Strange Aeons

Some of my more recent research has been into new, or at least improved, ways to make the web a more conversational place. In one sense it already is, we can link to stuff and we can comment on posts. This isn't always sufficient though. In reverse order, not all sites support commenting and even when they do comment support on websites can still be done better than it is right now as it can be difficult to mine, evaluate and analyse the insightful comments. I feel that a web conversation works better when it is a part of the Web, a strand of conversation that interleaves the various sites spread across the web leading a person, whether a spectator or part of the conversation, to engage in chance discovery. Linking is the primary structure of the web. Hyperlinks enable us to create uni-directional links between one site and another, so that when we wish to refer to another site, for example, when commenting in our own blog upon a claim made in another, we link to that originating source site. These links are only one way however, so I can refer to another site, but they have to respond to me if the conversation is to be reciprocal. This does not however stop any other web users from responding to my post in any way that they see fit. This gives us a very egalitarian system, with minimal structure, in which the web itself supports conversation - it is up to the conversation's participants to engage in it and make it happen. A nice example of this kind of web conversation occurred with the recent discussion between John Gruber of Daring Fireball and Joe Wilcox of Oddly Together. This spat basically boiled down to the suggestion by Joe that Daring Fireball should allow comments on posts because otherwise Joe has to respond to Daring Fireball posts on his own blog and John's response that "you write on your site; I write on mine". I think that that is exactly how it should be and that the web is a better place for it. I think that comments enable an immediacy of response but they don't guarantee that your response will stay in place. The only way to ensure that your response stays online is to post it in a place that you control with a link back to the original. Even better, if those two hadn't disagreed then I wouldn't have discovered this post about "Conversational Journalism" by Doreen Marchionni which to me echoes the argument that Alan Rusbridger made in his Hugh Cudlipp lecture (which I commented upon before). There are also many sites that are primarily designed to support conversation such as the Web 2.0 sites like Twitter but for the most part these lead us into a walled garden, with poor support for conversational threads and, due to the limitations of microblogging, the increased use of link shorteners which, whilst not breaking the links in the web, certainly weaken its structural integrity. From the perspective of the Twitter business this makes sense, the lock-in part anyway, you want to be the only game in town, but from the perspective of the web, this is a systemic weakness because a huge portion of web traffic, much of the web conversation, is occurring on a single site, a single point of failure. This is not to say that I am "against" Twitter, far from it, I like the idea of microblogging and to me it provides the basis, but not a complete solution, for enhanced distributed web conversation. It is just that I prefer open and distributed systems to closed and proprietary ones - mainly because when things break, and things always break, you can't fix them, you have to wait for someone else to do that. The next stages, as I see it, are twofold:

1. to provide aggregation mechanisms so that we can get different views on the conversations, for example, following it in whichever direction we like, instead of just following the uni-directional links, or exploring a particular sub-thread of the conversation.
2. to provide new tools to make it easier to engage in the conversation.

The first stage can be accomplished by building upon the Argument Interchange Format, at least in the guise offered by its second version which includes support for dynamic argumentation structures, including dialogues, rather than just static, monological argument. By utilising open, distributed, web-conversation repositories to store meta-information about the relationships between conversational utterances, much like trackbacks and pings enable a form of bi-directional linking, we can start to build new interfaces to engage with and track the web conversation. The second stage will be accomplished by the tool-builders, a process that is partly happening already with the advent of tools to enable Twitter users to tweet from wherever they want and with new and improved blogging and microblogging platforms such as Tumblr and Posterous making it easy to create short posts as a part of the web conversation.

Wow, where did the last month go? I am recently back from the second Argument Interchange Format (AIF) meeting which was held at the Dalmunzie Hotel in rural Scotland. The list of delegates to this meeting read like a who's who of online argumentation researchers - people who are developing argumentative tools which, in some way, communicate argumentative information between themselves and to their users. All in all the meeting went well. We achieved the aim of writing a first draft of a specification for the second AIF which includes fixes for many of the problems we have discovered since the original format was specified and introduces some new elements that many of us have noticed were glaringly omitted from the original. In terms of running and organising the event, there were some things that went really well such as:

• Ensuring that all delegates prepared a position statement beforehand so that rather than meeting and starting with "what are we going to talk about" we got straight to the job of discussing the next version of AIF.
• The first night over dinner we all wrote questions onto post it notes that framed the sorts of things that we were interested in. These were used to create a number of topic/discussion groups (which later became major organising elements of the paper draft) and whittle the 23 delegates into a manageable number of working groups with shared interests.

