2-Sep-84 21:50:00-PDT,14081;000000000000 Mail-From: LAWS created at 2-Sep-84 21:44:48 Date: Sun 2 Sep 1984 21:37-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #113 To: AIList@SRI-AI AIList Digest Monday, 3 Sep 1984 Volume 2 : Issue 113 Today's Topics: Humor - Eliza Passes Turing Test (again), AI Tools - Taxonomy Assistant & User Aids, Psychology - User Modeling, Speech Recognition - Separating Syllables, Conferences - Functional Languages ---------------------------------------------------------------------- Date: 29 Aug 84 18:22:02-PDT (Wed) From: decvax!minow @ Ucb-Vax.arpa Subject: Eliza Passes Turing Test (again) Article-I.D.: decvax.59 Readers of net.ai might enjoy this extract from "Computing Across America," Chapter 11: A High-tech Oasis in the Texas Sun, written by Stephen K. Roberts, published originally in (and Copyright 1984 by) Online Today, September 1984 (a CompuServ Publication). The Phantom SysOp (Austin SYSOP speaking) "Personally, I get a little tired of answering CHAT requests. That's why I brought up Eliza." "You mean..." He twinkled with wry humor. "You got it. It's the perfect Turing test. I have a second computer hooked up to my board system. When someone issues a CHAT request, it says 'Hello? Can I help you?' I changed all the messages so it emulates a sysop instead of a psychiatrist. Some people never do catch on." I groaned. "That's cruel!" (Transcribed by my collegue, John Wasser.) Martin Minow decvax!minow ------------------------------ Date: 31 Aug 1984 11:31:36 PDT From: Bill Mann Subject: reply to McGuire about taxonomy assistant (Reply to Wayne McGuire's comments on the need for a taxonomy assistant: ) I agree with the notion that representing and using conceptual relations effectively is one of the central problems of AI research. You say "It seems to me that in the knowledgebase management systems which I hope we will see developed in the near future will be embedded rich resources for evoking and representing taxonomies. Semantic nets provide an ideal scheme with which to do just that." How do we know that semantic nets are so good? Isn't this a complex unsolved problem, for which the effectiveness of semantic nets is still an open issue? I suspect that semantic nets are useful for these problems just as binary notation is useful. The representative power is there, but success depends not so much on the distinctive properties of nets as on the techniques that create and use the nets. I agree that they look promising. (Promises, promises.) You suggest that a taxonomy assistant might work by operating on the vocabulary of the domain, relating items. That sounds like another promising idea that might lead to a very powerful set of generalizations if it was tried. In the case that prompted all this, there is no recognized domain or literature. I have an experimental program which includes a specialized internal interface language having several hundred predefined operators. Doing a taxonomy is one way to increase my understanding of the interface language. So I would like to have a taxonomy assistant that did not have to presume a domain. Bill Mann ------------------------------ Date: 31 Aug 1984 12:20:08 PDT From: Bill Mann Subject: aids to the mind I've gotten several interesting replies to my inquiry about finding a "taxonomy assistant" that could help me in thinking about the organization of a collection of items. It raises a larger issue: What intellectual operations are worth developing programmed aids for? Nobody came up with a pointer to an existing program for the taxonomy task (except for something named PEGASUS, on the related topic of vocabulary construction; I need to check it out.) But still, there might be other sorts of programmed assistants out there. Here is a wish list for programmed assistants that could potentially be important for my lifestyle: RESOURCE ALLOCATION ASSISTANT: Given a supply or a flow of resources, help allocate them to particular uses. Budgeting, personal time allocation and machine scheduling are special cases. TIME ALLOCATION ASSISTANT: (a specialization, very important to me) Help work through allocation of my time so that feasible things get done, infeasible things don't get started, the things that get done are the important ones, things tend to get done on time, allocations get revised appropriately in the face of change, and the allocation provides suitable flexibility and availability to other people. I have in mind here much more than just the scratchpad-and-alarm-clock kind of time allocation resource. Those are fine as far as they go, but they don't go nearly deep enough. I want something that will ask me the pertinent questions when they are timely. EXPOSITORY WRITING ASSISTANT: In this case, my research on text generation has gone far enough to assure me that such a program is feasible. I have a systematic manual technique that works pretty well for me, that could be developed into an interactive aid. It would be very different from the sentence-critic sort of programs that are now emerging. NEGOTIATION ASSISTANT: There is a viewpoint and a collection of skills that are very helpful in bargaining to an agreement. A program could raise a lot of the right questions. *** That is just a starter list. What other sorts of assistants can we identify or conceive of? Other ideas can probably be developed from the problem-solving literature, e.g. Polya, Wickelgren and Lenat. This sort of thing could go far beyond the capabilities of autonomous AI programs. Often there are well known heuristics that are helpful to people but too indefinite for programs to apply; an assistant could suggest them. Proverbs are one sort. In sum, What do we want, and What do we have? Bill Mann ------------------------------ Date: 29 Aug 84 14:21:56-PDT (Wed) From: hplabs!hpda!fortune!amd!decwrl!dec-rhea!dec-bartok!shubin @ Ucb-Vax.arpa Subject: Replies to query for citations on user modeling Article-I.D.: decwrl.3473 I posted a request for some papers on modeling users and/or user behavior, and promised to post the results of my query. (The original posting was on or about 18 July 84). Here is a summary of the results; a line of hyphens separates one person's response from another. I haven't had to check all of them, and I may wind up with more references, which may be posted later. Any more suggestions are welcome. Thanks to all. ------ Elaine Rich, "Users are Individuals: Individualizing User Models" Int.J.Man-Machine Studies 18(3), March, 1983. Zog project at CMU Elliot Soloway at Yale -- modeling novice programmer behavior "The Psychology of Human-Computer Interaction" by Card, Moran, and Newell Current work at Penn by Tim Finin, Ethel Shuster, and Martha Pollock at UT at Austin by Elaine Rich Work on on-line assistance: Wizard program by Jeff Shrager and Tim Finin (AAAI 82) Integral help by Fenchel and Estrin Lisp tutor - John Anderson at CMU ------ Regarding users' models of computer systems: a. Schneiderman, B. and Meyer R. "Syntactic/Semantic Interactions in Programmer Behaviour: A Model and Experimental Results" Int. J. of Computer and Information Sciences, Vol 8, No. 3, 1979 b. Caroll, J.M., and Thomas, J.C. "Metaphor and the Cognitive Representation of Computing Systems" IEEE Trans. on Systems, Man, and Cybernetics, Vol SMC - 12, No. 2, March/April 1982. c. Anything from the CHI'83 conference -- Human Factors in Computing Systems sponsored by ACM. About Modelling the User: a. Card, Newell and Moran, a book whose title escapes me offhand -- it has a chapter entitled The human Information Processor. b. Rich, E. "Users are Individuals: Individualizing user Models" Int. J. Man-Machine Studies 18, 1983 -------- Peter Polson (U.COlorado) and David Kieras (U.Arizona) have a paper in this year's Cognitive Science Conference on a program that tests user interfaces by testing high-level descriptions of user behavior against expected system behavior. -------- There was a lot of work done at Xerox PARC in the late 70's on task times and such. They were interested in work relating to I/O device design (mice, etc.), as well as general models. Some very good task timing models came out of that work, I believe. ------- Take a look at the work of Elaine Rich at Texas (formerly CMU). ------- Chapter 6,The Psychology of Human-Computer Interaction,SK Card, SP Moran,A Newell ------- ...Some of the results of this are published in the 1983 AAAI Proceedings in the paper "Learning Operator Semantics by Analogy" by S. Douglas and T. Moran. "A Quasi-Natural Language Interface to UNIX"; S. Douglas; Proceedings of the USA-Japan Conference on Human-Computer Interaction; Hawaii; 18-20 Aug 84; Elsevier. ------------------------------ Date: 31 Aug 84 13:05:13-PDT (Fri) From: ihnp4!houxm!mhuxl!ulysses!burl!clyde!watmath!utzoo!dciem!mmt @ Ucb-Vax.arpa Subject: Re: Hearsay II question in AIList Digest V2 #110 Article-I.D.: dciem.1098 It turns out that even to separate the syllables in continuous speech you need to have some understanding of what the speaker is talking about! You can discover this for yourself by trying to hear the sounds of the words when someone is speaking a foreign language. You can't even repeat them correctly as nonsense syllables. I used to believe this myth myself, but my various visits to Europe for short (1-3 week periods, mostly) trips have convinced me otherwise. There is no point trying to repeat syllables as nonsense, partly because the sounds are not in your phonetic vocabulary. More to the point, syllable separation definitely preceded understanding. I HAD to learn to separate syllables of German long before I could understand anything (I still understand only a tiny fraction, but now I can parse most sentences into kernel and bound morphemes because I now know most of the common bound ones). My understanding of written German is a little better, and when I do understand a German sentence, it is because I can transcribe it into a visual representation with some blanks. (Incidentally, I also do some research in speech recognition, so I am well aware of the syllable segmentation problem. There do exist segmentation algorithms that correctly segment over 95% of the syllables in connected speech without any attempt to identify phonemes, let alone words or the "meaning" of speech. Mermelstein, now in Montreal, and Mangold in Ulm, Germany, are names that come to mind.) -- Martin Taylor {allegra,linus,ihnp4,floyd,ubc-vision}!utzoo!dciem!mmt {uw-beaver,qucis,watmath}!utcsrgv!dciem!mmt ------------------------------ Date: Wed 29 Aug 84 10:53:53-EDT From: Joseph E. Stoy Subject: Call For Papers CALL FOR PAPERS FUNCTIONAL PROGRAMMING LANGUAGES AND COMPUTER ARCHITECTURE A Conference Sponsored by The International Federation for Information Processing Technical Committees 2 and 10 Nancy, France 16 to 19 September, 1985 This conference has been planned as a successor to the highly successful conference on the same topics held at Wentworth, New Hampshire, in October 1981. Papers are solicited on any aspect of functional or logic programming and on computer architectures to support the efficient execution of such programs. Nancy, in the eastern part of France, was the city of the Dukes of Lorraine; it is known for its "Place Stanistlas" and its "Palais Ducal". "Art Nouveau" started there at the beginning of this century. There are beautiful buildings and museums and, of course, good restaurants. Authors should submit five copies of a 3000 to 6000-word paper (counting a full page figure as 300 words), and ten additional copies of a 300-word abstract of the paper to the Chairman of the Programme Committee by 31 January 1985. The paper should be typed double spaced, and the names and affiliations of the authors should be included on both the paper and the abstract. Papers will be reviewed by the Programme Committee with the assistance of outside referees; authors will be notified of acceptance or rejection by 30 April 1985. Camera-ready copy of accepted papers will be required by 30 June 1985 for publication in the Conference Proceedings. Programme Committee: Makoto Amamiya (NTT, Japan) David Aspinall (UMIST, UK) Manfred Broy (Passau University, W Germany) Jack Dennis (MIT, USA) Jean-Pierre Jouannaud (CRIN, France) Manfred Paul (TUM, W Germany) Joseph Stoy (Oxford University, UK) John Willliams (IBM, USA) Address for Submission of Papers: J.E. Stoy, Balliol College, Oxford OX1 3BJ, England. Paper Deadline: 31 January 1985. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ To receive a copy of the advance programme, return the following information to J.E. Stoy, Balliol College, Oxford OX1 3BJ, England or by electronic mail to JESTOY@UCL-CS.ARPA I plan to submit a paper: [ ] Subject: Name: Organisation: Address: ------------------------------ End of AIList Digest ******************** 5-Sep-84 09:44:18-PDT,16974;000000000001 Mail-From: LAWS created at 5-Sep-84 09:35:42 Date: Wed 5 Sep 1984 09:20-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #114 To: AIList@SRI-AI AIList Digest Wednesday, 5 Sep 1984 Volume 2 : Issue 114 Today's Topics: LISP - LISP for the Eclipse 250 with RDOS, Expert Systems - AGE Contact? & Programmed Assistants, Speech Understanding - Word Recognition, Philosophy - Now and Then, Seminars - Bay Area Computer Science, Conference - IJCAI-85 Call for Papers ---------------------------------------------------------------------- Date: 4 Sep 1984 9:00-EDT From: cross@wpafb-afita Subject: LISP for the Eclipse 250 with RDOS I recently joined a group here doing low level pattern recognition work applied to speech recognition and image processing. We have an Eclipse 250 running the RDOS operating system. We also have C (unix version 7 compatable). Does anyone out there know of a dialect of LISP that can be used with this system? Any suggestions? Please respond to the address listed below. Thanks in advance. Steve Cross cross@wpafb-afita ------------------------------ Date: 4 Sep 84 10:29 PDT From: Feuerman.pasa@XEROX.ARPA Subject: AGE: Who to contact? I'm interested in looking into AGE, which is quoted as being a "Stanford product." Does anyone have the name and phone number of who to contact to obtain things, such as manuals, users guides, etc. Thanks in advance. --Ken . ------------------------------ Date: Wed, 5 Sep 84 15:59 BST From: TONY HASEMER (on ALVEY at Teddington) Subject: programmed assistants In response to Bill Mann's list of desirable mechanised assistants, one of our graduate students urgently wants to know: if he drops everything else and writes a thesis-writing assistant, will he get a PhD for it? Tony Hasemer. ------------------------------ Date: 4 Sep 84 09:56 PDT From: Feuerman.pasa@XEROX.ARPA Subject: Understanding speech vs. hearing words The subject has come up about whether one need understand the semantics of an utterance before one can recognize words, or even syllables. While it seems a bit of research has been cited for both sides, I thought it would be interesting to offer an experience of mine for evidence: I was travelling in Italy, and it was that time of the evening again, time to find our daily ration of gelato (Italian ice cream)! Our search brought us into a bar of sorts, with Paul Simon's (I think it was Paul Simon) recording of "Slip Sliding Away" playing in the background. The bartender was singing along, only it didn't quite come out right. What he was singing was more like "Sleep Sliding Ayway" (all of the vowels being rather exagerated). I regret that I had no way of knowing whether he had seen the words written down before (which could account for some of his mis-pronunciations), but it was pretty clear that he had no idea of the meaning of what he was singing. --Ken. [It seems to me that the same sort of anecdote could be told of any child; they frequently store and repeat phrases that are to them merely nonsense (e.g., the alphabet, especially LMNOP). More to the point, a good first step in learning any new oral language is to listen to it, sans understanding, long enough to begin to identify syllables. This greatly simplifies later word drills since the student can then grasp the phonetic distinctions that the teacher considers important (and obvious). The implication for speech understanding is that it is indeed possible to identify syllables without understanding, but only after some training and the development of fairly sophisticated discriminant capabilities. -- KIL] ------------------------------ Date: Fri, 31 Aug 84 15:53 BST From: TONY HASEMER (on ALVEY at Teddington) Subject: Now and Then (Tony Hasemer challenges Norm Andrews' faith about cause and effect) You say: "logical proof involves implication relationships between discrete statements...causality assumes implication relationships between discrete events". Don't think me just another rude Brit, but:- > in what sense is a statement not an event A statement (as opposed to a record of a statement, which is a real-world object) takes place in the real world and therefore is an event in the real world. > what do you mean by "implication" This is the nub of all questions about cause and effect, and of course the word subsumes the very process it tries to describe. One can say "cause and effect", or "implication", or "logically necessary", and mean ALMOST the same thing in each case. They all refer to that same intangible feeling of certainty that a certain argument is valid or that event B was self-evidently caused by event A. > what do you mean by "relationship" Again, this is a word which presumes the existence of the very link we're trying to identify. May I suggest the following- The deductive logical syllogism (the prototype for all infallible arguments) is of the form All swans are white. This is a swan. Therefore it is white. Notice that the conclusion (3rd sentence) is only true iff the two premises (sentences 2 and 3) are true. And if you can make any descriptive statement beginning "All..." then you must be talking about a closed system. Mathematics, for example, is a set of logical statements about the closed domain of numbers. It is common, but on reflection rather strange, to talk about "three oranges" when each orange is unique and quite different from the rest. It is clear that we impose number systems on the real world, and logical statements about the square root of the number 3 don't tell us whether or not there is a real thing called the square root of three oranges. I'm saying that closed systems do not map onto the real world. Mathematics doesn't, and nor does deductive logic (you could never demonstrate, in practice, the truth of any statement about ALL of a class of naturally-occurring objects). On the contrary, the only logic which will in any sense "prove" statements about the real world (such as that the sun will rise tomorrow) is INDUCTIVE logic. Inductive logic and the principle of cause and effect are virtually synonymous. Inductive logic is fuzzy (deductive logic is two-valued), and bootstraps itself into the position of saying: "this must be true because it would be (inductively) absurd to suppose the contrary". There is no real problem, no contradiction, between the principle of cause and effect and deductive logic. There is merely a category mistake. The persuasive power of deduction is very appealing, but to try to justify an inductive argument (e.g. causality) by the criteria of deductive arguments is like trying to describe the colour red in a language which has no word for it. We just have to accept that in dealing with the real world the elegant and convenient certainties of the deductive system do not apply. The best logic we have is inductive: if I kick object A and it then screams, I assume that it screamed BECAUSE I kicked it. If repeated kicking of object A always produces the concomitant screams, I have two choices: either to accept the notion of causality, or to envisage the real world as being composed of a vast series of arbitrary possibilities, like billions of tossed pennies which only by pure chance have so far happened always to come down heads. Personally, I much prefer a fuzzy, uncertain logic to a chaos in which there is no logic at all! Belief in causality, like belief in God, is an act of faith: you can't hope to PROVE it. But whichever one chooses, it doesn't really matter: stomachs still churn and cats still fight in the dark. The very best solution to the problem of causality is to stop worrying about it. Tony. ------------------------------ Date: 04 Sep 84 1424 PDT From: Yoni Malachi Subject: Seminars - Abstracts for BATS [Forwarded from the Stanford bboard by Laws@SRI-AI.] The next Bay Area Theory Seminar (aka BATS) will be at Stanford, this Friday, 7 September. The talks (and lunch) will take place in Room 200-305. This is a room on the third floor of History Corner, the NE corner of the Stanford Campus Quadrangle. The schedule: 10:00am U. Vazirani (Berkeley): "2-Processor Scheduling in Random NC" 11:00am R. Anderson (Stanford): "A P-complete Problem and Approximations to It" noon: Lunch 1:00pm E. Lawler (Berkeley): "The Traveling Salesman Problem Made Easy" 2:00pm A. Schoenhage (Tuebingen, IBM San Jose): "Efficient Diophantine Approximation" ***************************************************************************** ABSTRACTS: 10:00am: U. Vazirani: "2-Processor Scheduling in Random NC" (joint work with D. Kozen and V. Vazirani) The Two-Processor Scheduling Problem is a classical problem in Computational Combinatorics, and several efficient algorithms have been designed for it. However, these algorithms are inherently sequential in nature. We give a randomizing poly-log time parallel algorithm (run on a polynomial number of processors). Interestingly enough, our algorithm for this purely combinatoric-looking problem draws on some powerful algebriac methods. The Two-processor Scheduling problem can be stated as follows: Given a set S of unit time jobs, and a partial order specifying precedence constraints among them, find an optimal schedule for the jobs on two identical processors. 11:00am: R. Anderson (Stanford): "A P-complete Problem and Approximations to It" The P-complete problem that we will consider is the High Degree Subgraph Problem. This problem is: given a graph G=(V,E) and an integer k, find the maximum induced subgraph of G that has all nodes of degree at least k. After showing that this problem is P-complete, we will discuss two approaches to finding approximate solutions to it in NC. We will give a variant of the problem that is also P-complete that can be approximated to within a factor of c in NC, for any c < 1/2, but cannot be approximated by a factor of better than 1/2 unless P=NC. We will also give an algorithm that finds a subgraph with moderately high minimum degree. This algorithm exhibits an interesting relationship between its performance and the time it takes. 1:00pm: E. Lawler (Berkeley): "The Traveling Salesman Problem Made Easy" Despite the general pessimism resulting from both theory and practice, the TSP is not necessarily a hard problem--there are many interesting and useful special cases that can be solved efficiently. For example, there is an efficient procedure for finding an optimal solution for the bottleneck TSP in the case that the distance matrix is "graded." This result will be used to show how to solve a problem of great practical importance to paperhangers: how to cut sheets from a long roll of paper so as to minimize intersheet wastage. Material for this talk is drawn from a chapter, by P. Gilmore, E.L. Lawler, and D.B. Shmoys, of a forthcoming book, The Traveling Salesman Problem, edited by Lawler, J.K. Lenstra, A.H.G. Rinnooy Kan, and D.B. Shmoys to be published by J. Wiley in mid-1985. 2:00pm: A. Schoenhage (Tuebingen, IBM San Jose): "Efficient Diophantine Approximation" Abstract: Given (a_1,...,a_n) in R^d (with d < n) and epsilon > 0, how to find a nontrivial x = (x_1,...,x_n) in Z^n of minimal Euclidean norm nu such that |x_1 a_1 + ... + x_n a_n| < epsilon holds. A weak version of this classical task (where epsilon and nu may be multiplied by 2^(cn) ) can be solved in time O(n^2 (d*n/(n-d) * log(1/epsilon))^(2+o(1))). The main tool is an improved basis reduction algorithm for integer lattices. ------------------------------ Date: Tue 4 Sep 84 09:27:09-PDT From: name AAAI-OFFICE Subject: IJCAI-85 Call for Papers IJCAI-85 CALL FOR PAPERS The IJCAI conferences are the main forum for the presentation of Artificial Intelligence research to an international audience. The goal of the IJCAI-85 is to promote scientific interchange, within and between all subfields of AI, among researchers from all over the world. The conference is sponsored by the International Joint Conferences on Artificial Intelligence (IJCAI), Inc., and co-sponsored by the American Association for Artificial Intelligence (AAAI). IJCAI-85 will be held at the University of California, Los Angeles from August 18 through August 24, 1985. * Tutorials: August 18-19; Technical Sessions: August 20-24 TOPICS OF INTEREST Authors are invited to submit papers of substantial, original, and previously unreported research in any aspect of AI, including: * AI architectures and languages * AI and education (including intelligent CAI) * Automated reasoning (including theorem proving, automatic programming,plan- ning, search, problem solving, commensense, and qualitative reasoning) * Cognitive modelling * Expert systems * Knowledge representation * Learning and knowledge acquisition * Logic programming * Natural language (including speech) * Perception (including visual, auditory, tactile) * Philosophical foundations * Robotics * Social, economic and legal implications REQUIREMENTS FOR SUBMISSION Authors should submit 4 complete copies of their paper. (Hard copy only, no electronic submissions.) * LONG PAPERS: 5500 words maximum, up to 7 proceedings pages * SHORT PAPERS: 2200 words maximum, up to 3 proceedings pages Each paper will be stringently reviewed by experts in the topic area specified. Acceptance will be based on originality and significance of the reported research, as well as the quality of its presentation. Applications clearly demonstrating the power of established techniques, as well as thoughtful critiques of previously published material will be considered, provided that they point the way to new research and are substantive scientific contributions in their own right. Short papers are a forum for the presentation of succinct, crisp results. They are not a safety net for long paper rejections. In order to ensure appropriate refereeing, authors are requested to specify in which of the above topic areas the paper belongs, as well as a set of no more than 5 keywords for further classification within that topic area. Because of time constraints, papers requiring major revisions cannot be accepted. DETAILS FOR SUBMISSION The following information must be included with each paper: * Author's name, address, telephone number and net address (if applicable); * Topic area (plus a set of no more than 5 keywords for further classification within the topic area.); * An abstract of 100-200 words; * Paper length (in words). The time table is as follows: * Submission deadline: 7 January 1985 (papers received after January 7th will be returned unopened) * Notification of Acceptance: 16 March 1985 * Camera Ready copy due: 16 April 1985 Contact Points Submissions should be sent to the Program Chair: Aravind Joshi Dept of Computer and Information Science University of Pennsylvania Philadelphia, PA 19104 USA General inquiries should be directed to the General Chair: Alan Mackworth Dept of Computer Science University of British Columbia Vancouver, BC, Canada V6T 1W5 Inquiries about program demonstrations (including videotape system demonstrations) and other local arrangements should be sent to the Local Arrangements Chair: Steve Crocker The Aerospace Corporation P.O. Box 92957 Los Angeles, CA 90009 USA Inquiries about tutorials, exhibits, and registration should be sent to the AAAI Office: Claudia Mazzetti American Association for Artificial Intelligence 445 Burgess Drive Menlo Park, CA 94025 USA ------------------------------ End of AIList Digest ******************** 7-Sep-84 10:47:53-PDT,13735;000000000001 Mail-From: LAWS created at 7-Sep-84 10:45:11 Date: Fri 7 Sep 1984 10:27-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #115 To: AIList@SRI-AI AIList Digest Friday, 7 Sep 1984 Volume 2 : Issue 115 Today's Topics: LISP - QLAMBDA & Common Lisp, Expert Systems - AGE Contact & Expository Writing Assistant, Books - Lib of CS and the Handbook of AI, AI Tools - Statistical Workstations and Time-Series Lisp, Binding - Jim Slagle, Speech Recognition - Semantics, Philosophy - Induction vs. Deduction & Causality, Seminars - A Calculus of Usual Values & Week on Logic and AI ---------------------------------------------------------------------- Date: Thu, 6 Sep 84 8:54:58 EDT From: "Ferd Brundick (VLD/LTTB)" Subject: QLAMBDA Does anyone have any information on a new LISP called QLAMBDA ?? It is a "parallel processor" language being developed by McCarthy at Stanford and is supposed to run on the HEP (Heterogeneous Element Processor). Since we have one of the original HEPs, we are interested in any information regarding QLAMBDA. Thanks. dsw, fferd Fred S. Brundick USABRL, APG, MD. ------------------------------ Date: 13 Aug 84 8:21:00-PDT (Mon) From: pur-ee!uiucdcsb!nowicki @ Ucb-Vax.arpa Subject: Re: Common Lisp - (nf) Article-I.D.: uiucdcsb.5500009 I am also interested in such info. We have Sun-2's running 4.2 and I am interested in obtaining Common Lisp for them. -Tony Nowicki {decvax|inuxc}!pur-ee!uiucdcs!nowicki ------------------------------ Date: Wed 5 Sep 84 17:28:19-CDT From: Charles Petrie Subject: AGE Call Juanita Mullen at (415)497-0474 for a good time in obtaining Stanford programs such as AGE. It'll cost you about $500. CJP ------------------------------ Date: 5 Sep 1984 14:08:13 PDT From: Bill Mann Subject: Clarification on the non-existence of the Expository Writing Assistant I've gotten several inquiries asking for the Expository Writing Assistant Program that I wished for in a previous message. Unfortunately, it doesn't exist. I'm convinced from studying text generation that we have ENOUGH TECHNICAL INFORMATION about the structure of text, the functions of various parts and how parts are arranged that such a program could be written. My own writing practise, which now in effect simulates such a program, indicates that the program's suggestions could be very helpful. An introduction to the text structures I have in mind was presented at the 1984 ACL/Coling conference at Stanford in July. The paper was entitled "Discourse Structures for Text Generation." Right now I have no plans to create the assistant. Sorry, folks. Bill Mann ------------------------------ Date: 4 Sep 84 16:36:13-PDT (Tue) From: ihnp4!houxm!vax135!cornell!uw-beaver!ssc-vax!adcock @ Ucb-Vax.arpa Subject: Re: Lib of CS intro offer: Handbook of AI Vols 1-3 for $5 Please note that the Handbook of AI is a REFERENCE book. It is not meant to be read from cover to cover. Also, this is the only books on AI that the Lib of CS sells. [I disagree with the first point. The Handbook is also an excellent tutorial, although it does lack illustrations. I enjoyed reading it cover to cover (although I admit to not having finished all three volumes yet). The second point is largely true, although they have offered The Brains of Men and Machines, Machine Perception, LISPcraft, and a few other related books. -- KIL] ------------------------------ Date: Fri 7 Sep 84 10:15:02-PDT From: Ken Laws Subject: Statistical Workstations and Time-Series Lisp (Tisp) Anyone interested in statistical workstations should look up the August IEEE Computer Graphics and Applications article "A Graphical Interface to an Economist's Workstation" by Thomas Williams of Wagner, Stott and Company, 20 Broad Street, New York, NY 10005. He describes a prototype for time-series analysis that was quickly assembled from standard Interlisp-D functions on the Xerox 1108. Apparently the economists of the International Monetary Fund took to it immediately, and Williams sees no problems in extending its capabilities to better support them. His company is also working on a workstation for professional securities traders. -- Ken Laws ------------------------------ Date: 5 Sep 1984 13:42-EDT From: Russ Smith Subject: Binding - Jim Slagle As of September 10, 1984 Dr. Slagle will have a new address: Professor James R. Slagle University of Minnesota 136 Lind Hall 207 Church Street, S.E. Minneapolis, MN 55455 (612) 373-7513 (612) 373-0132 slagle%umn-cs.csnet@csnet-relay.arpa (possibly...) ------------------------------ Date: 5 Sep 84 10:00:24-PDT (Wed) From: ihnp4!fortune!polard @ Ucb-Vax.arpa Subject: Re: Understanding speech versus hearing words Article-I.D.: fortune.4138 [Phonemes for the line-eater. -- KIL] Which hip was burned? Which ship was burned? Which chip was burned? and Which Chip was spurned? all sound the same when spoken at the speed of conversational speech. This is evidence that in order to recognize words in continuous speech you (and presumably a speech-recognition apparatus) need to understand what the speaker is talking about. There seem to be two reasons why understanding is necessary for word recognition in continuous speech: 1. The existence of homonyms. This is why "It's a good read." sounds the same as: "It's a good reed," and why the two sentences could not be distinguished without a knowledge of the context. 2. Sandhi, or sound changes at word boundaries. The sounds at the end of a word tend to blend into the sounds at the beginning of the next word in conversation, making words sound as if they ran into each other and making words sound differently than they would when said in isolation. The resulting ambiguities are usually resolved by context. Speech rarely occurs without some sort of context, and even then the first thing that usually happens is to establish a context for what is to follow. To paraphrase Edsgar Dijkstra: "Asking whether computers will understand speech is like asking whether submarines swim." -- Henry Polard (You bring the flames - I'll bring the marshmallows.) {ihnp4,cbosgd,amd}!fortune!polard ------------------------------ Date: Wed 5 Sep 84 10:54:11-PDT From: BARNARD@SRI-AI.ARPA Subject: induction vs. deduction Tony Hasemer's comments on causality and its relationship to inductive versus deductive logic are very well-taken. It's time for people in AI to realize that deduction is quite limited as a mode of reasoning. Compared to induction, the mathematical foundations of deduction are well-understood, and deductive systems are relatively easy to implement on computers. This no doubt explains its popularity in AI. The problem arises when one tries to extend the deductive paradigm from toy problems to real problems, and must confront exceptions, borderline cases, and, in general, the boggling complexity of the state space. While deduction proceeds from the general (axioms) to the specific (propositions), induction proceeds from the specific to the general. This seems to be a more natural view of human intelligence. By observing events, one recognizes correlations, and infers causality and other relationships. To be sure, the inferences may be wrong, but that's tough. People make mistakes. In fact, one of the weaknesses of deduction is that it does not permit one to draw conclusions that may be in error (assuming the axioms are correct), but that represent the best conclusions under the circumstances. Visual illusions provide good examples. Have you ever wondered why you see a Necker Cube as a cube (one of the two reversals), and not as one of the other infinite number of possiblities? Perhaps we learn of cubes through experience (an inductive explanation), but the effect also occurs with totally unfamiliar figures. A more general inductive explanation holds that we see the simplest possible figure (the Gestalt principle of Pragnanz). A cube, which has right angles and equal-length sides, is simpler than any of the other possiblilities. The concept of "simple" can be made precise: one description is simpler than another if it can be encoded more economically. This is sometimes called the principle of Occam's Razor or the principle of Minimum Entropy. Steve Barnard ------------------------------ Date: 6 Sep 84 07:39 PDT From: Woody.pasa@XEROX.ARPA Subject: Causality Food for thought: All the arguments for and against cause and effect and the workings of Causality have been based around the notion that the cause 'A' of an effect 'B' are time-related: we assume that for A to affect B, A must come before B in our perseption of time. But does this have to be the case? Mathematics (inductive and deductive logic) are time-independent identities; by assuming that Causality may be a time-dependent phenomina on the basis of time-independent arguments is at best wishful thinking. What's wrong with event A affecting event B in event A's past? You can't go back and shoot your own mother before you were born because you exist, and obviously you failed. If we assume the universe is consistant [and not random chaos], then we must assume inconsistancies (such as shooting your own mother) will not arise. It does not, however, place time constrictions on cause and effect. - Bill Woody Woody.Pasa@XEROX.Arpa [Until 7 September 1984] ** No net address ** [After 7 September 1984] ------------------------------ Date: Fri, 7 Sep 84 00:35:19 pdt From: syming%B.CC@Berkeley Subject: Seminar - A Calculus of Usual Values From: chertok@ucbkim (Paula Chertok) Subject: Berkeley Cognitive Science Seminar--Sept. 11 COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A SPEAKER: L.A. Zadeh Computer Science Division, UC Berkeley TITLE: Typicality, Prototypicality, Usuality, Dispositionality, and Common Sense TIME: Tuesday, September 11, 11 - 12:30pm PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 The grouping of the concepts listed in the title of this talk is intended to suggest that there is a close connection between them. I will describe a general approach centering on the concept of dispositionality which makes it possible to formulate fairly precise definitions of typicality and prototypicality, and relate these concepts to commonsense reasoning. These definitions are not in the classical spirit and are based on the premise that typicality and pro- totypicality are graded concepts, in the sense that every object is typical or prototypical to a degree. In addition, I will outline what might be called a calculus of usual values. ------------------------------ Date: Thu, 6 Sep 84 16:45:49 edt From: minker@maryland (Jack Minker) Subject: WEEK ON LOGIC AND AI WEEK of LOGIC and its ROLE in ARTIFICIAL INTELLIGENCE at THE UNIVERSITY OF MARYLAND OCTOBER 22-26, 1984 The Mathematics and Computer Science Departments at the University of Maryland at College Park are jointly sponsoring a Special Year in Mathematical Logic and Theoretical Computer Science. The week of October 22-26 will be devoted to Logic and its role in Artificial Intelligence. There will be five distinguished lectures as follows: Monday, October 22: Ray REITER "Logic for specification: Databases conceptual models, and knowledge representation languages" Tuesday, October 23: John McCARTHY "The mathematics of circumscription" Wednesday, October 24: Maarten VAN EMDEN "Strict and lax interpretations of rules in logic programming" Thursday, October 25: Jon BARWISE "Constraint logic" Friday, October 26: Lawrence HENSCHEN "Compiling constraint checking programs in deductive databases" All lectures will be given at: Time: 10:00 AM - 11:30AM Location: Mathematics Building, Room Y3206 The lectures are open to the public. If you plan to attend kindly notify us so that we can make appropriate plans for space. Limited funds are available to support junior faculty and graduate students for the entire week or part of the week. To obtain funds, please submit an application listing your affiliation and send either a net message or a letter to: Jack Minker Department of Computer Science University of Maryland College Park, MD 20742 (301) 454-6119 minker@maryland ------------------------------ End of AIList Digest ******************** 10-Sep-84 09:46:25-PDT,10580;000000000001 Mail-From: LAWS created at 10-Sep-84 09:44:06 Date: Mon 10 Sep 1984 09:37-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #116 To: AIList@SRI-AI AIList Digest Monday, 10 Sep 1984 Volume 2 : Issue 116 Today's Topics: AI Tools - FRL in Franz, Robotics - Symbolic Programming Query, Psychology - Memory Tests, Knowledge Representation - OPS5 Problem, LISP - Delayed Reply About muLISP, Speech Recognition - Syllables, Philosophy - Correction, Expert Systems - Personal Assistants, Seminar - Semantic Modulation ---------------------------------------------------------------------- Date: 5 Sep 84 10:08:15-PDT (Wed) From: decvax!mcnc!duke!ucf-cs!bethel @ Ucb-Vax.arpa Subject: Help : need a Full implementation of FRL. Article-I.D.: ucf-cs.1468 Does anyone have a full implementation of Minsky's FRL, running under Unix 4.2 and Franz Lisp ? If so would you please respond and let me know where you are. I would like to get the sources if they are available and not protected by company/university policy. Thanks in advance, Robert C. Bethel ...decvax!ucf-cs!bethel or ...duke!ucf-cs!bethel bethel.ucf-cs@Rand-Relay ------------------------------ Date: 3 Sep 84 12:35:53-PDT (Mon) From: hplabs!hao!denelcor!csu-cs!walicki @ Ucb-Vax.arpa Subject: prolog/lisp/robotics - query Article-I.D.: csu-cs.2619 I am looking for information on applications of symbolic computing (lisp, prolog) in the area of robotics. I do not have any specifics in mind; I am interested in any (even fuzzy) intersections of the abovementioned domains. Please respond by mail, and I will post a summary in net.ai. Jack Walicki Colorado State U. Computer Science Dept. (Fort Collins, CO 80523) {hplabs,hao}!csu-cs!walicki ------------------------------ Date: 9 Sep 84 18:11:40 PDT (Sunday) From: wedekind.es@XEROX.ARPA Subject: Memory tests Someone I know is looking for a battery of well-documented, self-administered memory tests. Does anyone know of an accessible source? thank you, Jerry ------------------------------ Date: Saturday, 8-Sep-84 18:35:50-BST From: O'KEEFE HPS (on ERCC DEC-10) Subject: OPS5 problem An MSc student came to me with a problem. He had a pile of OPS5 rules and was a bit unhappy about the means he had adopted to stop them looping. Each rule looked rather like (p pain77 (task ^name cert) (injury ^name injury6 ^cert ) (symptom ^name symptom9 ^present yes) -(done pain77) --> (make done pain77) (modify 2 ^cert (compute ....)) ) There were dozens of them. The conflict resolution rule of never firing the same rule on the same data more than once didn't help, as modify is equivalent to a delete and a make. What he actually wanted can be expressed quite neatly in Prolog: candidates(BestToWorst) :- setof(W/Injury, weight(Injury, W), BestToWorst). weight(Injury, MinusCertainty) :- prior_certainty(Injury, Prior), findall(P, pro(Injury, P), Ps), product(Ps, 1.0, P), findall(C, con(Injury, C), Cs), product(Cs, 1.0, C), MinusCertainty is -(1 - P + P*C*Prior). pro(Injury, Wt) :- evidence_for(Injury, Symptom, Wt), present(Symptom). con(Injury, Wt) :- evidence_against(Injury, Symptom, Wt), present(Symptom). product([], A, A). product([W|Ws], A, R) :- B is A*W, product(Ws, B, R). We managed to produce something intermediate between these two, it used evidence-for and evidence-against tables in working memory, and had just two hacky rules instead of the scores originally present. I did spot a way of stopping the loop without using negation, and that is to make the "certainty" held in the (injury ^name ^cert) WM elements a gensym whose value is the desired number, then as far as OPS5 is concerned the working memory hasn't been changed. Of course that makes patterns that use the number harder to write, and seems rather hacky itself. To come to the point, I have two questions about OPS5. 1) Is there a clean way of coding this in OPS5? Or should I have told him to use EXPERT? 