The 2011 meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT) was held alongside the banks of the Willamette River in Portland, Oregon in June. The conference maintains a very high bar for publication quality and provides an excellent opportunity to connect with colleagues working on various aspects of computational linguistics and human language technology.
Since the last time I attended the conference in Ann Arbor in 2005, the community had grown substantially. ACL 2011 received almost three times more submissions than ACL 2005 (1,146 vs. 423), while its continually low acceptance rate (18% in 2005, and 25% in 2011) reinforces the reputation of ACL as one of the most selective conferences in the world.
I found the culture in the community more grounded in empirical methodology and increasingly preoccupied with reproducible research: among all the accepted papers, 30 papers are accompanied by software packages and 35 papers are accompanied by a data set (these materials will be hosted on the ACL web site under http://www.aclweb.org/supplementals). In addition, interest in the statistical significance of results—a much needed companion to the empirical method—continues to grow. For a prime example, I encourage you to read Better Hypothesis Testing for Statistical Machine Translation: Controlling for Optimizer Instability by Jonathan H. Clark and colleagues at Carnegie Mellon.
As the internet is maturing and thus offering an ever-growing amount and variety of natural language text data, the interest in the community is decidedly shifting to problems of relevance in the real world. As a natural consequence, unsupervised methods for addressing established problems are at the forefront, as demonstrated in Dipanjan Das, Slav Petrov: Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections (Best Paper Award winner).
The plenary lectures were given by two distinguished invited speakers: Lera Boroditsky, who presented research on how the languages we speak shape the way we think, and David Ferrucci, who talked about DeepQA, a question answering system that gave an astonishing performance on Jeopardy!. The success of DeepQA solidly anchors the tremendous progress achieved in computable natural language processing and understanding. One question that arose in my mind after the talk was, “to what extent is DeepQA about emulating the natural language faculty in humans, versus making the computer an apt adjunct to humans seeking answers? Would a savvy user of a search engine be able to answer the same questions in a reasonable amount of time?”
Eugene Charniak, recipient of the ACL Lifetime Achievement Award, presented highlights from his past work and views on the future of computational linguistics. His concluding remarks stirred up long conversations in the hours following the closing ceremony—many of which carried on over a pint of local brew or a glass of Oregon wine, which was, of course, the best way to end a stimulating conference week. d
Since the last time I attended the conference in Ann Arbor in 2005, the community had grown substantially. ACL 2011 received almost three times more submissions than ACL 2005 (1,146 vs. 423), while its continually low acceptance rate (18% in 2005, and 25% in 2011) reinforces the reputation of ACL as one of the most selective conferences in the world.
I found the culture in the community more grounded in empirical methodology and increasingly preoccupied with reproducible research: among all the accepted papers, 30 papers are accompanied by software packages and 35 papers are accompanied by a data set (these materials will be hosted on the ACL web site under http://www.aclweb.org/supplementals). In addition, interest in the statistical significance of results—a much needed companion to the empirical method—continues to grow. For a prime example, I encourage you to read Better Hypothesis Testing for Statistical Machine Translation: Controlling for Optimizer Instability by Jonathan H. Clark and colleagues at Carnegie Mellon.
As the internet is maturing and thus offering an ever-growing amount and variety of natural language text data, the interest in the community is decidedly shifting to problems of relevance in the real world. As a natural consequence, unsupervised methods for addressing established problems are at the forefront, as demonstrated in Dipanjan Das, Slav Petrov: Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections (Best Paper Award winner).
The plenary lectures were given by two distinguished invited speakers: Lera Boroditsky, who presented research on how the languages we speak shape the way we think, and David Ferrucci, who talked about DeepQA, a question answering system that gave an astonishing performance on Jeopardy!. The success of DeepQA solidly anchors the tremendous progress achieved in computable natural language processing and understanding. One question that arose in my mind after the talk was, “to what extent is DeepQA about emulating the natural language faculty in humans, versus making the computer an apt adjunct to humans seeking answers? Would a savvy user of a search engine be able to answer the same questions in a reasonable amount of time?”
Eugene Charniak, recipient of the ACL Lifetime Achievement Award, presented highlights from his past work and views on the future of computational linguistics. His concluding remarks stirred up long conversations in the hours following the closing ceremony—many of which carried on over a pint of local brew or a glass of Oregon wine, which was, of course, the best way to end a stimulating conference week. d
Google, a platinum sponsor of ACL-HLT, provided a
surge of fun on the first few days of summer in Portland.
Posted by Ciprian Chelba, Research Scientist
Add a comment