Other things of course worked less well, for example,

• 23 computer scientists using Google Docs to collaboratively edit documents is fine. Except when you are in rural Scotland where your internet feed is provided via two-way satellite communications which are very quickly saturated, and suffer from fairly high latency anyhow.

Some lessons learned:

• Google Docs works quite well for collaboratively editing documents. Who knew? It even works well if you are using it to collaboratively edit LaTeX source, although obviously you don't get to compile it to anything useful or check source errors within Google Docs.
• Make sure that people aren't editing offline then copy-pasting into the Google Doc, as each time they do this they reintroduce the same errors that you just got finished fixing. (This one caused both myself and John to swear quite a bit)
• If you are working with LaTeX then don't forget to install it onto your laptop before you go as you then have to shell into a remote server which has a working LaTeX environment in order to compile the aforementioned LaTeX source into a PDF.
• Set up a local network using a small wireless access point and make some shared directories available on a small server such as a mac mini so that you are not transmitting all that data over the hotel's network all the time. We caused the Hotel to have to reset their router several times over the course of the meeting.
• Investigate collaborative software, such as Gobby, or a versioning system such as Mercurial that can run on the local server to keep track of the collaborative edits rather than relying on Google Docs. Although Google worked reasonable well I still had to go through the following steps to get a PDF generated:

1. Download the document from Google Docs as a txt file & save it as .tex file
2. Run the tex file through dos2unix
3. Get rid of any final non-printable ascii characters using the following:

   $ tr -cd '\11\12\15\40-\176' < file-with-binary-chars > clean-file

4. Compile as usual.

My favourite element of the new specification is the inclusion of support for dynamic argumentation. I had attended the meeting with a clear personal plan to ensure that the AIF2.0 include support for dialogue and I was assigned to the working group on dialogue with Bart Verheij, Raquel Mochales , and David Glasspool. One of the things that quickly became apparent was that thinking of the AIF in terms of monologic and dialogic argument was quite limiting and that other researchers had quite clear needs which wouldn't covered by an AIF which accounted for just arguments and dialogues. Instead what we developed was a model that saw the core AIF, version 1.0, as a model of static arguments and AIF2.0 as including both this as well as dynamic extensions that enable us to model not just dialogue but also other aspects of dynamic argument such as representing the order of argument elements for when analysing rhetorical and presentational aspects of an argument (thanks Raquel) and the history of the construction of an argument (thanks David).

or authority versus involvement as Alan Rusbridger describes it in his recent Hugh Cudlipp lecture in which he asks Does Journalism Exist? The us versus them refers to the idea that in the past information was held or restricted to certain authorities who controlled how that information was presented, consumed, and reused. More recently, this order has changed and the consumers are now as likely to also be the creators, the reader is no longer solely a passive consumer, but lives in

"a world in which many (but not all) readers want to have the ability to make their own judgements; express their own priorities; create their own content; articulate their own views; learn from peers as much as from traditional sources of authority."

This struck me as interesting and succinctly defines one of my more recent research activities which has revolved around designing interactions, and associated software infrastructure, to support online argumentation. By this I mean the necessary software and infrastructure to provide support for those people who want to:

• exercise their own judgement,
• create their own content,
• articulate their own views, and
• learn from their peers

just as Alan mentioned in his lecture. Although many of these activities can already be performed by those who are sufficiently able and motivated, the supporting technology is still rudimentary. Just as we are still trying to develop the best persuasive interfaces to influence behaviour, we are still trying to develop the best interfaces to support online argumentative interaction and thereby improve critical literacy. There are at least two approaches that we can take. One approach is via education, for example, my introductory undergraduate module in problem solving and critical thinking is exactly the kind of introductory course that should be a prerequisite for anybody who wants to be able to say that they have had an education. Not that my module is perfect, far from it, just that the nature of the module; teaching students how to discover the structure of arguments, to see where there are holes or errors in the reasoning, to recognise when a rhetorical trick is begin used against them; these are all the kinds of skills that members of a knowledge society should possess. At the moment this is also the kind of topic that is left to a kind of inate ability, some people are just good at arguing, and others are good at being mislead by those in the know. However education doesn't solve the problem of lack of explicit support in the infrastructure that we use to communicate. This leads directly to my second approach: building the tools that support online argument, enabling people to create their content, whilst also guiding and supporting the process of articulating viewpoints and exercising judgements. My initial prototype for an argublogging system was outlined in a CMNA workshop paper last summer and uses nothing more than the web simpliciter, some browser situated javascript, and an aggregation server, to begin the process of supporting web users who have just read something online that they agree or disagree with and want to respond. Of course it is easy to respond online, anyone can set up a blog in minutes and link back to the original. The problem is that ordinary links carry very little information other than the fact that one place links unidirectionally to another. Much of the contextual information about how you are responding is lost, for example, to know whether I am supportive or antagonistic with respect to the quote above from Rusbridger's lecture requires a reader to read and understand this post. Wouldn't it be nicer if that relationship was recorded? Even better, wouldn't it be great if you could see whether anybody else had responded to any aspects of this post elsewhere on the web? Ideally you would structure all of those quotes and responses into a single dialogue, gatherered from all of their locations across the web, possibly you might even want to visualise this dialogue, seeing the structure of the various arguments that make up the dialogue. Essentially this is what my argument blogging system does, enables web users to harvest textual quotes and respond to them within a structured dialogue. This structure is captured and stored in a web-accessible database (AIFDB), in an RDF language that reifies the Argument Interchange Format, and thus becomes a Semantic Web data source, ripe with all of the potential that that entails.

What?

In a previous post I mentioned the WWAW (pronounced WOW), or World Wide Argument Web, an unwieldy name for what is mostly referred to as simply the Argument Web. My Argument Blogging project, that I have posted about here, is a part of this Argument Web which is essentially a network of loosely coupled online applications and services that provide a web of argumentation resources.

One way to think of the Argument Web is as a way to work with information on the Web. If the Web/Web2.0 is basically a semi-structured mass of data oriented towards mark-up for human consumption, then we can see how there are different ways to work with that data that are not necessarily primarily human-oriented. For example, to do automated parsing of Web data we might look to Semantic Web technologies for structure and reasoning mechanisms useable by machines. For human-orientation we look to the Web (WWW) simpliciter or else the Web 2.0 if we want more interaction. If we wanted to work with the Web in terms of arguments and associated argumentative interactions such as dialoues and conversations then we would look to the Argument Web.

The Argument Web exists in relation with and in addition to the existing webs of data and simply provides an infrastructure and tools for eliciting, structuring, and storing data in terms of arguments and related concepts, and for subsequently interacting with that data.

Why?

There are a number of reasons that an argument web is interesting. To begin with, argumentation and dialogue protocols can be used to provide a good way to elicit knowledge. Dialogue protocols provide a good human oriented interaction mechanism for both eliciting data from users and for communicating it back to them at a pace and in an order of their choosing. Allied to the this is the fact that knowledge elicitation via dialogue games allows us to capture metadata about how the units of information relate to each other. This can be sufficiently well structured that the data can be reused in automated reasoning systems, for example, as knowledge bases within intelligent agents. This is an avenue that Chris and I explored in an IEEE Intelligent Systems Journal paper a couple of years ago [reed2007magtalo :: "Using Dialogical Argument as an Interface to Complex Debates"]. Additionally, this approach provides a way to begin building a truly large scale corpus of well structured, real-world argumentation which will be an invaluable resource for argumentation researchers. Obviously this is not an exhaustive list by any means but covers the central points that make it interesting to me.

