Support for (critical) questions

[christianvoigt]

Adding questions to Argdown was requested in #207 and previously in #129.

I am not fundamentally against this feature, but I have to think about the consequences that would have. For example, are questions simply statements connected to other statements via the "question" relation? From a theoretic standpoint that seems questionable (no pun intended). See here for the philosophy of questions.

If questions are neither statements nor arguments, what kind of additional relations would be needed to connect these new nodes to argument or statement nodes? Question and answer relations? How would answer relations look in the syntax? Can questions be attacked or supported? All these things are not yet clear to me, so I have hesitated to add this feature.

Another option would be to add "custom relation types" and "custom node types" to Argdown to give the users complete freedom over what additional relations and nodes they can use. This would fit Argdown's ideal of being "unopinionated", but at the moment I am not sure how easy it would be to implement that.

What you can do right now is transform each critical question of Walton's schemes into an "implicit" premise that you add to your premise-conclusion-structure.

So for example, if the critical question is "How credible is E as an expert source?", than you could add "E is credible as an expert source" as a premise to the argument. Other arguments could then attack or support this implicit assumption.

[mirkolenz]

Hi! Thank you for bringing up this topic. For our concrete use case, the following features are needed:

1. Annotate the critical questions as outlined by you.
2. Support schemes between the actual claims and premises that encode the type of inference.
3. Denote one node as being the major claim of the entire argument graph (in many cases, this will be the node at the top, but that is not the case in all circumstances).

The second point is quite important for me. Currently, Argdown only differentiates between supporting and attacking relationships. Using argumentation schemes, it becomes possible to specify the type of inference of supporting relationships. As an example, have a look at the following example that has been adapted from the Microtexts corpus (Peldszus and Stede 2016):

We currently rely on the Argument Interchange Format (AIF) in combination with OVA (http://ova.arg-tech.org/) to create such argument graphs. AIF differentiates between different nodes types: I-nodes (blue) for example contain the argumentative content, while S-nodes encode the scheme between I-nodes (or other S-nodes). Additionally, AIF supports other node types for rephrasing other I-nodes and so on. Supporting (at least some of) these features would enable whole new uses cases for Argdown.

The third point (specifying the major claim of an argument) is also not directly supported by OVA (we extended it to fit our use case), but could be accomplished by allowing users to specify metadata for each node.

I hope my explanations are understandable for you. For any questions, feel free to ask!

[christianvoigt]

This is a perfect opportunity to try out the new shareable Sandbox links!

Here is an example how you can produce a similar map in Argdown.

The colors can obviously be changed to your liking.

• use a #main tag for the main thesis, to assign a different color for it
• use tags for different kinds of arguments/inferences (#(positive consequences)), these can be visualized with different colors or even with icons (new feature!)
• since argument titles have to be unique, you have to add numbers to the titles ("Positive Consequences 1")

If you like, you can also add metadata, for example:

Alternative treatments should be subsidized in the same way as 
conventional treatments. {centralThesis: true}

I would personally prefer to put the text snippets into argument descriptions and then start to "reconstruct" the premise-conclusion structure based on these quotes. So that would lead to this visualization.

The logical reconstruction would then start like this (I only did this for the first argument, to give a quick impression).

• Note how we are now reformulating the original text snippet and adding an additional implicit premise.
• The relation is now automatically derived, because the conclusion of the argument simply is the main thesis
• I used tags to mark which premises are explicit in the source text, and which are implicit (you can also use metadata for that). This is visualized with different colors.
• I gave the argument a more descriptive title (while keeping the "positive consequences" tag to categorize the argument)
• I gave the main thesis the title "main thesis"

By changing the statement selection mode and related settings, you can choose, which statements should be visualized as nodes in the map. For example, you could decide that you only want to visualize the explicit premises as statement nodes (or maybe only the implicit premises) or only the statements that are attacked or supported by other arguments. This way you can keep the number of statements in the map low while still communicating the essential results of your reconstruction.

[mirkolenz]

Thank you so much for your detailed investigation and the detailed examples. I really appreciate your effort!

I plan to investigate a move from OVA to Argdown in early April and will see how far we can come using the methods you described here. I will keep you posted about the progress, as it would also include the ability to automatically transform graphs in the AIF format to Argdown (and vice versa), which may be of interest for others.

[christianvoigt]

Yes, let's keep in touch about that. AIF import/export would definitely be great! I talked about this feature with @ggbetz some time ago, I think there are some interesting problems to solve, but it should be doable. I will open another issue for that so that we can keep the different discussions separate.

[nathan-artist]

Above, @christianvoigt suggested:

use tags for different kinds of arguments/inferences (#(positive consequences)), these can be visualized with different colors or even with icons (new feature!)

and:

I would personally prefer to put the text snippets into argument descriptions and then start to "reconstruct" the premise-conclusion structure based on these quotes.

I have a comment on this that is relevant to the requirements of @mirkolenz:

Christian suggests: "put the text snippets into argument descriptions"; this could be restated as: Put the S-node into the I-node. This works because (if I am not mistaken) an S-node never exists without an I-node, so it is not necessary to display the S-node and I-node separately in the map. The argument mapping software Compendium and bCisive, do something similar, as I noted here: they display the edge/relation type as an icon within the node, not only as a line color (there is a "hybrid" icon when a node has multiple outgoing edges/relations). Icons work in Compendium and bCisive because they have only a small number of node/edge types. There are probably too many argumentation schemes to make an icon for each one.

(Edit: I removed an additional suggestion after I realized it was incorrect.)

[nathan-artist]

Above, @christianvoigt asked:

If questions are neither statements nor arguments, what kind of additional relations would be needed to connect these new nodes to argument or statement nodes? Question and answer relations? How would answer relations look in the syntax? Can questions be attacked or supported? All these things are not yet clear to me, so I have hesitated to add this feature.

As I noted here, there is literature about this. Some answers are suggested if we look at how IBIS implements questions. In IBIS:

1. There is a "question" node type.
2. There is a set of edge types only for questions, and these edges can connect out to any other node type.
3. The "position" node type (equivalent to a statement in Argdown, I guess) is the only other node type that can connect out to a question, and it connects via the "answers/responds to" edge type. Questions cannot be attacked or supported; they can only be questioned (by another question) or answered (by a position).

I am not suggesting that IBIS question rules are best for Argdown; I only provide them here as food for thought.

There are argumentation scheme templates for Compendium that use Compendium's question nodes to represent each scheme's critical questions. Part of the cognitive benefit of identifying an argumentation scheme is that each scheme comes with critical questions that serve as heuristics for adding more information to the map.

[mirkolenz]

Thanks for your suggestions, @nathan-artist!

Put the S-node into the I-node. This works because (if I am not mistaken) an S-node never exists without an I-node, so it is not necessary to display the S-node and I-node separately in the map.

Sadly this is not always the case. Consider the following example graph:

As you can see, two S-nodes are directly connected here. This is a "special case" in the AIF format and only defined for conflict S-nodes. This structure can be used to attack the inference between two statements.

This special case is the only reason why I am a bit hesitant to "merge" S-nodes and I-nodes.