2) As I mentioned, we did manage to do considerably better than his first attempt. But the thing that bothered me was that it hadn't occurred to him to use the WM for tables. The course he's in uses the Teknowledge(??) "OPS5 Tutorial" (the one with the Wine Advisor) and students seem to copy the Wine Advisor more or less blindly. Is there any generally available GOOD course material on OPS5, and if so who do we write to? Are there any moderate-size examples available? ------------------------------ Date: 10 May 84 11:33:00-PDT (Thu) From: hplabs!hp-pcd!hp-dcd!hpfcls!hpbbla!coulter @ Ucb-Vax.arpa Subject: Delayed Reply About muLISP Article-I.D.: hpbbla.4900001 It may not be what you are looking for, but there are several LISP implementations that run on CP/M. I bought muLISP which is distributed by MICROSOFT. It cost $200. Because of its larger address space, you should be able to get a more capable LISP for the IBM/PC, but it will cost more. The muLISP is fairly complete, although the only data type is integer (it can represent numbers up to 10**255). The DOCTOR (a.k.a. ELIZA) program is supplied with it and it runs. ------------------------------ Date: Fri, 7 Sep 84 17:44 EST From: Kurt Godden Subject: understanding speech, syllables, words, etc. Which hip was burned? Which ship was burned? which chip was burned? Which Chip was spurned? First of all, I disagree that all 4 sound 'the same' in conversational speech, esp. the last. The final [z] in 'was' gets devoiced because of the voiceless cluster that follows in 'spurned'. However, of course I do agree that often/usually context is necessary to DISAMBIGUATE, tho' not necessarily to understand in the first place. Since I am already writing this I might as well give my originally suppressed comments on the first person's statement that syllable identification requires understanding. I definitely do not agree with that claim. Others have mentioned learning a foreign language by first tuning the ear to the phonetics of the target language including that target's syllable types, and this is a point well taken. The notion of syllable is certainly different in different lgs, but apparently can be learned without understanding. The point is even clearer in one's native language. We have all heard Jabberwockish type speech and can clearly recognize the syllables and phonetic elements as 'English', yet we do so without any understanding. All this assumes that we know just what a syllable is, which we don't, but that's another argument and is not really suitable for ailist. -Kurt Godden ------------------------------ Date: 7 Sep 84 9:13:41-PDT (Fri) From: ihnp4!houxm!vax135!ariel!norm @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: ariel.751 > > From: TONY HASEMER (on ALVEY at Teddington) > > (Tony Hasemer challenges Norm Andrews' faith about cause and effect) > > You say: "logical proof involves implication relationships between > discrete statements...causality assumes implication relationships > between discrete events". > Hold on here! I, Norm Andrews, didn't say that! You are quoting someone going by the name "Baba ROM DOS" who was attempting to disprove my statement that "The concept of proof depends upon the concepts of cause and effect, among other things." Please don't assign other peoples' statements to me! I haven't time now to reply to any other part of your posting... Norm Andrews ------------------------------ Date: 6 Sep 84 7:14:39-PDT (Thu) From: decvax!genrad!teddy!mjn @ Ucb-Vax.arpa Subject: Personal Assistants Article-I.D.: teddy.391 FINANCIAL ASSISTANT I think this would be a good one to add to the list of personal assistants which would be valuable to have. It could be a great aid to budgeting and guiding investments. It should go beyond simple bookkeeping and offer advice (when it can). If conficts arise in where to spend money, it should be capable of asking questions to determine what you consider to be more important. Additional functionality might include analysis of spending patterns. Where does my money go? Such a question could be answered by this assistant. It might include gentle reminders if you are overspending, or not meeting a payment schedual, or forget something. ------------------------------ Date: 7 Sep 1984 15:04-EDT From: Brad Goodman Subject: Seminar - Semantic Modulation [Forwarded from the MIT bboard by SASW@MIT-MC.] David McAllester will give the next BBN AI Seminar at 10:30 AM on Wednesday September 12. The talk is in the 3rd floor large conference room at 10 Moulton St. Title and abstract follow. Semantic Modulation: A New General Purpose Inference Technique David McAllester Massachusetts Institute of Technology Semantic modulation is a general purpose inference technique based on the "modulation" of the interpretations of parameters which appear free in an assertional data base. A semantic modulation system includes a finite and fixed set Delta of formulae. By varying the interpretation of the free parameters in Delta it is possible to use the finite and FIXED data base Delta to perform a large set of inferences which involve reasoning about quantification. Semantic modulation is a way of reasoning with quantifiers that does not involve unification or the standard techniques of universal instantiation. Semantic modulation replaces these notions with the notion of a "binding premise". A binding premise is a propositional assumption which constrains the interpretation of one or several free parameters. ------------------------------ End of AIList Digest ******************** 12-Sep-84 10:11:27-PDT,15788;000000000001 Mail-From: LAWS created at 12-Sep-84 10:08:20 Date: Wed 12 Sep 1984 10:01-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #117 To: AIList@SRI-AI AIList Digest Wednesday, 12 Sep 1984 Volume 2 : Issue 117 Today's Topics: AI Tools - Expert-Ease, Expert Systems - Lenat Bibliography, Pattern Recognition - Maximal Submatrix Sums, Cognition - The Second Self & Dreams, Seminars - Computational Theory of Higher Brain Function & Distributed Knowledge ---------------------------------------------------------------------- Date: 10 Sep 1984 13:16:20-EDT From: sde@Mitre-Bedford Subject: expert-ease I got a flyer from Expert Systems Inc. offering something called Expert Ease which is supposed to facilitate producing expert systems. They want $125 for a demo version, so I thought to inquire if anyone out there can comment on the thing, especially since the full program is $2000. I'm not eager to buy a lemon, but if it is a worthwhile product, it might be justifiable as an experiment. Thanx in advance, David sde@mitre-bedford ------------------------------ Date: Tue, 11 Sep 84 19:16 BST From: TONY HASEMER (on ALVEY at Teddington) Subject: Lenat Please can anyone suggest any good references, articles etc. concerning Lenat's heuristic inferencing machine I'd be very grateful. Tony. [I can suggest the following: D.B. Lenat, "BEINGS: Knowledge as Interacting Experts," Proc. 4th Int. Jnt. Conf. on Artificial Intelligence, Tblisi, Georgia, USSR, pp. 126-133, 1975. D.B. Lenat, AM: An Artificial Intelligence Approach to Discovery in Mathematics as Heuristic Search, Ph.D. Dissertation, Computer Science Department Report STAN-CS-76-570, Heuristic Programming Project Report HPP-76-8, Artificial Intelligence Laboratory Report SAIL AIM-286, Stanford University, Stanford, California, 1976. D.B. Lenat, "Automated Theory Formation in Mathematics," 5th Int. Jnt. Conf. on Artificial Intelligence, Cambridge, pp. 833-42, 1977. D.B. Lenat and G. Harris, "Designing a Rule System That Searches for Scientific Discoveries," in D.A. Waterman and F. Hayes-Roth (eds.), Pattern-Directed Inference Systems, Academic Press, 1978. D.B. Lenat, "The Ubiquity of Discovery," National Computer Conference, pp. 241-256, 1978. D.B. Lenat, "On Automated Scientific Theory Formation: A Case Study Using the AM Program," in J. Hayes, D. Michie, and L.I. Mikulich (eds.), Machine Intelligence 9, Halstead Press (a div. of John Wiley & Sons), New York, pp. 251-283, 1979. D.B. Lenat, W.R. Sutherland, and J. Gibbons, "Heuristic Search for New Microcircuit Structures: An Application of Artificial Intelligence," The AI Magazine, Vol. 3, No. 3, pp. 17-33, Summer 1982. D.B. Lenat, "The Nature of Heuristics," The AI Journal, Vol. 9, No. 2, Fall 1982. D.B. Lenat, "Learning by Discovery: Three Case Studies in Natural and Artificial Learning Systems," in Michalski, Mitchell, and Carbonell (eds.), Machine Learning, Tioga Press, 1982. D. B. Lenat, Theory Formation by Heuristic Search, Report HPP-82-25, Heuristic Programming Project, Dept. of Computer Science and Medicine, Stanford University, Stanford, California, October 1982. To appear in The AI Journal, March 1983. D. B. Lenat, "EURISKO: A Program that Learns New Heuristics and Domain Concepts," Journal of Artificial Intelligence, March 1983. Also available as Report HPP-82-26, Heuristic Programming Project, Dept. of Computer Science and Medicine, Stanford University, Stanford, California, October 1982. -- KIL] ------------------------------ Date: Wed 12 Sep 84 01:50:03-PDT From: Ken Laws Subject: Pattern Recognition and Computational Complexity I have a solution to Jon Bentley's Problem 7 in this month's CACM Programming Pearl's column (September 1984, pp. 865-871). The problem is to find the maximal response for any rectangular subwindow in an array of maximum-likelihood detector outputs. The following algorithm is O(N^3) for an NxN array. It requires working storage of just over half the original array size. /* ** maxwdwsum ** ** Compute the maximum rectangular-window sum in a matrix. ** Return 0.0 if all array elements are negative. ** ** COMMENTS ** ** This algorithm scans the matrix, considering for each ** element all of the rectangular subwindows with that ** element as the lower-right corner. The current best ** window will either be interior to the previously ** processed rows or will end on the current row. The ** latter possibility is checked by considering the data ** on the current row added into the best window of each width ** for each lower-right corner element on the previous row. ** ** The memory array for tracking maximal window sums could ** be reduced to a triangular data structure. An additional ** triple of values could be carried along with globalmax ** to record the location and width of the maximal window; ** saving or recovering the height of the window would be ** a little more difficult. ** ** HISTORY ** ** 11-Sep-84 Laws at SRI-AI ** Wrote initial version. */ /* Sample problem. (Answer is 6.0.) */ #define NROWS 4 #define NCOLS 4 float X[NROWS][NCOLS] = {{ 1.,-2., 3.,-1.}, { 2.,-5., 1.,-1.}, { 3., 1.,-2., 3.}, {-2., 1., 1., 0.}}; /* Macro to return the maximum of two expressions. */ #define MAX(exp1,exp2) (((exp1) > (exp2)) ? (exp1) : (exp2)) main() { float globalmax; /* Global maximum */ float M[NCOLS][NCOLS]; /* Max window-sum memory, */ /* (triangular, 1st >= 2nd) */ int maxrow; /* Upper row index */ int mincol,maxcol; /* Column indices */ float newrowsum; /* Sum for new window row */ float newwdwsum; /* Previous best plus new window row */ float newwdwmax; /* New best for this width */ int nowrow,nowcol; /* Loop indices */ /* Initialize the maxima registers. */ globalmax = 0.0; for (nowrow = 0; nowrow < NCOLS; nowrow++) for (nowcol = 0; nowcol <= nowrow; nowcol++) M[nowrow][nowcol] = -1.0E20; /* Process each lower-right window corner. */ for (maxrow = 0; maxrow < NROWS; maxrow++) for (maxcol = 0; maxcol < NCOLS; maxcol++) { /* Increase window width back toward leftmost column. */ newrowsum = 0.0; for (mincol = maxcol; mincol >= 0; mincol--) { /* Cumulate the window-row sum. */ newrowsum += X[maxrow][mincol]; /* Compute the sum of the old window and new row. */ newwdwsum = M[maxcol][mincol]+newrowsum; /* Update the maximum window sum for this width. */ newwdwmax = MAX(newrowsum,newwdwsum); M[maxcol][mincol] = newwdwmax; /* Update the global maximum. */ globalmax = MAX(globalmax,newwdwmax); } } /* Print the solution, or 0.0 for a negative array. */ printf("Maximum window sum: %g\n",globalmax); } ------------------------------ Date: Sat 8 Sep 84 11:14:04-MDT From: Stan Shebs Subject: The Second Self The Second Self by Sherry Turkle is an interesting study of the relationship between computers and people. In contrast to most studies I've seen, this is not a collection of sensationalism from the newspapers combined with the wilder statements from various professionals. Rather, it is (as far as I know) the first thorough and scientific study of the influence of computers on human thinking (there's even a boring appendix on methodology, for those who are into details). The book starts out with analyses of young children's attitudes towards intelligent games (Merlin, Speak'n'Spell and others). Apparently, the children playing with these games spend a great deal of time discussing whether these games are actually alive or not, whether they know how to cheat, and so forth. The games manifest themselves as "psychological machines" rather than the ordinary physical machines familiar to most children. As such, they prompt children to think in terms of mental behavior rather than physical behavior, which is said to be an important stage in early mental development (dunno myself if psychologists hold this view generally). The theme of computers as "psychological machines" is carried throughout the book. Older children and adolescents exhibit more of a desire to master the machine rather than just to interact with it, but interviews with them reveal that they, too, are aware of the computer as something fundamentally different from an automobile, in the way that it causes them to think. Computer hobbyists of both the first (ca 1978) and later generations are interviewed, and one of them characterizes the computer as "a tool to think with". Perhaps the section of most interest to AIList readers is the one in which Turkle interviews a number of workers in AI. Although the material has an MIT slant (since that's where she did her research), and there's an excess of quotes from Pam McCorduck's Machines Who Think, this is the first time I've seen a psychological analysis of motives and attitudes behind the research. Most interesting was a discussion of "egoless thought" - although most psychologists (and some philosophers) believe that the existence of self-consciousness and an ego is a prerequisite to thought and understanding, there are many workers in AI who do not share this view. The resolution of this question will have profound effects on many of the current views in psychology. Along the same lines, Minsky gave a list of concepts common in computer science which have no analogies in psychology (such as the notions of "garbage collection" and "pure procedure"). I recommend this book as an interesting viewpoint on computer science in general and AI in particular. The experimental results alone are worth reading it for. Hopefully we'll see more studies along these lines in the future. stan shebs ------------------------------ Date: Wed 12 Sep 84 09:58:29-PDT From: Ken Laws Subject: Dreams A letter by Donald A. Windsor in the new CACM (September, p. 859) suggests that the purpose of dreams is to test our cognitive models of the people around us by simulating their behavior and monitoring for bizarre patterns. He claims that the "dream people" are AI programs that we construct subconsciously. -- Ken Laws ------------------------------ Date: 09/11/84 13:56:44 From: STORY Subject: Seminar - Computational Theory of Higher Brain Function [Forwarded from the MIT bboard by SASW@MIT-MC.] TITLE: ``A Computational Theory of Higher Brain Function'' SPEAKER: Leslie M. Goldschlager, Visiting Computer Scientist, Stanford University DATE: Friday, September 14, 1984 TIME: Refreshments, 3:45pm Lecture, 4:00pm PLACE: NE43-512a A new model of parallel computation is proposed. The fundamental item of data in the model is called a "concept", and concepts may be stored on a two-dimensional data structure called a "memory surface". The nature of the storage mechanism and the mode of communication which is required between storage locations renders the model suitable for implementation in VLSI. An implementation is also possible with neurons arranged in a two-dimensional sheet. It is argued that the model is particularly worthwhile studying as it captures some of the computational characteristics of the brain. The memory surface consists of a vast number of processors which are called "columns" and which operate asynchronously in parallel. Each processor stores a small amount of information and can be thought of as a simple finite-state transducer. Each processor is connected only to those processors within a small radius, or neighbourhood. As is usually found with parallel computation, the most important aspect of the model is the method of communication between the processors. It is shown in the talk how the function of the individual processors and the communication between them supports the formation and storage of associations between concepts. Thus the memory surface is in effect an associative memory. This type of associative memory reveals a number of interesting computational features, including the ability to store and retrieve sequences of concepts and the ability to form abstractions from simpler concepts. Certain capabilities taken from the realm of human activities are shown to be explainable within the model of computation presented here. These include creativity, self, consciousness and free will. A theory of sleep is also presented which is consistent with the model. In general it is argued that the computational model is appropriate for describing and explaining the higher functions of the brain. These are believed to occur in a region of the brain called the cortex, and the known anatomy of the cortex appears to be consistent with the memory surface model discussed in this talk. HOST: Professor Gary Miller ------------------------------ Date: Mon, 10 Sep 84 17:38:55 PDT From: Shel Finkelstein Reply-to: IBM-SJ Calendar Subject: Seminar - Distributed Knowledge [Forwarded from the Stanford bboard by Laws@SRI-AI.] IBM San Jose Research Lab 5600 Cottle Road San Jose, CA 95193 [...] Thurs., Sept. 13 Computer Science Seminar 3:00 P.M. KNOWLEDGE AND COMMON KNOWLEDGE IN A DISTRIBUTED Front Aud. ENVIRONMENT By examining some puzzles and paradoxes, we argue that the right way to understand distributed protocols is by considering how messages change the state of a system. We present a hierarchy of knowledge states that a system may be in, and discuss how communication can move the system's state of knowledge up the hierarchy. Of special interest is the notion of common knowledge. Common knowledge is an essential state of knowledge for reaching agreements and coordinating action. We show that in practical distributed systems, common knowledge is not attainable. We introduce various relaxations of common knowledge that are attainable in many cases of interest. We describe in what sense these notions are appropriate, and discuss their relationship to each other. We conclude with a discussion of the role of knowledge in a distributed system. J. Halpern, IBM San Jose Research Lab Host: R. Fagin Please note change in directions due to completion of new Monterey Road (82) exit replacing the Ford Road exit from 101. [...] ------------------------------ End of AIList Digest ******************** 13-Sep-84 22:19:58-PDT,14090;000000000001 Mail-From: LAWS created at 13-Sep-84 22:17:46 Date: Thu 13 Sep 1984 21:57-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #118 To: AIList@SRI-AI AIList Digest Friday, 14 Sep 1984 Volume 2 : Issue 118 Today's Topics: AI Tools - MACSYMA Copyright, Philosophy - The Nature of Proof, Robotics - Brian Reid's Robot Cook, Humor - Self-Reference & Seminar on Types in Lunches, Journals - Sigart Issue on Applications of AI in Engineering ---------------------------------------------------------------------- Date: 10 September 1984 15:29-EDT From: Paula A. Vancini Subject: MACSYMA Notice [Forwarded from the MIT bboard by Laws@SRI-AI.] TO: ALL MACSYMA USERS FROM: MIT Patent and Copyright Office DATE: August 31, 1984 SUBJECT: Recent Notices by Paradigm Associates Regarding MACSYMA Software Please be advised that the version of MACSYMA designated by Paradigm Associates in recent messages over this network as "DOE MACSYMA" is a version of MACSYMA copyrighted to MIT. "DOE MACSYMA" is an improper designation. MIT has delivered a copy of the MIT MACSYMA software to DOE, pursuant to MIT's contraactual obligations to DOE. Also be advised that Symbolics, Inc. is the only commercial company authorized by MIT to perform maintenance services on, or to make enhancements to, the MIT copyrighted versions of MACSYMA. MIT hereby disclaims any association with Paradigm Associates and has not granted Paradigm licensing rights to commercially make use of its copyrighted versions of the MACSYMA or NIL software. Queries to Hynes%MIT-XX@MIT-MC ------------------------------ Date: 10 Sep 84 14:33:25-PDT (Mon) From: decvax!genrad!teddy!rmc @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: teddy.403 I am not sure I agree that an inductive proof proves any more or less than a deductive proof. The basis of induction is to claim 1) I have applied a predicate to some specific cases within a large set (class) of cases. 2) I detect a pattern in the result of the predicate over those cases 3) I predict that the results of the predicate will continue following the pattern for the rest of the cases in the set. I state the proof pattern this way to include inductive arguments about natural world phenomena as well as mathematical induction. The proof is valid if the accepted community of experts agrees that the proof is valid (see for example various Wittgenstein and Putname essays on the foundations of mathematics and logic). The experts could be wrong for a variety of reasons. Natural law could change. The argument may be so complicated that everyone gets lost and misses a mistake (this has even happened before!) The class of cases may be poorly chosen. etc. The disagreement seems to be centered around a question of whether this community of experts accepts causality as part of the model. If it is, then we can use causality as an axiom in our proof systems. But it still boils down to what the experts accept. R Mark Chilenskas decvax!genrad!teddy!rmc ------------------------------ Date: 11 Sep 84 9:27:15-PDT (Tue) From: ihnp4!houxm!mhuxl!ulysses!allegra!princeton!eosp1!robison @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: eosp1.1106 Mark Chilenskas discussion of inductive proof is not correct for mathematics, and greatly understates the strength of mathematical inductive proofs. These work as follows: Given a hypothesis; - Prove that it is true for at least one case. - Then prove that IF IT IS TRUE FOR A GENERIC CASE, IT MUST BE TRUE FOR THE NEXT GENERIC CASE. For example, in a hypothesis about an expression with regard to all natural numbers, we might show that it is true if "n=1". We then show that IF it is true for "n", it is true for "n+1". By induction we have shown that the hypothesis is absolutely true for every natural number. Since true: n=1 => true for n=2, true: n=2 => true for n=3, etc. It is the responsibility of the prover to prove that induction through all generic cases is proper; when it is not, additional specific cases must be proved, or induction may not apply at all. Such an inductive proof is absolutely true for the logical system it is defined in, and just as correct as any deductive proof. When our perception of the natural laws change, etc., the proof remains true, but its usefulness may become nil if we perceive that no system in the real world could possibly correspond to the proof. In non-mathematical systems, it is possible that both deductive and inductive proofs will be seriously flawed, and I doubt one can try to prefer "approximate proofs" of one type over the other. If a system is not well-enough defined to permit accurate logical reasoning, then the chances are that an ingenious person can prove anything (see net.flame and net.religion for examples, also the congressional record). - Toby Robison (not Robinson!) allegra!eosp1!robison or: decvax!ittvax!eosp1!robison or (emergency): princeton!eosp1!robison ------------------------------ Date: Thu 13 Sep 84 09:14:34-PDT From: Ken Laws Subject: Inductive Proof - The Heap Problem As an example of improper induction, consider the heap problem. A "heap" of one speck (e.g., of flour) is definitely a small heap. If you add one speck to a small heap, you still have a small heap. Therefore all heaps are small heaps. -- Ken Laws ------------------------------ Date: Fri 7 Sep 84 09:40:42-CDT From: Aaron Temin Subject: Robot chef bites off too much [Forwarded from the UTexas-20 bboard by Laws@SRI-AI.] Our West Coast correspondent has returned with (among other things) an article from the San Jose Mercury News entitled "Robot cooks if it finds the beef" about Brian Reid's(TM){1} attempts to program a robot to cook beef Wellington. [...] Aaron {1} Brian Reid is a trademark of ScribeInc., Ouagadougou, Bourkina Fasso. [I have copied the following excerpts from the June 10 article. -- KIL] Robot cooks if it finds the beef by Kathy Holub Some professors will do anything for a theoretical exercise. Brian K. Reid, a food-loving assistant professor of electrical engineering at Stanford University, recently tried to simulate teaching a mindless robot how to cook beef Wellington, using Julia Child's 16-page recipe. He failed. Try telling a robot what "spread seasoning evenly" means. "You have to specify the number of grams (of seasoning) per square centimeter," he said, with a wry smile. It took him 13 hours and 60 pages of computer instructions just to teach the would-be automaton how to slice and season a slab of beef and put it safely in the oven. Julia Child takes only three pages to explain these simple steps. "Where I bogged down -- where I gave it all up and decided to go to bed -- was when I had to tell the robot how to wrap the beef in pastry," he said. But Reid, an excellent cook with a doctorate in computer science, was thrilled with the experiment, which involved only the computer program and not an actual robot. "It was exactly what I wanted," he said. "It showed that a cookbook does not tell the whole story, that there is a lot of information missing from the recipe" that human cooks provide without knowing it. The Wellington exercise, he believes, will help him reach his real goal: to teach a computer how to make integrated circuits with a computer "recipe" that doesn't depend on human judgement, memory or common sense. [...] He picked the recipe for his experiment, because it's the longest one in the book, involving 26 ingredients. Beef Wellington is a long piece of tenderlion that is baked twice, the second time in a light pastry crust that should turn golden brown. Forget telling the robot what "golden brown" means. "Every time I turned around I discovered massive numbers of things I was contributing without even thinking about it." For example, "Julia Child has, 'you slice the beef and season each piece separately'" before cooking, he said. "The meat must be cold or it won't hold its shape, but Julia doesn't tell you that. She assumes you know." For purposes of simplicity, Reid let the robot skip the slicing of mushrooms and onions and sauteeing them in butter "until done." "Cooking until done requires a great deal of knowledge. A robot doesn't know that fire [in the pan] isn't part of the process. It would happily burn the pan." But just telling the robot how to slice the meat, season it, reassemble it with skewers and put it in the oven was tricky enough -- like teaching a 3-year-old to fix a car. "You can't just say, 'Cut into slices,' Reid said. "You have to say, 'Move knife one centimeter to the east, cut.' And that assumes a sub-program telling th robot what 'cut' means." You can't tell a robot to slice 'across.' "Across what?" said Reid. "You can't tell a robot to eyeball something. You have to tell it to define the center of gravity of the beef, find the major axis of the beef and cut perpendicular to it." You also have to tell the robot how to find the beef, that is, distinguish it from the other ingredients, and when to stop slicing. These are standard problems in robotics. Other problems are not so standard. Reid forgot to specify that the skewers should be removed before the pastry shell is added. Julia may be forgiven for leaving this step out, but the robot trainer has tougher work. ------------------------------ Date: 9 September 1984 04:04-EDT From: Steven A. Swernofsky Subject: Humor in A.I.? I saw the following button at a science fiction convention: Q. Why did Douglas Hofstadter cross the road? A. To make this riddle possible. -- Steve ------------------------------ Date: 11 Sep 1984 14:52 EDT (Tue) From: Walter Hamscher Subject: Humor - Seminar on Types in Lunches [Forwarded from the MIT bboard by SASW@MIT-MC.] GENERALIZED TYPES IN GRADUATE STUDENT LUNCHES FIRST MEETING: Friday, Sept. 14, 1984, 12:00 noon PLACE: MIT AI Lab Playroom, 545 Tech. Sq., Cambridge, MA, USA ORGANIZER: Walter Hamscher, (walter@oz) An eating seminar about generalized cold cuts and spread-recognition; gluttonism, leftovers, and indigestion; related notions appearing in current and proposed lunches, such as volunteers, menus, and The Roosevelt Paradox ("There is no such thing as a free lunch") will be discussed. The slant will be toward identifying the underlying digestional problems raised by the desired menu features. For the first five minutes (during the visit of Prof. Gustav Fleischbrot, Univ. of Essen) we will present and discuss the papers below starting with the first two and concluding with the final two: 1. Burger, Chip N., ``The Nutritional Value of Pixels'', PROC. INT'L. CONF. 5TH GENERATION INGESTION SYSTEMS, Tokyo, to appear. Manuscript from Dept. of Computer Science, Univ. of Sandwich, 1984. 2. Burger, Chip N. and Gelly Muffin, ``A Kernel language for abstract feta cheese and noodles'', SEMANTICS OF FETA CHEESE: PROCEEDINGS, (eds.) Cream, MacFried and Potstick, Springer-Verlag, Lect. Notes in Comp. Sci. 173, 1-50, 1984. 3. MacDonald, Ronald, ``Noodles for standard ML'', ACM SYMP. ON LINGUICA AND LINGUINI, 1984. 4. Munchem, J. C., ``Lamb, D-Calories, Noodles, and Ripe Fruit'', Ph.D. Thesis, MIT, Dept. of EECS, September, 1984. Meeting time for the first five minutes is Fri. 12:00-12:05, and Friday 12:00-12:05 thereafter. Aerobics course credit can be arranged. ------------------------------ Date: Wednesday, 5 September 1984 23:28:30 EDT From: Duvvuru.Sriram@cmu-ri-cive.arpa Subject: Special Sigart Issue on Applications of AI in Engineering SPECIAL ISSUE ON APPLICATIONS OF AI IN ENGINEERING The April 1985 issue of the SIGART newsletter (tentative schedule) will focus on the applications of AI in engineering. The purpose of this issue is to provide an overview of research being conducted in this area around the world. The following topics are suggested: - Knowledge-based expert systems - Intelligent computer tutors - Representation of engineering problems - Natural language and graphical interfaces - Interfacing engineering databases with expert systems The above topics are by no means exhaustive; other related topics are welcome. Individuals or groups conducting research in this area and who would like to share their ideas are invited to send two copies of 3 to 4 page summaries of their work, preferably ongoing research, before December 1, 1984. The summaries should include a title, the names of people associated with the research, affiliations, and bibliographical references. Since the primary aim of this special issue is to provide information about ongoing and proposed research, please be as brief as possible and avoid lengthy implementation details. Submissions should be sent to D. Sriram (or R. Joobbani) at the following address or through Arpanet to Sriram@CMU-RI-CIVE. D. Sriram Design Research Center Carnegie-Mellon University Pittsburgh, PA 15213 Tel. No. (412)578-3603 ------------------------------ End of AIList Digest ******************** 16-Sep-84 15:58:32-PDT,15122;000000000000 Mail-From: LAWS created at 16-Sep-84 15:56:42 Date: Sun 16 Sep 1984 15:47-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #119 To: AIList@SRI-AI AIList Digest Sunday, 16 Sep 1984 Volume 2 : Issue 119 Today's Topics: LISP - VAX Lisps & CP/M Lisp, Philosophy - Syllogism Correction, Scientific Method - Induction vs. Deduction, Course - Logic Programming, Conference - Database Systems ---------------------------------------------------------------------- Date: Sun, 16 Sep 84 14:28 BST From: TONY HASEMER (on ALVEY at Teddington) Subject: Lisp on the VAX We have a VAX 11/750 with four Mb of memory, running NIL. We also have four Lisp hackers of several years' standing who are likely to write quite substantial programs. We have to decide whether to buy some extra memory, or to spend the money on Golden Common Lisp, which someone told us is much more effiecient than NIL. Can anyone please advise us? Thank you. Tony. ------------------------------ Date: 11 Sep 84 17:36:37-PDT (Tue) From: hplabs!sdcrdcf!sdcsvax!stan @ Ucb-Vax.arpa Subject: Lisp under CP/M Article-I.D.: sdcsvax.52 I recently purchased a copy of ProCode's Waltz Lisp for the Z80 and CP/M and found it to be a very good imitation of Franz Lisp. I downloaded some rather substantial programs I'ld written over the past two years and within 20 minutes had them up and running on my Kaypro. Surprisingly, there was little speed degradation unless there was a major amount of computations involved. All that was required (for my programs) were a few support routines to implement defun, terpri, etc. The manual is very complete and well written. (For example, it had examples of how to write defun in it.) Cost was just under $100.00, and well worth it. Now, if only my Kaypro could handle background processes like the VAX... Stan Tomlinson ------------------------------ Date: 11 Sep 84 11:06:09-PDT (Tue) From: hplabs!hao!seismo!rochester!rocksvax!rocksanne!sunybcs!gloria!colonel @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: gloria.535 >> All swans are white. >> This is a swan. >> Therefore it is white. >> >> Notice that the conclusion (3rd sentence) is only true iff the two >> premises (sentences 2 and 3) are true. A minor correction: "iff" does not belong here. The premises do not follow from the conclusion. -- Col. G. L. Sicherman ...seismo!rochester!rocksanne!rocksvax!sunybcs!gloria!colonel ------------------------------ Date: 14 Sep 84 09:01 PDT From: Feuerman.pasa@XEROX.ARPA Subject: Re: Inductive Proof - The Heap Problem At the risk of getting involved..... One thing bothers me about the inductive proof that all heaps are small. I will claim that that is NOT an inductive proof after all. The second requirement for a (mathematical) proof by induction states that one must show that P(n) implies P(n+1). I see nothing in the fact that one "speck" is small that NECESSARILY implies that two "specks" constitutes a small heap. One seems to conclude the fact that a two-speck heap is small from some sort of outside judgment of size. Thus, Small(1 Speck) does NOT imply Small(2 Specks), something else implies that. Lest we get into an argument about the fact that large for one could be small for another, I'll bring up another mathematical point: The Archimedian Principle. It basically says that given any number (size, number of specks, what have you), one can ALWAYS find a natural number that is greater. Applying that to the heap problem, given anyone's threshold of what constitutes a large heap and what constitutes a small heap, one can ALWAYS make a large heap out of a small heap by adding one speck at a time. I'll further note that one need not make that transition between small and large heaps a discreet number; as long as you can put a number on some sense of a large heap (regardless of whether that is the smallest large heap), you can always exceed it. For example, I will arbitrarily say that 10**47 specks in a heap makes it large. I don't have to say that 10**47 - 1 is small. Yet we will still be able to create a large heap (eventually). Now, anyone interested in speculating about what happens if someone's size function is not constant, but varies with time, mood, money in the bank, etc.? As further proof of my Archimedian Principle, we will note that I have just in fact turned a small heap/argument (Ken Laws' four line Heap Problem) into a large one (this message). --Ken ------------------------------ Date: Fri 14 Sep 84 14:30:14-PDT From: BARNARD@SRI-AI.ARPA Subject: induction vs. deduction The discussion of induction vs. deduction has taken a curious turn. Normally, when we speak of induction, we don't mean *mathematical induction*, which is a formally adequate proof technique. We mean instead the inductive mode of reasoning, which is quite different. Inductive reasoning can never be equated to deductive reasoning because it begins with totally different premises. Inductive reasoning involves two principles: (1) The principle of insufficient reason, which holds that in the absence of other information, the expectation over an ensemble of possibilities is uniform (heads and tails are equally probable). (2) The principle of Occam's razor, which hold that given a variety of theories about some data, the one that is "simplest" is preferred. (We prefer the Copernican model of the solar system to the Ptolemaic one, even though they both account for the astronomical data.) The relationship of time, causality, and induction has been investigated by the Nobel Laureate, Ilya Prigogine. The laws of classical physics, with one exception, are neutral with respect to the direction of time. The exception is the Second Law of Thermodynamics, which states that the entropy of a closed systems must increase, or equivalently, that a closed system will tend toward more and more disordered states. For a long time, physicists tried to prove the Second Law in terms of Newtonian principles, but with no success. Eventually, Boltzman and Gibbs explained the Second Law satisfactorily by using inductive principles to show that the probability of a system entering a disordered, high-entropy state is far higher than the converse. Prigogine proposes that random, microscopic events cause macroscopic events to unfold in a fundamentally unpredictable way. He extends thermodynamics to open systems, and particularly to "dissipative systems" that, through entropy exchange, evolve toward or maintain orderly, low-entropy states. Inductive reasoning is also closely connected with information theory. Recall that Shannon uses entropy as the measure of information. Brillouin, Carnap, and Jaynes have shown that these two meanings of entropy (information in a message and disorder of a physical system) are equivalent. Steve Barnard ------------------------------ Date: Wed 12 Sep 84 21:16:28-EDT From: Michael J. Beckerle Subject: Course Offering - Logic Programming [Forwarded from the MIT bboard by Laws@SRI-AI.] TECHNOLOGY OF LOGIC PROGRAMMING CS 270 Fall 1984 Professor Henryk Jan Komorowski Harvard University Aiken Computation Lab. 105 495-5973 Meeting: Mondays, Wednesdays - 12:30 to 2 PM, Pierce Hall 209 This year the course will focus on presenting basic concepts of logic programming by deriving them from logic. We shall study definite clause programs: - What they specify (the least Herbrand model). - How they are used: a logical view of the notion of query. - What computations of logic programs are: the resolu- tion principle, SLD-refutability, completeness and nega- tion by failure. This general background will serve as a basis to introduce a logic programming language Prolog and will be associated by a number of assignments to master specification programming. It will be followed by some implementation issues like in- terpreting, compiling, debugging and other programmer's sup- port tools. We shall then critically investigate a number of applications of Prolog to software specification, com- piler writing, expert system programming, embedded languages implementation, database programming, program transforma- tions, etc., and study language's power and limitations. The course will end with a comparison of definite clause programming to other formalisms, eg. attribute grammars, functional programming, rule based programming. Time per- mitting parallelism, complexity and other topics of interest will be studied. REQUIREMENTS A background in propositional logic, some fami- liarity with predicate calculus and general background in computer science (reasonable acquaintance with parsing, com- piling, databases, programming im recursive languages, etc.) is expected. WORKLOAD - one problem set on logic. - Two sets of Prolog assignments. - Mid-term mid-size Prolog single person project. - A substantial amount of papers to read: core papers and elected one-topic papers (the latter to be reviewed in sections). - Final research paper on individually elected topic (with instructor's consent). LITERATURE, REQUIRED PROGRAMMING IN PROLOG by Clocksin and Mellish. RESEARCH PA- PERS distributed in class. LITERATURE, OPTIONAL LOGIC FOR PROBLEM SOLVING, by Kowalski MICRO-PROLOG: LOGIC PROGRAMMING, by Clark and McCabe LOGIC AND DATABASES, edited by Gallaire and Minker IMPLEMENTATIONS OF PROLOG, edited by Campbell TENTATIVE PLAN 25 meetings - Introduction: declarative and imperative programming, the goals of Vth Generation Project. - Informal notions of: model, truth, provability. The syn- tax of predicate calculus, proof systems for predicate cal- culus completemess, soundness, models. - Transformation to clausal form, resolution and its com- pleteness. - Definite clause programs: * operational semantics * proof-theoretic semantics * fixed point semantics - Introduction to programming in Prolog. - Data structures. - Negation by failure and cut. - Specification programming methodology. - Advanced Prolog programming. - Algorithmic debugging. - Parsing and compiling in Prolog. - Abstract data type specification in Prolog. - Logic programming and attribute grammars, data flow analysis. - Interpretation and compilation of logic programs - Artificial intelligence applications: * metalevel programming * expert systems programming * Natural language processing - Alternatives to Prolog; breadth-first search, coroutines, LOGLISP, AND- and OR-parallelism. - Concurrent Prolog. - Relations between LP and functional programming. - LP and term rewriting. - Program transformation and derivation. - Object oriented programming. - Some complexity issures. - LP and databases. - Architecture for LP. ------------------------------ Date: Wed, 12 Sep 84 10:40:23 pdt From: Jeff Ullman Subject: Conference - Database Systems CALL FOR PAPERS FOURTH ANNUAL ACM SIGACT/SIGMOD SYMPOSIUM ON PRINCIPLES OF DATABASE SYSTEMS Portland, Oregon March 25-27, 1985 The conference will cover new developments in both the theoretical and practical aspects of database systems. Papers are solicited that describe original and novel research into the theory, design, or implementation of data- base systems. Some suggested but not exclusive topics of interest are: application of AI techniques to database systems, con- currency control, database and database scheme design, data models, data structures for physical database implementa- tion, dependency theory, distributed database systems, logic-based query languages and other applications of logic to database systems, office automation theory, performance evaluation of database systems, query language optimization and implementation, and security of database systems. You are invited to submit 9 copies of a detailed abstract (not a complete paper) to the program chairman: Jeffrey D. Ullman Dept. of Computer Science Stanford University Stanford, CA 94305 Submissions will be evaluated on the basis of significance, originality, and overall quality. Each abstract should 1) contain enough information for the program committee to identify the main contribution of the work; 2) explain the importance of the work, its novelty, and its relevance to the theory and/or practice of database management; 3) include comparisons with and references to relevant litera- ture. Abstracts should be no longer than 10 typed double- spaced pages (12,000 bytes of source text). Deviations from these guidelines may affect the program committee's evalua- tion of the paper. Program Committee Jim Gray Richard Hull Frank Manola Stott Parker Avi Silberschatz Jeff Ullman Moshe Vardi Peter Weinberger Harry Wong The deadline for submission of abstracts is October 12, 1984. Authors will be notified of acceptance or rejection by December 7, 1984. The accepted papers, typed on special forms or typeset camera-ready in the reduced-size model page format, will be due at the above address by January 11, 1985. All authors of accepted papers will be expected to sign copyright release forms. Proceedings will be distri- buted at the conference and will be available for subsequent purchase through ACM. The proceedings of this conference will not be widely disseminated. As such, publication of papers in this record will not, of itself, inhibit republi- cation in ACM's refereed publications. General Chairman: Local Arrangements Chairman: Seymour Ginsburg David Maier Dept. of CS Dept. of CS USC Oregon Graduate Center Los Angeles, CA 90007 19600 N. W. Walker Rd. Beaverton, OR 97006 ------------------------------ End of AIList Digest ******************** 19-Sep-84 09:35:47-PDT,13591;000000000000 Mail-From: LAWS created at 19-Sep-84 09:34:17 Date: Wed 19 Sep 1984 09:29-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #120 To: AIList@SRI-AI AIList Digest Wednesday, 19 Sep 1984 Volume 2 : Issue 120 Today's Topics: AI Tools - Micro Production Systems, Professional Societies - AI SIG in San Diego, Books - Publisher Info for The Second Self, Scientific Method - Swans & Induction, AI and Society - CPSR, Robotics - Kitchen Robots, Pattern Recognition - Maximum Window Sum, Course - Decision Systems, Games - Computer Chess Championship ---------------------------------------------------------------------- Date: 18 September 1984 1053-EDT From: Peter Pirolli at CMU-CS-A Subject: micro production systems [Forwarded from the CMU bboard by Laws@SRI-AI.] A friend of mine is looking for a production system language (however simple) that runs on an Apple (preferably) or any other micro. He basically wants to use the system to give some hands-on experience to fellow faculty members at a small university where main-frame resources are too scarce to run a full-blown production system. Any pointers to micro-based systems would be greatly appreciated. Send mail to pirolli@cmpsya or pirolli@cmua. ------------------------------ Date: 17 Sep 84 07:16 PDT From: Tom Perrine Subject: AI SIG in San Diego I have an off-net friend who is interested in starting (or finding) a Special Interest Group for AI in San Diego. It would appear that if ACM or IEEE knows about such a group, "they ain't talking." Is there anyone else in S.D. who would be interested in such a group? Please reply to me, not the Digest, of course. Please include name, address and a daytime phone. Thanks, Tom Perrine Logicon - OSD San Diego, CA ------------------------------ Date: Mon 17 Sep 84 12:41:04-MDT From: Stan Shebs Subject: Publisher Info for The Second Self I neglected to provide the details... The publisher is Simon & Schuster, ISBN is 0-671-46848-0, and LC number is QA76.T85 1984 (or something like that). The book is available in quite a few bookstores, including the big chains, so try there first. stan shebs ------------------------------ Date: 17 Sep 1984 08:47:02-EDT From: sde@Mitre-Bedford Subject: Swans: At least during the latter part of March, 1980, the statement, "all swans are white," was false; those familiar with Heinlein's "fair witness" concept will recognize the phrasing; I say it having witnessed black or near-black swans in Perth during the aforementioned time. Granting that the facts have little to do with the principle of the argument, I thought folks might nonetheless be amused. David sde@mitre-bedford ------------------------------ Date: 12 Sep 84 9:11:34-PDT (Wed) From: hplabs!tektronix!bennety @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: tektroni.3588 Toby Robison's comment on Mark Chilenska's discussion on inductive proof was quite apt -- however, we should note that induction is limited to statements on a countably infinite set. That is, induction can only work with integers. -bsy tektronix!bennety ------------------------------ Date: Mon, 17 Sep 84 11:01:17 PDT From: Charlie Crummer Subject: Uhrig's Stream of Consciousness in V2 #112 With regard to Werner's concern about unethical or immoral applications of AI: Computer Professionals for Social Responsibility (CPSR) is very concerned with this issue as am I. Please give me feedback on this. Perhaps the surest death-knell for the outrageous-dangerous stuff ("Intelligent Computers" that would make life-or- death decisions for the human race) is to require that they pass rigorous tests. If it is required that they actually work the way they are supposed to many of the systems will die a natural (and deserved) death. A comprehensive and rigorous top-down (parallel with the top-down design) testing program may be the answer. --Charlie ------------------------------ Date: 17 Sep 84 09:20:20 PDT (Monday) From: Hoffman.es@XEROX.ARPA Subject: AI in the kitchen, continued The article in V2,#118, "Robot cooks if it finds the beef", reminded me of the following: ".... John McCarthy, one of the founders of the field of artificial intelligence, is fond of talking of the day when we'll have 'kitchen robots' to do chores for us, such as fixing a lovely shrimp creole. Such a robot would, in his view, be exploitable like a slave because it would not be conscious in the slightest. To me, this is incomprehensible. Anything that could get along in the unpredictable kitchen world would be as worthy of being considered conscious as would a robot that could survive for a week in the Rockies. To me, both worlds are incredibly subtle and potentially surprise-filled. Yet I suspect that McCarthy thinks of a kitchen as ... some sort of simple and 'closed' world, in contrast to 'open-ended' worlds, such as the Rockies. This is just another example, in my opinion, of vastly under-estimating the complexity of a world we take for granted, and thus under-estimating the complexity of the beings that could get along in such a world. Ultimately, the only way to be convinced of these kinds of things is to try to write a computer program to get along in a kitchen...." Excerpted from a letter by DOUG HOFSTADTER in 'Visible Language', V17,#4, Autumn 1983. (In 1983, that periodical carried, in successive issues, an extensive piece by Knuth on his Meta-Font, a lengthy review by Hofstadter, and letters from both of them and from others.) --Rodney Hoffman ------------------------------ Date: 14 Sep 1984 16:36-EDT From: Dan Hoey Subject: Maximum window sum, in AIList V2 #117 Ken, Bentley's problem 7 asks for the complexity of the maximum subarray sum problem. I would advise you to call your algorithm a solution to the maximum subarray sum problem, rather than a solution to problem 7. You have given an upper bound for the complexity, but barring an equal lower bound problem 7 is still unsolved. I know of no lower bound larger than the size of the input. In case you're interested, here's another maximum subarray sum algorithm with the same time complexity, using less working storage. See the comments for a description of its working. Enjoy. Dan [The following is simpler, more efficient, and uses less auxilliary storage than the version I gave (although it does require buffering the full input array). I can't think of any improvement. -- KIL] /* ** maxsbasum ** ** Compute the maximum subarray sum in an array. In case all ** array elements are negative, the maximum sum is 0.0 ** for an empty subarray. ** ** COMMENTS ** ** Every subarray of an array is a full-height subarray of a ** full-width subarray of the array. ** ** This routine examines each of the O(NROWS^2) full-width ** subarrays of the array. A vector containing the sum of each ** column in the full-width subarray is maintained. The maximum ** full-height subarray sum of the full-width subarray corresponds ** to the maximum subvector sum of the vector of column sums, ** found in O(NCOLS) time using Kadane's algorithm. ** ** Running time is O(NROWS^2 NCOLS). Working storage for this ** program is dominated by the O(NCOLS) vector of column sums. ** ** HISTORY ** ** 16-Sep-84 Laws at SRI-AI ** Merged innermost two loops into one. ** ** 14-Sep-84 Hoey at NRL-AIC ** Cobbled this version together. ** Comm. ACM, September 1984; Jon Bentley ** published maximum subvector code (Pascal). ** Algorithm attributed to Jay Kadane, 1977. ** ** 11-Sep-84 Laws at SRI-AI ** Wrote another program solving the same problem. Parts of ** his program, from AIList V2 #117, appear in this program. */ /* Sample problem. (Answer is 6.0.) */ #define NROWS 4 #define NCOLS 4 float X[NROWS][NCOLS] = {{ 1.,-2., 3.,-1.}, { 2.,-5., 1.,-1.}, { 3., 1.,-2., 3.}, {-2., 1., 1., 0.