How?

The Argument Interchange Format, known as AIF, is a high level ontology of argumentation theoretic concepts that is used to communicate information about argument structure. This communication can occur between people, for example, argumentation researchers who discuss concepts such as Information Nodes( I-Nodes) and Scheme Node (S-Nodes) or who use the graph-based AIF ontology to visualise the structure of arguments and the reasoning contained therein. Communication of AIF concepts is however not restricted to people but has a number of computer implementations, for example, in RDF and OWL-DL meaning that argumentation data can be shared between programs.

The AIF-DB is a web-app for storing AIF data. It consists of a database and a RESTful interface for getting AIF data into and out of the database and also for searching it’s contents. The advantage of adopting a technology like the AIF-DB is that, unless you have special requirements over how your AIF data is stored, you have a ready made database and API that you can use from your application. This has simplified the process of building new argument software for the web and has been used as a core element in a number of new pieces of ArgumentWeb software including:

• OVA – An online agument analysis tool similar to Araucaria
• OVAView, the argument visualisation widget,
• The Argument Blogging software,
• ArgDB – An online corpus of analysed arguments which is the latest incarnation of the original AraucariaDB, the first large corpus of analysed arguments.

We have all read something online with which we have strongly agreed or disagreed. In many of these cases we might respond via comments, if the site supports commenting, or we might post something to our own website that links back to the original. Unfortunately this approach does not enable us to capture and reuse the argumentative structure inherent in the response and its content, we cannot easily automatically determine whether your response is supporting or attacking the original post. It requires a user to read and comprehend both your post and the one you link to and to then reconstruct the arguments. Additionally, once the original author has moved on the original site might disappear leaving a broken link, or they might edit the original post to reflect a different viewpoint.

If we look back to some of the original thinking on hypertext systems, the web was meant to be more of a two-way flow of information, where someone could upload some information, and others could respond. The web hasn’t developed liked that. Although we are bit-by-bit constructing a web with more user interaction (Web2.0), with more comprehensive commenting systems and tools for annotating webpages, we still do not have a consistent, web-wide interaction mechanism for two-way argumentative dialogue on the web. Correctly implemented, such a mechanism would be very important, transforming the web from its current, fairly static nature to a dynamic, two-way flow of highly-structured information. One area that is making some headway in this direction is in the emergence of technologies like MicroBlogging, Twitter, and all of the sites now including status updates, which are making it quick and easy to publish small amounts of information to the web. This information is subsequently federated and manipulated by the emerging real time web to offer facilities like real-time search, to find out what is happening and being talked about right now. Although these are great ways to use the web, there are a few unrealised opportunities, such as storing information that captures the fine-grained structure of interactions not just the content, which presupposes taking advantage of a second opportunity, that of providing a natural pervasive argument-oriented interface for engaging in fine-grained, online debate.

The problem of information structure and storage is that information is not stored in a way that makes it easy to automatically reuse (a problem that the Semantic Web is attempting to solve), particularly from an argumentative or dialogue oriented perspective. HTML and related technologies for example store information in ways that make it easy to parse for display, whereas formats like Atom and RSS, are oriented towards syndication in a time-ordered feed of information. None of the widely used technologies are oriented towards capturing the interactions between users, links for example, carry very little information, they do not indicate whether a response is supportive or antagonistic, or even that the response is actually a response. The only information that a link really carries is that a source has a relationship to another source. One way to structure information about the arguments is through one of the Argument Interchange Format (AIF) implementations. For example using the RDF AIF implementation offers a way to store structured argument information in such a way that it can be reused by other tools, for example, by argument visualisers or automated reasoners. Because this version of AIF is implemented using RDF it is a simple process to reuse this argument information within semantic web applications. A fully realised Argument Web based upon adoption of languages like the AIF also provides a useful bridge between the Web/Web2.0 and the Semantic Web.

