IRList Digest Friday, 14 November 1986 Volume 2 : Issue 59 Today's Topics: Query - Availability of Pisa Proceedings (1986 Int'l Conf. on R&D in IR)? - Use of neural networks for IR? Seminars - Schedule for Information Systems Research Group in Nova Scotia COGSCI - Learning by failing to explain CSLI - Head-Phrase Structure Grammar Theory and Research - Relation-changing Affixes and Homonymy News addresses are ARPANET: fox%vt@csnet-relay.arpa BITNET: foxea@vtvax3.bitnet CSNET: fox@vt UUCPNET: seismo!vtisr1!irlistrq ---------------------------------------------------------------------- Date: Wed, 29 Oct 86 13:16:38 pst From: Michael Shepherd Subject: a bunch of things Ed: ...I missed this year's sigir conference is Pisa, is it possible to purchase the Proceedings? ... [Note: I just called ACM. The order number is 606860 and the title is "Proc. 9th Annual International SIGIR Conf. on R&D in Information Retrieval." Cost is $18 ($24 nonmembers) and there is $3 shipping, from ACM Orders Department P.O. Box 64145 Baltimore MD 21264 They just received the copies in the warehouse yesterday! Regards, Ed] ------------------------------ Date: Fri, 7 Nov 86 11:57:29 -0100 From: Wyle Subject: RE: citations, conspectus ... I am building a neral network model and exploring the possibilities of using it to index text. Do any readers of IR list digest know of work in this area? ~~~~~~~~~~~~~~~~~~~~~~ Mitch Wyle ...!decvax!seismo!mcvax!cernvax!ethz!Wyle Institut fuer Informatik Wyle%ifi.ethz.cernvax. ETH / SOT 8092 Zuerich, Switzerland "Sic itur ad astra." [Note: in issue 28 was abstract of an article about using the Connection Machine for information retrieval. - Ed] ------------------------------ Date: Fri, 31 Oct 86 05:10:44 pst From: Michael Shepherd Subject: ir in Nova Scotia Ed: The following is a list of the seminars given in our Information Systems Research Group. If you want, I can periodically send you this information. ================== An informal "Information Systems Research Group" (ISRG) has been formed in Halifax, Nova Scotia, Canada. The purpose of this inter-disciplinary group is to provide a forum for people to present their research, discuss common interests, and to stimulate joint research efforts. As of November 28, 1986, four seminars will have been held. The following is a brief description of these seminars. Date: October 3, 1986 Title: Managing Views of Large Databases that are Tailored to Individual Users Speaker: Dr. Carl Hartzman Computing Science Division Dept. of Math., Stats., & Comp. Sci. Dalhousie University Halifax, Nova Scotia, Canada B3H 3J5 Abstract: Either for reasons of security or in order to avoid the necessity of dealing with large amounts of irrelevant data, users of large databases have restricted views of the database computed for them. These views are not permanently resident in the computer system. Thus, any update that the user may wish to make to the information appearing in his view must be translated to an update on the underlying database from which the view is computed, in such a way that the computation of the view yields exactly the updated information. Generally, if there are any translations that will accomplish this, there are more than one, each reflecting a different update policy. This profusion of translations yields the risk of putting wrong or inconsistent data into the database. Thus, database management systems have not typically permitted view updates. The question we examine here is the possibility of designing databases so that desired view updates have unique correct translations. Date: October 17, 1986 Title: Integrating User Profiles into Expert Systems for Bibliographic Retrieval Speaker: Dr. Michael Shepherd Computing Science Division Dept. of Math., Stats., & Comp. Sci. Dalhousie University Halifax, Nova Scotia, Canada B3H 3J5 Abstract: The purpose of the proposed research is to investigate knowledge representations for expert online bibliographic retrieval systems incorporating the profile-query relationship. The knowledge representations must be hospitable to user profiles and queries and permit the definition and implementation of a profile-query relationship. These knowledge representations must be able to represent the knowledge within a bibliographic retrieval system that permits online access to thousands and millions of documents. The representations must permit the user to query the knowledge- base as well as the database and the system must be able to make inferences from the knowledge-base, the profile, and the profile- query relationship. Date: November 14, 1986 Title: A Parallel Architecture for a Database Machine Speakers: Dr. W. Robertson and Dr. P. Bodorik School of Computer Science Technical University of Nova Scotia Halifax, Nova Scotia, Canada B3J 2X4 Abstract: This talk outlines the authors' approach to an architecture which is, in their opinion, eminently suitable for a database machine. An investigation of the characteristics of various interconnection topologies lead the authors to the conclusion that a dual-bus hypercube is probably the most suitable for a modular architecture for data base operations. The bottleneck at the mass storage interface is eliminated by the introduction of an interleaved memory structure at the interface. Under the assumption that a relation of interest is uniformly distributed over some number, m, of disc cylinders, and that all cylinders can be accessed concurrently, the disc to memory bandwidth is balanced. The maximum number of discs in the system is W**2, for a three cube, where W is the number of nodes connected to each bus of the hypercube. This structure has an advantage in that it requires a smaller number of connections per node and it is also easily extended to increase mass storage and/or computational capacity by increasing W and/or the hypercube dimension. The functional requirements of the