}}; /* Macro to return the maximum of two expressions. */ #define MAX(exp1,exp2) (((exp1) > (exp2)) ? (exp1) : (exp2)) main() { float MaxSoFar; /* Global maximum */ float ColSum[NCOLS]; /* Column sums of full-width subarray */ float MaxEndingHere; /* For Kadane's algorithm */ int lowrow,highrow; /* Bounds of full-width subarray */ int thiscol; /* Column index */ /* Loop over bottom row of full-width subarray. */ MaxSoFar = 0.0; for (lowrow = 0; lowrow < NROWS; lowrow++) { /* Initialize column sums. */ for (thiscol = 0; thiscol < NCOLS; thiscol++) ColSum[thiscol] = 0.0; /* Loop over top row of full-width subarray. */ for (highrow = lowrow; highrow < NROWS; highrow++) { /* Update column sum, find maximum subvector sum of ColSum. */ MaxEndingHere = 0.0; for (thiscol = 0; thiscol < NCOLS; thiscol++) { ColSum[thiscol] += X[highrow][thiscol]; MaxEndingHere = MAX(0.0, MaxEndingHere + ColSum[thiscol]); MaxSoFar = MAX(MaxSoFar, MaxEndingHere); } } } /* Print the solution. */ printf("Maximum subarray sum: %g\n",MaxSoFar); } ------------------------------ Date: Tue 18 Sep 84 15:09:58-PDT From: Samuel Holtzman Subject: Course - Decision Systems [Forwarded from the Stanford bboard by Laws@SRI-AI.] Course Announcement DECISION ANALYSIS AND ARTIFICIAL INTELLIGENCE Engineering Economic Systems 234 3 units Instructor: Samuel Holtzman Monday and Wednesday 2:00 to 3:15 pm Building 260, room 264 This course investigates the relationship between decision analysis and artificial intelligence in building expert systems for decision making in complex domains. Major topic areas include fundamentals of artificial intelligence (production systems, search, logic programming) and design of intelligent decision systems based on decision analysis (use of formal methods in decision making, represention and solution of decision problems, reasoning under uncertainty). The course will also cover programming in Lisp for students not familiar with the language. Course requirements include a sustantial project based on the concepts developed in the course. Prerequesites: EES 231 (Decision Analysis) or equivalent and familiarity with computer programming. For further information contact: Samuel Holtzman 497-0486, Terman 301 HOLTZMAN@SUMEX ------------------------------ Date: Mon Sep 17 17:15:08 1984 From: mclure@sri-prism Subject: Games - Computer Chess Championship [Forwarded from the SRI-AI bboard by Laws@SRI-AI.] The ACM annual North American Computer Chess Championship is a watering-hole for computer chess researchers, devotees, and ordinary chess players interested in what new improvements have been made in computer chess during the past year. Come see Ken Thompson and Belle seek out chess truth, chess justice, and the American Way! Watch David Levy wince as his chess program discovers innovations in chess theory unknown even to Grandmasters! Marvel at Bob Hyatt's Cray Blitz program as it slices through the opposition at many MIPS! See the tiny Spracklen program otherwise marketed as Prestige and Elite by Fidelity tally up points against the "big boys!" Gawk as ivory tower researchers such as Tony Marsland of University of Alberta try to turn obscure and obfuscating computer chess theory into tangible points against opposition! Watch in amazement as David Slate's NUCHESS program, a descendent of the famous Northwestern University Chess 4.5 program, tries to become the most "human-like" of chess programs! And strangest of all, see a chess tournament where the noise level is immaterial to the quality of play! The following information is from AChen at Xerox... 1) dates - 7-9 Oct, 1984 2) where - Continental Parlors at San Francisco Hilton 3) times - Sun 1300 and 1900, 7 Oct, 1984 Mon 1900, 8 Oct, 1984 Tue 1900, 9 Oct, 1984 4) who - Tournament director will be Mike Valvo four round Swiss-style includes Cray BLITZ, BELLE and NUCHESS. for more information: Professor M. Newborn School of Computer Science, McGill University 805 Sherbrooke Street West, Montreal Quebec, Canada H3A 2K6 note: this info can be found in July, 1984 issue of ACM Communications, page A21. Stuart ------------------------------ End of AIList Digest ******************** 19-Sep-84 21:57:28-PDT,12747;000000000000 Mail-From: LAWS created at 19-Sep-84 21:55:00 Date: Wed 19 Sep 1984 21:48-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #121 To: AIList@SRI-AI AIList Digest Thursday, 20 Sep 1984 Volume 2 : Issue 121 Today's Topics: Machine Translation - Aymara as Intermediate Language, Logic - Induction & Deduction, Linguistics - Pittsburghese, Expert Systems & Humor - Excuse Generation ---------------------------------------------------------------------- Date: 11 Sep 84 17:07:00-PDT (Tue) From: pur-ee!uiucdcs!uokvax!emjej @ Ucb-Vax.arpa Subject: Aymara as intermediate language? Article-I.D.: uokvax.900011 Yesterday in a local paper a news item appeared (probably AP or UPI) telling about a fellow in South America (Ecuador? Peru, perhaps?), named Ivan Guzman de Rojas, who seems to be having respectable success using a S. American Indian language, that of the Aymara Indians, as an intermediate language for machine translation of natural languages. The article seemed to indicate that Aymara is something of a pre-Columbian Loglan, near as I could tell. Any references to the literature concerning this would be greatly appreciated. (Send mail, I'll summarize to the net after a seemly interval.) James Jones uucp: ...!ctvax!uokvax!emjej or ...!{ctvax,mtxinu}!ea!jejones ------------------------------ Date: Wed, 19 Sep 84 14:10:13 pdt From: Stanley Lanning Subject: Nitpicking... "... induction is limited to statements on a countably infinite set." Well, that depends how you define induction. If you define it in the right way, all you need in a well-ordered set. Cardinality doesn't enter into it. Concerning the argument "All A are B, x is an A, therefore x is a B". It is not true that the conclusion is true only if the two assumptions are true. It is not even true that the argument is valid only if the assumptions are true. What is true is that we are guarenteed that the conclusion is true only if the assumptions are true. Thanks for your indulgence. -smL ------------------------------ Date: 17 September 1984 1419-EDT From: Lee Brownston at CMU-CS-A Subject: Pittsburghese figured out [Forwarded from the CMU bboard by Laws@SRI-AI.] The way Pittsburghers talk is a sure source of amusement for newcomers to this area. Most attention is devoted to diction, especially to the idioms. Although the latter are no more nor less illogical than any other idioms, they are easily identified and likely to be unfamiliar. Over the past couple years, I've been trying to figure out the system of phonology. I'm still working on the suprasegmentals, but I have some preliminary results on vowels and consonants that may be of some interest. As far as I can tell, the only consonantal departure from Standard American English is that the final 'g' is omitted from the present progressive to the extent that the terminal sound is the alveolar nasal rather than the palatal nasal continuant. This pronunciation is of course hardly unique to Pittsburgh. The vowels are much more interesting. The 'ow' sound is pronounced 'ah', as in 'dahntahn'. Confusion between, say, "down" and "don" is avoided since the 'ah' sound has already vacated: it is pronounced 'aw', as in 'Bawb signed the cawntract'. Similarly, 'aw' has gone to the greener pastures of 'or', as in 'needs worshed'. It appears that the chain ends here. As its discoverer, I shall call this phonological game of musical chairs "the great Pittsburgh vowel movement." ------------------------------ Date: Mon 17 Sep 84 23:38:41-CDT From: David Throop Subject: Humor - Excuse Generation TOWARDS THE AUTOMATIC GENERATION OF EXCUSES by David Throop The Great and Pressing Need for Excuses. There is a huge need in industry for excuses. A recent marketing survey shows that the biggest need in air transport is not for a service to get it there overnight, but for one that takes the blame for it being three weeks late. Every time there is a dockworkers' strike anywhere in the world, titans of commerce who are completely unaffected get on the phone. They then explain to all their customers that every order that is overdue is sitting on that dock, and that they couldn't help it. Then they grin. Because they've got a good excuse. And even the smallest industrial project needs a raft of excuses by the time it finishes. Computers have already helped with this need. Many problems that used to be blamed on the postal service, on the railroads and on telegraph operators are now routinely blamed on computers. "Your check is in the mail" has now been supplemented by "Our computer has been down, and we'll send you your money as soon as the repairman fixes it." Whenever a bridge collapses, specialized teams of system analysts are called in, in order to quickly blame the whole mess on a computer. But computers can do more than this. Computers have a place to play in the generation of excuses; actually coming up with the lies and evasions that keep our economy running. The Structure of Excuses There is a great size range in excuses. Many small excuses can be generated without any AI or other advanced techniques. And there will always be some really big FUBARS that will need humans to come up with appropriate excuses. But in between there is the somewhat stereotyped snafu that can be framed in some structure and has different excuse elements as slots. These are the half-assed excuses, the most fruitful field for knowledge engineering. Where It Came From It has been noted repeatedly in work on computer vision that a subject often does not have all of the necessary information to justify an observation, but that he makes it anyway and supplies some "excuse" to explain why some features are missing. The classic illustration of this problem is in envisioning a chair: the subject may only be able to see three of the legs but assumes a 4-legged chair. Indeed, Dr. Minsky presented such a chair at the AAAI in August. We interview the chair itself after the lecture, and asked it why it came with only three legs. The resulting string of excuses was impressive, and more robust than one might expect from a broken piece of furniture. These included: "I'm not registered with the local chairs' union, so they'd only let me up on stage if I took off one of my legs. "Accounting cut my travel allowance by 18%, so I had to leave my leg back in California. "This is just a demo chair that we put together for the conference. We have a programming team on the West coast that will have implemented another leg by October. "My secretary talked to somebody on the program committee who assured her that I wouldn't have to bring my own legs, and that there would be plenty of legs here in Austin. Then I go here and found they were overbooked. "I felt that three legs was adequate to demonstrate the soundness of the general leg concept, and actually implementing a fourth leg would have been superfluous." This underlined a central observation: making excuses is critical to perception, and is central to intelligence. I mean, think about. Sounding intelligent involves making gross generalizations & venting primitive prejudices and then making plausible excuses for them when they collide with reality. Any imaginable robot that understands the consequences of its action will want to weasel out of them. The 3 legged chair problem yielded a high number of palatable excuses. This toy problem shows the feasibility of generating large numbers of industrial-strength excuses. This goal would free humans from having to justify their actions, leaving them more time to spend on screwing things up. That, after all, seems to be what they are best at. How It Works A user makes request via SNIVEL (Stop-Nagging,-I'm-Verifying-an-Excuse Language), a user-friendly system that nods, clucks sympatheticly, encourages the user to vent his hostility & frustration, and has a large supply of sympathetic stock answers for lame excuses: "You poor dear, I know you were trying as hard as you could. "Well, you can't be blamed for trusting them. "I can certainly see how you couldn't get your regular work done after an emotional shock like that." The program then begins to formulate an excuse appropriate to the problem. Many problems can be recognized trivially and have stock excuses. These can be stored in a hash table and supplied without any search at all: "The dog vomited on it, so I threw it out. "It's in the mail. "I thought you LIKED it when I did that. "Six debates would probably bore the public. "I have a headache tonight. "I trusted in the advice of my accountant/lawyer/broker/good-time mama." If the problem is more complex, SNIVEL enters into a dialog with the user. Even if he wants to take responsiblity for his share of the problem, SNIVEL solicits the user, getting him to blame other people and explain why it wan't REALLY his fault. A report may be late getting to a client, for instance; it may ask what last minute changes the client had requested, and what kinds of problems the user had with a typing pool. SNIVEL shares records with the personnel file, so that it can quickly provide a list of co-workers' absences that problably slowed the whole process down. It has a parsing alogrithm that takes the original work order and comes with hundreds of different parses for each sentence, demonstrating that the original order was ambiguous and caused a lot of wasted effort. One of the central discoveries of AI has been that problems that look easy are often very hard. Proving this rigorously is a powerful tool: it provides the excuse that almost any interesting problem is too hard to solve. So of course we're late with the report. Theoretical Issues Not all the work here has focused on immediate payoffs. We have studied several theoretical issues involved with excuses. We've found that all problems can be partitioned into: 1) Already Solved Problems for which excuses are not needed. 2) Unsolved Problems 3) Somebody Else's Problem We concentrate on (2). We've shown that this class is further dividable. Of particular interest is the class of unsolved problems for which the set of palatable excuses is infinite. These problems never need to actually be solved. We can generate research proposals, programs and funds requests indefinitely without ever having to produce any results. We just compute the next excuse in the series and go on. Remaining problems It is easiest to generate excuses when the person receiving the excuse is either a complete moron or really couldn't care less about the whole project. Fortunately, this is often the case and can be the default assumption. But is often useful to model the receiver of the excuse. We can than calulate just how big a whopper he's likely to swallow. It is, of course, not necessary that the receiver believe the excuse, just that he accepts it. The system is not ready yet able to model why anyone would accept the excuse "Honestly, we're just friends, there's nothing between us at all." But our research shows that most people accept this excuse, and almost no one believes it. The system still has problems understanding different points of view. For instance, it cannot differentiate why "My neighbors were up drinking and fighting and doing drugs and screaming all night, so I didn't get any sleep at all," is a reasonable excuse for being late to work, but "I was up drinking and fighting and doing drugs and screaming all night, so I didn't get any sleep at all," is not. Finally, the machine is handicapped by its looks. No matter how brilliantly it calculates a good excuse, it can't sweep back a head of chestnut hair, fix a winning smile on its face, and say with heartfelt warmth, "Oh, thank you SO much for understanding..." And that is so much of the soul of a truly good excuse. ------------------------------ End of AIList Digest ******************** 20-Sep-84 23:20:22-PDT,12447;000000000000 Mail-From: LAWS created at 20-Sep-84 23:16:28 Date: Thu 20 Sep 1984 23:08-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #122 To: AIList@SRI-AI AIList Digest Friday, 21 Sep 1984 Volume 2 : Issue 122 Today's Topics: AI Tools - Production Systems on Micros, Logic - Deduction & Induction, Project - Traffic Information System, Seminar - Common Sense Thinking, Seminar Series - Theories of Information & NCARAI Series ---------------------------------------------------------------------- Date: Thu 20 Sep 84 11:13:30-CDT From: CMP.BARC@UTEXAS-20.ARPA Subject: Production Systems on Apple The only thing I have seen for Apple is micro-PROLOG + APES (Augmented PROLOG for Expert Systems), marketed in the U.S. by Programming Logic Systems, Inc., 31 Crescent Drive, Milford, CT 06460 (203-877-7988). I have no experience with the system, but the brochures I have seen and the price make it attractive. Micro-PROLOG runs on Apple II with a Z80 card and full l.c. keyboard, on the IBM PC and PC Jr., and on various configurations of Obsborne, Kaypro II, HP 150, TRS 2000, Xerox 820, among others. CP/M 80 systems require at least 48K RAM, while PC/MS DOS needs 128K. APES reportedly runs on any system which supports micro-PROLOG, but the order form lists only PC/MS DOS and CP/M 86 versions (for Apricot, Sirius and IBM PC compatible). APES requires a minimum memory configuration of 128K. In today's inflated market, the license fees of $295 each or $495 for both are not too outrageous. Clark and McCabe's book is included. The only other systems I've heard about are Expert-Ease and M.1 for the IBM PC and TI's Personal Consultant for their Professional Computer. These go for $2000, $12,500 and $3000 each. The literature and reviews of Expert-Ease make it look like a joke (a friendly interface to a table), but neither media has been able to give an example of the system's inductive capabilities. Expert- Ease appears to be able to form rules from examples, but the people writing the brochures and reviews don't seem to be able to understand or convey this. saw M.1 and the Personal Consultant demoed at AAAI. Both are Emycin clones, minus a lot of the frills (and thus, perhaps, minus the bugs). The Personal Consultant seemed more impressive. It is supposedly written in IQLISP, but does not appear to transport to non-TI computers running IQLISP. All of these products seem way overpriced, as university research has made them fairly simple engineering projects. In the case of the Personal Consultant, none of the academics who did the research seem connected with the project. I imagine that Teknowledge (M.1) has some of Emycin's designers on staff, and know that Michie is involved with Expert-Ease. Dallas Webster (CMP.BARC@UTexas-20) ------------------------------ Date: 15 Sep 84 18:16:54-PDT (Sat) From: decvax!mcnc!akgua!psuvax1!simon @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: psuvax1.1140 ....induction (in mathematics) can deal only with integers. (approximate quote). So what else do you expect a formal system to deal with? The only reasonable answer would be "small finite sets (that are equivalent to subsets of integers). Sure, there are non-denumerable sets that are interesting - but only to sufficiently abstract mathematicians. I do not see useful computer systems worrying about large cardinals, determinacy or the continuum. janos simon ------------------------------ Date: 20 Sep 84 17:30-PDT From: mclure @ Sri-Unix.arpa Subject: deduction vs. induction The recent claim in AILIST that 'deduction proceeds from the general (axioms) to the specific (propositions), induction proceeds from the specific to the general.' is not correct. A lucid definition and comparision of both can be found in: LOGIC AND CONTEMPORARY RHETORIC by Kahane Stuart ------------------------------ Date: Wed, 19 Sep 84 23:01:24 BST From: "Dr. A. Sloman" Subject: Project - Traffic Information System [Edited by Laws@SRI-AI.] An Intelligent Collator and Condenser of Traffic Information The Cognitive Studies Programme, at Sussex University, UK, now has an AI/Natural Language project to build a traffic information system. The project is concerned with a system which processes and integrates reports from the police about traffic accidents. It must also make decisions about which motorists are to be informed about these accidents, by means of broadcasts over an (eventually) nationwide cellular radio network. A significant part of the project will involve investigating to what extent unrestricted natural language input can be handled, and how the obvious problems of unexpected and ungrammatical input can be overcome. It will also be necessary to encode rules about intelligent broadcasting strategies for traffic information. A dedicated workstation (probably SUN-2/120) will be provided for the project, as well as access to network facilities and other computing facilities at Sussex University (mostly VAX-based). For information about the project, and/or about the large and growing AI group at Sussex University, please contact Chris Mellish, Arts Building E, University of Sussex, BRIGHTON BN1 9QN, England. Phone (0273)606755 - if Chris is not in ask for Alison Mudd. (Contact via netmail is not convenient at present.) Aaron Sloman ------------------------------ Date: Wed, 19 Sep 84 15:49:20 pdt From: chertok%ucbkim@Berkeley (Paula Chertok) Subject: Seminar - Common Sense Thinking BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A TIME: Tuesday, September 25, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 SPEAKER: John McCarthy, Computer Science Department, Stanford University TITLE: What is common sense thinking? ABSTRACT: Common sense thinking includes a certain collection of knowledge and certain reason- ing ability. Expert knowledge including scientific knowledge fits into the framework provided by common sense. Common sense knowledge includes facts about the conse- quences of actions in the physical and psychological worlds, facts about the pro- perties of space, time, causality and physi- cal and social objects. Common sense rea- soning includes both logical deductive and various kinds of non-monotonic reasoning. Much common sense knowledge is not readily expressible in words, and much that can be usually isn't. The lecture will attempt to survey common sense knowledge and common sense reasoning. It will be oriented toward expressing the knowledge in languages of mathematical logic and expressing the reasoning as deduction plus formal non-monotonic reasoning. ------------------------------ Date: Wed 19 Sep 84 19:55:18-PDT From: Dikran Karagueuzian Subject: Seminar Series - Theories of Information [Forwarded from the CSLI Newsletter by Laws@SRI-AI.] PROJECT ACTIVITIES FOR PROJECT F-1: THEORIES OF INFORMATION The notions information and of informational content are central to much of the work done at CSLI and are emerging as central notions in philosophy, computer science, and other disciplines. Thus we need mathematically precise and philosophically cogent accounts of information and the forms it takes. The F-1 project will hold a series of meetings on various CSLI researchers' approach to the notion of information. The emphasis will be on gaining a detailed understanding of the theories that are being developed and discussing issues in ways that will be helpful in making further progress. Those interested should attend the meetings regularly to help develop a working group with a shared body of knowledge. For this reason, we will not make it a practice to announce individual meetings, which will occur approximately bi-weekly, Tuesdays at 3:15, in the Ventura Seminar Room. The first meeting will be on October 2, when Jon Barwise will speak for a bit about the nature and prospects for a theory of information, followed by Fernando Pereira and/or Stan Rosenschein who will talk about the current state of situated automata theory. ---John Perry ------------------------------ Date: 20 Sep 84 15:26:51 EDT From: Dennis Perzanowski Subject: Seminar Series - Fall AI Seminar Schedule at NCARAI U.S. Navy Center for Applied Research in Artificial Intelligence Naval Research Laboratory - Code 7510 Washington, DC 20375-5000 FALL SEMINAR SERIES Monday, 24 Sept. 1984 Professor Hanan Samet Computer Science Department University of Maryland College Park, MD "Overview of Quadtree Research" Monday, 15 Oct. 1984 Professor Stefan Feyock Computer Science Department College of William and Mary Williamsburg, VA "Syntax Programming" Monday, 22 Oct. 1984 Professor Andrew P. Sage Computer Science Department George Mason University Fairfax, VA "Alternative Representations of Imprecise Knowledge" Monday, 5 Nov. 1984 Professor Edwina Rissland Department of Computer and Information Sciences University of Massachusetts Amherst, MA "Example-based Argumentation and Explanation" Monday, 19 Nov. 1984 Mr. Kurt Schmucker National Security Agency Office of Computer Science Research Ft. Meade, MD "Fuzzy Risk Analysis: Theory and Implication" The above schedule is a partial listing of seminars to be offered this year. When future dates and speakers are confirmed, another mailing will be sent to you. Our meetings are usually held on the first and third Monday mornings of each month at 10:00 a.m. in the Conference Room of the Navy Center for Applied Research in Artificial Intelligence (Bldg. 256) located on Bolling Air Force Base, off I-295, in the South East quadrant of Washington, DC. A map can be mailed for your convenience. Please note that not all seminars are held on the first and third Mondays this fall due to conflicting holidays. Coffee will be available starting at 9:45 a.m. for a nominal fee. IF YOU ARE INTERESTED IN ATTENDING A SEMINAR, PLEASE CONTACT US BEFORE NOON ON THE FRIDAY PRIOR TO THE SEMINAR SO THAT A VISITOR'S PASS WILL BE AVAILABLE FOR YOU ON THE DAY OF THE SEMINAR. NON-U.S. CITIZENS MUST CONTACT US AT LEAST TWO WEEKS PRIOR TO A SCHEDULED SEMINAR. If you would like to speak, be added to our mailing list, or would like more information, contact Dennis Perzanowski. [...] ARPANET: DENNISP@NRL-AIC or (202) 767-2686. ------------------------------ End of AIList Digest ******************** 23-Sep-84 11:42:57-PDT,15654;000000000000 Mail-From: LAWS created at 23-Sep-84 11:38:35 Date: Sun 23 Sep 1984 10:58-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #123 To: AIList@SRI-AI AIList Digest Sunday, 23 Sep 1984 Volume 2 : Issue 123 Today's Topics: AI Tools - OPS5, Expert Systems - Computer Program Usage Consultant, Literature - Introductory Books & IEEE Computer Articles, LISP - VMS LISPS, Logic - Induction and Deduction & Causality, Humor - Slimy Logic Seminar, Seminar - Analysis of Knowledge, Course & Conference - Stanford Logic Meeting ---------------------------------------------------------------------- Date: 21 Sep 84 13:24:47 EDT From: BIESEL@RUTGERS.ARPA Subject: Info needed on OPS5 Any information on compilers/interpreters for the OPS5 language on VAXen will be appreciated. I'm particularly interested in relatively short reviews and/or introductions to the language; a tutorial would be nice. If any of this stuff is available online I'd like to FTP it. Thanx in advance. Biesel@rutgers.arpa ------------------------------ Date: 17 Sep 84 15:28:05-PDT (Mon) From: hplabs!tektronix!uw-beaver!ssc-vax!alcmist @ Ucb-Vax.arpa Subject: Computer Program Usage Consultants? Article-I.D.: ssc-vax.99 I am working on an expert system to advise users setting up runs of a complex aerodynamics program. The project is sort of like SACON, only we're trying to do more. Does anyone know of work in progress that I should know about? I am interested in any work being done on 1. Helping users set up appropriate inputs for a sophisticated analytical or simulation program, 2. Diagnosing problems with the output of such a program, or 3. Interpreting large volumes of numerical output in a knowledgeable fashion. I am looking for current work that people are willing to talk about. Pointers to literature will be appreciated, even though our library is doing a literature search. Please reply by mail! I will send a summary of responses to anybody who wants one. Fred Wamsley Boeing Computer Services AI Center UUCP: {decvax,ihnp4,sdcsvax,tektronix}!uw-beaver!ssc-vax!alcmist ARPA: ssc-vax!alcmist@uw-beaver.ARPA ------------------------------ Date: 15 Sep 84 10:38:00-PDT (Sat) From: pur-ee!uiucdcs!convex!graham @ Ucb-Vax.arpa Subject: "introductory" book on AI?? Article-I.D.: convex.45200003 I would like to learn more about the AI field. I am almost "illiterate" now. I have a PhD in CS from Illinois and 26 years experience in system software such as compilers, assemblers, link-editors, loaders, etc... Can anyone cite a good book or books for the AI field which is comprehensive is tutorial, in the sense that it includes the motivation behind the avenues in AI that it describes, and includes a good bibliography to other works in the field? [Previous AIList discussion on this subject seems to have found Winston's new "Artificial Intelligence" and Elaine Rich's "Artificial Intelligence" to be good textbooks. The three-volume Handbook of AI is also excellent. Older texts by Nils Nilsson and by Bertram Raphael ("The Thinking Computer") still have much to offer. Other recent books cover LISP, PROLOG, and AI programming techniques, as well as expert systems and AI as a business. -- KIL] ------------------------------ Date: Fri 21 Sep 84 10:08:00-PDT From: Ken Laws Subject: Knowledge Engineering Article The September issue of IEEE Computer is devoted to AI systems, with emphasis on the man-machine interface. It's well worth reading. Frederick Hayes-Roth's article seems to be an excellent introduction to knowledge engineering. (The title is "The Knowledge-Based Expert System: A Tutorial," but it is not really an expert-systems overview.) The article by Elaine Rich on natural-language interfaces is also excellent. There are other articles on smart databases, tutoring systems, job-shop control, and decision support systems. There is also an article on a declarative parameter-specification system for Schlumberger's Crystal system. I found the article hard to follow, and I have strong doubts about the desirability of building a domain-independent parameter parser, then using procedural attachment in the parameter declarations to hack in runtime dependencies and domain-specific intelligent behavior. Even if this is to be done, the base program should have the option of requesting parameters only as (and if) they are needed, and should be able to create or alter the declarative structures dynamically at the time the parameters are requested. Given such a system, the declarative structures are simple a convenient way of passing control options to the user-query subroutine. Most of the procedual knowledge belong in the procedural code, not in declarative structures in a separate knowledge base. -- Ken Laws ------------------------------ Date: Sat, 22 Sep 84 14:48:59 EDT From: Gregory Parkinson Subject: VMS LISPS We run Yale's T on VMS and like it a lot. According to our benchmarks it runs (on the average) a little faster than DEC's Common Lisp. The T compiler gets rid of tail recursion which speeds things up a bit, and is about 40 times faster when dealing with labels. Subjectively, working with CL after working with T feels like driving a 76 Caddie Eldorado (power windows, seats, brakes, steering, etc.) after getting used to a Honda CRX. They both get you where you're going, but there's something about the Honda that makes you feel like you're really driving...... Greg Parkinson Cognitive Systems, inc. ------------------------------ Date: 21 Sep 84 3:42:03-EDT (Fri) From: hplabs!hao!seismo!mcvax!vu44!tjalk!dick @ Ucb-Vax.arpa Subject: Proof by induction, fun & entertainment Article-I.D.: tjalk.338 Claim: All elements of an array A[1..n] are equal to its first element. Proof by induction: Starting case: n = 1. Proof: Obvious, since A[1] = A[1]. Induction step: If the Claim is true for n = N, it is true for n = N + 1. Proof: All elements of A[1..N] are equal (premise), and since A[2..N+1] is an array of length N, all its elements are equal also. A[N] is in both (sub-)arrays, so A[1] = A[N] and A[N] = A[N+1] -> A[1] = A[N+1] which makes all of A[1..N+1] equal. End of proof of induction step The starting case and the induction step together prove the Claim. End of proof by induction Courtesy of Dick Grune Vrije Universiteit Amsterdam the Netherlands [ *** Spoiler *** The flaw, of course, is in the statement that "A[N] is in both (sub-)arrays". (I point this out to avoid a flood of mail supplying the answer.) -- KIL] ------------------------------ Date: Fri, 21 Sep 84 08:36 CDT From: Boebert@HI-MULTICS.ARPA Subject: More on induction and deduction More on induction and deduction, along with much other interesting and entertaining discussion, can be found in Proofs and Refutations by Imre Lakatos Cambridge ------------------------------ Date: Fri 21 Sep 84 10:32:39-PDT From: BARNARD@SRI-AI.ARPA Subject: induction vs. deduction In reply to the claim that my statement 'deduction proceeds from the general (axioms) to the specific (propositions), induction proceeds from the specific to the general.' is not correct (according to Kahane, LOGIC AND CONTEMPORARY RHETORIC), see Aristotle, BASIC WORKS OF ARISTOTLE, ed. by R. McKeon, Random House, 1941. ------------------------------ Date: 18 Sep 84 5:54:04-PDT (Tue) From: hplabs!hao!seismo!umcp-cs!chris @ Ucb-Vax.arpa Subject: Re: Causality Article-I.D.: umcp-cs.16 (Apply :-) to entire reply) > What's wrong with event A affecting event B in event A's past? You >can't go back and shoot your own mother before you were born because you >exist, and obviously you failed. If we assume the universe is >consistant [and not random chaos], then we must assume inconsistancies >(such as shooting your own mother) will not arise. It does not, >however, place time constrictions on cause and effect. Who says you can't even do that? Perhaps your existence is actually just a probablility function. If P(existence) becomes small enough you'll just disappear. Maybe that explains all those mysterious disappearances (``He just walked around the horses a moment ago...'') In-Real-Life: Chris Torek, Univ of MD Comp Sci (301) 454-7690 UUCP: {seismo,allegra,brl-bmd}!umcp-cs!chris CSNet: chris@umcp-cs ARPA: chris@maryland ------------------------------ Date: 17 Sep 84 18:21:16-PDT (Mon) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: Now and Then Article-I.D.: wdl1.424 Having spent some years working on automatic theorem proving and program verification, I am occasionally distressed to see the ways in which the AI community uses (and abuses) formal logic. Always bear in mind that for a deductive system to generate only true statements, the axioms of the system must not imply a contradiction; in other words, it must be impossible to deduce TRUE = FALSE. In a system with a contradiction, any statement, however meaningless, can be generated by deductive means. It is difficult to ensure the soundness of one's axioms. See Boyer and Moore's ``A Computational Logic'' for a description of a logic for which soundness can be demonstrated and a program which generates inductive proofs based on that logic. The Boyer and Moore approach works only for mathematical objects constructed in a specific and rigorous manner. It is not applicable to ``real world reasoning.'' There are schemes such as nonmonotonic reasoning which attempt to deal with contradictions. These are not logical systems but heuristic systems. Some risk of incorrect results is accepted in exchange for the ability to ``reason'' with non-rigorous data. A clear distinction should be made between mathematical deduction in rigorous spaces and heuristic problem solving by semi-logical means. John Nagle ------------------------------ Date: 20 Sep 1984 10:44 EDT (Thu) From: Walter Hamscher Subject: Humor & Seminar - Slimy Logic [Forwarded from the MIT bboard by SASW@MIT-MC.] The Computer Aided Conceptual Art Laboratory and Laboratory for Graduate Student Lunch presents SLIMY LOGIC or INDENUMERABLY MANY TRUTH-VALUED LOGIC WITHOUT HAIR by Lofty Zofty The indenumerably many-valued logics which result from the first stage of slime-ification are so to speak "non-standard" logics; but slimy logic, the result of the second stage of slime-ification, is a very radical departure indeed from classical logics, and thereby sidesteps many fruitless preoccupations of logicians such as completeness, consistency, axiomatization, and proof. In this talk I attempt to counter Slimy Logic's low and ever-declining popularity by presenting a "qualitative" view of slimy logic in which such definitions as 2 very true = true and -3/2 not very pretty false = false by the qualitative (i.e. so even people who don't carry around two calculators can understand them) definitions: very true = true and not very pretty false = ugly false I will then use this "qualitative" slimy logic to very nearly prove very much that Jon Doyle is probably not very right about nearly extremely many things. HOSTS: Robert Granville and Isaac Kohane Refreshments will be served Moved to the Third Floor Theory Group Playroom ------------------------------ Date: 20 September 1984 13:30-EDT From: Kenneth Byrd Story Subject: Seminar - Analysis of Knowledge [Forwarded from the MIT bboard by SASW@MIT-MC.] DATE: Wednesday, September 26, 1984 TIME: Refreshments, 3:45pm Lecture, 4:00pm PLACE: NE43-453 TITLE: ``A MODEL-THEORETIC ANALYSIS OF KNOWLEDGE'' SPEAKER: Dr. Joseph Y. Halpern, IBM, San Jose Understanding knowledge is a fundamental issue in many disciplines. In computer science, knowledge arises not only in the obvious contexts (such as knowledge-based systems), but also in distributed systems (where the goal is to have each processor know something, as in Byzantine agreement). A general semantic model of knowledge is introduced, to allow reasoning about statements such as "He knows that I know whether or not she knows whether or not it is raining." This approach more naturally models a state of knowledge than previous proposals (including Kripke structures). Using this notion of model, a model theory for knowledge is developed. This theory enables one to interpret such notions as a "finite amount of information" and "common knowledge" in different contexts. This is joint work with Ron Fagin and Moshe Vardi. HOST: Professor Silvio Micali ------------------------------ Date: Mon 17 Sep 84 09:01:21-PDT From: Jon Barwise Subject: Course & Conference - Stanford Logic Meeting Logic, Language and Computation Meeting The Association for Symbolic Logic (ASL) and the Center for the Study of Language and Information (CSLI) are planning a two-week summer school and meeting, July 8-20, 1985, at Stanford University. The first week (July 8-13) will consist of a CSLI Summer School, with courses on various topics, including PROLOG, LISP, Complexity Theory, Denotational Semantics, Generalized Quantifiers, Intensional Logic, and Situation Semantics. The second week (July 15-20) will be an ASL meeting with invited lectures (in Logic, Natural Language, and Computation), symposia (on "Logic in Artificial Intelligence", "Types in the Study of Computer and Natural Languages", and "Possible Worlds"), and sessions for contributed papers. Those interested should contact Ingrid Deiwiks, CSLI, Ventura Hall, Stanford, CA 94305 (ph 415-497-3084) before November 1, with an indication as to whether they would like to make a reservation for a single or shared room and board in a residence hall, and for what period of time. A more detailed program will be available in November. The program committee consists of Jon Barwise, Solomon Feferman, David Israel and William Marsh. ------------------------------ End of AIList Digest ******************** 23-Sep-84 22:06:48-PDT,14148;000000000001 Mail-From: LAWS created at 23-Sep-84 22:04:18 Date: Sun 23 Sep 1984 21:57-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #124 To: AIList@SRI-AI AIList Digest Monday, 24 Sep 1984 Volume 2 : Issue 124 Today's Topics: Algorithms - Demonstration Idea Wanted, Machine Translation - SIGART Special Issue, Natural Language - A Generalized Phrase Structured Grammar in Prolog, Expert Systems & Logic Programming - Kastner's Preference Rules in Prolog ---------------------------------------------------------------------- Date: 23 Sep 84 19:30:30 PDT (Sun) From: Mike Brzustowicz Subject: Demonstration Idea wanted A non-network friend of mine needs to demonstrate to a class the importance of detailed specifications. He has been trying to find a task which is easy to do but hard to describe, so that half of the class can write descriptions which the other half will follow literally and thereby fail to accomplish the described task. Anyone have any ideas other than tying shoelaces or cooking beef wellington? (Many people don't wear laced shoes and the facilities available aren't up to cooking :-)). Thanks! -Mike ------------------------------ Date: Thu, 20 Sep 84 20:41 EST From: Sergei Nirenburg Subject: SIGART Special Section on Machine Translation ACM SIGART SPECIAL SECTION ON MACHINE TRANSLATION AND RELATED TOPICS A special section on MT and related work is planned for an early 1985 issue of the SIGART Newsletter. The purpose of the section is: 1. To update the knowledge of the new paradigms in MT in the AI community 2. To help MT workers to learn about developments in AI that can be useful for them in their projects 3. To provide the MT community with updated information about current, recent and nascent MT projects 4. To help identify major topics, results and, especially, directions for future research. Contributions are solicited from MT workers, as well as all workers in AI, theoretical, computational and applied linguistics and other related fields who feel that their work has a bearing on MT (machine-aided human translation; automatic dictionary management; parsing and generating natural language; knowledge representation for specialized domains; discourse analysis; sublanguages and subworlds, etc., etc.) A detailed questionnaire to help you in preparing a response is available from the guest editor, Sergei Nirenburg Department of Computer Science Colgate University Hamilton NY 13346 USA (315) 824-1000 ext. 586 nirenburg@umass If you know of people interested in MT-related activities who are not on a net, please let them know about this call. The deadline for submissions is DECEMBER 1, 1984. Electronic submissions are welcome ------------------------------ Date: Thursday, 13-Sep-84 18:49:25-BST From: O'Keefe HPS (on ERCC DEC-10) Subject: Availability of a GPSG system in Prolog [Forwarded from the Prolog Digest by Laws@SRI-AI.] This message is composed of extracts from the ProGram manual. ProGram is a suite of Prolog programs that are intended to permit the design, evaluation, and debugging of computer realizations of phrase structure grammars for large fragments of natural languages. The grammar representation language employed is that known as GPSG (Generalized Phrase Structure Grammar). A GPSG grammar, as far as ProGTram is concerned, has up to nine components as follows: 1. Specification of feature syntax. 2. Immediate dominance rules (ID rules). 3. Metarules which operation on the ID rules. 4. Linear precedence rules (LP rules). 5. Feature coefficient default values. 6. Feature co-occurrence restrictions. 7. Feature aliasing data. 8. Root admissibility conditions. 9. A lexicon. All the major conventions described in the GPSG literature are implemented, including the Head Feature Convention, the Foot Feature Principle (and hence slash categories &c), the Control Agreement Principle, the Conjunct Realisation Principle, lexical subcategorisation and rule instantiation incorporating the notion of privilege. All the major parts of the grammar interpreter code are written in standard Prolog (Clocksin&Mellish). Installation of the system should be fairly simple on any machine of moderate size which supports Prolog. AVAILABILITY 1. The manual is "University of Sussex Cognitive Science Research Paper 35 (CSRP 035) and can be ordered from Judith Dennison, Congitive Studies Programme, Arts E, University of Sussex, Falmer, Brighton BN1 9QN, for 7.50 pounds including postage. 2. ProGram is aprt of the standard Sussex POPLOG system and is included, without extra charge, in all academic issues and updates of the POPLOG system. POPLOG is available to UK academic users for the sum of 500 pounds (special arrangements apply to holders of SERC AI grants who have a VAX running UNIX). Existing UK academic POPLOG users can obtain a free update of the POPLOG system which will include ProGram. POPLOG runs on VAXes under VMS and UNIX, and on Bleasdale BDC 680as under UNIX. [RAOK: The Bleasdale is a 68000, POPLOG is on SUNs too by now.] Non-educational customers (UK & overseas) who want ProGram with POPLOG should order it through System Designers Ltd, Systems House, 1 Pembroke Broadway, Camberley, Surrey GU15 3XH. This company makes POPLOG available to educational institutions in the USA for 995 dollars. 