IRList Digest Sunday, 1 Sep 1985 Volume 1 : Issue 5 Today's Topics: Query - Bibliographies on Representation of Knowledge Discussion - Teaching Law & AI/DB/IR Techniques Announcement - Report on Univ. Regina Workshop on Adaptive Inf. Proc. - Fast DB Support in Mu-Prolog References - Recent Articles, Bibliography Call for Papers - Intelligent Tutoring Systems, J. AI Jobs - AI Work in Education, Expert System Design ---------------------------------------------------------------------- Date: Mon, 8 Jul 85 00:00:19 cdt From: Mark Turner Subject: representation of knowledge [Copied from AIList Digest Volume 3 : Issue 90 - Ed] I am gathering for my students a bibligraphy of works on representation of knowledge. I am particularly concerned with cognitive psychology, artificial intelligence, philosophy, linguistics, and natural language processing. I would appreciate receiving copies of bibliographies others may already have on-line. Mark Turner Department of English U Chicago 60637 >ihnp4!gargoyle!puck!mark ------------------------------ From: ihnp4!utzoo!lsuc!dave@UCB-VAX Date: 7 Jul 85 13:10:14 CDT (Sun) Subject: AI techniques for teaching tax law? [Forwarded from AI-ED. Also could be of interest to legal IR people. - Ed] ... I've developed a fully-operating CAI course on Canadian income tax law which is used to teach Ontario's law students. This course uses no AI techniques at all. I believe that within certain domains (such as tax) and subject to certain limitations, one can design a surprisingly "intelligent" CAI system without AI. My CAI runs with an interpreter from disk files, and it's trivial to modify the disk files to catch wrong student answers which have been flagged by the system as unrecognized. After plenty of iteration with test students, therefore, I now have a system which "recognizes" and correctly responds to almost every imaginable wrong answer which students will try for a given question. Sample: ++++++++++ Computer: [long fact situation] How much tax is payable by X? Student: $1,500 Computer: Sorry, that's wrong. You're miscalculating the capital gain. Remember to take the cost base into account, and don't forget that only one-half of the gain is taxable! ++++++++++ Having said that, I do think AI has a role to play in my course and I'd like to explore using it in future lessons. In particular, I'd like to develop a mini-simulation using Prolog, and let the student play with variables or with input and output values and examine the results. Does anyone out there use Prolog models for CAI in this kind of way? The other advantage AI techniques would have over my present code is that questions could have randomly-different numbers instead of being the same each time through. However, it must still be possible to easily write tailored answers. In the above example, I would need a simple way to indicate "1500" as "(a+b)*d", where the correct answer is known to be "(a+b-c)*d/2". It must also be possible to generate only those fact situations (i.e., numbers) which are (a) easy for students to compute the answers to, and (b) unambiguous in terms of different answers - that is, 1500 in the example above must not be able to reflect two different chains of wrong reasoning. Perhaps, given these limitations, it would be better to hand-code a set of possible values for a, b, c and d rather than have the computer generate them. Any ideas out there? Has anyone done this kind of thing? Dave Sherman The Law Society of Upper Canada Toronto ------------------------------ From: "V.J. Raghavan" Date: Wed, 14 Aug 85 15:47:30 cst Subject: new wkshp rprt WORKSHOP REPORT A workshop was organized and conducted on June 10-11, 1985 at the University of Regina, Regina, Canada. It was entitled "Workshop On Foundations Of Adaptive Information Processing" and was sponsored by the Faculty of Graduate Studies and Research and the Computer Science Department of our University. The impetus for the workshop came as result of the following eventualities: (i) There has been a great surge of interest in the recent years on investigating research progress in AI and their implications for computer science research in general, (ii) In a number of sub-disciplines of computer science such as Information Retrieval, Pattern Recognition, Expert systems and Decision Support, there is a striking similarity in the manner in which the problems of interest are formulated, (iii) There are several faculty members in the computer science department at University of Regina whose research interests overlap information retrieval, AI, expert systems and so on, (iv) The ACM - SIGIR annual conference was in Montreal, Canada in the first week of June and several individuals, coming to that conference, had expressed a desire to visit University of Regina, and (v) The dean of our Graduate Studies and Research faculty was enthusiastic in providing financial support for this venture. The total attendance at the workshop was 20, including faculty members and graduate students of our department and eight guests who agreed to participate upon our invitation. There were, in all, twelve presentations each lasting for approximately forty minutes. Four of the twelve presentations were by faculty members of our computer science department. There was ample opportunity for informal discussions and exchange of ideas. This contributed greatly to the success of the workshop. The participants had many varied backgrounds. This was in line with the main emphasis of the workshop, which was to compare and contrast the various problem formulations in the disciplines represented by the participants and to discuss models and theoretical foundations upon which solutions to these problems are based. Following is a list of individuals who participated and the titles of their presentations. * Research into fuzzy