Our approach to tool-building will be incremental. We will try to design a succession of components so that they are both useful in their own right and convenient to adopt. To that aim, we will first focus on a decentralized approach, which we believe is more likely to be adopted at first, as it does not depend on tool adoption by other parties in the discussion. This runs the risk that the first stages of the tools may favour debate over deliberation; as we introduce server-side logic, on the other hand, it will focus on useful coordination mechanisms which should tip the scale back towards constructive deliberation.
Step 1: Ontology and microformat
The first component will implement a model for explicit identification of argument structure in text. Following ClaiMaker, we intend to model the discussion more than its object. Following IBIS [http://kmi.open.ac.uk/publications/index.cfm?trnumber=kmi-05-18]Simon J. Buckingham Shum, Albert M. Selvin, Maarten Sierhuis, Jeffrey Conklin, Charles B. Haley, Bashar Nuseibeh. «Rationale Management in Software Engineering», Hypermedia Support for Argumentation-Based Rationale: 15 Years on from gIBIS and QOC, 111–132. Computer Science Editorial. Springer-Verlag, 2006. and SDSS [http://www.hicss.hawaii.edu/HICSS_35/HICSSpapers/PDFdocuments/CLCSC03.pdf]Murray Turoff, Starr Roxanne Hiltz, Hee-Kyung Cho, Zheng Li, Yuanqiong Wang. «Social Decision Support Systems (SDSS)». IEEE, , Proceedings of the 35th Hawaii International Conference on System Sciences, 2002., we focus on discursive moves that move towards a decision.
- Like ClaiMaker [http://kmi.open.ac.uk/publications/pdf/KMI-TR-05-9.pdf]Victoria Uren, Simon J. Buckingham Shum, Michelle Bachler, Gangmin Li. «Sensemaking Tools for Understanding Research Literatures: Design, Implementation and User Evaluation». Int. Jnl. Human-Computer Studies, Sept 2005. and ClaimSpotter [http://kmi.open.ac.uk/publications/pdf/kmi-04-29.pdf]Bertrand Sereno, Simon J. Buckingham Shum, Enrico Motta. «ClaimSpotter: An Environment to Support Sensemaking with Knowledge Triples». Proceedings of the Intelligent User Interfaces Conference, San Diego, CA, USA, dec 2004. kmi. , we distinguish free text from claims identified within them. The ontology of relations might also follow their work.
- Primary claims (identified within the source text) are distinct from secondary claims (about the argument in the source text.)
- Following Nelson [/]Theodor Holm Nelson. Literary Machines. Mindful Press, Sausalito, California, 1980., we view transclusion as an essential to text references.
Technical aspects:
- Identification of primary claims in a document can be expressed using microformat within that text or in another document.
- This microformat would refer to the source text using XLink [http://www.w3.org/TR/xlink/]/XPointer [http://www.w3.org/TR/xptr-framework/]. We would extend this to allow use of robust text references [http://www.eecs.berkeley.edu/Pubs/TechRpts/2000/CSD-00-1091.pdf]T. Phelps R. Wilensky. «Robust Hyperlinks Cost Just Five Words Each». UCB//CSD-00-1091, January 2000. , as defined by R. Phelps: i.e. references should not be affected by simple edition of the target text.
- We intend to use RDFa [http://www.w3.org/TR/xhtml-rdfa-primer/]RDFa Primer 1.0: Embedding RDF in XHTML within the microformat to express the semantic content of the claim
- As argumentation structure is itself a matter of debate, our ontology representation should allow for extension points, though the extensibility mechanisms should be left for a later stage.
Step 2: In-browser representation of claims
We intend to design a visual representation of claims identified within documents, including those that are identified from outside the base document. Because text ranges will overlap, we are thinking of using markers in the margin (where rollover would highlight the ranges as well as show the claim), coupled with a condensed linear representation of all markers found in the text. We are still investigating visual representation of secondary claims. (We are not inclined to use free graph representations, as we find these tend to get cluttered.)
Technical aspects:
We will develop a javascript/CSS library that can collect references to a given text and display appropriate graphical markers to make the argumentation steps more explicit. This library will probably impose requirements on the structure of the source text.
Step 3: In-browser affordances for argument construction
A bookmarklet should allow to easily create a robust link on a document. The construction of claim microcontent should proceed from there.
Technical aspects:
- We are targetting X/HTML documents. Extensions to other data types would be desirable, but outside the scope of our project.
- The most general approach would be for the bookmarklet to start a CGI script on an external server or on the user's machine
- Firefox extensions might also be built
- Server-side tools should be integrated in major blogging platforms
This will allow bloggers to annotate their own text, of course, but also (optionally) to integrate annotations to their text that are defined (as microcontent) in comments or trackbacks.
Step 4: Representation of claims on external sources
Our initial design will require a text to include references to the javascript library so the claims can be displayed; a server-side application should allow to transform arbitrary texts so references to them can be displayed. A mechanism for transclusion should also be defined at this point (probably in the form of a javascript library).
