The Association for Computers and the Humanities (ACH) and the Association for Literary and Linguistic Computing (ALLC) had their annual joint conference in Athens, Georgia, this past week, 29 May - 2 June 2003. This was a little early: many members of the associations prefer dates in June or July, since some universities, particularly in Europe, are still in session at this time of year. When asked about this last year, though, Bill Kretzschmar of the University of Georgia, who did the local organization this year, simply said well, yes, we could almost certainly get the facility in July if we wanted it — but trust me on this, you really, really don't want to spend a week in Athens, Georgia, in the middle of July.

ACH and ALLC have been around since the 1970s, and the conference series has been going even longer; this is definitely the oldest established permanent floating venue for people interested in applying computers to humanities disciplines to gather. Some themes remain perennial (stylostatistics, authorship studies, and, alas, character set problems), and others change (computer-assisted language learning is less prominent now than when I started attending in the mid-1980s, and cultural and media studies have become prominent), but this conference remains, year after year, my favorite academic conference.

The conference started Thursday evening (after the conclusion of tutorials and of the meetings of the executive committees of the two associations) with a keynote address by John Maeda of the MIT Media Lab. He did not give the kind of academic lecture one normally expects at an academic conference, and it seems to have taken most people a little while to get used to his delivery and choice of subject matter. This is, perhaps, understandable; he began with a long discussion of tofu manufacture by way of introducing himself (his father ran — runs? — a tofu factory), before giving us a tour of some things he has designed: boxes which react to sounds made by the user, a drawing tool in which the mouse never moves, but the window does, a pencil that breaks down and hangs the way a computer does (a reaction to the dictum “The computer is nothing but a pencil”), closeup photographs of salt and pepper, programs to count the number of salt crystals in a packet, photographs of exhibits of his work, a two-minute demonstration of a graphics programming language, and on and on. Once I stopped worying about how it related to computing in the humanities, I enjoyed it a lot.

His pages at http://acg.media.mit.edu/people/maeda/ and http://www.maedastudio.com/ capture his voice and manner very well.

Among the more memorable sessions for me was the first paper session of the conference, in which people from the University of Kentucky, all working with the Anglo-Saxonist Kevin Kiernan, reported on tools they are building for image + transcription work; Kenneth Hawley reported on things from the philological / Anglo-Saxonist point of view and gave a user's overview of the tools. Jerzy Jaromczyk gave an architectural overview and described the IDE framework they are using, and Ionut Emil Iacob popped the hood and talked about the database substrate they are developing and how it deals with overlapping hierarchies. They seem to making many of the right decisions on how to do things. Not all the right decisions, I think: they are using home-brew DTDs which are going to make it hard for their tools to be useful for people who don't share Kevin Kiernan's particular views on what is most interesting about manuscripts and transcription. They over-estimate, I think, the fear of angle brackets to be expected from scholars — no one who has mastered Anglo-Saxon paleography is likely to find mastering XML syntax an over-taxing assignment. And their solution to overlapping textual phenomena, which effectively recreates SGML CONCUR, is going to fall short at some point, for the same reasons that CONCUR falls short of being a full solution to its problem space. I pointed them to work at Bergen on overlap and to Murata Makoto's old article on storing both physical and logical hierarchies in a single file format — Murata-san doesn't like his solution any more, but his article has a very good view of the problem. They got some rather sharp questions (sharp, but not really hostile) from Peter Robinson of De Montfort University and me (“Well, that's what I call ‘tough love’”, drawled Daniel Pitti from behind me, as Peter drilled into them on their choice of Eclipse), but the Kentuckians reacted superbly, taking the questions as a quick way into hard and useful questions. At one question, Alexander Dekhtyar (who had co-authored the paper with Iacob) popped up to answer, and gave as good as he got. Good content, pointed substantive discussion. It was a wonderful opening to the main part of the conference.

