IRList Digest Saturday, 7 January 1989 Volume 5 : Issue 5 Today's Topics: Discussion - IR collections and copyright releases Call for Papers - Bar-Ilan Symposium on the Foundations of AI COGSCI - Bootstrapping one-sided learning - Reason maintenance, Bilingual lexicon, Information content News addresses are Internet: fox@vtopus.cs.vt.edu BITNET: foxea@vtcc1.bitnet (replaces foxea@vtvax3) ---------------------------------------------------------------------- Date: Wed, 14 Dec 88 23:09 EST From: LEWIS@UMass.bitnet Subject: ir test collections and copyright Ed: I was recently at the Workshop on Evaluation of Natural Language Processing Systems, and there was considerable interest there in obtaining of IR test collections as sources of text. In particular, Mitch Marcus (formerly of Bell Labs, now at U Penn) is starting up a Treebank project to collect and tag a wide variety of corpora. He's very interested in obtaining some of the IR test collections, but wants to make sure he either has a copyright release or very good reason to believe the text is in the public domain. I said I'd look into this for him, and thought that you might be familiar with the legal situation on test collections, as a result of putting together the CD-ROM. Could you fill me in on this? Best, Dave ; BITNET: lewis@umass ; INTERNET: lewis@cs.umass.edu [Note:Well, I believe copyright law allows people to set up bibliographic collections on their own, so the Time collection you sent us from Cornell is probably fine. ACM allows use of the CACM collection, the ISI collection was obtained by going to library and keying in and so is probably OK as a bibliographic collection. LISA from Peter Willett is with their permission. INSPEC is not on my disc since we had a special license arrangement. Tefko Saracevic from Rutgets obtained permissions for his collection. Other data on Virginia Disc One has various permissions from different sources. Hope this helps, Ed. PS please pass on to M. Marcus and let me have his email address.] ------------------------------ From: "Prof. Yaacov Choueka" Date: Wed, 4 Jan 89 18:59:13 +0200 Subject: AI Symposium Bar-Ilan Symposium on the Foundations of Artificial Intelligence 19-21 June 1989 Sponsored by the Research Institute for the Mathematical Sciences Bar-Ilan University, Ramat Gan, Israel Symposium Chair: Martin Golumbic Organizing Chair: Ariel Frank Program Committee: Yaacov Choueka (Bar-Ilan University) Rina Dechter (Technion) Ariel Frank (Bar-Ilan University) Martin Golumbic (IBM Israel Scientific Center) David Harel (Weizmann Institute) Daniel Lehmann (Hebrew University) Judea Pearl (UCLA) Uri Schild (Bar-Ilan University) Micha Sharir (New York University) Jonathan Stavi (Bar-Ilan University) Bar-Ilan University, through its Center for Applied Logic and Artificial Intelligence (CALAI) of the Research Institute for the Mathematical Sciences, is pleased to announce its first "Symposium on the Foundations of Artificial Intelligence" to be held June 19-21, 1989. The Symposium will be international in scope, with invited lectures by several leading researchers from Israel and abroad. Although a small meeting is anticipated, with selected speakers and no parallel sessions, an attempt will be made to open attendance to all interested research scientists. The Bar-Ilan Symposium on the Foundations of Artificial Intelligence is intended to become a bi-annual event which will focus on a range of topics of concern to scholars applying quantitative, combinatorial, logical, algebraic and algorithmic methods to AI areas as diverse as decision support, automatic reasoning, knowledge-based systems, machine learning, computer vision, and robotics. These include applied logicians, algorithms and complexity researchers, AI theorists, and applications specialists using mathematical methods. By sponsoring such symposia, we hope to influence the spawning of new areas of applied mathematics and the strengthening of the scientific underpinnings of artificial intelligence. .................... INVITED LECTURES ...................... Ron Rivest (MIT) will lecture on "Recent Developments in Machine Learning Theory". Joe Halpern (IBM Research) will lecture on "Reasoning about Knowledge and Probability". Additional invited speakers will be announced at a later date. .................... CALL FOR PAPERS ....................... High quality research papers are solicited for consideration by the program committee to be presented at the Symposium. Submissions of extended abstracts of 4-10 pages or full papers must arrive by 15 March 1989 and should be sent in triplicate to: Prof. Martin Golumbic IBM Israel Scientific Center Technion City Haifa, Israel Decisions on presentations will be made on or before 15 April 1989. .................. REFEREED PROCEEDINGS .................... At the conclusion of the Symposium, all participants are invited to submit full length papers which will be refereed according the usual standard of the best professional journals, and those accepted will be published in a separate, special issue of the Annals of Mathematics and Artificial Intelligence as a permanent record of the Symposium. For further information on the Symposium and to receive additional announcements, contact Dr. Ariel Frank, BISFAI-89 Department of Mathematics and Computer Science Bar-Ilan University Ramat Gan, ISRAEL (email: ariel@bimacs.bitnet) ------------------------------ Date: Tue, 13 Dec 88 11:15:45 EST From: Peter de Jong Subject: Cognitive Science Calendar Reply-To: cog-sci-request@WHEATIES.AI.MIT.EDU cog-sci-calendar digest [Note: Extract - Ed.] Date: Mon, 12 Dec 88 15:27:37 EST From: menke (Baiba Menke) Subject: CRCT Colloquium Series Presents: COLLOQUIUM Bootstrapping One-sided