Step 5: Cross-site argument reference
It should be possible to obtain from a site a list of claims made in other sites that refer to claims made on that site, so as to allow peer-to-peer argumentation. This touches ideas of data ownership.
Technical aspects:
- Each claim within a site should be identifiable with a URI.
- That URI can be transformed in a standard URL which provides a RESTful API for claim manipulation.
- Getting that URL provides an RDFa representation of related claims, whether they come from the site or from comments/trackbacks.
- Posting to that URL allows to add a trackback source.
(Whether the server application chooses to store trackback sources or the claim URIs they provide will be implementation-dependent.)
- Getting the URL may allow for presentation variants (possibly using XSLT, and/or graphical libraries.)
- The server should provide appropriate redirection messages when URIs change, and the documents should contain a meaningful timestamp
Step 6: Identity and community
Claims are made by identified individuals, who are answerable for their site's contents.
However, it should be possible to define private sub-communities. Those would especially allow a "safe space" to examine drafts of controversial issues among trusted peers before they are presented to a larger community. (This is inspired by the School for Peace [http://sfpeace.org ]Rabah Halabi, Ed. Israeli and Palestinian Identities in Dialogue: The School for Peace Approach findings.)
Technical aspects:
- Use the openid [http://openid.net/] identity infrastructure
- Private sub-communities impacts the peer-to-peer architecture described above
- In a SDSS, identity should be complemented with an authentication infrastructure to allow for meaningful votes
Step 7: Claim equivalence and voting
Participants can express support for claims. Support can be tallied, yielding a vote. Conversely, expressing an objection should always take the form of expressing support for a claim that contradicts the original claim. Only support exists as a simple relationship in this ontology.
Votes on secondary claims are especially important, as they are used, among other things, to determine whether some primary claims can be identified as similar. If two claims from different authors are agreed to be equivalent (by the authors, or by an overwhelming majority of participants if one of the authors is not available to comment) then support for those options may be combined.
Note: As this system is to be used for social decision making, we must always assume that the claim base will be polluted by spurious claims. One way to get around this is that the original authors of source texts might always override identification of primary claims based in their work, or secondary claims based on such primary claims. Overriding must also be founded: If the argument goes the way of Godwin's law, and one of my opponent compares one of my claims to some of those attributed to Hitler (secondary claim of similarity) then contesting that secondary claim should take the form of showing where my positions and Hitler's differ. However, my doing so (or my endorsing someone else's doing so) is definitive.
Technical aspects:
- Support can be qualified along more than one axis, as in Delphi [http://web.njit.edu/~turoff/Papers/CDSCMC/CDSCMC.htm]Murray Turoff, Starr Roxanne Hiltz, Michael Bieber, Jerry Fjermestad, Ajaz Rana. «Collaborative Discourse Structures in Computer Mediated Group Communications». Journal of Computer-Mediated Communication, 4(4), jun 1999. .
- The widgets representation of argumentation steps should allow a reader to express support in a single-click operation
(This assumes that the reader has an application installed on his-her openid server that can tally the URLs of claims supported. Such an application should be easy to install and require low bandwidth. It must also allows readers to conveniently shift their support at a latter date and warn the claim application.)
Step 8: Representation of overall status
A representation should be derived to detect claims which are controversial (i.e. have support and whose counter-arguments are also well-supported); conversely, detect claims or primary sources which have not received adequate attention.
Technical aspects:
Vote propagation might use techniques equivalent to swarm propagation [http://arxiv.org/abs/cs/0609034]Marko A. Rodriguez. «Social Decision Making with Multi-Relational Networks and Grammar-Based Particle Swarms.». in ,40th Hawaii International International Conference on Systems Science (HICSS-40 2007), CD-ROM / Abstracts Proceedings, 3-6 January 2007, Waikoloa, Big Island, HI, USA. IEEE Computer Society, 2007, p.39..
Step 9: Reputation and completeness
In the case of isolated claims or un-analysed sources, "experts" could gain (visible) reputation through a process of extensive (uncontroversial) review of obscure claims.
Uncontroversial review means that the experts will draw parallels between a new claim and existing claims that are not the subject of controversy (as above). Note that the expert is free to also make controversial secondary claims on the new claim, as long as it is explicitly distinct from the classification activity.
Step 10: Basic inference
The application might be enhanced to derive inferential support. A form of liquid democracy [http://fc.antioch.edu/~james_green-armytage/vm/proxy.htm] may be implemented by delegating support to an "expert" on any claim pertaining to a certain issue: the additional support would be added to the expert's support to sub-issues.
The inference application is quite complex, and resource-intensive; support may eventually be added for peer-to-peer argument propagation, distributed storage, distributed computing of support inference, etc.
Further research
Finally, much experimentation is possible, and desirable, with the social dynamics of argument evaluation: What is the equivalent of a plenary? Should there be a randomization or rotation of which new arguments get presented for comments first? etc.