At the EMNLP-CoNLL 2007 conference in Prague, the best paper award went to "Modelling Compression with Discourse Constraints," by James Clarke and Mirella Lapata of the University of Edinburgh.
James got to give his talk to the full conference and was allowed 10 minutes of questions instead of the usual 5. The very end of the question period was hijacked by performance art:
Hal Daumé III, session chair: Jason?
Jason Eisner, program chair: Thanks for a very nice, uh, presentation. What I'm sure many in the audience are wondering — but may be reluctant to ask — is about the impact of this work. On global climate change.
James Clarke, speaker (taken aback): On global climate change.
Jason: Well yes — it's kind of a crisis? I apologize for asking, but I feel we can't ignore our personal responsibility, especially as scientists.
James (smiling slightly): Well, believe it or not, I was prepared for that question ... because our Ethics Committee asked the same thing. It's true we burned a lot of cycles on our experiments. But we did them in the Scottish winter, using them as our only source of heat. We were actually fairly cold. We also conserved resources by reusing the same data across experiments. Oh, and we purchased carbon offsets.
Jason (who's been half-heartedly nodding throughout): Okay, good, that's — that's great for you. But I was asking what you'd expect were this method to be widely adopted ...?
James (very earnestly): Look, we both agree this is an important issue. But I don't know what to say. Compression is inherently an environmentally friendly technology. Once the documents are compressed, it reduces the energy requirements of all future language processing.
Hal: Okay guys, that's enough. Next question.
Jason (sitting down): We'll take it offline.
Dan Bikel, audience member (who's been working himself up to this): Thank you. Dan Bikel, Dystopia Laboratories. I'm frankly quite outraged at the entire exchange that just occurred. I had no idea that this conference had been co-opted by the green liberal elite. I mean, I thought we were in Prague, not Hollywood.
[pontificates] Look, every day I drive my SUV to work and go to my office. Right next door is a room filled with large racks of machines cooled by an even larger air conditioner. And we use copious amounts of fresh drinking water to cool the CPUs themselves. All of this is just fine. As long as people are willing to pay for it, as researchers —as scientists — we have the God-given right to burn cycles, no matter how expensive the algorithm or the cluster on whch it's run.
In fact, while we're at it, Jason, your "Eisner and Satta" parsing algorithm is just the kind of "liberalism gone amok" that I'm talking about. I use an n5 parsing algorithm, even though your n3 algorithm exists. Know why? Because I can.
Jason (rising, exasperated): Dan, this wasn't even a parsing talk! How can anyone talk about generation today and ignore all future generations? ...
Hal: Okay, guys, I'm going to cap your CO2 emissions right now. Next session starts at 11:15.
At least, that was the script, drafted after Dan, Jason, and James riffed on the idea for 30 minutes at a conference coffee break a few days earlier, and decided we had to actually do it. We thought we'd better script it in advance to keep it short and to help us keep a straight face.
However, there was some improvisation at the actual event (during which most of the audience was cracking up). For example, James suggested that the question would be "simmering throughout the whole conference," and acted just flustered enough at first that a few people had to ask, "He was in on the joke, right?"
Several audience members noted afterward that Jason was throwing stones in a glass house, as editor of the massive 1220-page proceedings volume. He'd just been waving it around (to the extent he could) during his opening remarks, crowing about the growth of the field without acknowledging the shrinking of the forest. Perhaps James should have counterattacked and driven home the document compression message ...
|Jason Eisner - email@example.com|