone about religion? You have the interchange format as your natural language but you'll never be able to understand how they derive their conclusions. Is it not the same here? The AIF can illustrate where there are differences but are you ever going to convince a Toulminist that the true way is first ordered tree structured arguments? Without the AIF they can't even identify where the discrepancies exist. The AIF is just removing one variable from the potential variables from which confusion arises."

    "It was pointed out that in the opus graph, you might show a preference between the two rules that lead to the contradicting claims, instead of over the claims, as is shown.  We need to be clear about the reason that we show the latter and not the former - I think it's to do with the context - i.e. the query flies(opus).

Is it to do with context? Is it not just due to the vagaries of the user? This is a point that I think is lost on me. We just want to be able to interchange the representations of argument networks (ArgNet) right? Isn't whether a preference is drawn between one pair of nodes of the other a concern at a higher level. Let's say we have two visualisation tools. Tool A does not allow preferences between conclusions (for whatever reason), only premises and Tool B doesn't make that restriction. Tool B's designer draws the "opus" argumentation network as seen. Now, if the AIF has done its job, Tool A should be able to parse the network in that it sees there is a preference between the two conclusions. However because of its restrictions, it rejects the argNet of Tool B. At least it knows why its rejecting that argNet.I think the AIF should be broad enough to describe both."

AIF bias toward a Dungian approach: (This was an issue raised that I believe just needs a counter example or a confession.)

Jarred McGinnis

---

Now you can scan emails quickly with a reading pane. Get the new Yahoo! Mail.