extensions of information retrieval D. Kraft, Louisiana State University * Problems of introducing a Boolean structure into probabilistic retrieval A. Bookstein, University of Chicago * Adaptive Boolean information retrieval T. Radecki, Louisiana State University and Tech. University of Wroclaw * On metric data models and associated search strategies L. Goldfarb, Univ. of New Brunswick * A unified model for information retrieval M. Wong, Univ. of Regina * Effectiveness of genetic algorithm for document redescription M. Gordon, The Univ. of Michigan * Cluster analysis and genetic algorithm in machine learning L.A. Rendell, Univ. of Illinois at Urbana-champaign * A system for detecting "deep" and "shallow" concepts in simple definitions M. Janta-Polczynski, Univ. of Regina * A concept-learning information retrieval system - basic ideas W. Ziarko, Univ. of Regina * A generalized retrieval facility for management decision support J.S. Deogun, Univ. of Nebraska-Lincoln * Two axioms for performance evaluation of information retrieval systems P. Bollmann, Tech. University of Berlin * Some thoughts on adaptive clustering for information retrieval V.V. Raghavan, Univ. of Regina Dr. Jeff Sampson of Univ. of Alberta was to give a presentation entitled "Genetic algorithms - A class of adaptive search procedures". But to our great disappointment and deep sorrow, we learned that he passed away, the week earlier, while travelling in France. It is expected that a proceedings of the workshop would be published in the near future. In addition, a summary of each presentation is due to appear over next several issues of the SIGIR Forum. Workshop Chairman Vijay Raghavan University of Regina ------------------------------ Date: 05 Jul 85 13:45:05 +1000 (Fri) From: John Shepherd Subject: New Deductive Database System [Forwarded from PROLOG Digest Volume 3 : Issue 30 - Ed] A System for Very Large Deductive Databases using a Superimposed Codeword Indexing Scheme ===== This note is to announce the (near) availability of a deductive database system suitable for dealing with very large databases of Prolog rules. The indexing scheme used by the system is based on the method of two-level superimposed codewords as described in [1], which allows partial match retrieval. Superimposed codeword schemes provide a very efficient method of retrieving records from large databases in only a small number of disk accesses. Further, the access method can be tuned so that the ratio of "false matches" can be reduced by an arbitrary amount (with a corresponding increase in storage costs). Unlike many earlier systems, this system supports the storage and retrieval of completely general Prolog terms, including functors and variables, and it is even possible to store Prolog rules in the database. The system is in the final stages of development under Berkeley Unix (4.2BSD) and has already been interfaced to the MU-Prolog system[2,3]; it will be incorporated into release 3.2db of MU-Prolog which will be available soon for Unix and VMS. It is being developed as part of the Machine Intelligence Project at the University of Melbourne on a Pyramid 90x which was loaned to the project by Pyramid Technology in Australia. The figures given below are taken from the Pyramid running version 2.3.1 of the OSx operating system (in the Berkeley universe) with one 400 Mb disk. Preliminary tests, on a database of mail transfer pathways through Usenet containing one million facts, have been very encouraging. To store these facts, which have an average length of 60 bytes, required just over 80Mb, which means a storage overhead of about 30%. In the present system, with a one-million record database indexed on three attributes, the rate of insertion is six records per CPU second. The rate of insertion could be significantly increased if the system were run as a single-user batch- type system without locking controls. Specifying just two of the fields (each record contains four fields), retrieved on average just 3 records for a query which had only one correct answer. The system can achieve a record retrieval rate of around 1000 records per second for a query on highly clustered records, to about 50 records per second for a query on unclustered records, even on this large database; for smaller databases, even faster rates are achievable. A query with complete information, required on average 1.1 retrievals, and required 4 disk accesses (excluding overheads from the Unix file system). This system overcomes some of the limitations of the Unix file system. For example, it overcomes the limit of twenty open files per process by caching on Unix file descriptors, thus allowing several database relations to be accessed simultaneously. The system also provides data buffering to reduce the number of file opens and data reads. The processing time of logic programs such as the "ancestor" relation can be minimised by this feature. We would be interested in hearing from other groups who are developing similar systems. For further information on this system, contact Dr. K. Ramamohanarao (Rao) or John Shepherd at the following addresses: HardMail: Department of Computer Science University of Melbourne Parkville, Victoria, 3052 AUSTRALIA SoftMail: UUCP: {seismo,ukc,prlb2}!munnari!jas {decvax,eagle,pesnta}!mulga!jas (SLOW) ARPA: munnari!jas@seismo.ARPA CSNET: jas@munnari.oz ("jas" can be substituted by "rao") Also, Dr. Rao will be attending the Logic Programming Symposium in Boston, and would be willing to discuss the system there. [1] R.Sacks-Davis and K.Ramamohanarao "A Two Level Superimposed Coding Scheme for Partial Match Retrieval", Information Systems, v.8, n.4, 1983 [2] By way of comparison, this system eliminates a number of restrictions which were associated with the deductive database system provided with release 3.1 of