nodes and the bus protocol are currently under investigation. Date: November 28, 1986 Title: A Framework and Prototype for Intelligent Information Retrieval Researchers: Dr. Vijay V. Raghavan Center for Advanced Computer Studies University of Southwestern Louisiana Lafayette, Lousiana 70504-4330 Dr. Ravi S. Sharma Dept. of Mathematics and Computing Sciences St. Francis Xavier University Antigonish, Nova Scotia, Canada B2G 1C0 Speaker: Dr. Ravi S. Sharma Abstract: Information Retrieval (IR) is the process of selectively disseminating relevant information stored among a variety of information items. A scheme for efficient (in terms of computer resources used) and effective (in terms of recall and precision) information retrieval is proposed. This draws from rough set theory and an adaptive clustering strategy in order to achieve some amount of self-organization in the database. The relevance judgements of weighted queries are used as the basis for influencing the classification of documents, thus enabling user input to direct clustering and allow the clustering sub-system to learn. An evaluation metric based on the theory of rough sets is used as the clustering criterion, which is enhanced in an iterative fashion. The clustering algorithm begins with an arbitrary placement of the documents in the collection on a one-dimensional linear space and, as each query arrives, uses the relevance judgements to generate movement along the line so that "similar" documents are moved closer together and "dissimilar" ones moved further apart from each other. The precise definitions of what is meant by similar and dissimilar are also based on ideas from rough set theory. After a number of queries are processed, the initial positions change as the system "learns" the users' profiles. The classification, however, is re-clustered only when this movement is deemed significant as defined by a variation of a common measure for determining the similarity between classifications. The general method of clustering places boundaries on gaps so as to optimize the clustering criterion. Efficient and effective retrieval is sought by using a learning scheme known as retrieval-by-example. The prototype experimental implementation of the scheme indicates remarkable potential, producing recall and precision results of between 40 to 80 per cent. ------------------------------ Date: Mon, 20 Oct 86 03:10:09 edt From: DEJONG%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU Subject: Cognitive Science Calendar [Extract - Ed] Date: Friday, 17 October 1986 15:04-EDT From: JHC at OZ.AI.MIT.EDU subject: revolving seminar (for real this time) Thursday, 23 October 4:00pm Room: NE43-8th floor playroom LEARNING BY FAILING TO EXPLAIN Robert Joseph Hall MIT Artificial Intelligence Laboratory Explanation-based Generalization depends on having an explanation on which to base generalization. Thus, a system with an incomplete or intractable explanatory mechanism will not be able to generalize some examples. It is not necessary, in those cases, to give up and resort to purely empirical generalization methods, because the system may already know almost everything it needs to explain the precedent. Learning by Failing to Explain is a method which exploits current knowledge to prune complex precedents and rules, isolating their mysterious parts. This paper describes two techniques for Learning by Failing to Explain: Precedent Analysis, partial analysis of a precedent or rule to isolate the mysterious new technique(s) it embodies; and Rule Re-analysis, re-analyzing old rules in terms of new rules to obtain a more general set. ------------------------------ Date: Mon, 20 Oct 86 03:11:12 edt From: EMMA@CSLI.STANFORD.EDU Subject: CSLI Calendar, October 16, No. 3 [Extract - Ed] HPSG Theory and HPSG Research Ivan Sag October 23, 1986 This seminar presents an overview of the central ideas under development by members of the CSLI HPSG project. Head-Driven Phrase Structure Grammar is an information-based theory of the relation between syntactic and semantic structure. The syntactic concepts of HPSG evolved from Generalized Phrase Structure Grammar (GPSG) in the course of the last few years through extensive interaction with members of the CSLI FOG project. HPSG integrates key ideas of GPSG with concepts drawn from Kay's Functional Unification Grammar and Categorial Grammar and incorporates certain analytic techniques of Lexical-Functional Grammar. The semantic concepts of HPSG are a hybrid of Situation Semantics and the theory of thematic roles. Current HPSG theory embodies a number of important design properties: monotonicity, declarativeness and reversibility, yet current HPSG analyses require extensions of such standard frameworks as PATR-II. Current research ideas will be surveyed, as well as ongoing work on the hierarchical structure of the HPSG lexicon. ------------------------------ Date: Fri, 24 Oct 86 02:18:08 edt From: EMMA@CSLI.STANFORD.EDU Subject: CSLI Calendar, October 23, No. 4 [Extract - Ed] Relation-changing Affixes and Homonymy Abdelkader Fassi-Fehri October 28, 12:30, Trailer Classroom, CSLI Of special relevance to a natural theory of affixation are the following questions: a) What is the exact nature of the changes that a lexical unit undergoes as the result of an affixation process (role or argument reduction or increase, valency reorganization, etc.), and which level of representation is the most appropriate to state these changes? b) Given that languages use different systems of homonymic forms of affixes to express different valencies (or the same valency organized in different ways), is there a possible account which will predict which homonymy affixation would be natural, and which one would be accidental? We propose a theory of lexical organisation that answers these questions. ------------------------------ END OF IRList Digest ********************