3. Academic users of other Prolog systems can obtain a magnetic tape in UNIX "tar" format of the Prolog code of the ProGram system free, together with a copy of "The Program Manual", provided they pay the tape, postage, package, and handling costs (35 pounds). Copies can be ordered from Alison Mudd, Cognitive Studies Programme, Arts E, University of Sussex, Falmer, Brighton BN1 9QN A cheque for 35 pounds made payable to "The University of Sussex" should be enclosed with the order. I have no connection with POPLOG, ProGram, or (save a recent visit when I picked up the ProGram manual and saw PopLog running on its home ground) with the University of Sussex. Just to make sure you realise what ProGram is and isn't, it IS meant to be a convenient toolkit for *developing* a GPSG grammar, it is NOT meant to be the world's most efficient parser. The manual warns you that "in general, automatic exhaustive parsing with more than a few rules tends to be slow". You shouldn't need to know any Prolog in order to use ProGram. ------------------------------ Date: Friday, 14-Sep-84 21:20:02-BST From: O'Keefe HPS (on ERCC DEC-10) Subject: Interpreting Kastner's Preference Rules in Prolog [Forwarded from the Prolog Digest by Laws@SRI-AI. This is a declarative specification of an expert-system interpreter. -- KIL] I've always been quite impressed by the "EXPERT" stuff being done at Rutgers, and when I read Kastner's thesis Kastner, J.K. @i"Strategies for Expert Consultation in Therapy Planning." Technical Report CMB-TR-135, Department of Computer Science, Rutgers University, October 1983. (PhD thesis) I decided to write an interpreter for his rules in Prolog as an exercise. The first version just came up with the answer, that's the stuff that's commented out below. The second version left behind information for "explanations": chosen(Answer, Reason, WouldHaveBeenPreferred) Answer was the answer, Reason was the text the rule writer gave to explain his default ordering of the treatments, and WouldHaveBeenPreferred are the treatments we'd have preferred in this ordering if they hadn't been contraindicated despite(Answer, Contraindications) means that Answer was contraindicated by each of the problems listed, but it was still picked because the preferred choices had worse problems. rejected(Treatment, Contraindications) means that Treatment was rejected because it had the problems listed. Every treatment will be rejected or chosen. Note: in these two facts the Contraindications are those which were checked and found to be applicable, less severe ones may not have been checked. (This is a feature, the whole point of the code in fact.) You'll have to read Kastner's thesis to see how these rules are used, but if you're interested in Expert Systems you'll want to read it. Why have I sent this to the [Prolog] Digest? Two reasons. (1) someone may have a use for it, and if I send it to the library it'll sink without trace. (2) I'm quite pleased with the "no-explanations" version, but the "explanations" version is a bit of a mess, and if anyone can find a cleaner way of doing it I'd be very pleased to see it. I guess I still don't know how best to do data base hacking. A point which may be interesting: I originally had worst/6 binding its second argument to 'none' where there were no new contraindications. The mess which resulted (though it worked) reminded me of a lesson I thought I'd learned before: it is dangerous to have an answer saying there are no answers, because that looks like an answer. All the problems I had with this code came from thinking procedurally. :- op(900, fx, 'Is '). :- op(899, xf, ' true'). :- compile([ 'util:ask.pl', % for yesno/1 'util:projec.pl', % for project/3 'prefer.pl' % which follows ]). % File : PREFER.PL % Author : R.A.O'Keefe % Updated: 14 September 1984 % Purpose: Interpret Kastner's "preference rules" in Prolog :- public go/0, og/0. :- mode prefer(-, +, +, +), pass(+, +, +, -), pass(+, +, +, +, +, +, -), worst(+, -, +, +, -, -), chose(+, +, +), forget(+, +), compare_lengths(+, +, -), evaluate(+). prefer(Treatment, Rationale, Contraindications, Columns) :- pass(Columns, [], Contraindications, Treatment), append(Pref1, [Treatment=_|_], Columns), !, project(Pref1, 1, Preferred), assert(chosen(Treatment, Rationale, Preferred)). pass([Tr=Tests|U], Cu, Vu, T) :- worst(Tests, Rest, Cu, Vu, Cb, Vb), !, pass(U, [Tr=Rest], Cu, Vu, Cb, Vb, T). pass([T=_|U], C, _, T) :- chose(T, U, C). pass([], [T=_], _, _, C, _, T) :- !, chose(T, [], C). pass([], B, _, _, Cb, Vb, T) :- reverse(B, R), pass(R, Cb, Vb, T). pass([Tr=Tests|U], B, Cu, Vu, Cb, Vb, T) :- worst(Tests, Rest, Cu, Vu, Ct, Vt), compare_lengths(Vt, Vb, R), ( R = (<), C1 = Ct, V1 = Vt, B1 = [Tr=Rest], forget(B, Cb) ; R = (=), C1 = Cb, V1 = Vb, B1 = [Tr=Rest|B] ; R = (>), C1 = Cb, V1 = Vb, B1 = B, assert(rejected(Tr,Ct)) ), !, % moved down from worst/6 for "efficiency" pass(U, B1, Cu, Vu, C1, V1, T). pass([T=_|_], B, _, _, C, _, T) :- chose(T, B, C). worst([Test|Tests], Tests, C, [X|V], [X|C], V) :- evaluate(Test), !. worst([_|Tests], Rest, Cu, [_|Vu], Ct, Vt) :- worst(Tests, Rest, Cu, Vu, Ct, Vt). evaluate(fail) :- !, fail. evaluate(Query) :- known(Query, Value), !, Value = yes. evaluate(Query) :- yesno('Is ' Query ' true'), !, assert(known(Query, yes)). evaluate(Query) :- assert(known(Query, no)), fail. chose(Treatment, Rejected, Contraindications) :- assert(despite(Treatment, Contraindications)), forget(Rejected, Contraindications). forget([], _). forget([Treatment=_|Rejected], Contraindications) :- assert(rejected(Treatment, Contraindications)), forget(Rejected, Contraindications). compare_lengths([], [], =). compare_lengths([], _, <). compare_lengths( _, [], >). compare_lengths([_|List1], [_|List2], R) :- compare_lengths(List1, List2, R). /*---------------------------- % Version that doesn't store explanation information: prefer(Treatment, Rationale, Contraindications, Columns) :- pass(Columns, 0, [], Treatment). pass([], _, [T=_], T) :- !. pass([], _, B, T) :- reverse(B, R), pass(R, 0, [], T). pass([Tr=Col|U], I, B, T) :- worst(Col, 1, W, Reduced), !, ( W > I, pass(U, W, [Tr=Reduced], T) ; W < I, pass(U, I, B, T) ; W = I, pass(U, I, [Tr=Reduced|B], T) ). pass([T=_|_], _, _, T). % no (more) contraindications worst([], _, none, []). worst([Condition|Rest], Depth, Depth, Rest) :- evaluate(Condition), !. worst([_|Col], D, W, Residue) :- E is D+1, worst(Col, E, W, Residue). ---------------------------------------------------------------*/ antiviral(Which) :- evaluate(full_therapeutic_antiviral_dose_recommended), prefer(Which, efficiacy, [pregnancy, resistance, severe_algy, mild_algy ], [ ftft =[fail, rtft, at3, at1 ], fvira=[fail, rvira, av3, av1 ], fidu =[preg, ridu, ai3, ai1 ] ]). go :- antiviral(X), write(X), nl, pp(chosen), pp(despite), pp(rejected). og :- abolish(chosen, 3), abolish(despite, 2), abolish(known, 2), abolish(rejected, 2). ------------------------------ End of AIList Digest ******************** 26-Sep-84 00:05:39-PDT,13064;000000000001 Mail-From: LAWS created at 26-Sep-84 00:03:53 Date: Tue 25 Sep 1984 23:57-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #125 To: AIList@SRI-AI AIList Digest Wednesday, 26 Sep 1984 Volume 2 : Issue 125 Today's Topics: Expert Systems - Foster Care Placements, LISP - Franz Lisp Help, Inductive Proof - The Heap Problem, Machine Translation - Natural Languages as Interlinguas, Seminars - Semantic Modulation & SOAR Intelligent System Administrivia - Current Distribution List ---------------------------------------------------------------------- Date: Sun, 23 Sep 84 22:34 EST From: Ed Fox Subject: Expert System for Foster Care Placements One of my students has begun a project as described below. We are wondering if there are any similar projects that people would be willing to let us know about. Many thanks, Ed Fox. This expert system will provide assistance to social workers charged with finding suitable substitute care placements for children who cannot continue to live with their families. The system's rules will be based on expert input from social workers and an analysis of a social service agency's case records to determine the constellation of child, natural family, and substitute caregivers' characteristics and environmental factors which have been associated with successful placements in the past. Users will be asked for descriptive information about the child for whom a placement is being sought and his/her family. System output will provide the social worker with a description(s) of substitute care settings which can be expected to suit the needs of the particular child and contribute to a successful placement. ------------------------------ Date: 25 Sep 1984 07:59:09-EDT From: kushnier@NADC Subject: Help- Franz Lisp Help! Does anyone have a good practical guide to Franz LISP running under UNIX on a VAX ? Is there a way to list the LISP environment when running the interpreter or do you have to go in and out using the Unix editors? Can you save the LISP envirnment to an editor file while you are in LISP? P.S. I have the Franz LISP manual, but I haven't translated it to English yet. P.S.S I haven't even figured out what language it's written in....... Ron Kushnier kushnier@nadc.arpa [I'm not sure what's possible under Berkeley Unix (if that's what you have) since I'm using a VAX EUNICE system. Our people have rigged the EMACS editor so that it can be called from Franz, provided that you load and then suspend EMACS before starting up Franz. Interpreted functions can thus be edited and newly edited functions can be run; special editor macros facilitate this. 4.1BSD Unix lacks the interprocess mechanisms needed to support this (LEDIT), although EMACS process windows running Franz are possible; 4.2BSD may be more flexible. To examine your environment while in Franz, use the pp (pretty-print) command. You can certainly save an environment; check out the dumplisp and savelisp commands. For a readable Franz tutorial get Wilensky's new LISPcraft book. -- KIL] ------------------------------ Date: 19 Sep 84 14:42:49-PDT (Wed) From: ihnp4!houxm!mhuxj!mhuxn!mhuxl!ulysses!allegra!princeton!eosp1!robison @ Ucb-Vax.arpa Subject: Re: Inductive proof -- the heap problem Article-I.D.: eosp1.1131 BUT! Human beings continually reason inductively on tiny amounts of info, often two or even one case! We have some way of monitoring our results and taking back some iof the inductions that were wrong. AI has to get the hang of this some day... --- Toby Robison ------------------------------ Date: Mon, 24 Sep 84 22:28 EST From: Sergei Nirenburg Subject: natural languages as interlinguas for MT Re: using a natural language as an interlingua in a machine translation system A natural language and an MT interlingua have different purposes and are designed differently. An interlingua should be ambiguity-free and should facilitate automatic reasoning about the knowledge encoded in it. A natural language is designed to be used by truly intelligent speakers and hearers, so that a lot of polysemy, homonymy, anaphoric phenomena, even outright errors can be put up with -- because the understander is so sophisticated. Brevity is at a premium in natural language communication, not clarity. The most recent attempt to use a language designed for humans as an MT interlingua is the Dutch researcher A. Witkam's attempt in his DLT machine translation project. He plans to use Binary-Coded Esperanto (BCE) as the interlingua in a planned multilingual MT system. An analysis of the approach shows that in reality the system involves two complete (transfer-based) translation modules: 1) Source language to BCE; and 2) BCE to Target language. Of many points of criticism possible let me mention just that this approach (in effect, double transfer) has nothing to do with AI methods. If transfer is used, it is not clear why an interlingua should be involved at all. For some more discussion see Tucker and Nirenburg, "Machine Translation: A Contemporary View", in the 1984 issue of the Annual Review of Information Science and Technology. At the same time, it would be nice to see a technical discussion of the system by Guzman de Rojas -- is any such thing available? Sergei ------------------------------ Date: Mon, 24 Sep 1984 15:30 EDT From: WELD%MIT-OZ@MIT-MC.ARPA Subject: Seminar - Semantic Modulation [Forwarded from the MIT bboard by Laws@SRI-AI.] The AI revolving seminar this week is by David McAllester: SEMANTIC MODULATION: A Relevance Based Inference Technique The Reasoning Utility Package RUP provides a set of propositonal inference mechanisms for constructing inference-based data bases and reasoning systems. This talk will present new inference mechanisms which can be incorporated into the RUP architecture. These inference mechansisms reason about quantified formula using a technique based on the "modulation" of the interpretation of free parameters. By modulating the interpretation of free parameters it is possible to perform a wide variety of quantificational inferences without ever "consing" new formulae. The semantic modulation inference mechanism relies on a notion of relevance in propositional reasoning: when a formula is proven one can determine a subset of premises relevant to the proof. The relevant subset is usually smaller than the set of premises actually used in the proof. Semantic modulation is also closely related to the notions of "inheritance" and "virtual copy" used in semantic networks. Time: 2:00PM Wednesday Sept. 26 (THIS Wednesday) Place: 7th Floor Playroom ------------------------------ Date: Tue 25 Sep 84 11:09:13-PDT From: Paula Edmisten Subject: Seminar - SOAR Intelligent System [Forwarded from the Stanford SIGLUNCH distribution by Laws@SRI-AI.] DATE: Friday, September 28, 1984 LOCATION: Chemistry Gazebo, between Physical and Organic Chemistry TIME: 12:05 SPEAKER: John Laird, Xerox Corp. ABSTRACT: SOAR: An Architecture for General Intelligence I will present recent progress in developing an architecture for general intelligence, called Soar. In Soar, all problem solving occurs as search in a problem space and all knowledge is encoded as production rules. I will describe the Soar architecture and then present three demonstrations of its generality and power. 1. Universal Subgoaling: All subgoals are created automatically by the architecture whenever the problem solver is unable to carry out the basic functions of problem solving (so that all subgoals in Soar are also meta-goals). All the power of Soar is available in the subgoals, including creating new subgoals, making Soar a completely reflective problem solver. 2. A Universal Weak Method: The weak methods emerge from knowledge about a task instead of through explicit representation and selection. 3. R1-Soar: Although Soar was designed for general problem-solving, it is also effective in the knowledge-intensive domains of expert systems. This will be demonstrated by a partial implementation of the R1 expert system in Soar. Soar also has a general learning mechanism, called Chunking. Paul Rosenbloom will present this aspect of our work at the SIGLunch on October 5. ------------------------------ Date: Tue 25 Sep 84 14:08:12-PDT From: Ken Laws Reply-to: AIList-Request@SRI-AI Subject: Current Distribution List SIGART has recently been publishing names of companies involved in AI, which started me wondering just where AIList goes. The following are organizations that I mail to directly, as nearly as I can figure out from the net names. In some cases the digest goes to numerous campuses, departments, or laboratories; in others it goes to a single individual. AIList also goes to numerous sites through indirect remailings, particularly through Usenet redistribution. If anyone would like to add to my list, please send a brief message to AIList-Request@SRI-AI.ARPA. GOVERNMENT AND MILITARY: Admiralty Surface Weapons Establishment Air Force Institute of Technology Data Automation Army Armament Research and Development Command Army Aviation Systems Command Army Communications Electronics Command Army Engineer Topographic Laboratory Army Materiel Systems Analysis Activity Defense Communications Engineering Center National Aeronautics and Space Administration National Library of Medicine National Research Council Board on Telecomm.-Comp. Applications National Science Foundation Naval Air Development Center Naval Intelligence Processing System Support Activity Naval Ocean Systems Center Navel Personnel Research and Development Center Navel Research Laboratory Navel Surface Weapons Center Norwegian Defence Research Establishment LABORATORIES AND RESEARCH INSTITUTES: Aerospace Medical Research Laboratory Brookhaven National Laboratory Center for Seismic Studies Center for Studies of Language and Information Jet Propulsion Laboratory Lawrence Berkeley Laboratory Lawrence Livermore Labs Los Alamos National Laboratory MIT Lincoln Laboratory NASA Ames Research Center Norwegian Telecommunication Administration Research Institute Oak Ridge National Laboratory Sandia National Laboratories USC Information Sciences Institute CORPORATIONS AND NONPROFIT ORGANIZATIONS: ACM SIGART Advanced Computer Communications Advanced Information and Decision Systems Bolt Beranek and Newman Inc. Compion Corp. Digital Equipment Corp. Ford Aerospace and Communications Corp. GTE Laboratories General Motors Research Hewlett-Packard Laboratories Honeywell, Inc. Hughes Research IntelliGenetics International Business Machines Kestrel Institute Linkabit Litton Systems Logicon, Inc. Marconi Research Centre, Chelmsford Northrop Research Center Perceptronics Philips Rome Air Development Center SRI International Science Applications, Inc. Software A&E Tektronix, Inc. Texas Instruments The Aerospace Corporation The MITRE Corporation The Rand Corporation Tymshare Xerox Corporation UNIVERSITIES: Boston University Brandeis University Brown University California Institute of Technology Carnegie-Mellon University Clemson University Colorado State University Columbia University Cornell University Georgia Institute of Technology Grinnell College Harvard University Heriot_Watt University, Edinburgh Louisiana State University Massachusetts Institute of Technology New Jersey Institute of Technology New York University Oklahoma State University Rice University Rochester University Rutgers University St. Joseph's University Stanford University State University of New York University College London University of British Columbia University of California (Berkeley, Davis, UCF, UCI, UCLA, Santa Cruz) University of Cambridge University of Delaware University of Edinburgh University of Massachusetts University of Michigan University of Minnesota University of North Carolina University of Pennsylvania University of South Carolina University of Southern California University of Tennessee University of Texas University of Toronto University of Utah University of Virginia University of Washington University of Wisconsin Vanderbilt Virginia Polytechnic Institute Yale University ------------------------------ End of AIList Digest ******************** 26-Sep-84 22:52:37-PDT,10082;000000000000 Mail-From: LAWS created at 26-Sep-84 22:49:35 Date: Wed 26 Sep 1984 22:44-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #126 To: AIList@SRI-AI AIList Digest Thursday, 27 Sep 1984 Volume 2 : Issue 126 Today's Topics: AI & Business - Literature Sought, Expert Systems - Critique Pointer & Teknowledge's M.1 ---------------------------------------------------------------------- Date: 20 Sep 84 10:29:09-PDT (Thu) From: hplabs!sdcrdcf!trwrb!trwspp!jensen @ Ucb-Vax.arpa Subject: AI for Business Article-I.D.: trwspp.582 I hope that I can obtain a list of resources that apply AI techniques to business. Such resources would include: research bulletins, software, books, and conferences. Awhile back, I recall an AI for Business Summary being offered, perhaps one of you still has a copy lying around on disk. I will pass on submissions to requesters, via mail rather than a net posting. Thank you very much for your assistance. James Jensen [I believe that Syming%B.CC@Berkeley is keeping an AI for Business summary, as well as a list of interested individuals. This is still a suitable topic for AIList, of course. -- KIL] ------------------------------ Date: Wed, 26 Sep 1984 10:25 EDT From: Chunka Mui Subject: request for info: commercialization of ai Has anyone seen a report entitled "Commercial Applications of Expert Systems"? The author is Tim Johnson and it is put out by a company in London named OVUM. I'm wondering what perspective the report is written from and whether or not it is worth tracking down. Replies can be sent directly to me at Chunka%mit-oz@mit-mc if general interest in the topic does not exist. Thanks, Chunka Mui ------------------------------ Date: Wed 26 Sep 84 03:44:35-CDT From: Werner Uhrig Subject: A "populist's" view - Jerry Pournelle comments. in Popular Computing, Nov 84, p 59, Jerry writes in his column THE MICRO REVOLUTION about ARTIFICIAL EXPERTS: The computer as diagnostician has definite limits. worth reading as Jerry (love him or hate him) is a sharp and insightful 'populist' (consider this a compliment), who tries to bridge the gap between experts and academia and is doing a credible job at it. If you keep a folder with informative articles about AI, especially with emphasis on medical applications, you'll want to add this one. ------------------------------ Date: Tue 25 Sep 84 20:06:36-PDT From: JKAPLAN@SRI-KL.ARPA Subject: Clarification Regarding Teknowledge's M.1 Product I recently learned that an article by John Dvorak criticizing our M.1 product in the San Francisco Chronicle 7/29/84 was reproduced and distributed to the AIlist. This article presented a distorted and factually incorrect picture of the Teknowledge product. The author made no attempt to contact us for information prior to publishing the article and as far as we know, has not seen the product. The article appears to be based solely on information from a brochure, and hearsay. Based on the tone and content of the article, it was apparently written primarily for entertainment value, and so we decided it would not be fruitful to draft a formal reply. However, the AIlist might be interested in a response. [I added a note to the original article requesting such a response. -- KIL] First about M.1 - M.1 is a knowledge engineering tool that enables technical professionals without prior AI experience to build rule-based consultation systems. It is designed for rapid prototyping of large-scale applications, as well as building small-scale systems. The product includes a four-day hands-on course, extensive documentation, sample systems, training materials, one year of "hot-line" support, and maintenance. M.1 contains a variety of advanced features. Some of interest to the AIlist types include: certainty factors; a multi-window interactive debugging environment; explanation facility; list processing; single- and multi-valued attributes; variables; dynamic overlays of the knowledge base during consultations; presupposition checking; and automatic answer "completion". However, the system was carefully designed so that it can be learned incrementally, i.e. the beginner doesn't have to understand or use these features. An initial CPU costs $12,500 (not $12,000 as stated in the article), which includes training. Additional licenses costs $5,000 with training, and $2500 without. Strategically, M.1 fills a gap between mainframe- or lisp machine-based tools for AI professionals, and a variety of less sophisticated systems available to hobbyists. Turning to the article - Dvorak makes basically three points: 1. The program is overpriced for personal computer software. 2. The program gives bad advice about wine. 3. Expert systems are too complex to run on micros, at least with M.1. Let me respond briefly to each point. 1. M.1 is not targeted to "personal computer owners" the way Wordstar and VisiCalc are. M.1 is not intended, nor is it suitable for, mass distribution. While M.1 can be used effectively without a graduate degree in artificial intelligence, it is still quite a distance from business productivity tools (such as Lotus 1-2-3) for non-technical computer users. Rather, it is a tool for technical professionals. We decided to host the system on the IBM Personal Computer rather than the VAX or other environments because (a) we believed this would be more convenient for our target customers, and (b) it was technically possible without compromising the product. M.1 is priced consistent with similar systems that run on the IBM Personal Computer, such as CAD/CAM tools, or modelling and simulation packages. These systems typically appeal to a specialized audience, and come with extensive training and support (as does M.1). Our customers and the trade press understand the value of and rationale for such systems. Some members of the popular and business press do not. When we receive inquiries from these latter groups, we explain the product positioning and provide appropriate references and data points. We did not have this opportunity with Mr. Dvorak. 2. M.1 comes with a variety of sample knowledge systems, that illustrate various M.1 features and suggest potential areas of application. Skipping past extensive consultations in the M.1 brochure with a Bank Services Advisor and a Structural Analysis Consultant, Mr. Dvorak reprints an edited transcript of a sample system that provides Wine Advice, in an attempt to ridicule the quality of the product. In our brochure, the purpose of the brief wine advisor example is to illustrate that the user's preferences can be taken into account in a consultation, and that the user can change his or her mind part way through a consultation. Initially, the user specifies a preference for red wine, despite the fact that the meal contains poultry. The M.1 knowledge base naturally recommends a set of red wines. Mr. Dvorak's version of the consultation stops at this point. In the balance of the consultation, the user changes to moderately sweet white wines, and is advised to try chardonnay, riesling, chenin blanc, or soave. While it may occasionally provide controversial advice, the wine advisor sample systems was reviewed by two California wine experts before release, who felt that its advice was quite reasonable. 3. Regarding Mr. Dvorak's final point, he is simply wrong. Micros in general, and M.1 in particular, are powerful enough to solve high value knowledge engineering problems. Approximately 200 knowledge base entries (facts and rules) can be loaded at any one time, and can be overlayed dynamically if larger knowledge bases are required, making the only practical limit the amount of disk storage. Through the use of variables and other representational features, the language is more concise and powerful than most of its predecessors. Practical systems such as the Schlumberger Dipmeter Advisor and the PUFF system at the Pacific Medical Center in San Francisco use knowledge bases that could fit easily within the M.1 system without overlays. For pedagogical purposes, we reimplemented a subset of SACON, a system originally developed at Stanford University using EMYCIN, as a sample system. SACON provides advice to structural engineers on the use a complex structural analysis Fortran program. Our sample system demonstrates that M.1 has sufficient functionality at reasonable speed to accomplish this task. (The current version does NOT contain the entire original knowledge base - time and project resource constraints precluded our doing a complete translation. It includes all questions and control rules, which account for about 50% of the original system, but only about half of the judgmental rules, using no overlays. The reimplementation can run the standard consultation examples from the SACON literature.) AIlist readers may be interested to know that M.1 has been selling very well since its introduction in June. Our customers have been extremely pleased with the system - many have prototyped serious applications in a short period of time after taking the course, and at a cost far below their available alternatives. For more serious reviews of M.1, may I refer you to Rosann Stach Manager of Corporate Development and Public Relations Teknowledge Inc 525 University Ave Palo Alto, CA 415-327-6600 Jerry Kaplan Chief Development Officer Teknowledge ------------------------------ End of AIList Digest ******************** 27-Sep-84 23:57:43-PDT,11170;000000000000 Mail-From: LAWS created at 27-Sep-84 23:55:36 Date: Thu 27 Sep 1984 23:49-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #127 To: AIList@SRI-AI AIList Digest Friday, 28 Sep 1984 Volume 2 : Issue 127 Today's Topics: Computer Music - Mailing List, Expert Systems - Windows, Machine Translation - Natural Languages as Interlinguas, Natural Language - Idioms, Logic - Induction and Deduction, Seminar - Anatomical Analogy for Linguistics ---------------------------------------------------------------------- Date: 26 September 1984 1043-EDT From: Roger Dannenberg at CMU-CS-A Subject: Computer Music Mailing List [Forwarded from the CMU bboard by Laws@SRI-AI.] If you are interested in announcements pertaining to computer music (such as the one you are reading), send mail to Dannenberg@cmu-cs-a and I'll put you on my mailing list. First announcement: there will be a seminar on Monday, October 8, from 11 to 1 with pre-presentations of 3 talks from the 1984 International Computer Music Conference. Please let me know if you plan to attend. ------------------------------ Date: Thu 27 Sep 84 10:09:16-MDT From: Stan Shebs Subject: Windows and Expert Systems Has anyone else become bothered by the recent apparent equation between window packages and expert system tools? The recent spiel on Teknowledge's M.1 takes care to mention that it provides windows (along with other features). However, other vendors (for instance all of those at the recent AAAI) seem to emphasize their window and menu capabilities at the expense of actual reasoning capacity. Recent papers on expert systems at both AAAIs and IJCAIs include the obligatory picture of a screen with all the capabilities being shown at once (even if they're not really related to the paper's content). What's going on? Does a window system really have something substantial to offer expert systems development? If so, what is it? Ultra-high bandwidth for display, so that the system doesn't have to decide what the user wants to see - it just shows everything? Do people get entranced by all the pretty pictures? Ease of managing multiple processes (what expert system tools can even employ multiple communicating processes)? We've got zillions of machines with window systems around here, but they seem supremely irrelevant to the process of expert system development (perhaps because I tend to regard a system that requires only low-bandwidth communication to be more inherently intelligent - it has to do more inference to supply missing information). Can anyone give a solid justification for windows being an essential part of an expert systems tool? (Please no one say anything about it being easier to sell tools with flashy graphics...) stan shebs ------------------------------ Date: 26 Sep 1984 09:33-PDT (Wednesday) From: Rick Briggs Subject: natural languages as interlinguas for MT Sergia Nirenburg's statement that "a natural language and an MT interlingua have different purposes and are designed differently" is false and reveals an incorrect premise underlying much linguistic and AI research. There is a natural language which was spoken between 1000 B.C. and 1900 A.D. which was used amongst a scientific community, and which was ambiguity free(in some senses syntax-free) and which fascilitated automatic inference. Instead of saying "John gave Mary a book" these scientists would say "there was a giving event, having as agent John, who is qualified by singularity...etc". I have shown this well-developed system to be equivalent to certain semantic net systems, and in some cases the ancient language is even more specific. The language is an obscure branch of Indo-Iranian of which there are no translations, but the originals are extant. Natural languages CAN serve as interlingua. Rick Briggs briggs@riacs ------------------------------ Date: Thu 27 Sep 84 10:58:36-CDT From: David Throop Subject: Re: Having no crime rate & other text curiosities Continuing the consideration of texts that contain mistakes but are still comprehensible: Another example, this from the Summer '84 issue of Foreign Affairs (p 1077): "In nine months... the [Argentine] peso fell in value by more than 400 percent." ------------------------------ Date: 9 Sep 84 10:06:00-PDT (Sun) From: hplabs!hp-pcd!hpfclk!fritz @ Ucb-Vax.arpa Subject: Re: Inductive Proof - The Heap Problem Article-I.D.: hpfclk.75500005 As an example of improper induction, consider the heap problem. A "heap" of one speck (e.g., of flour) is definitely a small heap. If you add one speck to a small heap, you still have a small heap. Therefore all heaps are small heaps. -- Ken Laws That's a little like saying, "The girl next to me is blonde. The girl next to her is blonde. Therefore all girls are blonde." (Or, "3 is a prime, 5 is a prime; therefore all odd numbers are prime.") An observation of 2 (or 3, or 20, or N) samples does *not* an inductive proof make. In order to have an inductive proof, you must show that the observation can be extended to ALL cases. [I disagree with Gary's analysis of the flaw. I didn't say "if you add one speck to a one-speck heap", I said that you could add one speck to a (i.e., any) small heap. -- KIL] Mathematician's proof that all odd numbers are prime: "3 is a prime, 5 is a prime, 7 is a prime; therefore, by INDUCTION, all odd numbers are prime." Physicist's proof: "3 is a prime, 5 is a prime, 7 is a prime,... uhh, experimental error ... 11 is a prime, 13 is a prime, ...." Electrical Engineer's proof: "3 is a prime, 5 is a prime, 7 is a prime, 9 is a prime, 11 is a prime..." Computer Scientist's proof: "3 is a prime, 5 is a prime, 7 is a prime, 7 is a prime, 7 is a prime, 7 is a prime, 7 is a prime, ..." Gary Fritz Hewlett Packard Co {ihnp4,hplabs}!hpfcla!hpfclk!fritz ------------------------------ Date: Wed 26 Sep 84 10:42:28-MDT From: Stan Shebs Subject: Re: Induction There's another name for "induction" on one case: generalization. Lenat's AM and the Boyer-Moore theorem prover are both capable of doing generalizations, and there are probably others that can do it also. Not too hard really; if you've set up just the right formalism, generalization amounts to easily-implemented syntactic mutations (now all we need is a program to come up with the right formalisms!) stan shebs ------------------------------ Date: 17 Sep 84 9:03:48-PDT (Mon) From: hplabs!pesnta!scc!steiny @ Ucb-Vax.arpa Subject: Re: induction vs. deduction Article-I.D.: scc.156 A point about logical induction that has not come up is what Charles Sanders Peirce (who coined the term "pragmatism") argued that one could never prove anything inductively. We believe that any human will die eventurally and we reason that is so inductively. We do not, however, have records on every human that has ever existed, and humans that are still alive offer no evidence to support the statement "all humans die". Peirce (being pragmatic), did not think we should throw away the principle just because we can't prove anything with it. He suggested renaming it "reduction" (and renaming deduction "abduction"). This would leave the word "induction" available to those special cases where we do have all the evidence. -- Don Steiny - Personetics @ (408) 425-0382 109 Torrey Pine Terr. Santa Cruz, Calif. 95060 ihnp4!pesnta -\ fortune!idsvax -> scc!steiny ucbvax!twg -/ ------------------------------ Date: Wed, 26 Sep 84 17:27:31 pdt From: chertok%ucbkim@Berkeley (Paula Chertok) Subject: Seminar - Anatomical Analogy for Linguistics BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A TIME: Tuesday, October 2, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 SPEAKER: Jerry Sadock, Center for the Advanced Study in the Behavioral Sciences; Linguistics Department, University of Chicago TITLE: Linguistics as Anatomy ABSTRACT: The notion of modularity in linguistic sys- tems is often supported by invoking an ana- tomical metaphor in which the various sub- systems of the grammar are the analogues of the organs of the body. The primitive view of anatomy that is employed supposes that the organs are entirely separate in internal structure, nonoverlapping in function, shar- ply distinguished from one another, and entirely autonomous in their internal opera- tion. There is a great deal of suggestive evidence from language systems that calls many of these assumptions into question and indi- cates that there are transmodular `systems' that form part of the internal structure of various modules, that there is a good deal of redundancy of function between grammati- cal components, that the boundaries of the modules are unsharp, and that the workings of one module can be sensitive to the work- ings of another. These facts do not speak against either the basic notion of modular- ity of grammar or the anatomical analogy, but rather suggest that the structure of grammatical systems is to be compared with a more sophisticated view of the structure of physical organic systems than has been popu- larly employed. The appropriate analogy is not only biologi- cally more realistic, but also holds out the hope of yielding better accounts of certain otherwise puzzling natural language phenomena. ------------------------------ End of AIList Digest ******************** 1-Oct-84 10:19:13-PDT,11726;000000000000 Mail-From: LAWS created at 1-Oct-84 10:16:27 Date: Mon 1 Oct 1984 10:08-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #128 To: AIList@SRI-AI AIList Digest Monday, 1 Oct 1984 Volume 2 : Issue 128 Today's Topics: Education - Top Ten Graduate Programs, Natural Language - ELIZA source request, AI Tools - OPS5 & VMS LISPs & Tektronix 4404 AI Machine, Bindings - Syntelligence, AI Survey - Tim Johnson's Report, Expert Systems - John Dvorak's Column & Windows, Knowledge Representation - Generalization, Machine Translation - Natural Languages as Interlingua ---------------------------------------------------------------------- Date: 22 Sep 84 0:39:37-PDT (Sat) From: hplabs!sdcrdcf!sdcsvax!daryoush @ Ucb-Vax.arpa Subject: Top Ten Article-I.D.: sdcsvax.79 What are the top ten graduate programs in AI? MIT is first I suppose. --id ------------------------------ Date: 24 Sep 84 13:41:12-PDT (Mon) From: hplabs!hpda!fortune!amd!dual!zehntel!zinfandel!berry @ Ucb-Vax.arpa Subject: Humor - Top Ten Article-I.D.: zinfande.199 What are the top ten graduate programs in AI? -- Karyoush Morshedian To the best of my knowledge, NO AI program has ever graduated from an accredited degree-granting institution , though I do know of a LISP program that's a Universal Life Church minister..... Berry Kercheval Zehntel Inc. (ihnp4!zehntel!zinfandel!berry) (415)932-6900 ------------------------------ Date: 26 Sep 84 18:21:17-PDT (Wed) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: Top Ten Article-I.D.: wdl1.437 The Stanford PhD program probably ranks in the top 10. (The MS program is much weaker). ------------------------------ Date: 29 Sep 84 17:39:34-PDT (Sat) From: hplabs!hao!seismo!umcp-cs!koved @ Ucb-Vax.arpa Subject: Re: ELIZA source request Article-I.D.: umcp-cs.171 I would also like a copy of ELIZA if someone could send it to me. Thanks. Larry koved@umcp-cs or koved@maryland.arpa Spoken: Larry Koved Arpa: koved.umcp-cs@CSNet-relay Uucp:...{allegra,seismo}!umcp-cs!koved ------------------------------ Date: 26 Sep 84 18:21:31-PDT (Wed) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: Info needed on OPS5 Article-I.D.: wdl1.438 OPS5 runs in Franz Lisp on the VAX, and can be obtained from Charles Forgy at CMU. It can be obtained via the ARPANET, but an agreement must be signed first. ------------------------------ Date: 26 Sep 84 18:21:46-PDT (Wed) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: VMS LISPS Article-I.D.: wdl1.439 And then, there is INTERLISP-VAX, the Bulgemobile of language systems. ------------------------------ Date: 27 Sep 84 10:12:10-PDT (Thu) From: hplabs!tektronix!orca!iddic!rogerm @ Ucb-Vax.arpa Subject: Tektronix 4404 AI Machine Article-I.D.: iddic.1822 For information on the 4404 please contact your nearest Tektronix AIM Sales Specialist; Tektronix Incorporated. Farwest: Jeff McKenna 3003 Bunker Hill Lane Santa Clara, CA 95050 (408) 496-496-0800 Midwest: Abe Armoni PO Box 165027 Irving, TX. 75016 (214) 258-0525 Northwest: Gary Belonzi 482 Bedford St. Lexington, MA. 02173 (617) 861-6800 Southeast: Reed Phillips Suite 104 3725 National Drive Raleigh, NC. 27612 (919) 782-5624 This posting is to relieve tekecs!mako!janw from fielding responses that she doesn't have time to answer after her initial posting several weeks ago. Thank you. ------------------------------ Date: Fri 28 Sep 84 13:29:11-PDT From: Margaret Olender Subject: NEW ADDRESS FOR SYNTELLIGENCE [Forwarded from the SRI bboard by Laws@SRI-AI.] Syntelligence is pleased to announce their new Headquarters at 100 Hamlin Court P.O. Box 3620 Sunnyvale, CA 94088 408/745-6666 Effective September 1, 1984. ------------------------------ Date: Thu 27 Sep 84 08:18:14-PDT From: C.S./Math Library Subject: Tim Johnson's Report On AIList today, I saw where someone was asking about the report Tim Johnson did on Commercial Applications of AI. It was produced by Ovum Ltd. in England and is available for about $350 from a place in Portola Valley. I have the address at home and can send that to you later. The report covers AI research and applications in the USA and UK but also covers the larger research projects worldwide. It is a well written and researched report. Harry Llull ------------------------------ Date: 28 Sep 1984 15:15:09-PDT From: smith%umn.csnet@csnet-relay.arpa Subject: John Dvorak as an information source I assume that the John Dvorak who wrote the critique of M.1. is the same one that writes a weekly column in InfoWorld. He is not what I would consider a reliable source of technical information about computers. His columns usually consist of gossip and unsupported personal opinion. What he writes can be interesting but I like to see facts once in a while, too. I've read exactly one good column of his -- it was about computer book PUBLISHING rather than about computers or software. He looks to me like a talented individual who spends too much time out of his league, but is respected for it anyway. This is common in the 'popular' computer media these days, I guess. Rick. ------------------------------ Date: Fri 28 Sep 84 10:44:28-PDT From: Tom Dietterich Subject: Re: Windows and Expert Systems Reply to Sheb's flames: No there is no direct relationship between window systems and expert systems. However, the goal of these vendors is to sell software systems that make it easy to CONSTRUCT, DEBUG, and USE expert systems. We know that high bandwidth between programmer and program makes it easier to construct and maintain a program. Similarly, high bandwidth (properly employed) makes it easier to use a program. The goal is to reduce the cognitive load on the user/programmer, not to strive for maximizing the cognitive load on the program. Good software is 90% interface and 10% intelligence. --Tom ------------------------------ Date: Fri 28 Sep 84 11:04:58-PDT From: Tom Dietterich Subject: Generalization Reply to Shebs' other flame: "Induction...Not too hard really;" Shebs comments are very naive. Of course it isn't too hard to construct a MECHANISM that sometimes performs inductive generalizations properly. However, every mechanism developed thus far is very ad hoc. They all rely on "having the right formalism". In other words, the programmer implicitly tells the program how to generalize. The programmer communicates a set of "biasses" or preferences through the formalism. Many of us working in inductive learning suspect that general techniques will not be found until we have a THEORY that justifies our generalization mechanisms. The justification of induction appears to be impossible. Appeals to the Principle of Insufficient Reason and Occam's Razor just restate the problem without solving it. In essence, the problem is: What is rational plausible inference? When you have no knowledge about which hypothesis is more plausible, how do you decide that one hypothesis IS more plausible? A justification of inductive inference must rely on making some metaphysical assertions about the nature of the world and the nature of knowledge. A justification for Occam's razor, for example, must show why syntactic simplicity necessarily corresponds to simplicity in the real world. This can't be true for just any syntactic representation! For what representations is it true? --Tom ------------------------------ Date: Fri 28 Sep 84 14:58:48-PDT From: Bill Poser Subject: Natural languages as MT interlingua I would like to hear more about the language mentioned by briggs@riacs as a natural language suitable for use as an MT interlanguage. Specifically, what is it called and where is it documented? Where did he publish his demonstration that it is equivalent to certain kinds of semantic nets? I would also be interested to hear in what sense he means that it is a natural language. Virtually all known natural languages are ambiguous, in the sense that they contain sentences that are ambiguous, but that does not mean that they cannot be used unambiguously. An example is the use of English in mathematical writing-it is possible to avoid ambiguity entirely by careful choice of syntax and avoidance of anaphora. I wonder whether briggs' language is not of the same sort-a natural language used in a specialized and restricted way. Bill Poser (poser@su-csli,poser@su-russell) ------------------------------ Date: Fri, 28 Sep 84 15:14:41 PDT From: "Dr. Michael G. Dyer" Subject: Natural Languages A recent comment was made that natural languages can serve as an interlingua. I disagree. There's an ancient language used by scientists to communicate that's called "mathematics"... but is that a "natural" language? Natural languages have certain features, namely, ambiguity, reference to complex conceptualizations regarding human affairs, and abbreviated messages (that is, you only say a tiny bit of what you mean, and rely on the intelligence of the listener to combine his/her knowledge with the current context to reconstruct everything you left out). If that ancient language spoken by Iranian scientists was unambiguous and unabbreviated, then it's probably about as "natural" as mathematics is as a language. Then, also, there's LOGLAN, where, when you say (in it) "every sailor loves some woman", you specify whether each sailor has his own woman or whether everyone loves the same woman. Fine, but I'd hate to have to use it as an everyday "natural" language for gettting around. Natural languages are complicated because people are intelligent. The job of AI NLP researchers is to gain insight into natural languages (and the cognitive processes which support their comprehension) by working out mappings from natural languages into formal systems (i.e., realizable on stupid machines). It's hard enough mapping NL into something unambiguous without mapping it into a language that itself must be parsed to remove ambiguities and to resolve contextual references, etc. It's conceivable that a system could parse by a sequence of mappings into a sequence of slightly more formal (i.e., less "natural") intermediate languages. But then disambiguation, etc., would have to be done over and over again. Besides, people don't seem to be doing that. Natural languages and formal languages serve different purposes. English is currently used as an "interlingua" by the world community, but that is using the term "interlingua" in a different sense. The interlingua we need for NLP research should not be "natural". ------------------------------ End of AIList Digest ******************** 2-Oct-84 09:27:47-PDT,15108;000000000000 Mail-From: LAWS created at 2-Oct-84 09:22:53 Date: Tue 2 Oct 1984 09:18-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #129 To: AIList@SRI-AI AIList Digest Tuesday, 2 Oct 1984 Volume 2 : Issue 129 Today's Topics: Bindings - Syntelligence Address Correction, Induction - Induction on One Case, Machine Translation - Sanskrit, Humor - Onyx BC8820 Stone Block Reader, Seminar - Learning in SOAR, Conference - Knowledge-Based Command and Control ---------------------------------------------------------------------- Date: 01 Oct 84 1144 PDT From: Russell Greiner Subject: Syntelligence: Address Correction Syntelligence, an AI company specializing in building expert systems for business applications, has just moved. Its new address and phone number are Syntelligence 1000 Hamlin Court [not 100] PO Box 3620 Sunnyvale, CA 94088 (408) 745-6666 Dr Peter Hart, its president, can also be reached as HART@SRI-AI.arpa. (This net address should only be used for professional (e.g., AAAI related) reasons.) ------------------------------ Date: Mon 1 Oct 84 14:10:23-MDT From: Stan Shebs Subject: Re: Induction on One Case (My my, people seem to get upset, even when I think I'm making noncontroversial statements...) It wasn't clear whether Tom Dietterich (and maybe others) understood my remark on induction. I was merely pointing out that "induction on one case" is indistinguishable from "generalization". Simple-minded generalization IS easy. Suppose I have as input a Lisp list (A B), (presumably the first in a stream), and I tell my machine to create some hypotheses about what it expects to see next. Possible hypotheses are: (A B) - the machine expects to see (A B) forever (?X B) - the machine expects to see 2nd element B (A ?X) - similarly (?X ?Y) - 2-element lists Since these are lists, presumably one could get more elaborate... (?X ?Y optional ?Z) ... And end up with "the most general hypothesis": ?X All of these patterns can be produced just by knowing how to form Lisp lists; I don't think there's any hidden assumptions or biases (please enlighten me if there are). I would say that in general, one can exhaustively generate all hypotheses, when the domains are completely specified (i.e. a pattern like ( B) for the above example has an undefined entity "or" which has nothing to do with Lisp lists; one would have to extend the domains in which one is operating). Generating hypotheses in a more reasonable order is completely domain-dependent (and no general theory is known). Getting back to the example, all of the hypotheses are equally plausible, since there is only one case to work from (unless one wants to arbitrarily rank these hypotheses somehow; but none can be excluded at this point). I agree that selecting representations is very hard; there's not even any consensus about what representations are useful, let alone about how to select an appropriate one in particular cases. (Have I screwed up anywhere in this? I really wasn't intending to flame...) stan shebs ------------------------------ Date: 1 Oct 1984 16:01-PDT (Monday) From: Rick Briggs Subject: Sanskrit In response to the flood of messages I recieved concerning the ambiguity-free natural language, here is some more information about it. The language is a branch of Sastric Sanskrit which flourished between the 4th century B.C and 4th century A.D., although its beginnings are somewhat older. That it is unambiguous is without question. (I am writing two papers, one for laymen and one for those with AI background). A more interesting question is one posed by Dr. Michael Dyer, that is "is it a natural language?". The answer is yes, it is natural and it is unambiguous. It would be difficult to call a language living and spoken for over a millenium with as rich a literature as this langauge has anything but a natural language. The problem is that most (maybe all) of us are used to languages like English (one of the worst) or other languages which are so poor as vehicles of transmission of logical data. We have assumed that since all languages known have ambiguity, that it is a necessary property of natural languages, but there is no reason to make this assumption. The complaint that it is awkward to speak with the precision required to rule out ambiguity is one based on (I would guess) the properties of Engish or other common Indo-European languages. If one were to take a specific formulation such as a semantic net and "read" it in English the result is a cumbersome mass of detail which nobody would be willing to use in ordinary communication. However, if one were to take that same semantic net and translate it into the language I am studying you get (probably) one very long word with a series of affixes which convey very compactly the actual meaning of the semantic net. In other words, translations from this language to English are of the same nature as those from a semantic net to English (hence the equivalence to semantic nets), one compact structure to a long paragraph. The facility and ease with which these Indians communicated indicates that it is possible for a natural language to serve all purposes of artificial languages based on logic. If one could say what one wishes to say with absolute clarity (although with apparent redundancy) in the same time and with the same ease as you say part of what you mean in English, why not do so? And if a population actually got used to talking in this way there would be much more clarity and less confusion in our communication. Sastric Sanskrit allows you to say WHAT YOU MEAN without effort. The questions "Can you elaborate on that?" or "What exactly are you trying to say?" would simply not come up unless the hearer wished to go to a deeper level of detail. This language was used in much the same way as language found in technical journals today. Scientists would communicate orally and in writing in this language. It is certainly a natural language. As to how this is accomplished, basically SYNTAX IS ELIMINATED. Word order is unimportant, speaking is thus comparable to adding a series of facts to a data-base. What interests me about this language is: 1) Many theories derived recently in Linguistics and AI were independently in use over a thousand years ago, without computers or any need to eliminate ambiguity except for precise thinking and communication 2) A natural language can serve as a mathematical (or artificial language) and thus the dichotomy between the two is false. 