FLAIRS-96 Special Track Real-World Natural Language Understanding Natural language processing (NLP) has been a central topic of study in artificial intelligence and cognitive science for decades. Recently, there has been much interest in developing robust natural language processing systems. These systems go beyond the micro-worlds of limited complexity that many early systems dealt with and confront the issue of building scaled-up systems capable of dealing with "real" language. We call this capability "Real-World Natural Language Understanding" (RNLU). Two aspects of RNLU set it apart from traditional NLP, as denoted by the terms "real-world" and "understanding". First, RNLU concentrates on real-world texts (as opposed to researcher-generated sentences). Language processing is situated in the world of real texts, allowing RNLU to leverage off certain characteristics of such texts which provide a basis for scale-up. These characteristics include: o Redundancy. This enables the reader to be less than perfect as information is likely to appear in multiple locations. o Active engagement of the text by the reader. This means that the reader engages the text to the fullest level possible, given the reader's interest, cognitive load, and current goals. This also allows the reader to skim or skip material that is not relevant to the reader's goals. o Narrative agreement. The author of a text is assumed to be attempting to communicate a set of ideas to the reader. The text provides numerous affordances for the goal of comprehension. The other important characteristic of RNLU is "understanding." RNLU strives for some high level of comprehension from the texts being read. Exactly what this level needs to be varies from task to task, but simple syntactic parsing or keyword-based information extraction is usually insufficient for real-world natural language communication at a level comparable to human language abilities. RNLU exploits a range of information present in natural language texts, as well as background information known to the author and reader, in order to achieve a suitable depth of comprehension of the text. This track will bring together researchers in several areas, including traditional NLP, statistical NLP and situated NLP, with a common goal of building AI systems that can achieve real comprehension of actual texts. Multiple disciplines and perspectives will be represented, providing a unique opportunity to exchange ideas and advance the field. The track will highlight theoretical research in natural language processing as well as practical attempts to build scaled-up natural language processing systems. Finally, insofar as RNLU defines both the promise of NLP as well as its next challenge, this track will also help to better define the boundaries and potential of RNLU. Call for papers Interested researchers are invited to send an extended abstract of no more than 2000 words to kennethm@cc.gatech.edu. Accepted authors will be asked to submit a five-page paper detailing the work. Format of the track The track will be run as two technical sessions of presentations, each followed by an invited speaker acting as a general discussant. Group discussion will be strongly encouraged. The sessions will be organized either as talk sessions or as focussed poster sessions, depending on the number of accepted submissions. In either case, after the presentations, authors will sit on a panel for a general discussion led by the invited speaker. Important Dates o Oct 16, 1995 -- Abstract submission deadline o December 1995 -- Notices mailed to accepted/rejected authors o Mar 18, 1996 -- Final (camera ready) papers due o May 20 - 22, 1996 -- Conference dates Organizing Committee o Ashwin Ram, Georgia Institute of Technology (co-chair) http://www.cc.gatech.edu/cogsci/faculty/ram.html o Kenneth Moorman, Georgia Institute of Technology (co-chair) http://www.cc.gatech.edu/cogsci/students/Ai-students/kennethm/moorman.html o Eugene Charniak, Brown University http://www.cs.brown.edu/people/ec/home.html o Scott Huffman, Price Waterhouse Technology Centre http://ai.eecs.umich.edu/people/huffman/huffman.html o Yael Ravin, T.J. Watson Research Center o Chris Riesbeck, Northwestern University http://www.ils.nwu.edu/~e_for_e/people/riesbeck.html o Stuart C. Shapiro, SUNY Buffalo http://www.cs.buffalo.edu/pub/WWW/faculty/shapiro/shapiro.html HTML sites to note For more information on FLAIRS-96, see: FLAIRS-96 home page http://www.cis.ufl.edu/~ddd/FLAIRS/FLAIRS-96/index.html For more information on this special track, see: Real-world natural language understanding home page http://www.cc.gatech.edu/cogsci/conferences/flairs-rnlu-96.html