ELSNET joins two research and development communities, each with its own professional society, and Iíve just returned from the ACL/EACL 2001, held in Toulouse, July 6-11. It was the 10th European Chapter Meeting and the 39th ACL Meeting. Of course ACL is the Association for Computational Linguistics, the professional society for the language side of ELSNETís work, The Toulouse meeting was an impressive event, and here are some impressions.
The scientific papers were interesting both for their content as well as for indications they give of where the authors and referees now see the field. The trend toward a statistical approach to the classical questions of parsing and disambiguation continues, infused with new attention to linguistic detail. In this vein, there were papers on including sensitivity to grammatical heads in parsing, XML tools, maximizing conditional vs. joint probability, speech applications, and recovering dependency relations. There was new work on old applications such as machine translation, speech understanding, and question answering as well as papers on newer applications such as recognition and diagnosis of learnersí errors, text summarization, and information extraction on email. There were invited talks on linguistic search in audio files and on the use of statistical grammars in genetics. The prize for best paper was shared between Eugene Charniakís paper on "Immediate-Head Parsing for Language Models" and a paper entitled "Fast Decoding and Optimal Decoding for Machine Translation" by Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu and Kenji Yamada. There were tutorials on There were eleven workshops organized by ACL special interest groups (SIGís) on generation, dialogue and learning but mostly by ad hoc groups, on topics ranging from data-driven MT, open-domain question-answering, and knowledge management to collocations, evaluation, and Arabic language processing. It was a high-energy place!
The meeting of the professional society is also interesting for the overview of whatís going on outside the conference. Computational Linguistics, the journal of the ACL, and the most important archive of the fieldís results, continues to attract so many submissions that only about 20% can be accepted. There are special interest groups for large corpora (SIGDAT), lexica, parsing, generation, learning (SIGNLL), dialogue, semantics, phonology, multimedia, and each of these held special-purpose meetings in the last year or has published a CFP for a meeting soon. A new group devoted to Chinese Language Processing has just gotten started.
The ACL is embarking on a project to digitize all of the publications from its near forty years --- not only the annual conference proceedings and annual and biannual chapter proceedings, but also all of the many SIG publications it distributes --- 40K pages! COLINGís publications are also distributed by ACL, and COLINGís organizing committee has indicated that it will cooperate in this effort.
There were several indications that European organizational initiatives are breaking new ground and even setting trends. In an EACL panel on strategic funding for language technology, Joseph Mariani (French ministry) stood out for his clear vision separating fundamental research, technology development and applications, while Bernd Reuse (German BMBF) disarmed skepticism about the (putatively missing) economic impact of language by calming pointing to 20 products and eight companies that have spun off from Verbmobil. Nino Varile of the Directorate General XIII of the European Commission foresees a place for end users and marketing closely integrated with research and development, and this seems to be unique among all the funding agencies. I was surprised that no one mentioned a further point at which the European Commission has been innovative in its strategy for language technology, and that is in creating a forum where academic and industrial R&D can exchange ideas and experiences in strategically inspired research --- ELSNET. I know of no parallel at a national level anywhere.
The upcoming 40th anniversary of the ACL is also impressive. It might still be a bit arrogant to call computational linguistics a profession even now, putting it in a class alongside law, medicine and civil engineering. Whoever had the chutzpah to call this a profession 39 years ago must have been a daring soul indeed. But not crazy. In his Turing lecture, John Hopcroft recounted that, when he arrived in Stanford to teach computer science in the early sixties, no one could say very exactly what they wanted him to teach. Only years later did he recognize this as vision, a confidence in a strategic direction even before all the confirming details have crystallized.