3) There are methods for translating "regular" Sanskrit into Sastric Sanskrit from which much could be learned from NLP research. 4) The possibilities of this language serving as interlingua for MT. There are no translated texts and it takes Sanskrit experts a very long time to analyze the texts, so a translation of a full work in this language is a way off. However, those interested can get a hold of "Vaiyakarana-Siddhanta-Laghu-Manjusa" by Nagesha Bhatta. Rick Briggs NASA Ames ------------------------------ Date: Thu, 27 Sep 84 16:05:37 edt From: Walter Hamscher Subject: Onyx BC8820 Stone Block Reader [Forwarded from the MIT bboard by Laws@SRI-AI.] Professor Petra Hechtman of the Archaeology Dept has an Egyptian tombstone written in Hieroglyphs on an Onyx C8002 system running ONYX IV.II that he needs to read. The Onyx system that the block was written with has died (legend has it that it is archived in the temple of Tymsharin). He needs to get the data off the rock soon so that the exact date of Graduate Student Lunches can be calculated (the most recent prediction fixes the date of the next "bologna eclipse" as Friday the 28th at noon in the Third Floor Playroom, hosted by David "Saz" Saslov and Mike "Mpw" Wellman). According to Data Gene-rock, the original Filer was 1/4 cubit, 6250 spd (strokes per digit), 90 RAs, up to 10K BC. Anyone who has, knows of, or has chips off the original device that might be able to decipher the stone, please contact Prof. Hechtman at x5848, or at /dev/null@mit-htvax. ------------------------------ Date: Mon 1 Oct 84 10:25:14-PDT From: Paula Edmisten Subject: Seminar - Learning in SOAR [Forwarded from the Stanford SIGLUNCH distribution by Laws@SRI-AI.] DATE: Friday, October 5, 1984 LOCATION: Chemistry Gazebo, between Physical and Organic Chemistry TIME: 12:05 SPEAKER: Paul S. Rosenbloom Assistant Professor ABSTRACT: Towards Chunking as a General Learning Mechanism Chunks have long been proposed as a basic organizational unit for human memory. More recently chunks have been used to model human learning on simple perceptual-motor skills. In this talk, I will present recent progress in extending chunking to be a general learning mechanism by implementing it within a general problem solver. Combining chunking with the SOAR problem-solving architecture (described by John Laird in the SigLunch of September 28) we can take significant steps toward a general problem solver that can learn about all aspects of its own behavior. The combination of a simple learning mechanism (chunking) with a sophisticated problem-solver (SOAR) yields: (1) practice speed-ups, (2) transfer of learning between related tasks, (3) strategy acquisition, (4) automatic knowledge-acquisition, and (5) the learning of general macro-operators of the type used by Korf (1983) to solve Rubik's cube. These types of learning are demonstrated for traditional search-based tasks, such as tic-tac-toe and the eight puzzle, and for R1-SOAR (a reformulation of a portion of the R1 expert system in SOAR). This work has been pursued in collaboration with John Laird (Xerox PARC) and Allen Newell (Carnegie-Mellon University). ------------------------------ Date: 24 Sep 1984 18:13-EDT From: ABN.CJMERRICK@USC-ISID.ARPA Subject: Conference - Knowledge-Based Command and Control SYMPOSIUM & EXHIBITION ON "ARTIFICIAL INTELLIGENCE" TO BE HELD IN KANSAS CITY, MISSOURI "THE ROLE OF KNOWLEDGE BASED SYSTEMS IN COMMAND & CONTROL" SPONSORED BY: KANSAS CITY CHAPTER OF AFCEA OCTOBER 17-19, 1984 The Kansas City Chapter of the Armed Forces Communications and Electronics Association is proud to announce that it is sponsoring its Second Annual Symposium and Exhibition to discuss the applicability of artificial intelligence and knowledge based systems to command and control requirements, in both the military and commercial environments. The Symposium will be enhanced by the presence of hardware and software exhibits, representing advances in technology related to the theme. Highlights of the Symposium will include noted individuals such as Dr. Joseph V. Braddock of the BDM Corporation addressing user perspectives of utilizing knowledge based systems to fulfill command and control needs. Dr. Robert W. Milne of the Air Force Institute of Technology will address AI technology and its application to command and control. A luncheon presentation will be given by Lieutenant General Carl E. Vuono, Commander, Combined Arms Center, Fort Leavenworth and Deputy Commander, Training and Doctrine Command. General Donn A. Starry (Ret), Vice President and General Manager, Space Missions Group of Ford Aerospace and Communications Corporation will be the guest speaker following the evening meal on Thursday. The Symposium and Exhibition will be held over a three-day period commencing with an opening of the exhibit area and a cocktail and hors d'oeuvres social on October 17, 1984. Technical sessions will begin at 8:00 a.m. on October 18. The format of the technical presentation will consist of two high intensity panel discussions, a session in which pertinent papers will be presented and two guest lectures. ABBREVIATED AGENDA WEDNESDAY, 17 OCTOBER 1984 1200-1700 Check in & Registration 1700-1900 Welcome Social & Exhibits Open THURSDAY, 18 0CTOBER 1984 0800-1145 SESSION I - Panel Discussion: "Status and Forecast of of AI Technology as it applies to Command and Control" Panel Moderator: Mr. Herbert S. Hovey, Jr. Director, U.S. Army Signals Warfare Laboratory Vint Hill Farms Station Warrenton, Virginia 22186 1145-1330 Luncheon/Guest Speaker: Lieutenant General Carl E. Vuono Commander, U.S. Army Combined Arms Center Deputy Commander, Training and Doctrine Command Fort Leavenworth, Kansas 66207 1330-1700 SESSION II - Presentation of Papers 1700-1830 Social Hour 1830-2030 Dinner/Evening Speaker: General Donn A. Starry (Ret) Vice President & General Manager Space Missions Group of Ford Aerospace and Communications Corporation FRIDAY, 19 OCTOBER 1984 0800-1200 SESSION III - Panel Discussion: "User Perspectives of Pros and Cons of Knowledge Based Systems in Command and Control" Panel Moderator: Brigadier General David M. Maddox Commander, Combined Arms Operations Research Activity Fort Leavenworth, Kansas 66027 To make reservations or for further information, write or call: AFCEA SYMPOSIUM COMMITTEE P.O. Box 456 Leavenworth, Kansas 66048 (913) 651-7800/AUTOVON 552-4721 MILITARY POC IS: CPT (P) CHRIS MERRICK CACDA, C3I DIRECTORATE FORT LEAVENWORTH, KANSAS 66027-5300 AUTOVON: 552-4980/5338 COMMERCIAL: (913) 684-4980/5338 ARPANET: ABN.CJMERRICK ------------------------------ End of AIList Digest ******************** 3-Oct-84 11:04:26-PDT,13138;000000000000 Mail-From: LAWS created at 3-Oct-84 11:02:08 Date: Wed 3 Oct 1984 10:56-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #130 To: AIList@SRI-AI AIList Digest Wednesday, 3 Oct 1984 Volume 2 : Issue 130 Today's Topics: Games - Chess Program, Pattern Recognition - Minimal Spanning Trees, Books - Tim Johnson's Report, Academia - Top Graduate Programs, AI Tools - OPS5 & Windows, Games - Computer Chess Tournament & Delphi Game ---------------------------------------------------------------------- Date: Tue, 2 Oct 84 21:46:14 EDT From: "David J. Littleboy" Subject: Chess Request I would like to acquire a state of the art chess program, preferably better than USCF 1500, to run on a 68000 based machine (an Apollo). Something written in any of the usual languages (C, Pascal) would probably be useful. Since I intend to use it as an opponent for the learning program I am building, I would also like the sources. I am, of course, willing to pay for the program. Any pointers would be greatly appreciated. Alternatively, does anyone know of a commercial chess machine with an RS-232 port? Thanks much, David J. Littleboy Littleboy@Yale ...!decvax!yale!littleboy By the way, the basic theoretical claim I start from is that the "problem space" a chess player functions in is determined not so much by the position at hand, as by the set of ideas, plans, and experiences he brings to bear on that position. Thus I view chess as a planning activity, with the goals to be planned for deriving from a player's experiences in similar positions. ------------------------------ Date: 2 Oct 1984 11:25-cst From: "George R. Cross" Subject: MST distributions [Forwarded from the SRI bboard by Laws@SRI-AI.] I am interested in references to the following problem: Suppose we have n-points uniformly distributed in a subset S contained in p-dimensional Euclidean space R^p: 1.What is the distribution of the largest length of the Minimum Spanning Tree (MST) over the n-points? Assume Euclidean distance is used to define the edge weights. 2.What is the distribution of the length of edges in the MST? 3.What is the distribution of the size of the maximal clique? Asymptotic results or expected values of these quantities would be interesting also. We expect to make use of this information in cluster algorithms. Thanks, George Cross Computer Science Louisiana State University CSNET: cross%lsu@csnet-relay ------------------------------ Date: Tue 2 Oct 84 09:54:13-PDT From: C.S./Math Library Subject: Tim Johnson's Report The Commercial Application of Expert Systems Technology by Tim Johnson is a 1984 publication from Ovum Ltd., 14 Penn Road, London N7 9RD, England. It is also available from IPI, 164 Pecora Way, Portola Valley, Ca. 94025 and sells for $395. The report is 382 pages and primarily covers expert systems research in the USA and UK although it also describes some of the larger research projects worlwide. Harry Llull, Stanford University Math/CS Library ------------------------------ Date: 29 Sep 84 20:19:50-PDT (Sat) From: decvax!ittvax!dcdwest!sdcsvax!daryoush @ Ucb-Vax.arpa Subject: Re: Top Ten Article-I.D.: sdcsvax.149 Stanford is defintely one of the 3 best, if not THE best. --id ------------------------------ Date: 3 Oct 84 11:41:55 EDT From: BIESEL@RUTGERS.ARPA Subject: OPS5 info summary. Thanks are due to all the folks who responded to my request for information on OPS5. What follows is a summary of this information. There are at least three version of OPS5 currently available: 1) DEC Compiler QA668-CM in BLISS, available to 2 and 4 year degree granting institutions for $1000. Documentation: AA-GH00A-TE Forgy's Guide AA-BH99A-TE DEC's User Guide 2)Forgy's version (Charles.Forgy@CMU-CS-A), running under Franz Lisp on VAXen. A manual is also available from the same source. 3)A T Lisp version created by Dan Neiman and John Martin at ITT (decvax!ittvax!wxlvax!martin@Berkeley). This version is also supported by some software tools, but cannot be given away. For costs and procedures contact John Martin. Short courses on OPS5 are available from: Smart System Technology 6870 Elm Street McLean, VA 22101 (703) 448-8562 Elaine Kant and Lee Brownston@CMU-CS-A, Robert Farrell@Yale and Nancy Martin at Wang Labs are writing a book on OPS5, to be published this Spring by Addison_Wesley. Regards, Pete ------------------------------ Date: Mon 1 Oct 84 14:35:13-MDT From: Stan Shebs Subject: Summary of Window Responses I got several replies to my question about the relation between windows and expert systems. The consensus seemed to be that since an expert system development environment is like a programming environment, and since PEs are known to benefit from having multiple windows available, windows are an important part of expert system tools. Incidentally, the issue of graphics is orthogonal - graphics is useful in a great number of applications (try describing the weirder geologic formations in words!), although perhaps not all. I have a little trouble with both assumptions. I looked in my nifty collection of reprints, "Interactive Programming Environments" (Barstow, Shrobe, and Sandewall, eds., pub. by McGraw-Hill), and found no research supporting the second assertion. Its main support appeared to be anecdotal. My own anecdotal experience is that even experienced users spend an inordinate amount of clock time trying to do something right, but are not aware of just how much time they're taking (pick a menu item, oops!, undo, try again, then search all over the screen for 5 chars of text, then go through an elaborate sequence of ops to grab those chars, paste them in the wrong place when your mouse hand jiggles, delete, and try again, etc). It's interesting to note that Winograd's two papers (from 1974 and 1979) talk about all kinds of things that a PE should have, but with no mention of graphics anywhere. The first assertion appears to be true, and is a sad comment on the sophistication of today's expert system tools. If expert system environments are just PEs, why not just supply PEs? What's the important difference between a Lisp stack backtrace and a rule system backtrace? Why can't today's expert system tools at least provide a TMS and some detailed explanation facilities? Why hasn't anybody included some meta-level knowledge about the tool itself, as opposed to supplying an inscrutable block of code and a (possibly correct) user's manual? I don't understand. It seems as though the programming mentality reigns supreme (if you don't understand that remark, go back and carefully reread Winograd's 1979 paper "Beyond Programming Languages" (in CACM, and reprinted in the abovementioned book). stan shebs ------------------------------ Date: Tue Oct 2 12:24:29 1984 From: mclure@sri-prism Subject: reminder of upcoming computer chess tournament in San Francisco This is a reminder that this coming Sunday (Oct 7) will herald the beginning of the battle of the titans at the San Francisco Hilton "continental parlors" room at 1pm. Cray Blitz the reigning world champion program will attempt to squash the vengeful Belle. Nuchess, a perennial "top-finishing contender" and descendent of Chess 4.5, wants a piece of the action and would be very happy to see the Belle/Cray Blitz battle cause both to go up in a puff of greasy, black smoke, leaving Nuchess as the top dog for the entire year. It promises to be as interesting as it is every year. You don't have to be a computer-freak or chess-fanatic to enjoy the event. Come on by for a rip-roaring time. Stuart ------------------------------ Date: Sun Sep 30 16:02:03 1984 From: mclure@sri-prism Subject: Delphi 15: cruncher nudges bishop The Vote Tally -------------- The winner is: 14 ... Ne8 There were 16 votes. We had a wide mixture. The group seemed to have difficulty forming a plan. Many different plans were suggested. The Machine Moves ----------------- Depth Move Time for search Nodes Machine's Estimate 8 ply h3 6 hrs, 4 mins 2.18x10^ +4% of a pawn (P-KR3) Humans Move # Votes BR ** -- BQ BN BR BK ** 14 ... Ne8 4 ** BP ** -- BB BP BP BP 14 ... Rc8 3 BP ** -- BP -- ** -- ** 14 ... Nh5 3 ** -- ** WP BP -- ** -- 14 ... Nd7 2 -- ** -- ** WP ** BB ** 14 ... Qd7 2 ** -- WN -- WB WN ** WP 14 ... Nxe4 1 WP WP -- ** WQ WP WP ** 14 ... Qb6 1 WR -- ** -- WR -- WK -- Prestige 8-ply The machine's evaluation turned from negative to slightly positive. Apparently it likes this position somewhat but still considers the position even. The Game So Far --------------- 1. e4 (P-K4) c5 (P-QB4) 11. Be2 (B-K2) Nxe2 (NxB) 2. Nf3 (N-KB3) d6 (P-Q3) 12. Qxe2 (QxN) Be7 (B-K2) 3. Bb5+(B-N5ch) Nc6 (N-QB3) 13. Nc3 (N-QB3) O-O (O-O) 4. o-o (O-O) Bd7 (B-Q2) 14. Be3 (B-K3) Ne8 (N-K1) 5. c3 (P-QB3) Nf6 (N-KB3) 15. h3 (P-KR3) 6. Re1 (R-K1) a6 (P-QR3) 7. Bf1 (B-KB1) e5 (P-K4) 8. d4 (P-Q4) cxd4 (PXP) 9. cxd4 (PXP) Bg4 (B-N5) 10. d5 (P-Q5) Nd4 (N-Q5) Commentary ---------- BLEE.ES@XEROX 14 ... Ne8 as 14 ... Nh5?; 15. h3 B:f3 (if 15 ... Bd7?; 16. N:e5 and white wins a pawn) 16. Q:f3 Nf6 (now we've lost the bishop pair, a tempo and the knight still blockades the f pawn and the white queen is active...) (if 16 ... g6?; 16. Bh6 Ng7; 17. g4 and black can't support f5 because the light square bishop is gone) while 14 ... Nd7?; 15. h3 Bh5; 16. g4 Bg6; and black has trouble supporting f5. I expect play to proceed: 15. h3 Bd7 16. g4 g6 17. Bh6 Ng7 18. Qd3 f5 (at last!) 19. g:f5 g:f5 JPERRY@SRI-KL In keeping with the obvious strategic plan of f5, I vote for 14...N-K1. N-Q2 looks plausible but I would rather reserve that square for another piece. SMILE@UT-SALLY 14 ... Nh5. Paves the way for f5. Other possibility is Qd7 first. Either way I believe f5 is the key (as it often is!). REM@MIT-MC I'm not much for attacking correctly, so let's prepare to double rooks: 14. ... Q-Q2 (Qd7) (It also helps a K-side attack if somebody else can work out the details.) VANGELDER@SU-SCORE 14. ... Nxe4 (vote) In spite of what the master says, White can indefinitely prevent f5 by h3, Bd7, g4. Will the computer find this after Ne8 by Black? Stronger over the board is 14 ... Nxe4. If 15. Nxe4 f5 16. N/4g5 f4 and Black regains the piece with advantage. The majority will probably not select this move, which may be just as well, as attack-by-committee could present some real problems. Nevertheless, the computer presumably saw and examined several ply on this line and it would be interesting to see what it thinks White's best defense is. An alternate line for White is 15. Nxe4 f5 16. N/4d2 e4 17. h3 Bh5 18. Bd4 Bg4!? 19. Nxe4 fxe4 20. Qxe4 Bxf3 21. gxf3 Rf4. There are many variations, but most are not decisive in 8 ply, so the computer's evaluation function would be put to the acid test. ACHEN.PA@XEROX 13 ... Nh5 (keep up the pressure) this might provoke 14 g3 Bd7, either 15 Nd2 or h4 to start a counter attack. the black is hoping to exchange the remaining knight with queen's bishop 16 ... Nf4 then maybe attempt to encircle the white with Qb6 attacking the weakside behind the pawns. (note: if 13 ... Nh5 can't 14 ... f5 for the obvious reason) Solicitation ------------ Your move, please? Replies to Arpanet: mclure@sri-prism, mclure@sri-unix or Usenet: ucbvax!menlo70!sri-unix!sri-prism!mclure ------------------------------ End of AIList Digest ******************** 5-Oct-84 10:02:25-PDT,20707;000000000000 Mail-From: LAWS created at 5-Oct-84 09:56:45 Date: Fri 5 Oct 1984 09:50-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #131 To: AIList@SRI-AI AIList Digest Friday, 5 Oct 1984 Volume 2 : Issue 131 Today's Topics: Linguistics - Sastric Sanskrit & LOGLAN & Interlinquas ---------------------------------------------------------------------- Date: Wed, 3 Oct 1984 23:55 PDT From: KIPARSKY@SU-CSLI.ARPA Subject: Sanskrit has ambiguity and syntax Contrary to what Briggs claims, Shastric Sanskrit the same kinds of ambiguities as other natural languages. In particular, the language allows, and the texts abundantly exemplify: (1) anaphoric pronouns with more than one possible antecedent, (2) ambigous scope of quantifiers and negation, (3) ellipses, (4) lexical homonymy, (5) morphological syncretism. Even the special regimented language in which Panini's grammar of Sanskrit is formalized (not a natural language though based on Sanskrit) falls short of complete unambiguity (see Kiparsky, Panini as a Variationist, MIT Press 1979). The claim that Sanskrit has no syntax is also untrue, even if syntax is understood to mean just word order: rajna bhikshuna bhavitavyam would normally mean "the beggar will have to become king", bhikshuna rajna bhavitavyam "the king will have to become a beggar" --- but in any case, there is a lot more to syntax than word order. ------------------------------ Date: Wed, 3 Oct 84 01:23:07 PDT From: "Dr. Michael G. Dyer" Subject: Sastric Sanskrit Re: Rick Briggs' comments on a version of Sastric Sanskrit. Well, I AM incredulous! Imagine. The entire natural language processing problem in AI has already been solved! and a millenium ago! All we need to do now is publish a 'manual' of this language and our representational problem in NLP is over! Since this language can say anything you want, and "mean exactly what you say" and "with no effort", and since it is unambiguous, it sounds like my problems as an NLP researcher are over. I DO have a few minor concerns (still). The comment that there are no translations, and that it takes sanskrit scholars a "very long time" to figure out what it says, makes it sound to me like maybe there's some complex interpretations going on. Does this mean that a 'parser' of some sort is still needed? Also, I'd greatly appreciate a clearer reference to the book (?) mentioned. Who is the publisher? Is it in English? What year was it published? How can we get a copy? Another problem: since this language has an "extensive literature" does that include poetry? novels? Are the poems unambiguous? are there plays on words? metaphor? (Can you say the equivalent of "Religion is the opiate of the masses"? and if not, it that natural? if not, then how are analogical mappings formed?) satire? humor? puns? exaggeration? fantasy? does the language look like a bunch of horn clauses? (most of the phenomena in the list above involve AMBIGUITY of context, beliefs, word senses, connotations, etc. How does the literature avoid these features and remain literature?) Finally, Yale researchers have been arguing that representational systems for story understanding requires explict conceptual structures making use of scripts, plans, goals, etc. Do such constructs (e.g. scripts) exist explicity in the language? does its literature make use of idioms? e.g. "John drove Mary [home]" vs "John drove Mary [to drink]" Also, why is English "worse" than other languages? Chinese has little syntax and it's ambiguous. Latin has very free word order with prefixes and suffixes and it's ambiguous. Both rely heavily on context and implicit world knowledge. Early work by Schank included representing a Mayan dialect (i.e. Quiche') in Conceptual Dependency. Quiche seems to have features standard to other natural languages, so how is English worse? In the book "Reader over Your Shoulder", Graves & Hodge have a humorous piece about some town councilmen trying to write a leash law. No matter how they state it, unhappy assumptions pop up. e.g. "No dogs in the park without a leash" seems to be addressed to the dogs. "People must take their dogs into the park on a leash" seems to FORCE people to drag there dogs into the park (and at what hour?) even if they don't want to do so. etc etc what about reference? does sastric sanskrit have pronouns? what about IT? does IT have THEM? etc if so, how does it avoid ambiguous references? how many different types of pronouns does it have (if any)? Let's have some specific examples. E.g. does it have the equivalent of the word "like"? Before you answer "yes", there's a difference between "John likes newsweek" and "John likes chocolate" In one case we want our computer to infer that John likes to "eat" chocolate (not read it) and in the other case that he likes to read newsweek (not eat it). Sure, I COULD have said "John likes to eat chocolate" specifically. but I can abbreviate that simply to "x likes " and let the intelligent listener figure out what I mean. When I say "John likes to eat chocolate" do I mean he enjoys the activity of eating, or that he feels better after he's eaten? When I say "John likes to eat chocolate but feels terrible afterwards" I used the word "but" because I know it violated a standard inference on the part of the listener. Natural languages are "expectation-based". Does this ancient language require the speaker to explicitly state all inferences & expectations? Like I said already, if this ancient language really does what is claimed, then we should all dump the puny representational systems we've been trying to invent and extend over the last decade and adopt this ancient language as our final say on semantics. Recent work by Layman Allen (1st Law & Technology conference) in normalizing American law shows that the logical connectives used by lawyers are horribly ambiguous. Lawyers use content semantics to avoid noticing these logical ambiguities. Does this brand of sanskrit have a text of ancient law? What connectives did they use? Maybe the legal normalization problem has also already been solved. Did they have a dictionary? If so, can we see some of the entries? How do the dictionary entries combine? No syntax AT ALL? Loglan adds suffixes onto everything and it's plenty awkward. It has people who write poems in it and other "literature" but you can probably pack all loglanners who "generate" loglanese into a single phone booth. Just how many ancient scholars spoke this sanskrit? I look forward to more discussion on this incredible language. -- A still open-minded but somewhat skeptical inquirer ------------------------------ Date: Thursday, 4-Oct-84 23:59:06-BST From: O'KEEFE HPS (on ERCC DEC-10) Subject: An Unambiguous Natural Language? There was a recent claim in this digest that a "branch of Sastric Sanskrit" was an unambiguous natural language. There are a number of points I'd like to raise: (a) If there are no translated texts, and if it takes a very long time for an expert in "ordinary" Sanskrit to read untranslated texts, it seems more than likely that the appearance of being free from ambiguity is an illusion due to our ignorance. (b) Thanks for the reference. But judging by the title you need to know a lot more about Indian languages to read it than most of the readers of this digest, and without knowing the publisher one would have to be thoroughly at home with the literature to even find it. (c) It's news to me that Sanskrit wasn't an Indo-European language. The Greek-English dictionary I have a copy of keeps pointing to Sanskrit roots as if the two languages were related, but what do they know? If Sastric Sanskrit is an Indo-European language, it is astonishing that it alone is unambiguous. It's especially astonishing when the one non-Indo-European language of which I have even the sketchiest acquaintance (Maaori) isn't unambiguous either and when no-one seems to be claiming that Japanese or Chinese or any other common living language is unambiguous. (d) Dead languages are peculiarly subject to claims of perfection. Without a living informant, we cannot tell whether our failure to discover another reading means there isn't one or whether it just means that we're ignorant of a word sense. I suppose this is point (a) again. (e) If a language permits metaphor, it is ambiguous. The word for "see" in ordinary Sanskrit is something like "oide", and I'm told that it can mean "understand" as well as "perceive with the eye". Do we KNOW that the Sastric Sanskrit words for "see", "grasp", and so on were NEVER employed with this meaning? (f) We're actually dealing with an ambiguous term here: "ambiguous". The following definition is the only one I can think of which is not dependent on some "expert's" arbitrary choice: a sentence S in a text is ambiguous if taking into account assumed common knowledge and the context supplied by the rest of the text there is some natural language L such that S has at least two incompatible translations in L. Here's an example: there are four people in a room, A, B, C, D. This is the beginning of the text, and nothing else in the text lets us judge these points, and we've never heard of A,B,C,D before. A says to D: "we came from X." I assume we know exactly what place X is. Now, does A mean that A,B,C and D all came from X? (reminding D) A,B,C came from X? A and D came from X? (he knows B and C are listening) A and one of B and C came from X? We need to distinguish between dual and plural number, and between inclusive first person and exclusive first person. If the language L marks the gender of plural subjects, we may need to know in the case of A and (B or C but not both) which of B and C was intended. Now consider A mentioning to D "that table", assuming that there are several tables in the same room, all of the same sort. We need to know whether the table he is indicating is near D (it can't be near A or he'd say "this table") or whether it is distant from both A and D. Does the branch of Sanskrit in question make all these distinctions? Can every tense in it be translated to a unique English tense? Does it have no broad colour terms such as the "grue" present in several languages? Failing that, by what criterion IS it unambiguous? {What's a better definition of ambiguity? This one strikes most people I've offered it to as too strong.} (g) Absence of syntax is no guarantee of unambiguity. Consider the phrase "blackbird". It doesn't matter how we indicate that black modifies bird, the source of ambiguity is that we don't know whether the referent is some generic bird that happens to be black (a crow, say), or whether this phrase is used as the name of a species. In English you can tell the difference by prosody, but that doesn't work to well with long-dead languages, and if you thought it always meant turdus merula you might never find anything in the fixed stock of surviving texts to reveal the mistake. (h) What evidence is there that this language was spoken? Note that if a text in this language quotes someone as speaking in it, that still isn't evidence that the language was spoken. I've just been reading a book set in Greece, with Greek characters, but the whole thing was in English... Are there historians writing in other languages who say that the language was spoken? (i) There is another ambiguous term: "natural" language. Is Esperanto a natural language? Is Shelta? The pandits were nobody's fools, after all, Panini invented Backus-Naur form for the express purpose of describing Sanskrit, and I am not so contemptuous of the ancient Indians as to say that they couldn't do a better job of designing an artificial language than Zamenhof did. I'm not saying the language isn't unambiguous, just that it's such a startling claim that I'll need more evidence before I believe it. ------------------------------ Date: 3 Oct 84 12:57:24-PDT (Wed) From: hplabs!sdcrdcf!sdcsvax!sdamos!elman @ Ucb-Vax.arpa Subject: Re: Sanskrit Article-I.D.: sdamos.17 Rick, I am very skeptical about your claims that Sastric Sanskrit is an unambiguous language. I also feel you misunderstand the nature and consequences of ambiguity in natural human language. | The language is a branch of Sastric Sanskrit which flourished |between the 4th century B.C and 4th century A.D., although its |beginnings are somewhat older. That it is unambiguous is without |question. Your judgment is probably based on written sources. The sources may also be technical texts. All this indicates is that it was possible to write in Sastric Sanskrit with a minimum of ambiguity. So what? Most languages allow utterances which have no ambiguity. Read a mathematics text. |The problem is that most (maybe all) of us are used |to languages like English (one of the worst) or other languages which |are so poor as vehicles of transmission of logical data. I think you have fallen victim to the trap of the egocentrism. English is not particularly less (or more) effective than other languages as a vehicle for communicating logical data, although it may seem that way to a native monolingual speaker. | The facility and ease with which these Indians communicated |indicates that it is possible for a natural language to serve all |purposes of artificial languages based on logic. How do you know how easily they communicated? I'm serious. And how easily do you read a text on partial differential equations? An utterance which is structurally ambiguous may not be the easiest to read. |If one could say what one wishes to say with absolute clarity (although |with apparent redundancy) in the same time and with the same ease as |you say part of what you mean in English, why not do so? And if a |population actually got used to talking in this way there would be |much more clarity and less confusion in our communication. Here we come to an important point. You assume that the ambiguity of natural languages results in loss of clarity. I would argue that in most cases the structural ambiguity in utterances is resolved by other (linguistic or paralinguistic) means. Meaning is determined by a complex interaction of factors, of which surface structure is but one. Surface ambiguity gives the language a flexibility of expression. That flexibility does not necessarily entail lack of clarity. Automatic (machine-based) parsers, on the other hand, have a very difficult time taking all the necessary interactions into account and so must rely more heavily on a reliable mapping of surface to base structure. | As to how this is accomplished, basically SYNTAX IS ELIMINATED. |Word order is unimportant, speaking is thus comparable to adding a |series of facts to a data-base. Oops! Languages may have (relatively) free word order and still have syntax. A language without syntax would be the linguistic find of the century! In any event, the principal point I would like to make is that structural ambiguity is not particularly bad nor incompatible with "logical" expression. Human speech recognizers have a variety of means for dealing with ambiguity. In fact, my guess is we do better at understanding languages which use ambiguity than languages which exclude it. Jeff Elman Phonetics Lab, Dept. of Linguistics, C-008 Univ. of Calif., San Diego La Jolla, CA 92093 (619) 452-2536, (619) 452-3600 UUCP: ...ucbvax!sdcsvax!sdamos!elman ARPAnet: elman@nprdc.ARPA ------------------------------ Date: Friday, 5 Oct 1984 10:15-EDT From: jmg@Mitre-Bedford Subject: Loglan, properties of interlinguas, and NLs as interlinguas There has been a running conversation regarding the use of an intermediate language or interlingua to facilitate communication between man and machine. The discussion lately has focused on whether or not it is possible or even desirable for a natural language (i.e., one which was made for and spoken/written by humans in some historical and cultural context) to serve in this role. At last glance it would seem to be a standoff between the cans and cannots. It might be interesting to see if a consensus can at least be reached regarding what an interlingua might be like and therefore whether any natural languages or formal ones for that matter would fit or could be made to fit the necessary form. It would seem that a candidate language would possess a fair sample of the following characteristics (feel free to add to or modify this list): 1) small number of grammar rules--to reduce the trauma of learninng a new language, simplify parsing program, and generally speed up the works 2) small number of speech sounds--to ease learning, and, if well chosen, improve the distinction between sounds and thus the apprehensibil- ity of the spoken language 3) phonologically consistent--for similar reasons as 2) above 4) relative freedom from syntactic ambiguity--to ease translation activities and provide an experimental tool for exploring ambiguity in NLs and thought 5) graphologically regular/consistent with phonology--to ease the transition to the interlingua by introducing no new characters and only simple spelling rules 6) simple morphology--to improve the recognizability of words and word types by limiting the structures of legal words to a few and making word construction regular 7) resolvability--to aid in machine and human information extraction, particularly in noisy environments, by combining well-chosen phonology and morphology 8) freedom from cultural or metaphysical bias--to avoid introducing unintended effects due to specific built-in assumptions about the universe that may be contained within the language 9) logical clarity--to ensure the ability to construct the classical logical connections important to semantically and linguistically useful expressions 10) wealth of metaphor--to allow this linguistic feature to be studied and provide a creative tool for expression These features were selected to try to characterize the intent of a hypothetical designer of an interlingua. Possibly no product could fully merge all the features without compromising unacceptably some of the desir- able traits. If this list appears unacceptable, make suggestions and/or additions and deletions until a workable list results. It is likely that no current or historical natural language would combine a sufficient number of the above features to stand out as an obvious choice to use as interlingua. Simplicity, regularity, ease of learning, ease of information extraction, lack of syntactic ambiguity, and the rest are the earmarks of a constructed language. It remains to be seen that a so-constructed language can be used by humans to express unrestrictedly the full range of human thought. In response to Dr. Dyer's comment about loglan, I can testify that it is not all that hard to get around in. It is a "foreign" language, however, and thus takes some learning and getting used to. It does have several of the features that an interlingua would. Only experience will ultimately reveal whether it is "natural" enough to be useful for exploring the rela- tionship between thought and language and formal enough to be machine- realizable. -Michael Gilmer jmg@MITRE-BEDFORD.ARPA ------------------------------ End of AIList Digest ******************** 7-Oct-84 09:52:21-PDT,12693;000000000000 Mail-From: LAWS created at 5-Oct-84 10:30:02 Date: Fri 5 Oct 1984 10:19-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #132 To: AIList@SRI-AI AIList Digest Saturday, 6 Oct 1984 Volume 2 : Issue 132 Today's Topics: Bindings - Query about D. Hatfield, Applications - AI and Business, AI Literature - List of Sources, Academia - Top Graduate Programs, Conference - Fifth Generation at ACM 84, AI Tools - OPS5 & YAPS & Window Systems & M.1, Scientific Method - Induction, Seminar - Natural Language Structure ---------------------------------------------------------------------- Date: Wed, 3 Oct 1984 15:54 EDT From: MONTALVO%MIT-OZ@MIT-MC.ARPA Subject: Query about D. Hatfield Wed., Aug. 29 Computer Science Seminar at IBM-SJ 10:00 A.M. WYSIWYG PROGRAMMING D. Hatfield, IBM Cambridge Scientific Center Host: D. Chamberlin This message appeared some time ago. [Can someone provide] any pointers to the speaker, D. Hatfield? Does he have any papers on the same subject? Thanks. Fanya Montalvo, MIT, AI Lab. ------------------------------ Date: 3 Oct 84 8:39:05-PDT (Wed) From: hplabs!sdcrdcf!sdcsvax!noscvax!bloomber @ Ucb-Vax.arpa Subject: Re: AI for Business Article-I.D.: noscvax.641 I would also be interested in pointers to books or articles that emphasize the business (preferably practical) uses of AI. Thanks ... Mike -- Real Life: Michael Bloomberg MILNET: bloomber@nosc UUCP: [ihnp4,akgua,decvax,dcdwest,ucbvax]!sdcsvax!noscvax!bloomber ------------------------------ Date: Wed, 3 Oct 84 00:05 CDT From: Jerry Bakin Subject: Keeping up with AI research I am interested in following trends and research in AI. What do active AI'ers feel are the important journals, organizations and conferences? Thanks, Jerry Bakin -- Bakin@HI-Multics [I have sent Jerry the list of journals and conferences compiled by Larry Cipriani and published in AIList V1 N43. In short, AI Magazine AISB Newsletter Annual Review in Automatic Programming Artificial Intelligence Behavioral and Brain Sciences Brain and Cognition Brain and Language Cognition Cognition and Brain Theory Cognitive Psychology Cognitive Science Communications of the ACM Computational Linguistics Computational Linguistics and Computer Languages Computer Vision, Graphics, and Image Processing Computing Reviews Human Intelligence IEEE Computer IEEE Transactions on Pattern Analysis and Machine Intelligence Intelligence International Journal of Man Machine Studies Journal of the ACM Journal of the Assn. for the Study of Perception New Generation Computing Pattern Recognition Robotics Age Robotics Today SIGART Newsletter Speech Technology IJCAI International Joint Conference on AI AAAI American Association for Artificial Intelligence TINLAP Theoretical Issues in Natural Language Processing ACL Association of Computational Linguistics AIM AI in Medicine MLW Machine Learning Workshop CVPR Computer Vision and Pattern Recognition (formerly PRIP) PR Pattern Recognition (also called ICPR) IUW Image Understanding Workshop (DARPA) T&A Trends and Applications (IEEE, NBS) DADCM Workshop on Data Abstraction, Databases, and Conceptual Modeling CogSci Cognitive Science Society EAIC European AI Conference Would anyone care to add a list of organizations? -- KIL] ------------------------------ Date: Wed, 3 Oct 84 13:31:08 From: Bob Woodham Subject: Top Graduate Programs I cannot resist offering my contribution but first three comments: 1. A strict linear ordering is rather meaningless so I've simply listed schools alphabetically within two broad categories. 2. Not surprisingly, given my location, I've expanded things to all of North America. There are good programs outside the continent but I'm not qualified to comment. 3. If your favourite school is missing, let that indicate my ignorance rather than a slight. Since this is roughly the advice I give our own students, I'd like to hear more. Category I: Major Strength in all Areas of AI (alphabetic order) CMU, MIT, Stanford Category II: Major Strength in at least one Area of AI, adequate overall (alphabetic order) Illinois, McGill, Penn, Rochester, Rutgers, Texas (at Austin), Toronto, UBC, Yale There are other schools with strengths, or emerging strengths, that are worth considering. Thankfully, I'm already beyond the requested number of ten. Any of the above schools could be an excellent choice, depending on the particular area of interest. ------------------------------ Date: 3 Oct 1984 14:24-PDT From: scacchi%usc-cse.csnet@csnet-relay.arpa Subject: ACM 84 Just a short note to point out that at the 1984 ACM Conference in San Francisco has a number of sessions on AI and Fifth Generation technologies. In particular, there are at least three sessions that focus on the broader social consequences that might arise from the widespread adoption and use of AI systems. The three sessions include: 1. "The Workplace Impacts of Fifth Generation Computing -- AI and Office Automation" on tuesday (9 Oct 84) morning 2. "Social and Organizational Consequences of New Generation Technology" on tuesday afternoon. 