MU-Prolog. That system implemented the database manager as a separate process from the Prolog interpreter, communicating via Unix pipes. The present system is designed as a library package which is compiled into the host system; it could be incorporated fairly easily into most Prolog interpreters, or, in fact, into any systems that wished to perform partial match retrieval. The use of pipes in the old MU-Prolog system placed severe limitations (because of Unix file descriptor limitations) on the number of transactions (queries) which could be active concurrently; the new system has eliminated this restriction. Finally, this system lifts the restriction that only ground facts could be stored in the database, by allowing the storage of arbitrary Prolog terms (including rules). [3] L.Naish "MU-Prolog 3.2db Reference Manual", Technical Report, Department of Computer Science, University of Melbourne, 1985. ------------------------------ Date: 10 Aug 1985 02:25-EST From: leff%smu.csnet@csnet-relay.arpa Subject: Recent Articles, bibliography ... [Extracted from messages in AIList Digest Volume 3: #109 - Ed] %A Barbara Kellam-Scott %T Harvard Law School Computerizes the Paper Chase %J Hardcopy %D JUL 1985 %P 19 %V 14 %N 7 %K DEC %X Harvard Law School is automating their Legal Services Clinics. They have plans to include an expert system to assist lawyers in handling these cases. Digital Equipment has contributed to this program. %A Clara Y. Cuadrado %A John L. Cuadrado %T Prolog Goes to Work %J MAG4 %P 151-159 %K Symbolics Al Despain Yale Patt Berkely %X Al Despain and Yale Patt of Berkeley have achieved 425,000 LIPS using a custom designed processor. Symbolics has achieved 100,000 LIPS using custom microcode. Discusses general issues of Prolog in the contex of a maze traversing system. Also discusses the Japanese Fifth Generation project. %A F. Bouille %T The 'HBDS' Database Model Kernel of a Structured Data Base System. Making Databases Work %J IEEE Proceedings of Trends and Applications %D 1984 %P 324-331 %A V. P. Kobler %T Overview of Tools For Knowledge Base Construction %J International Conference on Data Engineering %I IEEE %C Los Angeles, Ca %D 1984 %P 282-285 %A C. Maioli et al. %T Prototypes of Expert Systems for a Friendly Man Machine Interaction. User Terminals for Information/Communication Systems %J 31st International Congress on Electronics. Proceedings %D 1984 %P 35-42 %A Z. L. Rabinovich %T Machine Intelligence and Fifth Generation Computer Structures %J Cybernetics %V 20 %N 3 %D MAY-JUN 1985 %P 426 %A W. Dilger %A W. Womannn %T The METANET: A Means for the Specification of Semantic Networks as Abstract Datatypes %J International Journal of Man-Machine Studies %V 21 %N 6 %D DEC 1984 %P 463 %A Michael B. First %A Lynn J. Soffer %A Randolph A. Miher %T QUICK (quick Index to Caduceus Knowledge) Using the Internish/Cadaceus Knowledge Base as an Electronic Textbook of Medicine %J Computers and Biomedical Research %V 18 %N 2 %D APR 1985 %P 137 ------------------------------ From: Ken Laws Date: Fri 9 Aug 85 13:48:26-PDT Subject: JAI Issue on Intelligent Tutoring Systems >From CACM, August 1985: [Forwarded from AI-ED. - Ed] Papers on intelligent tutoring systems are sought for a special issue of the Journal of Artificial Intelligence. Topics appropriate for this issue include knowledge representations tailored for use in an Intelligent Tutoring System (ITS); architectures for ITSs; methods for building student models; methods for diagnosing student's bugs and misconceptions; tutoring strategies; the use of natural language; design of the human-computer interface; case studies of ITSs; and psychological research relevant to the constructions of ITSs. Manuscripts should be submitted by February 15, 1986, to one of the guest editors: Elliot Soloway, Department of Computer Science, Yale University, P.O. Box 2158, New Haven, CT 06520; or William Clancy, Stanford Knowledge Systems Laboratory, 701 Welch Road, Building C, Palo Alto, CA 94304. ------------------------------ From: dartvax!creare!pbb%creare%dartmouth.csnet@CSNET-RELAY Date: Fri, 2 Aug 1985 09:02 est Subject: AI Work in ED [Forwarded from AI-ED. - Ed] AI Work in ED at Creare The following describes ongoing work at Creare (KREE-`ARE-EE) in the field of AI in education and ends with a call for consultant help. We at Creare are writing a phase II proposal for the design of an Expert System which will be used for discovering and describing scientific giftedness in students, primarily in the Junior High age group. The Expert System will try to detect the presence of giftedness by matching computer administered test results against knowledge based "feature models". We will develop a set of knowledge bases to be used in the adaptive testing environment where the selection of questions and the interpretation of answers are dynamically compared to a set of rule based models of gifted scientific thinking. The administration of questions will be controlled by another knowledge base which has the goal of efficiently determining which model best conforms to the testing results. These models of giftedness do not really exist yet, and we are not aware of any "experts" currently in this field. We hope that we can develop expertise in this area as our models are augmented and refined. We are working with education and testing specialists to achieve this goal. We are currently looking for help with the overall design of the the proposed Expert System. Experienced knowledge-base system designers are being sought to help review our design and to become part of the project should Creare be awarded the phase II grant. Anyone who might be interested and qualified, particularly those reasonably close to New Hampshire, should contact: Phil Bowman Creare, Inc. Box 71 Hanover, NH 03755 (603) 643-3800 dartvax!creare!pbb