Learning Professor Manfred Warmuth University of California, Santa Cruz Visiting Harvard University Wednesday, December 14, 1988 4:00 PM Aiken Lecture Hall 101 (Tea in Aiken Main Lobby at 3:30) Abstract This introduces a new framework for constructing learning algorithms. Our methods involve a master algorithm which uses learning algorithms for any intersection closed concept classes as subroutines. This master algorithm is capable of learning any concept class whose members can be expressed as the nested differences of concepts from an intersection closed class. Steven Salzberg from Harvard has been using a version of our algorithm (discovered independently) for predicting, among other things, breast cancer data. In some cases, his method outperforms the best previously known learning algorithms. However his results are only empirical. The crux of the type of research that we present here is that using the methodology of computational learning theory (which is rooted in the earlier works of Vapnik and others) we can give efficient algorithms and in some cases prove their optimality. We can show that our algorithm is optimal or nearly optimal with respect to several different criteria. These criteria include: the number of examples needed to produce a good hypothesis with high confidence, the worst case total number of mistakes made, and the expected number of prediction errors made in the first $t$ trials. We also present and analyze an algorithm for the intersection closed class consisting of submodules of $Z^k$. This algorithm learns with an absolute mistake bound within a $\log \log n$ factor of the theoretical optimum. Applications of this algorithm include a learning version of the word problem for finitely generated abelian groups, and learning a subset of the permutation invariant regular languages for which the minimum DFA consistency problem cannot be approximated by any polynomial.\\[\medskipamount] Host: Professor L.G. Valiant ------------------------------ Date: Thu, 15 Dec 88 09:52:33 EST From: Peter de Jong Subject: Cognitive Science Calendar Reply-To: cog-sci-request@WHEATIES.AI.MIT.EDU cog-sci-calendar digest [Note - Extracts - Ed.] Date: Tue, 13 Dec 88 12:00:36 EST From: reiter@harvard.harvard.edu Subject: Harvard AI Colloquim - Drew McDermott A General Mechanism for Reason Maintenance Professor Drew McDermott Department of Computer Science Yale University Monday, December 19, 1988 4 PM Aiken Computation Laboratory 101 (Tea at 3:30 pm Aiken Main Lobby) Abstract Several sorts of reason-maintenance (aka ``truth'' maintenance) systems have been built, distinguished by whether negation is permitted, whether context switching requires relabeling, how contradiction is handled, and whether nonmonotonicity is allowed. Several technical and technological issues must be solved in order to combine the features of all of these systems. Here is one solution: Let dependencies be clauses (as McAllester does); allow literals of the form ``Lp'' (p is definitely true) to provide for nonmonotonicity; label literals with boolean combinations of assumptions (deKleer), allowing assumptions to be marked as absent (McDermott). The resulting system automatically mimics Doyle's mechanism for dependency-directed backtracking. If you want nogoods too, it can be shown that nonmonotonic premises never enter into them, so that nogoods correspond to classical clauses. These get added to the clause network, thus subsuming some of deKleer's special propagation rules under standard McAllester-style boolean propagation. The new clauses never add ``odd loops'' that would break the nonmonotonic mechanism. The talk will include a discussion of practical applications. Host: Prof. Barbara Grosz ------------------------------ Date: Wed 14 Dec 88 17:53:45-EST From: Marc Vilain Subject: BBN AI Seminar -- Igal Golan BBN Science Development Program AI Seminar Series Lecture AN ACTIVE BILINGUAL LEXICON FOR MACHINE TRANSLATION Igal Golan IBM Scientific Center, Haifa, Israel (GOLAN%ISRAEARN.BITNET@CUNYVM.CUNY.EDU) BBN Labs 10 Moulton Street 2nd floor large conference room 10:30 am, Tuesday December 20 An approach to the Transfer phase of a Machine Translation system is presented, where the bilingual lexicon plays an active role, guiding Transfer by means of executable descriptions of word senses. The means for lexical sense specification are, however, general enough, and can in principle apply to other system architectures, e.g. in the Generation phase if Transfer is intentionally kept minimal. The active lexicon is the only system component which is exposed to users and can serve to linguistically control Transfer effects. A unified approach to lexicon creation and maintenance is proposed, which contains means to gradually refine sense specification and tailor the definitions to specific text domains. The underlying linguistic principles, the nature of sense distinction required for translation, and the formal structure of the lexicon are discussed. ----------------------------- Date: Wed 14 Dec 88 17:54:34-EST From: Marc Vilain Subject: BBN AI Seminar -- David Israel BBN Science Development Program AI Seminar Series Lecture VARIETIES OF CONTENT: INFORMATIONAL VS. SEMANTIC; PURE VS. INCREMEMENTAL David Israel SRI International (ISRAEL@Warbucks.AI.SRI.COM) BBN Labs 10 Moulton Street 2nd floor large conference room 10:30 am, Thursday December 22 In this talk, I will present an informal exposition of a theory of information content due to John Perry and myself, and apply some of the notions and distinctions central to that theory to some issues about the semantics of singular reference in natural language. ------------------------------ END OF IRList Digest ********************