3. "Social Implications of Artificial Intelligence" on wednesday afternoon. If you are able to attend the ACM 84 conference and you are interested in discussing or learning about social analyses of AI technology development, then you should try to attend these sessions. -Walt- (Scacchi@Usc-cse via CSnet) ------------------------------ Date: 2 Oct 84 16:03:48-PDT (Tue) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: obtaining OPS-5 Article-I.D.: wdl1.458 OPS-5 is obtained from Charles Forgy at CMU, reached at the following address. Do not contact me regarding this. Forgy, Charles L. (CLF) CHARLES.FORGY@CMU-CS-A Carnegie-Mellon University Computer Science Department Schenley Park Pittsburgh, Pennsylvania 15213 Phone: (412) 578-3612 ------------------------------ Date: Wed, 3 Oct 84 23:39:58 edt From: mark@tove (Mark Weiser) Subject: ops5 and yaps. For those of you interested in ops5, don't forget YAPS. Yaps was described by Liz Allen of Maryland at the '83 AAAI. Yaps, yet another production system, uses Forgy's high speed short cuts for left hands sides which fall into ops5's limited legal lhs, but yaps also allows fully general left hand sides. Yaps second advantage over ops5 is that it is imbedded in the Franz lisp flavors system (also from Maryland), so that one can have several simultaneous yaps objects and send them messages like add-a-rule, add-object-to-database, etc. For more information, mail liz@maryland. Spoken: Mark Weiser ARPA: mark@maryland CSNet: mark@umcp-cs UUCP: {seismo,allegra}!umcp-cs!mark ------------------------------ Date: 1 Oct 84 18:21:18-PDT (Mon) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: Windows and Expert Systems Article-I.D.: wdl1.451 I've noticed this lately too; I've also seen the claim that ``windows were developed ten years ago by the AI community'', but the early Alto effort at PARC, which I saw demonstrated in 1975 by Allen Kay, was not AI-oriented; they were working primarily on improved user interfaces, including window systems. John Nagle ------------------------------ Date: 30 Sep 84 8:30:02-PDT (Sun) From: decvax!mcnc!unc!ulysses!burl!clyde!watmath!water!rggoebel@Ucb-Vax.arpa Subject: Re: Clarification Regarding Teknowledge's M.1 Product Article-I.D.: water.20 I've just read what amounts to an advertisement for Teknowledge's M.1 software product. I can't believe there isn't something to be criticized in a product that comes from such an infant technology? I'd be interested to know what's wrong with M.1? Will Teknowledge give it away to universities to teach students about expert systems? Is SRI-KL using M.1 for anything (note origin of original message)? On a lighter note, what is novel about a software system that supports ``variables?'' Randy Goebel Logic Programming and Artificial Intelligence Group Computer Science Department University of Waterloo Waterloo, Ontario, CANADA N2L 3G1 UUCP: {decvax,ihnp4,allegra}!watmath!water!rggoebel CSNET: rggoebel%water@waterloo.csnet ARPA: rggoebel%water%waterloo.csnet@csnet-relay.arpa [I am not aware of any SRI use of M.1, nor do I know of anyone at SRI who has a financial interest in it. Many people around the country have mailboxes on systems where they once worked or otherwise have incidental access; I assume that is the case here. An SRI group has recently come out with its own micro-based expert system toolkit, SeRIES-PC, a PROSPECTOR derivative. -- KIL] ------------------------------ Date: 1 Oct 84 22:21:20-PDT (Mon) From: hplabs!hpda!fortune!wdl1!jbn @ Ucb-Vax.arpa Subject: Re: Re: Clarification Regarding Teknowle Article-I.D.: wdl1.453 I'd like to see them offer a training version of the program for $50 or so which allowed, say, a maximum of 50 rules, enough to try out the system but not enough to implement a production application. This would get the tool (and the technology) some real exposure. John Nagle ------------------------------ Date: Wed 3 Oct 84 00:05:12-PDT From: Tom Dietterich Subject: re: Induction Well I guess I don't understand Stan Shebs' point regarding induction very well. I agree with everything he said in his message: It is indeed possible to generate all possible generalizations of some fact within some fixed, denumerable domain of discourse. The problem of induction is to infer PLAUSIBLE beliefs from a finite set of examples. Shebs is correct in saying that from any finite set of examples, a very large (usually infinite) set of generalizations can be generated. He is also correct in saying that--in the absence of any other knowledge or belief--all of these generalizations are equally plausible. The problem is that in common-sense reasoning, all of these generalizations are not equally plausible. Some seem (to people) to be more plausible than others. This reflects some hidden assumptions or biases held by people about the nature of the common sense world. ------------------------------ Date: Thu, 4 Oct 84 15:17:51 pdt From: chertok%ucbkim@Berkeley (Paula Chertok) Subject: Seminar - Natural Language Structure BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A TIME: Tuesday, October 9, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 SPEAKER: Gilles Fauconnier, Linguistics Dept, UC San Diego & University of Paris TITLE: Roles, Space Connectors & Identification Paths ABSTRACT: Key aspects of natural language organization involve a general theory of connections linking mental constructions. Logical and structural analyses have overlooked this important dimension, which unifies many superficially complex and disparate phenomena. I will focus here on the many interpretations of descriptions and names, and suggest a reassessment of notions like rigidity, attributivity, or ``cross-world identification.'' ------------------------------ End of AIList Digest ******************** 8-Oct-84 10:17:47-PDT,16005;000000000000 Mail-From: LAWS created at 8-Oct-84 10:07:46 Date: Mon 8 Oct 1984 09:42-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #133 To: AIList@SRI-AI AIList Digest Monday, 8 Oct 1984 Volume 2 : Issue 133 Today's Topics: Bindings - John Hosking Query, Workstations - Electrical CAD/CAE & TI LISP Machine, AI Tools - Graph Display, Expert Systems - Liability, Humor - Theorem Proving Contest, Comments - Zadeh & Poker, Seminar - First Order Logic Mechanization ---------------------------------------------------------------------- Date: Saturday, 6-Oct-84 2:12:41-BST From: O'KEEFE HPS (on ERCC DEC-10) Subject: References wanted Anyone know where I can find anything by John Hosking, now of Auckland University New Zealand? Said to be in expert systems/knowledge representation field. ------------------------------ Date: 3 Oct 84 17:08:44-PDT (Wed) From: hplabs!intelca!qantel!dual!amd!turtlevax!ken @ Ucb-Vax.arpa Subject: Electrical CAE software/hardware Article-I.D.: turtleva.541 We've been gathering information about CAD/CAE for electrical/computer engineering and have been deluged with a foot's worth of literature. No on makes the entire package of what we want, which includes schematic entry, hierarchical simulation, timing verification, powerful functional specification language, finite-state machine generator, PAL primitives, PLA and PROM high-level language specification compiling down to JEDEC format, driver for a Data I/O or more dependable PROM/PAL programmer, transient and frequency analysis (SPICE works well here), symbolic, analytical, and graphical mathematics, etc. We've accepted the fact that we will need to get several packages of software, but are prepared to buy no more than 1 extra piece of hardware, if we can't get software to run on our VAX or Cadlinc workstations. Has anyone used any of the available products? Does anyone have any recommendations? Following is a list of suppliers of CAE tools of some sort, for which I managed to get some literature, and is in no way guaranteed to be complete: Altera Assisted Technology Avera Corporation Cad Internet, Inc. Cadmatics Cadnetix Cadtec CAE Systems Calma Chancellor Computer Corporation Control Data Daisy Design Aids, Inc. Futurenet GenRad HHB Softron Inference Corp. Intergraph Interlaken Technology Corp. Mentor Metalogic, Inc. Metheus Mirashanta Omnicad Corp. Phoenix Racal-Redac Signal Technology, Inc. Silvar-Lisco Step Engineering Symbolics Teradyne Valid Vectron Verstatec Via Systems VLSI Technology, Inc. -- Ken Turkowski @ CADLINC, Palo Alto, CA UUCP: {amd,decwrl,flairvax,nsc}!turtlevax!ken ARPA: turtlevax!ken@DECWRL.ARPA ------------------------------ Date: Fri 5 Oct 84 16:23:15-PDT From: Margaret Olender Subject: TI LISP MACHINE [Forwarded from the SRI-AI bboard by Laws@SRI-AI.] Texas Instruments invites ACM attendees (and AIC-ers) to see the new TI LISP machine demo-ed at the San Francisco Hilton 333 O'Farrel Street Imperial Suite Room #1915 Monday, October 8, 1984 5:00pm - 8:00pm Refreshments and hors d'oeuvers. Bring your ACM badge for admission. ...margaret ------------------------------ Date: Sat 6 Oct 84 23:56:50-PDT From: Scott Meyers Subject: Wanted: info on printing directed graphs I am faced with the need to come up with an algorithm for producing hardcopy of a directed graph, i.e. printing such a graph on a lineprinter or a V80 plotter. Rather than just plopping the nodes down helter-skelter, I will have an entry node to the graph which I will place at the far left of the plot, and then I will want to plot things so that the edges generally point to the right. If anyone has solved this problem or can give me pointers to places where it has been solved, or can offer any other assistance, I would very much like to hear from you. Thanks. Scott [Scott could also use a routine printing graphs top to bottom if that is available. -- KIL] ------------------------------ Date: Sun, 7 Oct 84 13:47:09 pdt From: Howard Trickey Subject: printing graphs [Forwarded from the Stanford bboard by Laws@SRI-AI.] I did a program that takes a graph description and produces a TeX input file which in turn produces a reasonably nice looking graph on the Dover (\special's are used to draw lines at arbitrary angles; I can use Boise by specifying only rectilinear lines, but it doesn't look as good). There's no way to use it as is for the output devices mentioned in the previous message, but the algorithms I used may be of interest. There can be different types of nodes, each drawn with a user-specified TeX macro. The graph description says which nodes there are and of what type, and what edges there are. Edges go to and from symbolically specified points on nodes. The output looks best when the graph is acyclic or nearly acyclic, since that's what my graphs are so I didn't spend time on other cases. The program isn't robust enough or easy enough to use for general use, but I can point people to it. If you need the capability badly enough, it's not too difficult to get used to. It's written in Franz Lisp. Howard Trickey ------------------------------ Date: 3 Oct 84 12:46:11-PDT (Wed) From: decvax!cwruecmp!atvax!ncoast!rich @ Ucb-Vax.arpa Subject: AI decision systems - What are the risks for the vendor? Article-I.D.: ncoast.386 The rapid advance of Artificial Intelligence Software has caused me to wonder about some of the possible legal problems. SITUATION: We are a software vendor that develops an AI software package. this package has been tested and appears to be correct in design and logic. Additionally, the package indicates several alternative solutions as well as stating that there could be alternatives that are overlooked. What risk from a legal standpoint does the developer/vendor have to the user IF they follow the recommendation of the package AND the decision is proven to be incorrect several months later? I would appreciate your opinions and shall post the compiled responses to the net. From: | the.world!ucbvax!decvax!cwruecmp! Richard Garrett @ North Coast Xenix | {atvax!}ncoast!rich 10205 Edgewater Drive: Cleveland, OH |................................... (216) 961-3397 \ 44102 | ncoast (216) 281-8006 (300 Baud) ------------------------------ Date: Sat 6 Oct 84 14:01:30-PDT From: Ken Laws Subject: Liability Just as damning as using an incompetent [software] advisor is failing to use a competent one. If a doctor's error makes you a cripple for life, and if he had available (and perhaps even used) an expert system counceling a better course of treatment, is he not guilty of malpractice? Does the doctor incur a different liability than if he had used/not used a human consultant? The human consultant would normally bear part of the liability. Since you can't sue an expert system, do you sue the company that sold it? The programmer? The theoretician who developed the algorithm? I'm sure there are abundant legal precedents for all of the above. For anyone with the answers to the above, here's an even more difficult problem. Systems for monitoring and interpreting electrocardiograms are commonly adjusted at the "factory" to match the diagnostic style of the purchasing physician. Suppose that the doctor requests that this be done, or even does it himself. Suppose further that he is incompetent at this type of diagnosis (after all, he's buying a system to do it for him), and that customization to match his preferences can be shown to degrade the performance of the software. Is he liable for operating the system at less than full capability? I assume so. Is the manufacturer liable for making the adjustment, or for providing him the means of doing it himself? I would assume that also. What are the relative liabilities for all parties? -- Ken Laws ------------------------------ Date: 4 Oct 1984 09:51 EDT (Thu) From: Walter Hamscher Subject: GSL sponsored Theorem Proving Contest [Forwarded from the MIT bboard by Laws@SRI-AI.] DATE: Friday, 5 October, 12 noon PLACE: 3rd Floor Playroom HOST: Reid Simmons REAGAN vs. MONDALE THEOREM PROVING CONTEST To help the scientific community better assess this year's presidential candidates, GSL (in conjunction with the Laboratory for Computer Research and Analysis of Politics) proudly presents the first Presidential Theorem Proving Contest. The candidates will have 10 minutes to prepare their proofs, 10 minutes to present, and then 5 minutes to criticise their opponents' proofs. A pseudorandom number generator will be used to determine the order of presentation. The candidates will be asked to prove the following theorem: * Let (a + a + a ...) be a conditionally convergent series. 1 2 3 Show by construction that there exists a rearrangement of the a such that i lim (a + ... a ) = 0. n -> inf 1 n Note: To increase public interest in this contest, the theorem will actually be phrased in the following way: Let (deficit + deficit + deficit ...) be a 1980 1981 1982 series with both positive and negative terms. Rearrange the terms so that: lim (deficit + ... deficit ) = $ 0.00 year -> inf 1980 year ------------------------------ Date: 2 Oct 84 21:50:35-PDT (Tue) From: hplabs!ames!jaw @ Ucb-Vax.arpa Subject: Re: Humor & Seminar - Slimy Logic Article-I.D.: ames.548 This B-Board article [on slimy logic] is a master parody, right down to the "so to speak" mannerism. Thanks for the entertainment! I took a couple of courses from Professor Zadeh at Berkeley in the 70s, not just in Fuzzy Logic, but also formal languages, where we all struggled with LALR(1) lookahead sets. The fuzzy controversy was raging then, with Prof. William Kahan, numerical analyst, being Zadeh's arch-enemy. Kahan was a natural devil's advocate, himself none too popular for raving on, in courses on data structures, a bit muchly about the way CDC 6400 Fortrash treated roundoff of the 60th bit. Apparently, there's some bad blood over the size of Zadeh's grants (NSF?) for his fuzzy baby. They both have had tenure for years, so maybe a pie-throwing contest would be appropriate. Anyway, looks like the fuzzy stuff is now making the rounds at MIT. Zadeh, who ironically wrote the book on linear systems (circa 1948), at least got the linguistics department hopping with the fuzzies, influencing the Lakoffs (George, mainly) to trade in their equally ad hoc transformational grammars for fuzzy logic. Kinda soured me on natural language theory, too. I mean, is there life after YACC? Old Lotfi has left an interesting legacy via his children. Zadeh's daughter, I understand is a brilliant lawyer. One son, after getting his statistics Ph.D. at 20 or so, claims to have draw poker figured out. Bluffing is dealt with by simple probability theory. As I remember, "Winning Poker Systems" is one of those "just-memorize-the-equivalent-of- ten-phone-numbers-for-instant-riches" books. He worked his way through school with funds won in Emeryville poker parlors. Not too shabby, but not too fuzzy, either ... -- James A. Woods {ihnp4,hplabs,philabs}!ames!jaw (jaw@riacs.ARPA) [Dr. Zadeh also invented the Z-transform used in digital signal processing and control theory. -- KIL] ------------------------------ Date: 5 Oct 84 18:31:33-PDT (Fri) From: hplabs!hao!seismo!rochester!rocksanne!sunybcs!gloria!colonel @ Ucb-Vax.arpa Subject: Re: fuzzy poker Article-I.D.: gloria.578 One son, after getting his statistics Ph.D. at 20 or so, claims to have draw poker figured out. ... When I was working with the SUNY-Buffalo POKER GROUP, we managed to verify some of N. Zadeh's tables with hard statistics. Anybody who's interested can find some of our results in Bramer's anthology _Computer Game-Playing: Theory and Practice_ (1983). -- Col. G. L. Sicherman ...seismo!rochester!rocksanne!rocksvax!sunybcs!gloria!colonel ------------------------------ Date: 05 Oct 84 1318 PDT From: Carolyn Talcott Subject: Continuing Seminar - FOL & First Order Logic Mechanization [Forwarded from the Stanford bboard by Laws@SRI-AI.] Seminar on FOL: a mechanized interpretation of logic presented by Richard Weyhrauch Time: 4:15 to 6:00 Date: Alternate Tuesdays begining October 9 Place: Room 252 Margret Jacks Hall The topic of this seminar is a description of FOL, a collection of structures that can be used to provide a mechanized interpretation of logic. We will present specific examples of interest for logic, philosophy and artificial intelligence to illustrate how the FOL structures give formal solutions, or at least shed light on, some classical problems. We will also describe the details of FOL, a computer program for constructing these structures. This provides a link between logic and AI. Mechanization is an alternative foundation to both constructive and classical logic. I have always found constructive foundations unconvincing. Taken by itself, it fails to explain how we can understand classical semantics well enough to make the distinction. Even more -- a philosophically satisfactory account of reasoning must explain why in the comparatively well behaved case of mathematical foundations the classical arguments carry conviction for practising mathematicians. On the other hand the use of set theoretic semantics also seems to require infinite structures to understand elementary arguments. This conflicts with the simple observation that people understand these arguments and they are built from only a finite amount of matter. Mechanization provides a semantics that is both finitist and at the same time allows the use of classical reasoning. ------------------------------ Date: Sat, 6 Oct 84 13:56:04 pdt From: Vaughan Pratt Subject: FOL seminar [Forwarded from the Stanford bboard by Laws@SRI-AI.] On the other hand the use of set theoretic semantics also seems to require infinite structures to understand elementary arguments. This conflicts with the simple observation that people understand these arguments ... In my day it was not uncommon for students to reason about all the reals in a finite amount of time - in fact it was even required for exams, where you only had three hours. Whatever has modern mathematics come to? ... and they [people] are built from only a finite amount of matter. By weight and volume, yes, but with elementary particles breeding like rabbits one sometimes wonders about parts count. Now here's a problem spanning particle physics and number theory: if there exists such a thing as an elementary particle, and if there are a fixed finite number of them in an uncharged hydrogen atom at absolute zero, is that number prime? -v ------------------------------ End of AIList Digest ******************** 8-Oct-84 23:11:41-PDT,19482;000000000000 Mail-From: LAWS created at 8-Oct-84 23:06:43 Date: Mon 8 Oct 1984 23:03-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #134 To: AIList@SRI-AI AIList Digest Tuesday, 9 Oct 1984 Volume 2 : Issue 134 Today's Topics: Seminars - AI Control Design & Fault Diagnosis & Composite Graph Theory, Lectures - Logic and AI, Program - Complexity Year at MSRI ---------------------------------------------------------------------- Date: Mon 8 Oct 84 09:31:31-PDT From: Ken Laws Subject: Seminar - AI Control Design From the IEEE Grid newsletter for the SF Bay Area: Some very exciting new ideas on the role of expert systems in control design for AI will be presented at the Oct. 25 meeting of the Santa Clara Valley Control Systems, Man and Cybernetics Society. The talk, by Dr. Thomas Trankle and Lawrence Markosian of Systems Control Technology, will report work in progress to develop an AI system that implements a linear feedback control designer's expert knowledge. This AI system is a planning expert system written in LISP, and has knowledge of linear control design rules and an interface with a control CAD package. The LISP code represents the design rules as operators that have goals, preconditions, and side effects. Higher-level operators or "scripts" represent expert design procedures. The control design process takes the form of a recursive goal-directed search, aided by the expert designer's heuristics. Cocktails at 6:30 pm, dinner ($11) at 7:00, presentation at 8:00. Rick's Swiss Chalet, 4085 El Camino, Palo Alto Reservations by Oct. 24, Council Office, (415) 327-6622. ------------------------------ Date: Mon 8 Oct 84 09:48:09-PDT From: Paula Edmisten Subject: Seminar - Reasoning About Fault Diagnosis with LES [Forwarded from the Stanford SIGLUNCH distribution by Laws@SRI-AI.] DATE: Friday, October 12, 1984 LOCATION: Chemistry Gazebo, between Physical and Organic Chemistry TIME: 12:05 SPEAKER: Walter Perkins Lockheed Palo Alto Research & Development ABSTRACT: Reasoning About Fault Diagnosis with LES The Lockheed Expert System (LES) is a generic framework for helping knowledge engineers solve problems in diagnosing, monitoring, designing, checking, guiding, and interpreting. Many of the ideas of EMYCIN were incorporated into its design, but it was given a more flexible control structure. In its first "real" application, LES was used to guide less-experienced maintenance personnel in the fault diagnosis of a large electronic signal-switching network. LES used not only the knowledge of the expert diagnostician (captured in the familiar form of "IF-THEN" rules), but also knowledge about the structure and function of the device under study to perform rapid isolation of the module causing the failure. In this talk we show how the topological structure of the device is modeled in a frame structure and the troubleshooting rules of the expert are conveniently represented using LES's case grammar format. We also explain how "demons" are used to setup an agenda of relevant goals and subgoals. The system was fielded in November 1983, and is being used by Lockheed technicians. A preliminary evaluation of the system will also be discussed. LES is being applied in a number of other domains which include design verification, satellite communication, photo-interpretation, and hazard analysis. Paula ------------------------------ Date: Sat 6 Oct 84 15:26:34-PDT From: Andrei Broder Subject: Seminar - Composite Graph Theory [Forwarded from the Stanford bboard by Laws@SRI-AI.] AFLB talk 10/11/84 - Joan Feigenbaum (Stanford): Recognizing composite graphs is equivalent to testing graph isomorphism In this talk I will explore graph composition from complexity theoretic point of view. Given two graphs G1 and G2, we construct the composition G = G1[G2] as follows: For each node in G2, insert a copy of G1. If two copies correspond to nodes that are adjacent in G2, then draw in all possible edges x -- y such that x is in one copy and y is in the other. A graph that can be expressed as the composition of two smaller graphs is called composite and one that cannot is called irreducible. Composite graphs have a great deal of structure and their abstract mathematical properties have been studied extensively. In particular, Harary and Sabidussi have characterized the relationships between the automorphism groups of G1 and G2 and the automorphism group of their composition. Graph composition has been used by Garey and Johnson and Chv\'atal to study NP-complete problems. Garey and Johnson used it to derive upper bounds on the accuracy of approximation algorithms for graph coloring. Chv\'atal showed that the Hamiltonian circuit problem remains NP-complete even if the input graph is known to be composite. In this talk, I consider what seems to be a more basic question about composite graphs; namely, how difficult are they to recognize? The main result I will give is that testing whether a graph is composite is equivalent to testing whether two graphs are isomorphic. In the proof that recognizing composite graphs is no harder than testing graph isomorphism, I will give an algorithm that either declares a graph irreducible or finds a non-trivial decomposition. This distinguishes graph- decomposition from integer-factorization, where primality-testing and factoring are not known to have the same complexity. The inherent difficulty of the recognition problem for composite graphs gives some insight into why some difficult graph theoretic problems, such as Hamiltonian circuit, are no easier even if the inputs are known to be composite. Furthermore, assuming P does not equal NP, graph isomorphism is one of the most important problems for which neither a polynomial time algorithm nor a proof that there cannot be such an algorithm is known. Perhaps examining a problem that is equivalent to it will yield insight into the complexity of the graph isomorphism problem itself. For example, if all irreducible graphs have succinct certificates, then graph isomorphism is in Co-NP. If there is time, I will also show that for cartesian multiplication, another way to construct product graphs, the recognition problem is in P. This talk presents joint work with Alex Schaffer. ***** Time and place: October 11, 12:30 pm in MJ352 (Bldg. 460) **** ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Regular AFLB meetings are on Thursdays, at 12:30pm, in MJ352 (Bldg. 460). - Andrei Broder ------------------------------ Date: Mon, 8 Oct 84 15:00:32 edt From: minker@maryland (Jack Minker) Subject: Lectures - Logic and AI at Maryland, Oct. 22-26 FINAL ANNOUNCEMENT WEEK of LOGIC and its ROLE in ARTIFICIAL INTELLIGENCE at THE UNIVERSITY OF MARYLAND OCTOBER 22-26, 1984 The Mathematics and Computer Science Departments at the University of Maryland at College Park are jointly sponsor- ing a Special Year in Mathematical Logic and Theoretical Computer Science. The week of October 22-26 will be devoted to Logic and its role in Artificial Intelligence. The titles and abstracts of the five distinguished lectures that are to be presented are as follows: Monday, October 22: RAYMOND REITER (University of British Columbia) LOGIC FOR SPECIFICATION: DATABASES, CONCEPTUAL MODELS AND KNOWLEDGE REPRESENTATION LANGUAGES. AI systems and databases have a feature in common: they require representations for various aspects of the real world. These representations are meant to be queried and, in response to new information about the world, modified in suitable ways. Typically, these query and modification processes require reasoning using the underlying representa- tion of the world as premises. So, it appears natural to use a suitable logical language for representing the relevant features of the world, and proof theory for the reasoning. This is not the normal practise in databases and AI. The representations used assume a variety of forms, usu- ally bearing little or no resemblance to logic. In AI, examples of such representation systems include: semantic networks, expert systems, and many different knowledge representation languages such as KRL, KL-ONE, FRL. In data- bases, example representation systems are the relational data model, and various conceptual or semantic models like TAXIS and the entity-relationship model. The point of these representation systems is that they provide their users with computationally efficient ways of representing, and using the knowledge about an application domain. The natural role of logic in databases and AI is a language for specifying representation systems. On this view, one must distinguish between the abstract specification, using logic, of the knowledge content of a database or AI application, and its realization as a representation system. This distinction has pleasant consequences: 1. The logical specification provides a rigorous semantics for the representation system realizing the specification. 2. One can prove the correctness of representation systems with repect to their logical semantics. 3. By taking seriously the problem of logically speci- fying an application, one discovers some rich and fas- cinating epistemological issues e.g. the centrality of non-monotonic reasoning for representation systems. Tuesday, October 23: JOHN McCARTHY (Stanford University) MATHEMATICS OF CIRCUMSCRIPTION Circumscription (McCarthy 1980, 1984) is a method of non- monotonic reasoning proposed for use in artificial intelli- gence. Let A(P) be a sentence expressing the facts "being taken into account", where P stands for a "vector" of predi- cates regarded as variable. Let E(P,x) be a wff depending on a variable x and the Ps. The circumscription of E(P,x) is a second order formula in P expressing the fact that P minimizes lambda x.E(P,x) subject to the facts A(P). The non-monotonicity arises, because augmenting A(P) sometimes reduces the conclusions that can be drawn. Circumscription raises mathematical problems similar to those that arise in analysis in that it involves minimization of a functional subject to constraints. However, its logical setting doesn't seem to permit direct use of techniques from analysis. Here are some open questions that will be treated in the lecture. 1. What is the relation between minimal models and the theory generated by the circumscription formula? 2. When do minimal models exist? 3. The circumscription formula is second order. When is it equivalent to a first order formula? 4. There are several variants of circumscription including successive circumscriptions and prioritized circumscription. What are the relations among these variants? References: McCarthy, John (1980): "Circumscription - A Form of Non-Monotonic Reasoning", Artificial Intelligence, Volume 13, Numbers 1,2, April. McCarthy, John (1984): "Applications of Circumscription to Formalizing Common Sense Knowledge". This paper is being given at the 1984 AAAI conference on non-monotonic reasoning and is being submitted for publication to Artificial Intelli- gence. Wednesday, October 24: MAARTEN VAN EMDEN (University of Waterloo) STRICT AND LAX INTERPRETATIONS OF RULES IN LOGIC PROGRAMMING The strict interpretation says only that is admit- ted which is explicitly allowed by a rule. The lax interpretation says only that is excluded which is explicitly disallowed. This distinction is impor- tant in mathematics and in law, for example. Logic programs also are susceptible to both interpreta- tions. We discuss the use of fixpoint techniques to determine Herbrand models of logic programs. We find that least fixpoints and least models correspond to the strict interpretation and characterize suc- cessful finite computations of logic programs. Greatest fixpoints and greatest models correspond to the lax interpretation and are closely related to negations inferred by finite failure and to terms con- structed by certain infinite computations. Thursday, October 25: JON BARWISE (Stanford University) CONSTRAINT LOGIC. Constraint Logic is based on a semantics that grew out of situation semantics, but on a syntax similar to that from first-order logic. The sematics is not car- ried out in set theory, as is usual in logic, but in a richer theory I call Situation Theory, a theory about things like situations, roles, conditions, types and constraints. While the syntax is not so unusual look- ing, the connection between the syntax and semantics is much more dynamic than is in traditional logic, since the interpretation assigned to a given *use* of some expression will depend on context, in particular, on the history of the "session". For example, vari- ables are interpreted as denoting roles, but different uses of a given variable x may denote increasingly constrained roles as a session proceeds. This is one feature that makes Constraint Logic interesting with regard to AI in general and with regard to non- monotonic logic in particular. Friday, October 26: LAWRENCE HENSCHEN (Northwestern University) COMPILING CONSTRAINT-CHECKING PROGRAMS IN DEDUCTIVE DATABASES. There are at least two kinds of formulas in the inten- sional database which should always be satisfied by the interpretations corresponding to the various states of the database -- definitions and integrity constraints. In our approach, formulas defining new relations are used in response to queries to compute portions of those defined relations; such formulas are therefore automatically satisfied by the underlying database state. On the other hand integrity con- straints may need to be checked each time the database changes. Of course, we believe there are significant advantages in being able to express integrity con- straints in a non-procedural way, such as with first order logic. However, reevaluating an entire first- order statement would be wasteful as normally only a small portion of the database needs to be checked. We present (resolution-based) techniques for developing from first-order statements efficient tests for classes of updates. These tests can be developed at database creation time, hence are compiled, and can be applied before a proposed update is made so that failure does not require backing out. Lectures will be given at: MWF 11:00 AM - 12:30 PM TTH 10:00 AM - 11:30 AM Location: Mathematics Building, 3rd Floor Room Y3206 The lectures are open to the public. If you plan to attend kindly notify us so that we can make appropriate plans for space. We regret that all funds available to support junior faculty and graduate students have been allo- cated. For additional information contact: Jack Minker Department of Computer Science University of Maryland College Park, MD 20742 (301) 454-6119 minker@maryland ------------------------------ Date: Mon, 8 Oct 84 15:24:48 pdt From: ann%ucbernie@Berkeley Subject: Program - Complexity Year at MSRI [Forwarded from the Univ. of Wisconsin by Udi@WISC-RSCH.] [Forwarded from the SRI bboard by Laws@SRI-AI.] COMPLEXITY YEAR AT MATHEMATICAL SCIENCES RESEARCH INSTITUTE A year-long research program in computational complexity will take place at the Mathematical Sciences Research Institute, Berkeley, California, beginning in August, 1985. Applications are solicited for memberships in the Institute during this period. The Institute will award eight or more postdoctoral fellowships to new and recent Ph.D.'s who intend to participate in this program. These fellowships are generally for the entire year, but half-year awards are also possible. It is hoped and expected that members at the more senior level will come with partial or full support from sab- batical leaves and other sources. Memberships for any period are possible, although, for visits of less than three months, Institute support is limited to awards to help offset living expenses. The Program Committee for the complexity year consists of Richard Karp and Stephen Smale (co-chairmen) and Ronald Graham. The program will emphasize concrete computational problems of importance either within mathematics and computer science or in the application of these disciplines to operations research, numerical computation, economics and other fields. Attention will be given both to the design and analysis of efficient algorithms and to the inherent computational complexity of problems. Week-long workshops are planned on topics such as complexity theory and operations research, complexity theory and numerical analysis, algebraic and number-theoretic computation, and parallel and distributed computation. Programs in Mathematical Economics and in Geometric Function Theory will take place concurrently with the Computational Complexity program. Address inquiries and applications to: Calvin C. Moore, Deputy Director Mathematical Sciences Research Institute 2223 Fulton St., Room 603 Berkeley, California 94720 Applicants' files should be completed by January 1, 1985. The Institute is committed to the principles of Equal Opportunity and Affirmative Action. ------------------------------ End of AIList Digest ******************** 10-Oct-84 11:09:11-PDT,13458;000000000001 Mail-From: LAWS created at 10-Oct-84 11:07:42 Date: Wed 10 Oct 1984 11:02-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #135 To: AIList@SRI-AI AIList Digest Wednesday, 10 Oct 1984 Volume 2 : Issue 135 Today's Topics: Expert Systems - NL Interfaces & Training Versions, AI Reports - Request for Sources & Computer Decisions Article, News - TI Lisp Machines & MCC, AI Tools - Printing Directed Graphs, Law - Liability for Expert Systems ---------------------------------------------------------------------- Date: 7 Oct 84 22:43:39-PDT (Sun) From: hplabs!sdcrdcf!trwrba!cepu!ucsbcsl!discolo @ Ucb-Vax.arpa Subject: Writing natural language/expert systems software. Article-I.D.: ucsbcsl.172 I will be writing an simple expert system in the near future and was wondering the advantages and disadvantages of writing something like that in Prolog or Lisp. I seem to prefer Prolog, even though I don't know either one very well yet. Are there any other languages out there which are available under 4.2BSD for this purpose? I would appreciate replies via mail. Thanks. uucp: ucbvax!ucsbcsl!discolo arpa: ucsbcsl!discolo@berkeley csnet: discolo@ucsb USMail: U.C. Santa Barbara Department of Computer Science Santa Barbara, CA 93106 GTE: (805) 961-4178 ------------------------------ Date: 9 Oct 84 3:42:10-PDT (Tue) From: hplabs!kaist!kiet!sypark @ Ucb-Vax.arpa Subject: Natural Language Processing Systems Article-I.D.: kiet.232 Please send me the informations about natural language processing systems which is machine translator or i/o interface for expert systems. What I want is as following. 1. Overview of the system 2. Source is available ? 3. How much price ? ------------------------------ Date: 9 Oct 84 09:14 PDT From: Feuerman.pasa@XEROX.ARPA Subject: Training version of Expert System Tools John Nagle brings up a good idea when talking about M.1. One major problem in trying to investigate various Expert System Building Tools is that they are very expensive just to buy to find out whether they actually lend themselves well to solving a problem. One never really can find out what it is like to use a system from a canned demo or user guides. The idea of having a training version (a stripped down version that doesn't allow full-sized applications) could give someone enough experience with the system to allow them to know what sorts of application a tool is good for. (Undoubtedly this would be viewed as a bad marketing ploy; why would anyone want to come up with a cheap system that would probably only keep someone from buying the full-fledged expensive version.) With that comment, I pessimistically ask: Does anyone know of any tool out there that has such a stripped down training version? --Ken . ------------------------------ Date: 9 Oct 1984 16:32:15 EDT (Tuesday) From: Charles Howell Subject: Various Technical Reports I would like to know what Technical Reports are available from some of the leading centers for research in AI and related fields (how's that for a broad topic?). Any addresses of Publications Offices (or whatever) that have a catalog and ordering / purchase information will be appreciated. Implicit in this request is a request for suggestions about what places are putting out interesting reports; any and all suggestions will be cheerfully accepted! I'll collect the answers and post them to the AIList if there is much response. Thanks, Chuck Howell Howell at MITRE ------------------------------ Date: Tue 9 Oct 84 22:54:27-PDT From: Ken Laws Subject: Computer Decisions Article I just ran across an AI-in-business article in the August issue of Computer Decisions. It features a roundtable of 14 consultants and business bigwigs. Phone numbers and reader service numbers are given for 18 AI vendors, and mention is made of an annual AI report -- AI Trends '84, a description of the technologies and profile of 50 key vendors by DM Data Inc., Scottsdale AZ, $195, (602) 945-9620. The article includes advice on getting started in AI (buy some Lisp machines, hire some hackers and AI experts, and expect some failures), a short glossary (including Lisp, a new language ...), and a short bibliography (including The Mythical Man-Month). -- Ken Laws ------------------------------ Date: Tue 9 Oct 84 15:19:30-CDT From: Werner Uhrig Subject: news from Austin: TI Explorer, MCC, and more [ from the Austin American Statesman, p. D6 - Oct 9, 84 ] TI Explorer finds new path ================================= Texas Instruments in Austin has landed a major business prize: a multi-million-dollar order for up to 400 of its highly sophisticated Explorer symbolic processing systems from the Laboratory for Computer Science at MIT. The computers will be bought over the next 2 years to establish the world's largest network of LISP machines involved in computer research. TI officials said the order is significant in view of the fact that only about 1,000 of the specialized computers are in existence. TI plans to deliver 200 machines in 1985 and 200 in 1986. Boing joins MCC as 19th member of the consortium ==================================================== ... paying a sign-up fee of $500,000. The cost for joining goes up to $1-million on Jan 1. There are 4 seperate research programs at MCC, with a combined annual budget of more than $50 million. Boing reportedly has joined only one program thus far, an effort to find new ways to connect complex computer chips with the equipment the chips are supposed to control, but is considering joining the other three as well. MCC's managers are especially eager for Boing to join the artificial intelligence program. They believe Boing's participation in that expensive program would draw other aerospace companies to it, spreading out the expense and making it a cheaper deal for everyone involved. Boing is the fourth major aerospace defense contractor to become an MCC member [following Rockwell, Lockheed, and Martin Marietta]. [ in other news: real estate prices and traffic jams are coming along nicely, thank you. the city is being sued by the state for polluting the river and trying to sue everyone connected with building 2 nuclear power reactors, which are WAY overdue and WAY over-budget, and not close to being finished. Austin is still trying to sell its 16% of the project, and given that nobody wants to buy it, is close to pushing for abandoning the whole project. So you really don't want to come here ..... (-: I don't make the news, only report it ] ------------------------------ Date: Tue 9 Oct 84 17:49:39-PDT From: PENTLAND@SRI-AI.ARPA Subject: TI's new Lisp Machines News about TI's new Lisp Machines: Timing figures, 1/60th of a second. Both TI and 3600 were 1Mword memory, 300Mbyte disk op TI 3600 comment --------------------------------------------------------------- bitblt 270 441 shows basic memory cycle time floating pt 23 17 //,* about the same, TI has 25 bit number cons 25-40 17-40 depends somewhat on paging paging 225-280 160-450 same transfer rate, seek time 50% more for TI create flavor instance 140 52 not fully microcoded yet send msg 52 21 not fully microcoded yet function call 31 16 not fully microcoded yet 32bit floating 33 17 includes consing in TI machine It appears that by the April delivery date, the TI will be the equal of a 3600. It is already much more than an LMI, Cadr or LM2 (I ran these benchmarks on an LM2, it was 1/2 to 1/5 the TI in all cases). Ask for the benchmark programs if you are interested in details. ------------------------------ Date: Mon, 8 Oct 84 16:58 CDT From: Jerry Bakin Subject: Re: Wanted: info on printing directed graphs Some friends of mine came up with such a program. I have included its first comment below. It is written in Pascal, somewhere; I have a version I rewrote (i.e., force translated) into Multics PL/I. If you can use either one, let me know. We do not support FTP, so if their is a wide demand for this, I may ask someone else to take it off my hands. There might be a small problem, they are currently selling a some of their software, I have to find out if this is a portion of that software. Even if it is, the following provided a source for more information. (* TRPmod - A routine to print N-ary trees on any character printer. This routine takes as input an arbitrary N-ary tree, some interface routines, and assorted printer parameters and writes a pictorial representation of that tree using an output routine provided in the call to treeprint. The tree is nicely formatted and is divided into vertical stripes that can be taped together after printing. Options exist to print the tree backwards or upside down if desired. The algorithm for treeprint originally appeared in "Pretty-Printing of Trees", by Jean G. Vaucher, Software-Practice and Experience, Vol. 10, pages 553-561 (1980). The algorithm used here has been modified to support N-ary tree structures and to have more sophisticated printer format control. Aside from a common method of constructing an ancillary data structure and some variable names, they are now very dissimilar. treeprint was written by Ned Freed, Kevin Carosso, and Douglas Grover at Harvey Mudd College. (714) 621-3219 (ask for the Mathlib Director) ------------------------------ Date: 6 Oct 84 8:51:42-PDT (Sat) From: decvax!mcnc!idis!cadre!geb @ Ucb-Vax.arpa Subject: re: liability for expert systems Article-I.D.: cadre.57 This is a subject that we are quite interested in as we develop medical expert systems. There has been no court case nor precedent nor law covering placement of blame in the cases of errors in expert systems. The natural analogy would be medical textbooks. As far as I know, no author of a textbook has been found liable for errors that resulted in mistreatment of a patient. Therefore, the logical liability should lie with the treating physician to properly apply the knowledge. Having said this, it is best to recognize that customs such as this were developed in a much different society of 100 years ago. Now every possible person in a case is considered fair game and undoubtedly until a court rules or legislation is passed, you must consider yourself at risk if you distribute an expert system. Unfortunately, there is no malpractice insurance available for programmers and you will find a clause in just about any other insurance that you might carry that states that the insurance you have doesn't cover any lawsuits stemming from the practice of your profession. Sorry. ------------------------------ Date: 10 October 1984 0854-PDT (Wednesday) From: bannon@nprdc (Liam Bannon (UCSD Institute for Cognitive Science)) Reply-to: bannon Subject: Liability and Responsibility wrt expert systems I was interested in the messages raising the issue of where responsibility lies if a person follows the advice of an AI system and it turns out to be wrong, or where the person disregards the computer system advice, but the system turns out to be right (AI Digest V2#133). I am not a lawyer or AI system builder, but I am concerned about some of the social dimensions of computing, and have been concerned about how expert systems might actually be used in the work environment. There have been few full-length papers on this topic, to my knowledge. One that I have found interesting is that by Mike Fitter and Max Sime "Creating Responsive Computers: Responsibility and Shared Decision-Making" which appeared in the collection H. Smith and T. Green (Eds.) Human Interaction with Computers (Academic Press, 1980). They point out "the possibility that a failure to use a computer might be judged negligent if, for example, a physician neglected to ask a question, the answer to which was crucial to a diagnosis, AND a computer system would have asked the question." This hinges on a famous 1928 case in the US, called the T.J. Hooper, where a tugboat owner was found negligent for not having radio sets on them, thus not hearing radio reports of bad weather which would have made them seek safety avoiding the loss of the barges which the tugs had in tow - this despite the fact that at that time radio was only used by one tugboat company! This raises a host of interesting questions about how expert systems could/should be used, especially in medicine, where the risks/benefits are highest. Comments? -liam bannon ------------------------------ End of AIList Digest ******************** 11-Oct-84 09:59:40-PDT,15257;000000000001 Mail-From: LAWS created at 11-Oct-84 09:56:05 Date: Thu 11 Oct 1984 09:52-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #136 To: AIList@SRI-AI AIList Digest Thursday, 11 Oct 1984 Volume 2 : Issue 136 Today's Topics: AI Tools LMI (Uppsala) Prolog & Kahn's DCG's, Law - Liabilities of Sofware Vendors, Games - Preliminary Computer Chess Results, Psychology - Distributed Intelligence, Linguistics - Sastric Sanskrit, Conference - Computational Linguistics Call for Papers ---------------------------------------------------------------------- Date: 8 Oct 84 13:53:26-PDT (Mon) From: hplabs!sdcrdcf!trwrba!logico!burge @ Ucb-Vax.arpa Subject: LMI (Uppsala) Prolog + Kahn's DCG's: User Experiences Article-I.D.: logico.124 Does anyone have any experiences to relate about "LM-Prolog", implemented in Zetalisp at the University of Uppsala by Ken Kahn and Mats Carlsson? And/or of the DCG and "Grammar Kit" that comes with it? (We've been using the DEC-11 implementation for several years, but now it's time to expand...) Also, our site is new to the net, and if anyone could send me previous items, it would help me find out what all has been happening out there...!! --John Burge [818] 887-4950 LOGICON, Operating Systems Division, 6300 Variel #H, Woodland Hills, Ca. 91367 ------------------------------ Date: Wed, 10 Oct 84 13:55:07 cdt From: "Walter G. Rudd" Subject: Liabilities of Sofware Vendors Maybe I am being naive or something, but I don't see why AI software should be different from any other when it comes to the liability of the vendor. My attorney has written me a boilerplate contract that contains a clause something to the effect that "vendor is not liable for third-party or consequential damages that result from the use of the product." Doesn't that take care of the problem? If not, maybe I had better find an expert attorney system. ------------------------------ Date: Wed 10 Oct 84 01:57:07-PDT From: Donald S. Gardner Subject: Preliminary computer chess results The computer chess championship is almost over and BELLE has severly bitten the dust. This special purpose hardware (with ~1600 integrated circuits and a PDP-11/23) first tied a program called Phoenix running on a VAX-11/780 and then was beat by NuChess running on a CRAY 1M. NuChess was the program previously called chess 4.7 and was the champion until 1980 when it was beaten by BELLE. The first place winner during the fourth round was declaired to be the program CRAY BLITZ running on a cluster of 4 (FOUR) CRAY's. This system checks in at 420 million instructions per second. Now, CRAY time costs appx $10,000 per hour per computer and each game lasts around 5 hours. This adds up to a cool $1M in computer time! Of course that is in "funny money", but still impressive. There was also a program from Canada which ran on 8 Data General computers (Novas and an Eclipse), two more CRAYs (80 mips each), two Amdahl computers (10 & 13 mips), one CDC Cyber 176 (35 mips) and a Burrough's 7800 (8 mips). ------------------------------ Date: 9 Oct 84 11:53:24 PDT (Tuesday) From: Jef Poskanzer Reply-to: SocialIssues^.PA@XEROX.ARPA Subject: Distributed Intelligence [Excerpted from Human-Nets Digest by Laws@SRI-AI.] By Erik Eckholm New York Times Computer buffs call it "flaming." Now scientists are documenting and trying to explaim the surprising prevalence of rudeness, profanity, exultation and other emotional outbursts by people when they carry on discussions via computer. [...] "It's amazing," said Kiesler. "We've seen messages sent out by managers - messages that will be seen by thousands of people - that use language normally heard in locker rooms." [...] in addition to calling each other more names and generally showing more emotion than they might face to face, people "talking" by computer took longer to agree, and their final decisions tended to involve more risks than those reached by groups meeting in person. [...] "This is unusual group democracy," said Sara Kiesler, a psychologist at Carnegie-Mellon. "There is less of a tendency for one person to dominate the conversation, or for others to defer to the one with the highest status." [...] ------------------------------ Date: 9 Oct 1984 11:09-PDT (Tuesday) From: Rick Briggs Subject: Sastric Sanskrit I would like to respond to recent criticisms concerning Sastric Sanskrit. Firstly, Kiparsky is confusing Sanskrit in general from Sastric Sanskrit. His example, "bhikshuna rajna..." is NOT Sastric Sanskrit but plain ordinary Classical Sanskrit. I did not mean to imply that lack of word order is a sufficient condition for unambiguity, only that it is an indication. As to Dr. Dyer's comments: Yes, a parser will be needed due to the difficulty with translations but this is due to the nature of what one translates into. In the case of English, the difference between the two languages creates the difficulty in translation, not inherent complexities in Sastric Sanskrit. The work I mentioned was edited by Pandit Sabhapati Sharma Upadhyaya in Benares, India and published recently(1963) by the Chowkhamba Sanskrit Series Office. Also, there is something like a concept of scripts in that subsets of discourse (possibly nested) are marked off("iti" clauses) and therefore the immediate context is defined. My comments about English stem from its lack of case. Languages like Latin are potentially capable of rendering logical formulation with less ambiguity since a mapping from its syntactic cases can be made to a set of "semantic cases", depending on how good the case system is. Sanskrit has 8(including the vocatice) and a correspondance(though not complete) is made between the cases of Classical Sanskrit and the "karakas" or auxiliary actions in grammatical Sastric Sanskrit. For example, the dative case is "usually" mapped onto the semantic case "recipient" but not always. The exceptions make up the extension from the commonly known language and the Sastra. An example is in order: "Caitra cooks rice in a pot" is expressed ordinarily in Sanskrit as "Caitra: sthaalyaam taNDulam pacati" (double vowels indicate length, capitals indicate retroflex) In Sastric Sanskrit: sthaliiniSTataNDulaniSTa: viklittijanakashcaitraabhinnaashrayako: vyaapaara: which translates into English: "There is an activity(vyaapaara:) , subsisting in the pot, with agency residing in one substratum not different from Caitra, which produces the softening which subsists in rice." The vocabulary is the same as in Classical Sanskrit, with the addition of terms such as "none other than", and "not different from". Syntax is eliminated in the sense that the sentence is read as "there is an abstract activity" with a series of "auxiliary activities" which "agree" semantically with vyaapaara:. Thus each agreement here ends with ah: which indicates its SEMANTIC agreement with the abstract activity. What I am saying is that each "karaka" is equivalent to a semantic net triple, which can be stored away as eg. "activity, agent, none other than Caitra" etc. Thirdly, the first two points of O'Keefe's have been addressed. Sanskrit is definitely Indo-European but its daughter languages inherited the verbal roots(dhatus) not the methodology of its grammarians. Even though no other(that I know of) natural language has found it worthwhile to pursue the developement of unambiguous languages for a thousand years or so, one parallel can be found: recent work in natural language processing. The difference is that THEY used it in ordinary communication and AI techniques have computer processing in mind. Even though the language is dead there are theoretical works which deal specifically with unambiguity. After reading these, even though you may argue that ambiguity exists (I'd like to see those arguments), you must concede that total precision and an escape from syntax and ambiguity was a primary aim of these scientists. I find that interesting in itself. It is a possible indication that we do actually think "in semantic nets" at some deep level. Point e) again is a confusion with regular Sanskrit. The example of 4 people in a room A,B,C,D would not be a problem in this language. Since precision is required in utterances(see the example above) one would simply not say "we came from X", you would say "there was an activity connected to a coming-activity, having as object X and having agency residing in none other that (we 2, we 3 etc.)