I also heard some papers on temporal modeling, which were interesting but which I didn't understand fully: Johanne Drucker of Virginia was very keen to allow “subjective inflections” of the timelines, but it was not clear to me what she meant by “inflections”. I kept looking for turning points in the time lines, local minima or maxima or the like, but I think she must have meant something more like body English. Jon Christoph Meister of Hamburg talked about “Tagging Time in Prolog”, but most of his remarks focused on the narratological aspects of timelines and their recognition; he showed only two lines of Prolog, and they were so small on the screen that I couldn't read them. On the plus side, he is using TEI feature structure markup for his results, which makes him the first person I've encountered in years who takes the feature structure tag set seriously. So I was glad I went, even though it meant missing an interesting session on ‘deep markup’ elsewhere at the same time, in which Wendell Piez apparently really pissed off a representative of the Orlando Project by observing that they are using a straightforward data model, so that what makes their work hard is not problems of data modeling but other things. Third-hand accounts suggest the Orlando people felt that meant Wendell thought their work was superficial and easy, and accordingly let him have it. The high point of the session on temporality, from my point of view, was a talk by David Saltz of Georgia about a project to provide a 3D simulation of a live vaudeville performance. He's a theater historian, not a technologist, so he talked mostly about the intellectual issues involved in simulations of performances. He was quite insightful on some of the contradictory impulses involved. No one wants to prettify the past and present it unrealistically, of course — but when the theater historians in the project proposed to include a black-face act, on the grounds that black-face was a very important part of vaudeville, the computer scientists all started to prepare to bail out of the project. I understand the historians' point of view, and I also sympathize with the computer scientists. We also are part of history. So in the end they substituted an ethnic monologuist, who does a sketch involving a Jew and an Irishman. No one, observed Saltz drily, seemed to have any issues with re-enacting these two stereotypes.

On Sunday and Monday, there was a wonderful series of talks on markup and markup-theoretical topics. Paul Caton of the Brown University Women Writers Project talked about “Theory in Text Encoding” and argued that the “theory” appealed to in earlier work by Allen Renear and others (including me) doesn't really qualify as “theory” — Paul is so smart and so well versed in contemporary literary theory that I sometimes find his arguments arduous to follow, but he speaks very clearly and differs from some literary theorists in persuading me that the task of following the argument is worth while. In a separate session, Allen Renear (formerly of Brown, but now at the University of Illinois) gave a talk in which he responded to a critique of XML launched a couple of years ago by Dino Buzzetti, a specialist in medieval philosophy at Bologna; unfortunately, Dino was not at the conference to respond to the response. Peter Robinson threw away his announced topic (Anastasia, an XML publication system he has built — he said “it's there, it works, there is documentation, go read about it if you want to know”) and talked instead about the need to change XML so that it doesn't require documents to be tree-structured. He seems to have re-invented Claus Huitfeldt's old Multi-Element Code System (MECS) using start- and end-tags in XML syntax. (I'd say he had reinvented Netscape-style tag salad, except that he really did try to make a serious argument for it.) The problem is that Peter seems to think it would be straightforward to continue document-grammar-based validation after abandoning trees, but I wasn't able to elicit any useful response to questions about how he thought that would work. Information on Anastasia is, indeed, on the Web, at http://www.sd-editions.com/anastasia/index.html; I'm not sure this is the page Peter was displaying, but it's what I've found. Somewhere along the way, I gave a talk about XML Schema 1.0 as a language for document grammars; the text of my slides is at http://www.w3.org/People/cmsmcq/2003/achallc/achallc2003.html. I don't know whether the SVG animations will make sense without oral explanations, but perhaps they will. For that matter, I don't know whether they made sense with the oral explanations; I was kind of rushed, and afterwards some colleagues were muttering about ten pounds of flour and five-pound sacks.

On Monday, this thread continued as Wendell Piez (of Mulberry Technologies) and Julia Flanders (of Brown and the WWP) and John Lavagnino (King's College London) gave talks in a session on “Ambiguity, Technology, and Scholarly Communication” — all were thought provoking, as I have come to expect of them, and Julia's text was, as always, full of the most astonishing and insights and memorable images. It took me a while to adjust to the fact that they were all using the term ambiguity to describe what I would have called vagueness or underspecification — it matters because while ambiguity in the narrow sense (a finite set of distinct alternate interprepretations or parse trees) may be present or absent, vagueness or underspecification is always and necessarily present in any representation: no representation is ever so complete that it captures every property of the thing represented, which means the presence of vagueness in representations is co-extensive with the presence of representations, and ought to be slightly less surprising than its absence would be.

Next year, the conference will be in Göteborg, Sweden, June 11-16 (http://www.hum.gu.se/allcach2004/); a call for papers will appear in late June.