The opportunity for increased interaction stems from the fact that many users of the web engage in arguments and dialogue a lot. Even so, support for argumentation on the web is very rudimentary. On many sites the extent of support is that you can reply to a post by commenting, possibly this is supplemented by some support for threading of comments so that a rudimentary dialogue can be had but that is far from a given. Faced with this many users resort to either copy & pasting a portion of text to indicate what specificatlly they are responding to, or using @username to respond to a given user and hoping that it is sufficiently clear as to which specific comment of the respondee is being resonded to. The @ nomenclature has been adopted on Twitter to allow extended dialogues to occur but these are very coarse grained and it can be difficult to follow a thread of conversation for more that a few tweets. Then again the aim of Twitter is to provide a venue for rapidly posting small updates, not finely nuanced conversations articulating differing points of view. One general way to support extended fine-grained dialogues is through the use of dialogue games. Simple, multiplayer turn-taking games that model the interactions between participants during dialogue. The moves in dialogue games usually correspond to a particular type of act within a dialogue so, for example, you might make a question move directed towards a particular participant, who in their next move responds to the question move with an answer. Alternatively, a participant might make an assertion move, asserting their position on an issue, to which other participants might respond with an array of moves, for example, agreeing with, disagreeing with, or challenging the position. Such games are modelled on everyday human interactions so form the basic of quite natural interaction mechanisms for people to use – an advantage in securing uptake amongst everyday web-users.

A system for distributed online argumentative dialogue would allow us to;

1. select any publically accessible section of text,
2. appropriately respond to it,
3. publically record our response wherever we desired so that others can continue the conversation, and
4. reuse this structured information in intelligent software.

Additionally, such a system should not rely on a central server, nor should it rely on the installation of a lot of additional client-side software, or the large scale uptake of software, or channeling of users to a particular site, or a requirement for site owners to install additionaly software. Rather, the system should rely on open distributed platforms and minimal opt-in to gain access to basic functionality.

ArgumentBlogging is a term used to describe a method for argumentative interaction on the Web which is implemented in an initial prototype system (initially published at the CMNA 9 workshop) that utilises a bookmarklet and a simple dialogue game to provide a way to capure argument content and store it using the AIF.

Floris sent me a link to this advert for a PhD student at the University of Amsterdam’s Leibniz Center for Law. The advert relates to a new FP7 project called IMPACT (not to be confused with Chris’ IMM-PACT EPSRC funded project) which aims to do the following:

IMPACT is an international project, partially funded by the European Commission under the 7th Framework programme. It will conduct original research to develop and integrate formal, computational models of policy and arguments about policy, to facilitate deliberations about policy at a conceptual, language-independent level. To support the analysis of policy proposals in an inclusive way which respects the interests of all stakeholders, research on tools for reconstructing arguments from data resources distributed throughout the Internet will be conducted. The key problem is translation from these sources in natural language to formal argumentation structures, which will be input for automatic reasoning.

What is of particular interest to me is the idea of building new tools for reconstructing arguments from data resources on the web – very similar to the argument blogging approach to online argumentation that I have been working on recently. The main difference here though is that the prospective IMPACT tools appear to be aimed at working with the web as a static resource whereas argument blogging is meant to be an active, user-centered activity, although one extension that I am looking at is to integrate active, “no I disagree because…” type arguments with “and here is a resource to prove it…” type arguments, which I see as being a mixture of both active and static argumentation. Also similar is Mark’s OVA software, an online argument analysis tool similar to Araucaria, although currently this can only be used to analyse a single web resource not collate and resuse arguments across multiple sources.

Interestingly this project will use the LKIF, understandable given the domain in which the project is to be executed, but it makes me wonder about the possibility of tools to translate between AIF & LKIF so that resources from one domain can be accessed by tools from the other and vice versa.

As promised here are the slides from my Argument Blogging presentation at CMNA 9 in Pasadena, California.

This Dilbert cartoon seemed apropo given that I am still working on my CMNA9 presentation of the Argument Blogging Project work. I managed to record a short film of Colin talking about his project at the School of Computing 09 degree show held in the Queen Mother Building at Dundee University. I am not a videographer but I think that I managed to capture the essence of Colin's implementation work on the project (or rather Colin managed to make a good presentation and I managed to hold the camera reasonably steadily and zoom in/out and focus on the relevant bits almost at the right times).