." The number would have to be specified. "Blackbird" would be specified as either "a color-event residing in a bird or "blackbird" would be taken as a primitive nominal. Lastly, Jeff Elman's criticisms. A comparison between mathematics and Satsra is not a fair one. Sastric texts have been written in the domains of Science, Law, Mathematics, Archery, Sex,Dance, Morality... I wonder how these texts could be written in mathematical formulisms; the Sastric language is, however, beautifully and elegently suitable for these texts (Sastra means basically "scientific"). I disagree with the statement that "Surface ambiguity gives the language a flexibility of expression. That flexibility does not necessarily entail lack of clarity." Even if ambiguity adds flexibility I do not see how it follows that clarity is maintained. If there are 4 people in the room and one says "we", that is less clear than the case where the language necessitates saying we 3. I also disagree with "...structural ambiguity is not particularly bad nor incompatible with 'logical' expression." Certainly ambiguity is a major impediment to designing an intelligent natural language processor. It would be very desireable to work with a language that allows natural flexibility without ambiguity. And I still maintain that the language is syntax free, word order or no word order. And maybe this is the linguistic find of the century. One last point about metaphor, poetry etc. As an example to illustrate these capabilities in Sastric Sanskrit, consider the "bahuvrihi" construct (literally "man with a lot of rice") which is used currently in linguistics to describe references outside of compunds. "Bahuvrihi" is itself an example, literally "bahu"-many "vrihi" rice. Much rice is taken here as he who posesses a lot of rice, and in Classical Sanskrit different case endings can make "bahu-vrihi" mean "he or she who wants a lot of rice" , "is on a lot of rice" etc. Aha! Ambiguity? Only in Classical, in Sastric Sanskrit the use of semantic cases instead of syntactic do not allow any ambiguity. Rick ------------------------------ Date: 8 Oct 1984 11:10:37 PDT From: Bill Mann Subject: Conference - Computational Linguistics Call for Papers CALL FOR PAPERS 23rd Annual Meeting of the Association for Computational Linguistics 8-12 July 1985 University of Chicago Chicago, Illinois This international conference ranges over all of computational linguistics, including understanding, generation, translation, syntax and parsing, semantics, natural language interfaces, speech understanding and generation, phonetics, discourse phenomena, office support systems, author assistance, translation, and computational lexicons. Its scope is intended to encompass the contents of an Applied Natural Language Processing Conference as well as one on Theoretical Issues in Natural Language Processing. In short, we are striving for comprehensiveness. The meeting will include presented papers, system demonstrations, and, on 8 July, a program of computational linguistics tutorials. Authors should submit, by 18 January 1985, 6 copies of an extended summary (6 to 8 pages) to William C. Mann, ACL85 Program Chairman, USC/ISI, 4676 Admiralty Way, Marina del Rey, CA 90292, USA; (213)822-1511; mann@isib. The summaries should describe completed work rather than intended work, and should indicate clearly the state of completion and validation of the research reported, identify what is novel about it, and clarify its status relative to prior reports. Authors will be notified of acceptance by 8 March 1985. Full length versions of accepted papers prepared on model paper must be received, along with a signed copyright release notice, by 26 April 1985. All papers will be reviewed for general acceptability by one of the two panels of the Program Committee identified below. Authors may designate their paper as either an Applications Paper or a Theory Paper; undesignated papers will be distributed to one or both panels. Review Panel for Applications Papers: Timothy Finin University of Pennsylvania Ralph Grishman New York University Beatrice Oshika System Development Corporation Gary Simons Summer Institute of Linguistics Jonathan Slocum MCC Corporation Review Panel for Theory Papers: Robert Amsler Bell Communications Research Rusty Bobrow Bolt Beranek and Newman Daniel Chester University of Delaware Philip Cohen SRI International Ivan Sag Stanford University Those who wish to present demonstrations of commercial, developmental, and research computer programs and equipment specific to computational linguistics should contact Carole Hafner, College of Computer Science, Northeastern University, 360 Huntington Avenue, Boston MA 02115, USA; (617)437-5116 or (617)437-2462; hafner.northeastern@csnet-relay. For planning purposes, we would like this information as early as possible, but certainly before 30 April. Local arrangements will be handled by Martha Evens, Computer Science Department, Illinois Institute of Technology, Chicago, IL 60616, USA; (312)567-5153 or (312)869-8537; evens@sri-ai. For other information on the conference, on the 8 July tutorials, and on the ACL more generally, contact Don Walker (ACL), Bell Communications Research, 445 South Street, Morristown, NJ 07960, USA; (201)829-4312; bellcore!walker@berkeley. Please note that the dates of the conference will allow people to attend the National Computer Conference, which will be held in Chicago the following week. ======================================================================== PLEASE POST PLEASE REDISTRIBUTE ------------------------------ End of AIList Digest ******************** 12-Oct-84 23:38:37-PDT,15064;000000000001 Mail-From: LAWS created at 12-Oct-84 23:35:03 Date: Fri 12 Oct 1984 23:28-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #137 To: AIList@SRI-AI AIList Digest Saturday, 13 Oct 1984 Volume 2 : Issue 137 Today's Topics: Fuzzy Logic - Query, AI Literature - The AI Report, AI Tools - OPS5 & LM-Prolog & VMS PSL 3.2, Lisp Machines - TI Explorers, Games - ACM Chess Tournament & Chess Planning, Seminar - Knowledge Based Software Development, Conference - AI Society of New England ---------------------------------------------------------------------- Date: 10 Oct 84 8:55:33-PDT (Wed) From: hplabs!intelca!qantel!dual!fortune!polard @ Ucb-Vax.arpa Subject: Fuzzy logic references wanted Article-I.D.: fortune.4472 Would anyone be kind enough to send me (or post) a list of readings that would serve as an introduction to fuzzy logic? Thank you, Henry Polard Henry Polard (You bring the flames - I'll bring the marshmallows.) {ihnp4,cbosgd,amd}!fortune!polard N.B: The words in this posting do not necessarily express the opinions of me, my employer, or any AI project. ------------------------------ Date: Thu 11 Oct 84 21:09:44-PDT From: ROBINSON@SRI-AI.ARPA Subject: Omission Your list of AI information resources omits a significant publication: The Artificial Intelligence Report published by Artificial Intelligence Publications. ------------------------------ Date: 11 Oct 84 13:38:27 EDT From: BIESEL@RUTGERS.ARPA Subject: Addendum to OPS5 list. Some readers of this list pointed out a couple of omissions on the OPS5 summary posted a few days ago; thanks are due them for the additional material. A version of OPS5 called OPS5E, running on the Symbolics 3600 is available from Verac, Inc. 10975 Torreyana Road, Suite 300 San Diego, CA 92121 Prices: $3000 object code, $10000 source, $1000 one year support. There is also a version for the Xerox D series machines (Dandelion, Dolphin, Dorado) available from Science Applications International Corp. 1200 Prospect St. P.O.Box 2351 La Jolla, CA 92038 (619) 454-3811 Price: $2000. ------------------------------ Date: 12 Oct 84 09:23 PDT From: Kahn.pa@XEROX.ARPA Subject: Re: LM-Prolog, Grammar Kit My experiences using LM-Prolog have been very positive but I am surely not an un-biased judge (being one of the co-authors of the system). (I am tempted to give a little ad for LM-Prolog here, but will refrain. Interested parties can contact me directly.) Regarding the Grammar Kit, the main thing that distinguishes it from other DCGs is that it can continuously maintain a parse tree. The tree is drawn as parses are considered and parts of it disappear upon backtracking. I have found this kind of dynamic graphic display very useful for explaining Prolog and DCGs to people as well as debugging specific grammars. ------------------------------ Date: Thu 11 Oct 84 07:16:44-MDT From: Robert R. Kessler Subject: PSL 3.2 for Vax VMS PSL 3.2 for Vax VMS We are pleased to announce that Portable Standard LISP (PSL) version 3.2 is now available for Vaxen running the VMS operating system. PSL is about the power, speed and flavor of Franz LISP or MACLISP, with growing influence from Common LISP. It is recognized as an efficient and portable LISP implementation with many more capabilities than described in the 1979 Standard LISP Report. PSL's main strength is its portability across many different systems, including: Vax BSD Unix, Extended Addressing DecSystem-20 Tops-20, Apollo DOMAIN Aegis, and HP Series 200. A version for the IBM-370 is in beta test, a Sun version is 90% complete and two Cray versions are being used on an experimental basis. Since PSL generates very efficient code, it is an ideal delivery vehicle for LISP based applications (we can also provide PSL reseller licenses for binary only and source distributions). PSL is distributed for the various systems with executables, all sources, an approximately 500 page manual and release notes. The release notes describe how to install the system and how to rebuild the various modules. We are charging $750 for the Vax/VMS version of PSL for Commercial Site licenses. Non-profit institutions and all other versions of PSL will not be charged a license fee. We are also charging a $250 tape or $350 floppy distribution fee for each system. PSL is in heavy use at Utah, and by collaborators at Hewlett-Packard, Rand, Stanford, Columbia and over 200 other sites. Many existing programs and applications have been adapted to PSL including Hearn's REDUCE computer algebra system and GLISP, Novak's object oriented LISP dialect. These are available from Hearn and Novak. To obtain a copy of the license and order form, please send a NET message or letter with your US MAIL address to: Utah Symbolic Computation Group Secretary University of Utah - Dept. of Computer Science 3160 Merrill Engineering Building Salt Lake City, Utah 84112 ARPANET: CRUSE@UTAH-20 USENET: utah-cs!cruse ------------------------------ Date: Thu, 11 Oct 84 03:52:11 pdt From: weeks%ucbpopuli.CC@Berkeley (Harry Weeks) Subject: TI Lisp Machines. The recent article from PENTLAND@SRI-AI has some interesting benchmark data. I am looking seriously at Lisp Machines for purchase in the near future, so I went around to the Xerox, Symbolics and LMI people at ACM 84. I was told by the LMI folks that they were OEMs for the TI machines. (The machines do look almost identical.) So I didn't chat with the TI folks -- perhaps a mistake. If LMI does OEM their machines to TI, why the difference in performance? Perhaps someone in the know can clarify this. If anyone out there with comparative experience in these various machines can say a few words on what they think are the relative merits of each vendor's product it would be quite helpful to prospective buyers. I came away with little substantive basis for comparison from talking with the salesmen. Most of them were high on pretension, low on comprehension and quite adept at parrying questions. As an incidental note, I found at the conference that Lisp and Prolog are now available under PRIMOS on Prime computers. A positive side- effect of the increased interest in AI is the widening spectrum of environments supporting AI languages, an important factor for soft- ware producers looking for a wide market. Harry Weeks (Weeks@UCBpopuli) P.S. I just happened to read the latest Datamation today [10/11] and it contains a news article which also provides some information on the TI machines. ------------------------------ Date: Thu 11 Oct 84 23:52:54-PDT From: PENTLAND@SRI-AI.ARPA Subject: TI Lispm Timings - Clarification Re: TI Lisp Machine timings People have criticized me for the recently circulated comparison of TI and Symbolics machines; mistaking the simple, rough timings I ran on the TI and Symbolics machines for serious benchmarks. I am surprised that anyone thinks that benchmarking a machine can be a simple as the comparison I did, which was limited by a need for extreme brevity. I therefore want to make clear that the timings I ran were ROUGH, QUALITATIVE measures of very limited portions of the machines performance, and bear only a VERY ROUGH, ORDER-OF-MAGNITUDE RELATIONSHIP TO THE TRUE PERFORMANCE of the machines. That is, there is NO warranty of accuracy for such simple tests. Serious benchmarking has yet to be done. Alex Pentland ------------------------------ Date: Fri 12 Oct 84 16:49:27-CDT From: CMP.BARC@UTEXAS-20.ARPA Subject: TI Explorers for MIT Mike Green of Symbolics told us that MIT's "multi-million-dollar order" is essentially a gift from TI to MIT. He said that MIT has confirmed this. Apparently, TI is donating 200 machines to MIT and giving them the option to buy another 200 at $28K each over the next two years. However, TI is working to get DARPA to pay for the second 200! If this is true, I just may "order" a few hundred myself. Dallas Webster CMP.BARC@UTexas-20.ARPA ------------------------------ Date: 11 Oct 84 12:06:18 EDT From: Feng-Hsiung.Hsu@CMU-CS-VLSI Subject: ACM Chess Tournament [Forwarded from the CMUC bboard by Laws@SRI-AI.] The following message was posted on usenet: The standings follow. Ties were broken by considering the sum of the opponents' scores. Since 'Bebe' and 'Fidelity X' deadlocked here, the sum of the opponents' opponents' scores were tallied. Deadheat again, so by fiat, Fidelity walked home with the second place trophy, as Bebe finished second at ACM '83. (At least, I think this is what happened, the groggy hardcore disbanding at 1 am). There were surprises, including a disappointing showing by Belle. I shall leave game commentary to the experts. Mike Valvo and Danny Kopec emceed the fourth round, and several other masters were in attendance, including former World Juniors champ Julio Kaplan. Blitz was running on a 420 MIP four-barrel Cray XMP-48, computing 100K nodes per second (Belle does 160K). Bebe is a custom bit-slice micro, with hardware assist for various functions. Fidelity is a commercial 6 MHz 6502, and International Software Experimental is David Levy's Apple II. Cray Blitz 2150 4 Fidelity X 1900 3 Bebe 1927 3 Chaos 1714 3 Belle 2200 2.5 Nuchess 2100 2 Phoenix 1910 2 Novag X 1970 2 Int. Soft. X 2022 (est) 2 Schach 2.7 N/A 1.5 Ostrich 1475 1 Awit 1600 1 Merlin N/A 1 Xenarbor N/A 0 ------------------------------ Date: Tue, 9 Oct 84 21:07:34 edt From: krovetz@nlm-mcs (Bob Krovetz) Subject: chess and planning A very nice paper on a program that uses planning in making chess moves is: "Using Patterns and Plans in Chess", Dave Wilkins, Artificial Intelligence, Vol. 14, 1980. The program is called PARADISE, and has found a mate that was 19 ply deep! -Bob (Krovetz@NLM-MCS) ------------------------------ Date: 11 Oct 1984 1306-EDT From: Scott Dietzen Subject: Seminar - Knowledge Based Software Development [Forwarded from the CMUC bboard by Laws@SRI-AI.] Friday, October 12 2:00 PM in Wean 5409 Knowledge Based Software Development in FSD Robert Balzer USC/ISI Our group is persuing the goal of an automation based software development paradigm. While this goal is still distant, we have embeded our current perceptions and capabilities in a prototype (FSD) of such a software development environment. Although this prototype was built primarily as a testbed for our ideas, we decided to gain insight by using it, and have added some administrative services to expand it from a programming system to a computing environment currently being used by a few ISI researchers for all their computing activities. This "AI operating system" provides specification capabilities for Search, Coordination, Automation, Evolution and Inter-User Interaction. Particularly important is evolution, as we recognize that useful systems can only arise, and remain viable, through continued evolution. Much of our research is focused on this issue and several examples will be used to characterize where we are today and where we are headed. Naturally, we have started to use these facilities to evolve our system itself. ------------------------------ Date: Thu, 11 Oct 84 17:43:05 edt From: Douglas Stumberger Subject: Conference - AI Society of New England The Sixth Annual Conference of the Artificial Intelligence Society of New England Oct. 26-27, 1984 It is time once again for our legendary annual AISNE meeting! In keeping with our time-honored tradition, we will have an invited speaker for Friday night, with panel discussions and talks by students on Saturday. Accommodations on Friday night will be informal. Bring a sleeping bag, and we can find you a place to stay. If you want us to find you a place, tell Doug Stumberger at Boston University how many bodies you have. Note: If you have a faculty representative at your institution, they can pass this information on to Doug for you in order to minimize long distance phone calls. (If you don't know who your faculty rep. is, it's probably the person who dis- tributed this announcement.) There is no admission charge, and no formal registration necessary, though if you need informal ac- comodations for Friday night, please let Doug know. The event will be held at: Department of Computer Science Boston University 111 Cumington Street Boston, MA The Program is: Friday, Oct. 26 8:00 pm. Invited Talk by David Waltz (Brandeis University) "Massively Parallel Models and Hardware for AI" 9:00 pm. Libational Social Hour Saturday, Oct. 27: 10:00 am. Panel discussion chaired by Elliot Soloway (Yale) "Intelligent Tutoring Systems" 11:30 am. Talks on Academic Research Projects (15 min. each) 12:30 pm. Lunch 2:00 pm. Panel discussion chaired by Michael Lebowitz (Columbia U.) "Natural Language - What Matters?" 3:30 pm. More Talks 4:30 pm. AISNE Business Meeting Program Coordinator: Local Coordinator: Wendy Lehnert Douglas Stumberger COINS Department of Computer Science University of Massachusetts 111 Cumington Street Amherst, MA 01003 Boston, MA 02215 413-545-3639 617-353-8919 csnet: lehnert@umass-cs csnet: des@bostonu bitnet: csc10304@bostonu ------------------------------ End of AIList Digest ******************** 14-Oct-84 19:47:18-PDT,16737;000000000000 Mail-From: LAWS created at 14-Oct-84 19:45:18 Date: Sun 14 Oct 1984 19:38-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #138 To: AIList@SRI-AI AIList Digest Monday, 15 Oct 1984 Volume 2 : Issue 138 Today's Topics: Metadiscussion - Citing AIList, AI - Definition, Linguistics - Mailing List & Sastric Sanskrit & Language Evolution, Conference - SCAIS ---------------------------------------------------------------------- Date: 14 Oct 84 19:56:17 EDT From: Allen Subject: AILIST as a source of info.... Many recent AILIST discussions have fascinated me, and I'm sure that at some point in the near future I'll be using information presented here for a paper or two. Just exactly how do I credit an electronic bboard in a research paper? And who (i.e. moderator, author of info, etc.) do I give credit to? -Allen LUTINS@RU-BLUE [Certainly the author must be credited. I am indifferent as to whether AIList is mentioned since I consider the digest just a communication channel by which authors circulate their unpublished ideas. (You wouldn't cite Ma Bell or your Xerox copier.) This viewpoint is intended to avoid copyright difficulties. On the other hand, a reference to AIList might help someone look up the full context of a discussion. Does any librarian out there know a good citation form for obscure newsletters, etc., that the reader could not be expected to track down by name alone? -- KIL] ------------------------------ Date: 14 Oct 84 14:49:51 EDT From: McCord @ DCA-EMS Subject: Model for AI Applications Since the beginning, some intelligence, albeit explicit and highly focused, has been built into nearly every program written. This is obviously not the "artificial" intelligence we now talk, market, and sell. Surely, to be worthy of the title "artificial" intelligence, an AI application must exhibit some minimum characteristics such as a specified level of control over its environment, the ability to learn, and its transportability or adaptability to related applications. Has anyone developed a model of an AI application that may be used to discriminate between "programs" and "artificial" intelligence? Also, does anyone have any comments on Dr. Frederick Brook's (of The Mythical Man-Month fame) pragmatitic approach ("Intelligence Amplification (IA) Is Better Than Artificial Intelligence (AI)") to AI? ------------------------------ Date: Fri, 12 Oct 84 17:09:29 edt From: Douglas Stumberger Subject: natural language mailing list Does anyone know of a mailing list devoted solely to linguistics/computational linguistics? douglas stumberger csnet: des@bostonu bitnet: csc10304@bostonu ------------------------------ Date: Thu 11 Oct 84 18:15:59-MDT From: Uday Reddy Subject: Sastric Sanskrit Coming from India and having learnt a little bit of Sanskrit, let me make a few comments to add to Rick Briggs's claims. I do not know for a fact if Sastric Sanskrit is unambiguous. In fact, I have not heard of it before. But, its unambiguity seems plausible. First of all, as to the history of Sanskrit. It is an Indo-European language but it has an independent line of development from all the languages spoken outside the Indian subcontinent, i.e., all its daughters are spoken, to the best of my knowledge, only in the subcontinent. Not only its dhatus but its methodologies have been inherited by its daughters. Even the Dravidian languages (the other family of languages spoken in the subcontinent which are not daughters of Sanskrit) have been influenced by its methodologies. For example, the first formal grammar of my own mother tongue, which is not born of Sanskrit, was written in Sanskrit Panini-style. Strictly speaking, neither Sanskrit nor its daughters have a word order. The sophisticated case system makes it possible to communicate without word order. The subject and object are identifiable from their own cases independent of their position in a sentence. Incidentally, the cases are merely a convenience. The prepositions (which become suffixes in Sanskrit and its daughters) serve the same purpose, though they are more verbose. However, the role of various words in a sentence is not always independently identifiable. This leads to ambiguity rather than unambiguity. Kiparsky's example "rajna bhikshuna bhavitavyam" has BOTH the meanings "the beggar will have to become the king" and "the king will have to become the king" The latter meaning is normally understood, because it interprets the sentence in the word order "subject-object-verb" which is the most frequently used. This kind of unambiguity is more of an exception than the standard. I would say it occurs not more than 5% of the time in normal prose. It is resolved by resorting to the "natural" word order. Sastric Sanskrit is a subset of normal Sanskrit, i.e., every sentence of Sastric Sanskrit is also a sentence of normal Sanskrit. This also means that Sastric Sanskrit did not evolve naturally on its own, but was the result of probably hundreds of years of research to eliminate ambiguity in communication. It should be possible for the initiated and knowledgeable to dig up the research that went into the development of this subset. What seems to be important is whether an unambiguous subset of a language can be formed by merely imposing rules on how sentences can be formed. I am not convinced about that, but I cannot also say it is impossible. Ancient Indian scholars had a curious mixture of dogma and reason. One cannot take their claims at their face value. If an unambiguous subset of Sanskrit could be developed, it should also be possible for all the languages. What is special about Sanskrit is that the redundancy needed to disambiguate the language could be added in Sanskrit without substantial loss of convenience. In English, adding this redundancy leads to a lot of awkwardness, as Briggs's examples exemplify. Uday Reddy ------------------------------ Date: 12 Oct 84 09:30 PDT From: Kahn.pa@XEROX.ARPA Subject: Language Evolution This discussion of Sanskrit leads me to ask the question of why languages have evolved the way they have. Why have they moved away from case? Generalizing from the only example I know of (Old Norse to Modern Swedish) I wonder why distinctions that seem useful have disappeared. For example, Old Norse had singular, plural, and dual (when two people were involved). Why would such a distinction come into a language and then disappear hundreds of years later. Why did Sastric Sanskrit die? [Otto Jesperson (1860-1943), famous Danish linguist, studied such matters at a time when classical Greek and Latin were very much in vogue and modern languages with few cases, genders, tenses, and moods were considered retrogressive. He held the opposed view that English and Chinese were the most advanced languages, and that the superiority of modern languages stems from seven characteristics: 1) Shorter forms, easier and faster to speak. The Gospel of St. Matthew contains 39k syllables in Greek, 33k in German, 29k in English. 2) Fewer forms to burden memory. Gothic habaida, habaides, habaidedu, and 12 other forms map to just "had" in English. 3) Regular formation of words. 4) Regular syntactic use of words. 5) Flexible combinations and constuctions. Danish "enten du cller jeg har uret" is straightforward, whereas the inflected "either you or I am wrong" or "either you are wrong, or I" is awkward. 6) Lack of repetitious concord. Latin "opera virorum omnium bonorum veterum" expresses plurality four times, genitive case four times, and masculine gender twice; the English "all good old men's works" has no such repetition. 7) Ambiguity is eliminated through regular word order. Jesperson designed his own artificial language, Novial, after working on a modified (and never adopted) form of Esperanto called Ido. For more information, see Peter Naur's Programming Languages, Natural Languages, and Mathematics in the December 1975 issue of Communications of the ACM. -- KIL] ------------------------------ Date: Sat, 13 Oct 84 13:39:03 PDT From: Southern California AI Society Subject: Conference - SCAIS I noticed the announcement of AISNE on AIList. Since SCAIS is inspired by AISNE, it seems appropriate to announce it in AIList also. Here goes: ************************************************************************ 1ST MEETING OF SCAIS SCAIS -- Southern California Artificial Intelligence Society (Pronounced "skies".) The purpose of SCAIS is to help create an AI community spirit among AI researchers and research labs in the Southern California area. (As far south as San Diego and as far north as Santa Barbara, but probably concentrated in the greater LA area.) SCAIS is inspired by AISNE (AI Society of New England). AISNE meets at least once a year, at locations such as Yale, MIT, UMass, Stoneybrook, etc. in the New England area. (See prior AIList announcement of AISNE.) Our first SCAIS meeting is intended to give everyone an opportunity to meet other active AI researchers and graduate students in the area. Short talks on research projects will be given by students and AI lab project leaders, who will describe what AI research is going on in the area. In addition, we hope to generate a list of the names, phone numbers, net mailing addresses, and research interests of the attendees. If our first SCAIS meeting is successful, future meetings will then be held on a periodic basis at different sites. SCAIS is intended for serious AI researchers and graduate AI students who reside in S. Calif., who are working in the field and who are interested in learning about the research of others in the 'greater' LA area. SCAIS is NOT intended as a forum for industrial recruiting or for interested on-lookers. Attendance at our first SCAIS meeting is expected to be 100-150 people and is by invitation only. AI researchers in the S. Calif. area can request an invitation by contacting: SCAIS-REQUEST@UCLA-CS.ARPA or SCAIS-REQUEST@UCLA-LOCUS.ARPA (or ...!ucla-cs!scais-request on uucp). You should include your name, affiliation, address, net-address, phone number, and research area. ************************************************************************ (almost complete) AGENDA of 1st SCAIS Conference (Oct 29, 8:00am-7:00pm, California Room, UCLA Faculty Center) 8:00 - 8:30 Morning OPEN HOUSE at UCLA AI Lab & Demos 8:30 - 8:40 Michael Dyer -- Welcome and Overview of UCLA AI 8:40 - 10:15 SESSION #1 UCLA (75 min) ============== Sergio Alvarado (stu) -- "Comprehension of Editorial Text" Uri Zernik (stu) -- "Adult Language Learning" Erik Mueller (stu) -- "Daydreaming and Story Invention" Charlie Dolan (stu) -- "Reminding and Analogy" Judea Pearl -- "Learning Hidden Causes from Raw Data" Ingrid Zuckerman (stu) -- "Listener Model for Generation of Meta-Technical Utterances in Math Tutoring" Rina Dechter (stu) -- "Mechanical Generation of Heuristics for Constraint-Satisfaction Problems. Tulin Mangir -- "Applications of Expert Systems to CAD and CAT of VLSI" Vidal -- "Reconfigurable Logic Knowledge Representation and Architectures for Compact Expert Systems" Aerospace Corp. (20 min) ======================== Steve Crocker -- "Overview" Paul Mazaika -- "False Event Elimination" Ann Brindle -- "Automated Satellite Control" John Helly -- "Representational Basis for A Distributed Expert System" *Break* (coffee & danish) 10:15 - 10:30 10:30 - 11:50 SESSION #2 UC Irvine (60 min) ================== Pat Langley -- "Overview of UCI AI Research" Rogers Hall (stu) -- "Learning in Multiple Knowledge Sources" Student (w/ Rick Granger) -- " NEED TITLE " IBM (20 min) ============ John Kepler -- "Overview of IBM Scientific Center Activities in AI" Gary Silverman -- "The Robotics Project" Alexander Hurwitz -- "Intelligent Help for Computer Systems" 11:50 - 1:10 LUNCH (Sequoia Rooms 1,2,3 in Faculty Center) 1:10 - 2:40 SESSION #3 USC/ISI (90 min) ================ Kashif Chaudhry (stu) -- "The Advance Robot Programming Project" Shari Naberschnig (stu) -- "The Distributed Problem Solving Project" Yigal Arens -- "Natural Language Understanding Research at USC" Ram Nevatia -- "Overview of Computer Vision Research at USC" Dan Moldovan -- "Parallel Processing in AI" Jack Mostow -- "Machine Learning Research at ISI" Bill Mann -- "Natural Language Generation Research at ISI" Norm Sondheimer -- "Natural Language Interface Research at ISI" Tom Kaczmarek -- "Intelligent Computing Environment Research at ISI" Bob Neches -- "Expert Systems Research at ISI" Bob Balzer -- "Specification-Based Programming Research at ISI" *Break* 2:40 - 2:55 (coffee & punch) 3:00 - 4:20 SESSION #4 Hughes AI Center (20 min) ========================= D. Y. Tseng -- "Overview of HAIC Activities" JPL (20 min) ==================== Steven Vere -- "Temporal Planning" Armin Haken -- "Procedural Knowledge Sponge" Len Friedman "Diagnostics and Error Recovery" TRW (20 min) ============ Ed Taylor -- "AI at TRW" Rand Corp (20 min) ================== Phil Klahr -- "Overview of Rand's AI Research" "AI in Simulation" Henry Sowizral -- "Time Warp" "ROSIE: An Expert System Language" Don Waterman -- "Explanation for Expert Systems" "Legal Reasoning" Randy Steeb -- "Cooperative Intelligent Systems" *Break* 4:20 - 4:40 (coffee & punch) 4:40 - 6:00 SESSION #5 UC San Diego (20 min) ====================== Paul Smolensky -- "Parallel Computation: The Brain and AI" Paul Munro -- " Self-organization and the Single Unit: Learning at the Neuronal Level" SDC (20 min) ============= Dan Kogan -- "Intelligent Access to Distributed Data Management" Robert MacGregor -- "Logic-Based Knowledge Management System" Beatrice T. Oshika -- "User Interfaces: Speech and Nat. Lang." Cal State, Fullerton (10 min) ============================= Arthur Graesser -- "Symbolic Procedures of Question Answering" Rockwell Science Center (5 min) =============================== William Pardee -- "A Heuristic Factory Scheduling System" General Research Corp (5 min) ============================= Jim Kornell -- "Analogical Inferencing" Northrup (5 min) ================= Steve Lukasis -- 'NEED TITLE" Aerojet (5 min) =============== Ben Peake -- "NEED TITLE" Litton (5 min) ============== speaker -- "NEED TITLE" Logicon (5 min) =============== John Burge -- "Knowledge Engineering at Logicon" 6:00 - 7:00 GENERAL MEETING OF SCAIS MEMBERS SCAIS Panel & General Meeting possible themes: * Assessment - Where from here? * State of AI in S. Calif. * Organization of SCAIS * Future Hosting * Univ - Industry connections * Software - hardware community sharing * Arrival of IJCAI-85 in LA * LA AI Consortium/Institute ??? 7:00 - 7:30 Evening OPEN HOUSE at UCLA AI Lab & Demos (3677 Boelter Hall) > 7:30pm Interested parties may form groups and dine at various restaurants in Westwood Village ------------------------------ End of AIList Digest ******************** 17-Oct-84 11:02:53-PDT,18973;000000000000 Mail-From: LAWS created at 17-Oct-84 10:59:47 Date: Wed 17 Oct 1984 10:55-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #139 To: AIList@SRI-AI AIList Digest Wednesday, 17 Oct 1984 Volume 2 : Issue 139 Today's Topics: Seminars - Monotonic Processes in Language Processing & Qualitative Analysis of MOS Circuits & Knowledge Retrieval as Specialized Inference & Juno Graphics Constraint Language & PECAN Program Development System & Aesthetic Experience, Symposium - Complexity of Approximately Solved Problems, Course - Form and Meaning of English Intonation ---------------------------------------------------------------------- Date: Wed, 10 Oct 84 16:09:03 pdt From: chertok%ucbkim@Berkeley (Paula Chertok) Subject: Seminar - Monotonic Processes in Language Processing BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A TIME: Tuesday, October 16, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 SPEAKER: Martin Kay, Xerox Palo Alto Research Center; Center for the Study of Language and Infor- mation, Stanford University TITLE: Monotonic Processes in Language Processing ABSTRACT: Computation proceeds by manipulating the associations between (variable) names and values in accordance with a program of rules. If an association, once established, is never changed, then the process as a whole is monotonic. More intuitively, mono- tonic processes can add arbitrary amounts of detail to an existing picture so long as they never change what is already there. Monotonic processes underlie several recent proposals in linguistic theory (e.g. GPSG, LFG and autosegmental phonology) and in artificial intelligence (logic programming). I shall argue for seeking monotonic solu- tions to linguistic problems wherever possi- ble while rejecting some arguments fre- quently made for the policy. ------------------------------ Date: 15 Oct 1984 11:17 EDT (Mon) From: "Daniel S. Weld" Subject: Seminar - Qualitative Analysis of MOS Circuits [Forwarded from the MIT bboard by SASW@MIT-MC.] Wednesday October 17, 1984 4:00pm 8th floor playroom Brian C. Williams Qualitative Analysis of MOS Circuits With the push towards sub-micron technology, transistor models have become increasingly complex. The number of components in integrated circuits has forced designers' efforts and skills towards higher levels of design. This has created a gap between design expertise and the performance demands increasingly imposed by the technology. To alleviate this problem, software tools must be developed that provide the designer with expert advice on circuit performance and design. This requires a theory that links the intuitions of an expert circuit analyst with the corresponding principles of formal theory (i.e., algebra, calculus, feedback analysis, network theory, and electrodynamics), and that makes each underlying assumption explicit. Temporal Qualitative Analysis is a technique for analyzing the qualitative large signal behavior of MOS circuits that straddle the line between the digital and analog domains. Temporal Qualitative Analysis is based on the following four components: First, a qualitative representation is composed of a set of open regions separated by boundaries. These boundaries are chosen at the appropriate level of detail for the analysis. This concept is used in modeling time, space, circuit state variables, and device operating regions. Second, constraints between circuit state variables are established by circuit theory. At a finer time scale, the designer's intuition of electrodynamics is used to impose a causal relationship among these constraints. Third, large signal behavior is modeled by Transition Analysis, using continuity and theorems of calculus to determine how quantities pass between regions over time. Finally, Feedback Analysis uses knowledge about the structure of equations and the properties of structure classes to resolve ambiguities. ------------------------------ Date: 1 Oct 1984 13:27-EDT From: Brad Goodman Subject: Seminar - Knowledge Retrieval as Specialized Inference [Forwarded from the MIT bboard by SASW@MIT-MC.] On Thursday, October 11th, at 10:30 a.m., Alan Frisch, from the Cognitive Studies Programme, University of Sussex, Brighton, England and from the Department of Computer Science, University of Rochester, Rochester, New York, will speak at the 3rd floor large conference room at BBN, 10 Moulton Street in Cambridge. Knowledge Retrieval as Specialized Inference Artificial intelligence reasoning systems commonly employ a knowledge base module that stores a set of facts expressed in a representation language and provides facilities to retrieve these facts. Though there has been a growing concern for formalization in the study of knowledge representation, little has been done to formalize the retrieval process. This research remedies the situation in its study of retrieval from abstract specification to implementation. Viewing retrieval as a highly specialized inference process that attempts to derive a queried fact from the set of facts in the knowledge base enables techniques of formal logic to be used in abstract specifications. This talk develops alternative specifications for an idealized version of the retriever incorporated in the ARGOT natural language system, shows how the specifications capture certain intuitions about retrieval, and uses the specifications to prove that the retriever has certain properties. A discussion of implementation issues considers an inference method useful in both retrieval and logic programming. ------------------------------ Date: 15 October 1984 1240-EDT From: Staci Quackenbush at CMU-CS-A Subject: Seminar - Juno Graphics Constraint Language [Forwarded from the CMU bboard by Laws@SRI-AI.] Name: Greg Nelson Date: October 22, 1984 Time: 3:30 - 4:30 p.m. Place: WeH 5409 Title: "An Overview of Juno" Connect a computer to a marking engine, and you have a drawing instrument of unprecedented precision and versatility. Already some graphics artists have given up their T-squares and pens for the new world of raster displays, pointing devices, and laser printers. But they face a serious difficulty: to exploit the power and generality of the computer requires programming. We can't remove this difficulty, but we can smooth it by programming in the language of the geometry of images rather than in the low-level language of some particular representation for images. These considerations led to the design of Juno, an interactive and programmable graphics system. The first basic principle of Juno's design is that geometric constraints be the mechanism for specifying locations. For example, a Juno program might specify that points A, B, and C be collinear and that the distance from A to B equal the distance from B to C; the interpreter will solve these constraints by numerical methods. The second principle of the design is that the text of a Juno program be responsive to the interactive editing of the image that the program produces. For example, to create a program to draw an equilateral triangle, you don't type a word: you draw a triangle on the display, constrain it to be equilateral, and command Juno to extract the underlying program. ------------------------------ Date: Tue 16 Oct 84 09:46:34-PDT From: Susan Gere Subject: Seminar - PECAN Program Development System EE380/CS310 Computer Systems Laboratory Seminar Time: Wednesday, October 17, 4:15 p.m. Place: Terman Auditorium Title: PECAN: Program Development Systems that Support Multiple Views Speaker: Prof. Steven Reiss, C.S.D. Brown University This talk describes the PECAN family of program development systems. PECAN is a generator that is based on simple description of the underlying language and its semantics. Program development systems generated by PECAN support multiple views of the user's program. The views can be representations of the program, its semantics and its execution. The current program views include a syntax-directed editor, a Nassi-Schneiderman flow graph, and a declaration editor. The current semantic views include expression trees, data type diagrams, flow graphs, and the symbol table. Execution views include the interpreter control and a stack and data view. PECAN is designed to make effective use of powerful personal machines with high-resolution graphics displays, and is currently implemented on APOLLO workstations. ------------------------------ Date: Tue, 16 Oct 84 16:56:22 pdt From: chertok@ucbcogsci (Paula Chertok) Subject: Seminar - Aesthetic Experience BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A TIME: Tuesday, October 23, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 SPEAKER: Thomas G. Bever, Psychology Department, Columbia University TITLE: The Psychological basis of aesthetic experi- ence: implications for linguistic nativism ABSTRACT: We define the notion of Aesthetic Experience as a formal relation between mental representations: an aesthetic experience involves at least two conflicting represen- tations that are resolved by accessing a third representation. Accessing the third representation releases the same kind of emotional energy as the 'aha' elation asso- ciated with discovering the solution to a problem. We show how this definition applies to various artforms, music, literature, dance. The fundamental aesthetic relation is similar to the mental activities of a child during normal cognitive development. These considerations explain the function of aesthetic experience: it elicits in adult- hood the characteristic mental activity of normal childhood. The fundamental activity revealed by consid- ering the formal nature of aesthetic experi- ence involves developing and interrelating mental representations. If we take this capacity to be innate (which we surely must), the question then arises whether we can account for the phenomena that are usu- ally argued to show the unique innateness of language as a mental organ. These phenomena include the emergence of a psychologically real grammar, a critical period, cerebral asymmetries. More formal linguistic properties may be accounted for as partially uncaused (necessary) and partially caused by general properties of animal mind. The aspects of language that may remain unex- plained (and therefore non-trivially innate) are the forms of the levels of representa- tion. ------------------------------ Date: Mon 15 Oct 84 11:32:31-EDT From: Delores Ng Subject: Symposium - Complexity of Approximately Solved Problems SYMPOSIUM ON THE COMPLEXITY OF APPROXIMATELY SOLVED PROBLEMS APRIL 17-19, 1985 Computer Science Department Columbia University New York, NY 10027 SUPPORT: This symposium is supported by a grant from the System Development Foundation. SCOPE: This multidisciplinary symposium focuses on problems which are approximately solved and for which optimal algorithms or complexity results are available. Of particular interest are distributed systems, where limitations on information flow can cause uncertainty in the solution of problems. The following is a partial list of topics: distributed computation, approximate solution of hard problems, applied mathematics, signal processing, numerical analysis, computer vision, remote sensing, fusion of information, prediction, estimation, control, decision theory, mathematical economics, optimal recovery, seismology, information theory, design of experiments, stochastic scheduling. INVITED SPEAKERS: The following is a list of invited speakers. L. BLUM, Mills College C.H. PAPADIMITRIOU, Stanford University J. HALPERN, IBM J. PEARL, UCLA L. HURWICZ, University of Minnesota M. RABIN, Harvard University and Hebrew University D. JOHNSON, AT&T - Bell Laboratories S. REITER, Northwestern University J. KADANE, Carnegie-Mellon University A. SCHONHAGE, University of Tubingen R. KARP, Berkeley K. SIKORSKI, Columbia University S. KIRKPATRICK, IBM S. SMALE, Berkeley K. KO, University of Houston J.F. TRAUB, Columbia University H.T. KUNG, Carnegie-Mellon University G. WASILKOWSKI, Columbia University and University of Warsaw D. LEE, Columbia University A.G. WERSCHULZ, Fordham University M. MILANESE, Politecnico di Torino H. WOZNIAKOWSKI, Columbia University and University of Warsaw CONTRIBUTED PAPERS: All appropriate papers for which abstracts are contributed will be scheduled. To contribute a paper send title, author, affiliation, and abstract on one side of a single 8 1/2 by 11 sheet of paper. TITLES AND ABSTRACTS MUST BE RECEIVED BY JANUARY 15, 1985 PUBLICATION: All invited papers will appear in a new journal, JOURNAL OF COMPLEXITY, published by Academic Press, in fall 1985. REGISTRATION: The symposium will be held in the Kellogg Conference Center on the Fifteenth Floor of the International Affairs Building, 118th Street and Amsterdam Avenue. The conference schedule and paper abstracts will be available at the registration desk. Registration will start at 9:00 a.m. There is no registration charge. FOR FURTHER INFORMATION: The program schedule for invited and contributed papers will be mailed by about March 15 only to those responding to this account with the information requested below. If you have any questions, contact the Computer Science Department, Columbia University, or call (212) 280-2736. To help us plan for the symposium please reply to this account with the following information. Name: Affiliation: Address: ( ) I will attend the Complexity Symposium. ( ) I may contribute a paper. ( ) I may not attend, but please send program. ------------------------------ Date: Mon 15 Oct 84 22:43:20-PDT From: Bill Poser Subject: Course - Form and Meaning of English Intonation COURSE ANNOUNCEMENT Mark Liberman and Janet Pierrehumbert of AT&T Bell Laboratories will give a course sponsored by the Linguistics Department and the Center for the Study of Language and Information entitled: FORM AND MEANING OF ENGLISH INTONATION Place: Seminar Room, CSLI, Stanford University Dates: Monday 5 November - Saturday 17 November Hours: MWF 16:30-18:00 TTh 16:30-18:00 & 19:30-21:30 Sat 10:00-12:30 & 14:00-17:00 A brief description follows: (1) What Participants will learn to describe and interpret the stress, tune and phrasing of English utterances, using a set of systematically arranged examples, given in the form of transcripts, tapes and pitch contours. The class will also make use of an interactive real-time pitch detection and display device. We will provide a theory of English intonation patterns and their phonetic interpretation, in the form of an algorithm for generating synthetic F0 contours from underlying phonological representations. We will investigate the relation of these patterns to the form, meaning and use of the spoken sentences that bear them, paying special attention to intonational focus and intonational phrasing. Problem sets will develop or polish participants' skills in the exploration of experimental results and the design of experiments. (2) Who No particular background knowledge will be presupposed, although participants will have to acquire (if they do not already have) at least a passive grasp of many technical terms and concepts. Thus, it will be helpful to have had experience (for instance) with at least some of the terms "hertz" (not the car company), "fricative," "copula," "lambda abstraction," "gradient vector." Several kinds of people, from engineers through linguists and psychologists to philosophers, should find the course's contents interesting. However, we will angle the course towards participants who want to study the meaning and use of intonation patterns, and we hope that a significant fraction of the course will turn into a workshop on this topic. (3) Registration Pre-registration is not mandatory, but if you expect to attend it would be helpful if you would let Bill Poser (poser@su-csli) know. Stanford students wishing to take the course for credit may enroll for a directed reading with Paul Kiparsky or Bill Poser. ------------------------------ End of AIList Digest ******************** 17-Oct-84 22:43:05-PDT,16535;000000000001 Mail-From: LAWS created at 17-Oct-84 22:39:17 Date: Wed 17 Oct 1984 22:29-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #140 To: AIList@SRI-AI AIList Digest Thursday, 18 Oct 1984 Volume 2 : Issue 140 Today's Topics: Applications - Agriculture & Biofeedback, AI Tools - InterLisp-D DBMS & OPS5 & OPS5E & Verac & Benchmarks, Law - Liability of Software Vendors, Metadiscussion - List Citations ---------------------------------------------------------------------- Date: Tue, 16 Oct 84 11:45:49 cdt From: "Walter G. Rudd" Subject: AI applications in agriculture I would like to know of any work in applying AI techniques to improve agricultural production. Tou at Florida and Michalski at Illinois had some things going; what is the status of these projects? Is there anything else going on? Thanks in advance for any help you can give me. Walt Rudd Department of Computer Science 298 Coates Hall Louisiana State University Baton Rouge, Louisiana 70803 rudd@lsu ------------------------------ Date: 3-Oct-84 23:53 PDT From: William Daul / Augmentation Systems Div. / McDnD Subject: PC <--> Biofeedback Instrument Link (info wanted) A friend has asked me to see if I can uncover some information for him. So...here goes... He wants to connect an EEG biofeedback instrument to a personal computer (IBM or APPLE). He hasn't decided on which. 