I have a project in the works that I have had on the back burner for a while based around the idea of argument blogging - capturing the argumentative structure of interactions occurring online, particularly in blog posts and comments. In the meantime there are some similar technologies being developed elsewhere on the web. One of these technologies is encouraged commentary which uses JQuery to provide a nice interaction mechanism for dealing with comments on a blog. The basic idea is that you select the text you wish to respond to and a bubble fades into the screen next to the highlighted text offering you a button that you can click if you wish to respond. If you click the button then the highlighted text is copied into your comment and you get the opportunity to type in your response which then appears in the usual comment space near the bottom. Instructions for getting encouraged commentary working on your blog can be found over at Don't Trust This Guy. The system is nice, and automates some of the things that we have been doing to overcome the limitations of traditional comments, such as putting in the @user bit to indicate whose comment you are responding to. Another post about this new technology can be found over at readwriteweb. Whilst a good example of how to enhance the existing system by providing simple, natural, and intuitive tools to support the user I don't think that it goes far enough in supporting the kind of structural capture that we are looking for with the so-called nascent World Wide Argument Web (WWAW). I would like to be able to capture the actual argumentative structure that underpins these discussions and export the individual arguments as AIF and the dialogues as AIF+ so that they can be reused in argumentation specific tools like Araucaria. The approach I am taking is to specify dialogue games, using the Dialogue Game Description Language (DGDL), that give an appropriate range of performatives to associate with the act of commenting in blogs. We then use a similar technique to argument and dialogue capturing that we used in MAgtALO, where the argumentative relationships between statements are inferred from the types of moves that the players select in the dialogue game.

There are a now quite a few argumentation tools, a number of which are online such as ArgDF, Avicenna, MAgtALO. Many of these tools already use, or are planning to offer support for, arguments described using the argument interchange format (AIF). As these individual tools for working with argumentation are deployed they form individual points within an argumentation software ecosystem that I have alluded to before called the World Wide Argumentation Web (WWAW pronounced WOW). I think that more tools will be developed in the near future that will diversify this ecosystem making it easier to create and work with arguments online. The next step is to begin considering how these individual parts of the WWAW can be joined up to fulfill the networked promise implicit in the WWAW name. This is a question that has been asked before with respect to FOAF. How can/should multiple FOAF data sources be federated? Rather than suggesting a single means by which all of the AIF data resources are federated I think a healthier argumentation ecosystem would be constructed if there were multiple ways to join up the arguments in the WWAW. For example: 1. Autodiscovery - similar to RSS autodiscovery and FOAF autodiscovery so that an AIF description associated with a web page can be automatically found by applications that understand AIF. 2. Internal AIF Links - Within an AIF document there should be links to other AIF documents that are related. This would support spidering of the WWAW by starting with a source AIF document which links to others, which in turn link to yet more, ad infinitum. 3. Registries - Where links to distributed AIF documents & AIF repositories can be posted by their creators or discoverers. 4. Indexes & Search Engines - Created by search engines to enable AIF documents to be discovered & searched. Indexes work hand in hand with search engines and WWAW spiders to discover AIF documents, and possibly to add further value to them. Currently google should index any AIF document posted onto public web servers but there are also semantic web oriented search engines such as swoogle and indexes like sindice should also work, as well as yet to be developed AIF only search engines and indexes. 5. Repositories - Central locations where AIF documents can be posted and stored. Similar to the AraucariaDB, as of writing the only corpus of analysed argument available online. Analagous to tools like PTSW I think that whilst there is a manageably small number of individual AIF tools then these federation issues are moot, and once the AIF is as well known and used as FOAF or RSS then again the issue is moot (although it is pertinent with respect to how best to work with the various resources). It is in the middle ground when we are trying to scale up to a wide adoption of structured argumentation on the web that good tools for advertising, linking together, searching, discovering, and adding value to the AIF are particularly important because they mark the difference between a small interchange format used by a minority of enthusiasts, and a widely adopted strategy for adding more value to data on the web.