1. What are the necessary componets of such a system (hard disk, disk controller, etc)? 2. He wants to get a spectrum analysis (FFT) of the recordings, both real time and compressed. Does anyone know of existing software he could use? Emre Konuk MRI 555 Middlefield Rd. Palo Alto, CA. 94301 Tel: 415-321 3055 -- wk 415-856 0872 -- hm I suspect he would like to know if anyone knows of existing groups doing similar work. If you have information, you can send it to me "electronically" and I will pass it on to him. Thanks, --Bi// (WBD.TYM@OFFICE-2.ARPA) ------------------------------ Date: 15 Oct 84 16:55:43 PDT (Monday) From: Cornish.PA@XEROX.ARPA Subject: InterLisp-D based Database Management Systems I would like information on any Database Management Systems that are implemented in InterLisp-D. More generally, I'd like literature pointers to the issues of Database Management in AI. Thank you, Jan Cornish ------------------------------ Date: 14 Oct 1984 21:00-EST From: George.Wood@CMU-CS-G.ARPA Subject: Another OPS5 Version There is also a Common Lisp version of OPS5, running on VAX/VMS Common lisp, PERQ (Spice) Lisp, Data General's Common lisp for the MV 4000/8000/10000 series, and Symbolics 3600 in common lisp mode. This version was derived from Forgy's Franz Lisp Implementation by George Wood (GDW@CMU-CS-PS1) with help from Dario Giuse (Dario.Giuse@CMU-CS-SPICE) on the PERQ version and standardization. Sorry this missed the original call for information. ------------------------------ Date: 16 Oct 84 14:35 PDT From: Tom Perrine Subject: OPS5E and Verac Verac has moved. The new address is: Verac 9605 Scranton Rd. Suite 500 San Diego, CA 92121 Attn: Pete Paine (619)457-5550 I believe that you must already have OPS5 before you can get OPS5E, which is OPS5E(xtended). It runs on all of the Symbolics machines, and (now) also the TI Explorer. ------------------------------ Date: 15 October 1984 18:32-EDT From: George J. Carrette Subject: LMI, TI, and Lisp Benchmarks. Note: Comments following are due to George Carrette and Ken Sinclair, hackers at LMI, mostly covering specific facts which have been disclosed in previous announcements in "the trades." * As far as benchmarks are concerned we would suggest that people at least wait until RPG publishes his results, which we consider to be the most serious effort to honestly represent the speed capabilities of the various machines. * TI and LMI OEM arrangements. (1) LMI buys NuMachines on an OEM basis from TI. To these LMI adds the LAMBDA processor, software to support multiple LAMBDA and 68000 Unix processors to run together on the NuBus, sharing disks, ethernet, and other devices. (2) LMI has a license to build NuMachines. (3) It was a technology transfer agreement (license) between LMI and TI that led to the transfer of technology to TI which was the basis of the Explorer. (4) LMI has an OEM agreement to purchase Explorers from TI. To these we will add our own microcode, optimizing compiler, and other products to be announced. [Thank you very much for the reliable information. I'm afraid most of us don't keep up with the trade press, and messages like yours are a great help. A reader providing benchmarks a year ago (some of RPG's old benchmarks, in fact) was chastised for not waiting for RPG's report. At the time, I had never heard of RPG; I assume many other people still have not. If he hurries he may be able to benchmark the machines before the good citizens of Palo Alto start using them for doorstops. Meanwhile, I see no harm in someone publishing timing statistics as long as he offers to provide the code involved. One further note: the benchmarks recently published in AIList were originally circulated privately. It was at my request that they were made available to the list. I thank Dr. Pentland for letting me pass them along, and I regret any inconvenience he may have had as a result. -- KIL] ------------------------------ Date: Fri, 12 Oct 84 13:26:18 EDT From: Stephen Miklos Subject: Liability of software vendors > "Maybe I am being naive or something, but I don't see why > AI software should > be different from any other when it comes to the liability of the vendor. > My attorney has written me a boilerplate contract that contains a clause > something to the effect that "vendor is not liable for third-party or > consequential damages that result from the use of the product." > Doesn't that take care of the problem? If not, maybe I had better find > an expert attorney system." Afraid not. Product liability can jump over the middleman (here the doctor) and is not a contractually-based liability, thus contract terms between the software vendor and the doctor or hospital cannot prevent the liability from attaching. If the aggrieved party sued the doctor, the doctor could not turn around and sue the software vendor (due to the limitation of liability clause given above) but the aggrieved party could sue the software vendor directly and avoid the contract limitation (since he never signed any contract with the vendor). So much for standing to sue. As far as actual liability is concerned, it becomes dicy. Products Liability relies on a product being used in the normal way it is intended to be used causing some kind of injury. It seems to me that the cause of the injury is the doctor's reliance on the software, and therefore the doctor is the "proximate cause." If, however, the particular software product becomes widely used by doctors, the causation seems to shift. A reason for this might be that a single doctor trying out a new piece of technology is responsible for taking greater care to make sure it works than is a doctor who is doing what is accepted in the medical community. For instance, a medical malpractice charge can be avoided by proving that all the doctor's actions were such as would be recommended by the medical community in touch with the state of the art. So, an experimental medical program ought to be safe--the doctor is the guilty party for fooling around with experimental stuff while treating a patient (at least without getting a waiver). But an established program that has a deeply hidden bug in it is the stuff plaintiffs' fortunes are made on. By the way, you are not naive in assuming that an ai program will not be treated differently by the courts than a regular program. But what the AI program is trying to do--make judgments, diagnose illnesses, god knows what all else--will introduce the risk of injury. No one is going to be killed by a defective copy of Visi-calc. ****Disclaimer****--> I got my law degree back in '79, but I am not now, and never have been, a practising attorney in any jurisdiction. (I did pass the Connecticut Bar Exam.) These remarks are not to be construed as legal advice, and should not be relied on as such by anyone. These remarks are also not necessarily the opinions of my employer, or of Mario Cuomo, whom I have never met. Stephen J. Miklos Cognitive Systems New Haven, CT ------------------------------ Date: Mon 15 Oct 84 08:48:26-PDT From: C.S./Math Library Subject: AI List--Crediting Ideas From AI List My first reaction to the question about how to cite something from AI List is that it is an organized form of communication. That is, there are dates, volumes, numbers, an electronic place etc. To me, this is what distinguishes it from "just a communication channel" like the telephone or the xerox copier. I view AI List much closer to the journals but in electronic format. Therefore if I were to cite something from AI LIst, I would use the format for journal articles: author, possibly topic for title of comment, AI List for title; the number, volume, and date of the list; and one additional item, the electronic address. If these lists are going to be kept and can be looked up and referred to, I would recommend as complete a citation as possible. If AI List is viewed as more closely related to informal communications between researchers, then the format would be that which one uses when referrring to a conversation or personal letter. However to me that would indicate that another person would not have access to the primary discussion. Harry Llull, Mathematical and Computer Sciences Library, Stanford University. ------------------------------ Date: 15-Oct-84 14:10 PDT From: Kirk Kelley Subject: Re: AILIST as a source of info.... From: Allen Many recent AILIST discussions have fascinated me, and I'm sure that at some point in the near future I'll be using information presented here for a paper or two. Just exactly how do I credit an electronic bboard in a research paper? And who (i.e. moderator, author of info, etc.) do I give credit to? This reminds me of Ithiel de Sola Pool's lament in note 8 to a paragraph in his chapter on electronic publishing in Technologies of Freedom (Belknap Harvard 1983): "... The character of electronic publishing is illustrated by the problem of citing the information in this paragraph, which came from these interest group exchanges themselves. Shall I cite the Arpanet list as from Zellich at Office-3?" I am NOT an expert on obscure citations, so I can freely throw out the following suggestion using Allen Lutins' original query for an example. "12345" would be the message ID if any had been provided: Lutins, Allen, "AILIST as a source of info...." message 12345 of 14 Oct 1984 19:56 EDT, Lutins@RU-BLUE.ARPA or AIList Digest, V2 #138, 15 Oct 1984, AIList@SRI-AI.ARPA. -- kirk [Alas, the title of a message is not a good identifier. Many of the messages in the AIList mailbox have meaningless titles (e.g., Re: AIList Vol. 2, No. 136) or titles appropriate to some other bboard. Some even have no titles. I commonly supply another title as a service to readers and as aid to my own sorting of the messages. The title sent out to Arpanet readers may thus differ from the title Usenet readers see before I get the messages. -- KIL] ------------------------------ Date: 15 October 1984 2252-PDT (Monday) From: bannon@nprdc (Liam Bannon (UCSD Institute for Cognitive Science)) Reply-to: bannon Subject: citing information on electronic newsboards Allen Lutins query about how to cite information obtained from AIList interests me, as I have confronted this issue recently. I sent out a query on netnews on "computer-mediated social interaction" (it even got on this List) and received a no. of interesting replies. I just sent out a note on the "results" to net.followup, including quotations from several msgs sent to me. I don't identify authors explicitly, partly because of requests for anonymity. (I have however privately acknowledged the contributions, and certainly do not try to pass them off as being my own work.) I think this is ok for a net reply, but as I am writing a large paper on the topic, I have decided to explicitly ask all the people that I quote a) for permission to quote them, and b)for permission to include their names with the quotes. As to citing AIList, or net.general, or whatever, some of the msgs sent to me were also broadcast to a newsgroup, others were sent privately over the net to me, so I am unsure how to cite them. It is an interesting issue though, as if credit is not given properly for ideas that first appeared on the net, then there is a danger that people will be reluctant to share ideas on the net until after "official" publication, thus destroying the vitality of the net. I'll go ask some librarians to see if they have any thoughts. I would be interested in other people's opinions on the issue. -liam bannon (bannon@nprdc) ------------------------------ Date: Tue, 16 Oct 1984 14:08 EDT From: MONTALVO%MIT-OZ@MIT-MC.ARPA Subject: AILIST as a source of info.... [Certainly the author must be credited. ... ] I'm not a librarian but have had some experience in citing obscure reference. I think it can be cited just like a newsletter is cited, after all, it is a newsletter: citing author, title, newletter name, Vol., and No.; maybe method of publication (ARPANET). It is a form of publication, though informal, just like a newsletter. As for copyright, I don't see that there is any problem since none of the authors I've seen have ever copyrighted their material. I'm assuming it's fair game for copying, but that scientific (or literary) protocol would oblige us to credit authors. Fanya [The welcome message I send out to each new subscriber states: List items should be considered unrefereed working papers, and opinions to be those of the author and not of any organization. Copies of list items should credit the original author, not necessarily the AIList. The list does not assume copyright, nor does it accept any liability arising from remailing of submitted material. The phrase "working papers" (which is also used by the SIGART newsletter) is intended to mean that the author is not ready to officially publish the material and thus is not surrendering copyright. This might not hold up in court, but it does establish the context in which people have been submitting their material. I have not been as strict as some list moderators in protecting authors against unauthorized copying. (The Phil-Sci list is/was particularly strict about this.) I have treated AIList as just another bboard that happens to have a distributed readership. I have forwarded items to AIList from university bboards (as well as physical bboards), and I have no objection to similar copying in return. I would draw the line at some major journal or copyrighted book quoting directly from the list without at least asking the readership whether anyone objected to the copying. As I do not hold copyright, however, it really makes no difference where I draw the line. If someone copies material and the author sues, the resolution will be up to a judge. All that I can do is to clarify the intention that should be ascribed to submitters in the absence of other declarations. -- KIL] ------------------------------ End of AIList Digest ******************** 18-Oct-84 10:36:00-PDT,15673;000000000001 Mail-From: LAWS created at 18-Oct-84 10:33:29 Date: Thu 18 Oct 1984 10:28-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #141 To: AIList@SRI-AI AIList Digest Thursday, 18 Oct 1984 Volume 2 : Issue 141 Today's Topics: LISP - Common Lisp Flavors, AI Tools - OPS5 & Benchmarks, Linguistics - Language Evolution & Sastric Sanskrit & Man-Machine Language, AI - The Two Cultures ---------------------------------------------------------------------- Date: Thursday, 18 Oct 1984 06:11:24-PDT From: michon%closus.DEC@decwrl.ARPA (Brian Michon DTN: 283-7695 FPO/A-3) Subject: Common Lisp Flavors Is there a flavor package for Common Lisp yet? ------------------------------ Date: 18 Oct 84 02:26 PDT From: JonL.pa@XEROX.ARPA Subject: OPS5 & Benchmarks Two points, inspired by issue #140: 1) Xerox has a "LispUsers" version of OPS5, which is an unsupported transliteration from the public Franz version, of a year or so ago, into Interlisp-D. As far as I know, this version is also in the public domain. [Amos Barzilay and myself did the translation "in a day or so", but have no interest in further debugging/supporting it] 2) Richard Gabriel is out of the country at the moment; but I'd like to take a paragraph or two to defend his benchmarking project, and report on what I witnessed at the two panel discussions sessions it sponsored -- one at AAAI 83 and the other at AAAI 84. The latter was attended by about 750+ persons (dwindling down to about 300+ in the closing hours!). In 1983, no specific timing results were released, partly because many of the machines under consideration were undergoing a very rapid rate of development; in 1984, the audience got numbers galore, more perhaps than they ever wanted to hear. I suspect that the TI Explorer is also currently undergoing rapid development, and numbers taken today may well be invalid tomorrow (Pentland mentioned that). The point stressed over and over at the two panel sessions is that most of these benchmarks were picked to monitor some very specific facet of Lisp performance, and thus no single number could adequately compare two machines. In the question/answer session of 1983, someone tried to cajole some such simplistic ratio out of Dr Gabriel, and his reply is worth re-iterating "Well, I'll tell you -- I have two machines here, and on one of the benchmarks, they ran at the same speed; but on another one, there was a factor of 13 difference between them. So, now, which number do you want? One, or Thirteen?" One must also note that many of the more important facets for personal workstations were ignored -- primarily, I think because it's so hard to figure out a meaningful statistic to monitor for them, and partly because I'm sure Dick wanted to limit somewhat the scope of his project. How does paging figure into the numbers? if paging is factored out, then what do the numbers mean for a user who is frequently swapping? What about local area network access to shared facilities? What about the effects of GC? I don't know anyone who would feel comfortable with someone else's proposed mixture of "facets" into a whetstone kind of benchmark; it's just entirely possible that the variety of facet mixtures found in Lisp usage is much greater than that found in Fortran usage. [Nevertheless, I seem to remember that the several facets reported upon by Pentland are at the core of almost any Lisp (or, rather, ZetaLisp-like Lisp) -- function call, message passing, and Flavor creation -- so he's not entirely off the wall.] In summary, I'd say that both manufacturers and discerning buyers have benefited from the discussions brought about by the Lisp timings project; the delay on publication of the (voluminous!) numbers has had the good effect of reminding even those who don't want to be reminded that *** a single number simply will not do ***, and that "the numbers", without an understanding analysis, are meaningless. Several of the manufacturer's representatives even admitted during the 1984 panel sessions that their own priorities had been skewed by monitoring facets involved in the Lisp system itself, and that seeing the RPG benchmarks as "user" rather than "system" programs gave them a fresh look at the areas that needed performance enhancements. -- Jon L White -- ------------------------------ Date: 18 October 1984 0646-PDT (Thursday) From: mbr@nprdc Reply-to: mbr@NPRDC Subject: Re: Timings I along with about 8 million others heard RPG (Richard Gabriel) talk at AAAI this year and at the Lisp Conference both this year and 2 years ago, so the benchmarks are around. I dunno if he has the results on line (or for that matter what his net address is-- he was at LLL doing common lisp for the S1 last I heard), but someone in net land might know, and a summary could be posted to AIList mayhaps? Mark Rosenstein [Dr. Gabriel is on the net, but I will let him announce his own net address if he wishes to receive mail on this subject. -- KIL] ------------------------------ Date: 15 Oct 1984 09:40-EST From: Todd.Kueny@CMU-CS-G.ARPA Subject: Language Evolution - Comments For what its worth: Any language in use by a significant number of speakers is under constant evolution. When I studied ancient Greek only singular and plural were taught; dual was considered useful only for very old texts, e.g. Homer or before. The explanation for this was twofold: 1) as the language was used, it became cumbersome to worry about dual when plural would suffice. The number of endings for case, sex and so on is very large in ancient Greek; having dual just made things more cumbersome. 2) similarly, as ancient Greek became modern Greek, case to a large extent vanished. Why? Throughout its use, Greek evolved many special forms for words which were heavily used, e.g. to be. Presumably because no one took the time to speak the complete original form and so its written form changed. I pose two further questions: 1) Why would singular, dual, and plural evolve in the first place? Why not a tri and quad as well? Dual seems to be (at least to me) very unnatural. 2) I would prefer English to ancient Greek principally because of the lack of case endings and conjugations. It is very difficult to express certain new ideas, e.g. the concept of a word on its own with no sex or case, in such a language. Why would anyone consider case useful? -Todd K. ------------------------------ Date: 15 Oct 1984 09:52-PDT (Monday) From: Rick Briggs Subject: Re: Langauge Evolution Why do languages move away from case? Why did Sastric Sanskrit die? I think the answer is basically entropy. The history of language development points to a pattern in which linguists write grammars and try to enforce the rules(organization), and the tendency of the masses is to sacrifice elaborate case structures etc. for ease of communication. One of the reasons Panini codified the grammar of Sanskrit so carefully is that he feared a degeneration of the language, as was already evidenced by various "Prakrits" or inferior versions of Sanskrit spoken by servants etc. The Sanskrit word for barbarian was "mleccha" which means "one who doesn't speak Sanskrit"; culture and high civilization were equated with language. Similarly English "barbarian" is derived from the greek "one who makes noises like baa baa" i.e. who doesn't speak Greek. Current Linguistics has begun to actually aid this entropy by paying special attention to slang and casual usage(descriptive vs. prescriptive). Without some negentropy from the linguists, I fear that English will degenerate further. Rick Briggs ------------------------------ Date: Monday, 15-Oct-84 19:32:13-BST From: O'KEEFE HPS (on ERCC DEC-10) Subject: Sastric Sanskrit again Briggs' message of 9 Oct 84 makes things a lot clearer. The first thing is that Sastric Sanskrit is an artificial language, very like Fitch's "unambiguous English" subset (he is a philosopher who has a paper showing how this rationalised dialect is clear enough so you can do Natural Deduction proofs on it directly). One thing he confuses me about is case. How is having case a contribution to unambiguity? What is the logical difference between having a set of prepositions and having a set of cases? Indeed, most languages that have cases have to augment them with prepositions because the cases are just too vague. E.g. English has a sort of possessive case "John's", but when we want to be clear we have to say "of John" or "for John" or "from John" as the case may be. Praise of Latin is especially confusing, when you recall that (a) that language hasn't got a definite article (it has got demonstratives) and (b) the results of a certain church Council had to be stated in Greek because of that ambiguity. If you can map surface case to semantic case, surely you can map prepositions to semantic case? The second thing which Briggs makes clear is that Sastric Sanskrit is unbelievably long-winded. I do not believe that it can ever have been spontaneously spoken. The third thing is that despite this it STILL isn't unambiguous, and I can use his own example to prove it. He gives the coding of "Caitra cooks rice in a pot", and translates it back into English as "There is an activity(vyaapaara:), subsisting in the pot, with agency residing in one substratum not different from Caitra, which produces the softening which subsists in rice." Is Caitra BOILING the rice or STEAMING it? It makes a difference! Note that this doesn't prove that Sastric Sanskrit can't describe the situation unambiguously, only that it contains at least one ambiguous sentence. Then too, suppose I wanted to translate this into Greek. I need to know whether or not to use the middle voice. That is, is Caitra cooking the rice for HIMSELF, or for someone ELSE? Whichever choice I make in my translation, I run the risk of saying something which Briggs, writing Sastric Sanskrit, did not intend. So it's ambiguous. Now that Briggs has made things so much clearer, I would be surprised indeed if AI couldn't learn a lot from the work that went into the design of Sastric Sanskrit. Actually using their formalism for large chunks of text must have taught its designers a lot. Though if "blackbird" really is specified as "a colour- event residing in a bird" the metaphysical assumptions underlying it might not be immune to criticism. A final point is that we NEED languages which are capable of coding ambiguous propositions, as that may be what we want to say. If Briggs see Caitra cooking some rice in a pot, he may not KNOW whether it is for Caitra or for another, so if Briggs is going to tell me what he sees, he has to say something I may regard as ambiguous. Similarly, when a child says "Daddy ball", that ambiguity (give me the ball? bounce the ball? do something surprising with the ball?) may be exactly what it means to say; it may have no clearer idea than that it would like some activity to take place involving Daddy and the ball. A language which is incapable of ambiguous expression is suited only to describing mathematics and other games. ------------------------------ Date: 16 Oct 84 11:01:49-CDT (Tue) From: "Roland J. Stalfonovich" Subject: AI Natural Language Much has been said in the last few notes about old or forgotten human languages. This brings up an interesting point. Has anyone thought of making (or is there currently) a 'standard' language for AI projects? Not a programming language, but rather a communication language for interspecies communication ,man to machine-man (that is the whole hope of AI after all). Several good choices exist and have existed for several generations. The languages of Esperanto and Unifon are two good choices for study. Esperanto was devised around the turn of the century for the purpose of becoming the international language of the world. To these ends it has obviously failed. This does not however say that it is not without merit. It's advantages of an organized verb conjugation and easy noun and pronoun definition make it a good choice for an 'easily implemented' language. Unifon is a simplification of English. It involves the replacement of the 26 characters of the English alphabet by a set of 40 characters representing the 40 phonics (thus the name) of the English language. This would allow the implementation of the language for speech synthesis (a pet project of many research groups). There are many more languages, and I am sure that everyone has his or her own favorite. But for the criteria of being easily implemented on a computer in both the printed and spoken form, Esperanto and/or Unifon should be seriously considered. ------------------------------ Date: Mon 15 Oct 84 11:22:35-PDT From: BARNARD@SRI-AI.ARPA Subject: The Two Cultures of AI It seems to me that there are two quite separate traditions in AI. One of them, which I suppose includes the large majority of AI practitioners, is devoted to rule-based deductive methods for problem solving and planning. (I would include most natural language understanding work in this category, as well.) The other, which occupies a distinctly minority position, is concerned with models of perception --- especially visual perception. It is my experience that the followers of these two traditions often have trouble communicating. I want to suggest that this communication problem is due to the fundamental difference in the kinds of problems with which these two groups of people are dealing. The difference, put simply, is that "problem solving" is concerned with how to find solutions to well-posed problems effectively given a sufficient body of knowledge, while "perception" is concerned with how to go beyond the information given. The solution of a well-defined problem, once it is known, is known for certain, assuming that the knowledge one begins with is valid. Perception, on the other hand, is always equivocal. Our visual ability to construct interpretations in terms of invariant properties of physical objects (shapes, sizes, colors, etc.) is not dependent on sufficient information, in the formal logical sense. As a researcher in perception, I have to admit that I am often annoyed when problem-solving types insist that their formal axiomatic methods are universal in some sense, and that they essentially "define" what AI is all about. No doubt they are equally annoyed when I complain about the severe limitations of the deductive method as a model of intelligence, and relentlessly promote the inductive method. I'll end, therefore, with a plea for tolerance, and for a recognition that intelligence may, and in fact must, incorporate both "ways of knowing." ------------------------------ End of AIList Digest ******************** 19-Oct-84 10:02:03-PDT,15572;000000000001 Mail-From: LAWS created at 19-Oct-84 09:57:57 Date: Fri 19 Oct 1984 09:50-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #142 To: AIList@SRI-AI AIList Digest Friday, 19 Oct 1984 Volume 2 : Issue 142 Today's Topics: Applications - Biofeedback Instrument Link, LISP - Common Lisp Flavors, AI Tools - TI Expert System Development Tool & Benchmarks, Linguistics - Languages and Cases, Knowledge Representation - Universal Languages, Administrivia - Sites Receiving AIList & Net Readership ---------------------------------------------------------------------- Date: 18 Oct 84 14:28:34 EDT From: kyle.wbst@XEROX.ARPA Subject: Biofeedback Instrument Link The John F. Kennedy Institute For Handicapped Children (707 North Broadway, Baltimore, Maryland 21205 phone 955-5000) has done work in this area. Contact Lynn H. Parker, or Dr. Michael F. Cataldo. They have also published in things like Journal of Behavioral Medicine. Dr. D. Regan at Dalhousie University Department of Psychology Halifax, N.S. B3H 4J1 has also done a lot in this area including the real time Fourier analysis in a feedback loop. You can read about his work in the Dec. 1979 issue of Scientific American (Vol. 241, No. 6 around p. 144 as I recall). At Carnegie-Mellon University were some people with experience in this area. You may try contacting: A. Terry Bahill (Bioengineering ); Mark B. Friedman (Psychology and EE). They may also be able to put you in touch with a person they worked with about 4 years ago at the Pittsburgh Home for Crippled Children called Mata Loevner Jaffe. I think she left full time status at HCC and is now a professor at the University of Pittsburgh. If you want historical info, look in the literature for a system called PIAPACS (I forgot what the acronym stands for now) that was developed by LEar Siegler Co. in Michigan for test pilots at Edwards Air Force Base in California in the mid 1960's. And finally there is the historical work at Cambridge Air Force Research Labs in the early 1960's to put a man in a feedback loop to use amplitude modulation of the brain waves (alphas) to send morse code via a PDP-8 (to clean up the signals and do some limited pattern recognition) to a teletypewriter to transmit the first message "CYBERNETICS". Shortly thereafter, Barbara Brown (I'm not sure of the first name here) at the VA Hospital in Los Angeles used BFT techniques to have subjects control lights and small model railroad trains. Earle. P.S. The ultimate source of commercially available hardware and software in this area would be the TRACE Center at the University of Wisconsin at Madison. ------------------------------ Date: Thu, 18 Oct 1984 17:53 EDT From: Steven Subject: Common Lisp Flavors I am working on Flavors as part of the Spice Lisp project at CMU. Although a prototype system has been finished, we are currently in the process of redesigning the thing from the ground up in an attempt to make it more modular and portable [we've pretty much trashed the idea of a "white-pages" (manual-level) object-oriented interface for now]. Could be another month. -- Steve ------------------------------ Date: Thu, 18 Oct 84 11:43:53 pdt From: Stanley Lanning Subject: Expert System Development Tool from TI [From the October 1984 issure of Systems & Software magazine, page 50] TI AI tool prompts users to develop application With many companies now entering the artificial-intelligence business, the question, "Are there enough AI experts to write the programs?" has been raised. The answer is that Ph.D.s in AI are no longer needed to write expert systems because several expert-system-development tools are available, including one just introduced by Texas Instruments. To ensure that AI tools can be used by nonexperts, Texas Instruments has introduced a first-of-a-kind tool that prompts users for all information needed to develop an expert system. The Personal Consultant is a menu-and window-oriented system that devolps rule-based, backward-chaining expert systems on the TI Professional Computer under MS-DOS operating systems... ------------------------------ Date: 19 October 1984 12:07-EDT From: George J. Carrette Subject: LMI, TI, and Lisp Benchmarks. Glad to be of some help. The main problem I had with Pentland's note was the explanatory comments which were technically not as informative as they could have been. Let me take a moment to review them: (1) BITBLT. This result has more to do with the different amounts of microcode dedicated to such things and the micro instruction execution speed. Both the TI and 3600 have a simple and fast memory bus talking to similar dynamic ram technology. (On the other hand the LAMBDA has a cache and block/read capability) (2) FLOATING POINT. Unless TI has extensively reworked the seldomly used small-floating-point-number code from what LMI sent them, it is the case that small floats are converted into longs inside the microcode and then converted back. (2)(3) CONS & PAGING. ??? Would be more interesting to know how long a full-gc of a given number of mega-conses takes. That bears more on the real overall cost of consing and paging. (4) MAKE-INSTANCE. Could indeed be improved on both the TI and the 3600. People who need to make instances fast and know how usually resort to writing their own %COPY-INSTANCE, since overhead of system default MAKE-INSTANCE depends a lot on sending :INIT methods and other parsing and book-keeping duties. (5)(6) SEND/FUNCALL. These are full microcoded, although improvements are possible. There are some fundamental differences between the LMI/TI micro architecture and the 3600 when it comes to function calling though. In a "only-doing-function-calls-but-no-work" kind of trivial benchmark there are good reasons why a LMI/TI architecture will never equal a 3600 architecture. (7) 32bit floating. Similar comment as applied to small floats, there wasn't any 32-bit floating point number representation in the code before, the floating point numbers were longer than 32 bits total. Then there was a reference to "it is already much more than an LMI, CADR or LM2." First of all, the Explorer *is* an LMI product, and secondly the main product line based on the LMI-LAMBDA has some fundamentally different features including pageable microstore, lisp->microcode compiler, plenty of room for user loadable microstore, SMD disk interface, multiple-processor software support, physical memory cache, which can very strongly and materially change the performance of many applications interesting in both AI research and practice. If you need raw performance in simulation, vision research, array processing, the classic way to go is special microcode or special purpose hardware. The rule may be that simple operations (such as what one may find in trivial benchmarks) done many times call for specilization. The LAMBDA has better support for microcode development, (more statistics counters, micro history, micro stack, micro store, the possibility of doing lambda->lambda debug using multiple-processor lambda configuration, paging microcode good for patching during development) than any other lispmachine. Of course, it does have a high degree of microcode compatibility with the Explorer, which does suggest some possible ways to do things probably of interest more to applying technology than to pure get-it-up-the-first-time research. -gjc ------------------------------ Date: Thu 18 Oct 84 15:44:36-MDT From: Uday Reddy Subject: Languages and cases When we discuss why cases have disappeared, we should also consider why they have appeared. It is clear that they have appeared as "naturally" as they have disappeared. Which of these represents a rise in "entropy"? A reasonable explanation seems to be that cases have appeared for the sake of convenience and brevity. Before their proliferation, probably prepositions and suppositions were used. Eventually, the cases became such a burden that people moved away from their complexity. Don't we see the same trend in programming languages? Uday Reddy ------------------------------ Date: Fri 19 Oct 84 09:22:50-MDT From: Stan Shebs Subject: Universal Languages (again!) Before worrying about a universal language for man-machine communication, we need a universal mechanism for knowledge representation! After all, the external language cannot include concepts (words) for things that are not internally expressible. And while there have been numerous claimants for the status of a UKRL (Universal Knowledge Representation Language) (one of my own projects included), there are none that can really qualify, except perhaps on the basis of Turing-equivalence. Perhaps the best overall candidate is some kind of logical formalism, but as one makes the formalism more general, it seems to become more content-free. Seems to me (from examination of the literature) that the search for a UKRL was very active about 3-5 years ago, but that now everybody has given it up as being the wrong thing to look for (does anybody who was there disagree with this analysis?). These days, I'm inclined to believe that one might establish conditions for *sufficiency* in a KRL. There's the obvious condition that the KRL should be Turing-equivalent. Less obviously perhaps, the KRL should also have the means of automatically translating expressions written using that KRL to ones in some other KRL. Also, the KRL should have complete knowledge of itself (the second condition probably implies this). There may be other reasonable conditions (such as some condition stating that KRL expressions should have some explicit relation to things in the "real world"), but I think the three above should be a minimum. Notice that they also make the question of a *single* UKRL irrelevant. Two sufficiently powerful KRLs can translate themselves back and forth freely, so neither is more "universal" than the other. Notice also that any given KRL must have knowledge of at least one other KRL, in order to facilitate the translation process. When such KRLs are available, then we can profitably think about standard ways of communicating (to ease the poor humans' difficulty with handling 69 KRLs all at once!) stan shebs ps I haven't actually seen any research along these lines (although Genesereth and Mackinlay made some suggestive remarks in their AAAI-84 paper). Is anybody out there looking at KRL translation, or maybe something more specific, like OPS5 <-> Prolog? ------------------------------ Date: Thu 11 Oct 84 10:44:38-PDT From: Ken Laws Reply-to: AIList-Request@SRI-AI Subject: Sites Receiving AIList Readers at the following sites have responded to my Sep. 26 list of AIList recipients (Volume 2, No. 125), or have since signed up for the digest. (There are still many other sites, of course, particularly on Usenet. I have also had contact with individuals who receive the digest but cannot respond via the net.) Army Ballistic Research Laboratory Army Missile Command Defense Communications Agency DoD Computer Security Center Edwards Air Force Base Arthur D. Little, Inc. Battelle Northwest (Pacific Northwest Laboratory) Bell Communications Research Interactive Systems Corporation Lockheed Microelectronics and Computer Corporation Varian Associates Case Western Reserve University Dundee College of Technology, Scotland Indiana University Northeastern University Southern Methodist University Stockton State College University of California at San Diego University of Illinois at Urbana University of Waterloo Washington University in St Louis My apologies to any sites I previously misspelled, including Naval Personnel Research and Development Center Naval Research Laboratory Naval Surface Weapons Center University of Rochester -- Ken Laws ------------------------------ Date: Wed, 10 Oct 84 01:50:10 edt From: bedford!bandy@mit-eddie Subject: Net Readership [Forwarded from the Human-Nets digest by Laws@SRI-AI.] Date: Mon, 8 Oct 84 14:28 EDT From: TMPLee@MIT-MULTICS.ARPA Has anyone ever made an estimate (with error bounds) of how many people have electronic mailboxes reachable via the Internet? (e.g., ARPANET, MILNET, CHAOSNET, DEC ENET, Xerox, USENET, CSNET, BITNET, and any others gatewayed that I've probably overlooked?) (included in that of course group mailboxes, even though they are a poor way of doing business.) Gee, my big chance to make a bunch of order of magnitude calculations.... [...] USENET/DEC ENET: 10k machines, probably on the order of 40 regular users for the unix machines and 20 for the "other" machines so that's 100k users right there. [Rich Kulaweic (RSK@Purdue) notes 15k users on 40 Unix machines at Purdue, with turnover of several thousand per year. -- KIL] BITNET: something like 100 machines and they're university machines in general, which implies that they're HEAVILY overloaded, 100-200 regular active users for each machine - 10k users. [A news item in the latest CACM mentions 200 hosts at 60 sites, soon to be expanded to 200 sites worldwide. A BITNET information center is also being developed by a consortium of 500 U.S. universities, so I expect they'll all get nodes soon. -- KIL] Chaos: about 100-300 machines, 10 users per machine (yes, oz and ee are heavily overloaded at times, but then there's all those unused vaxen on the 9th floor of ne43). 1k users for chaosnet. I think that we can ignore csnet here (they're all either on usenet or directly on internet anyway...), so they count for zero. ARPA/MILNET: Hmm... This one is a little tougher (I'm going to include the 'real' internet as a whole here), but as I remember, there are about 1k hosts. Now, some of the machines here are heavily used (maryland is the first example that pops to mind) and some have moderate loads (daytime - lots of free hardware at 5am!), let's say about 40 regular users per machine -- another 10k users. I dare not give a guesstimate for Xerox. [Murray.PA@Xerox estimates 4000 on their Grapevine system. -- KIL] So it's something on the order of 100k users for the community. [...] Well, it could be 50k people, but these >are< order of magnitude calculations... [Mark Crispin (MRC@Score) notes that there are 10k addressable mailboxes at Stanford, but that the number of active users is perhaps only a tenth of this. Andy's final estimate might be inflated or deflated by such a factor. -- KIL] Now that I've stuck my neck out giving these estimates, I'm awaiting for it to be chopped off. andy beals bandy@{mit-mc,lll-crg} ------------------------------ End of AIList Digest ******************** 20-Oct-84 22:19:08-PDT,14557;000000000001 Mail-From: LAWS created at 20-Oct-84 22:13:42 Date: Sat 20 Oct 1984 22:07-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #143 To: AIList@SRI-AI AIList Digest Sunday, 21 Oct 1984 Volume 2 : Issue 143 Today's Topics: Programming Languages - Buzzwords, AI Tools - LISP Machine Benchmarks, Linguistics - Language Evolution & Sastric Sanskrit, Seminar - Transformational Grammar and AI, PhD Oral: Theory-Driven Data Interpretation ---------------------------------------------------------------------- Date: 19 October 1984 22:52-EDT From: Herb Lin Subject: buzzwords for different language types Could someone out there please tell me the usual catch phrases for distinguishing between languages such as C, Pascal, Ada on one hand and languages such as LISP on the other? Is it "structured" vs "unstructured"? List vs ?? Thanks. ------------------------------ Date: Fri 19 Oct 84 13:08:44-PDT From: WYLAND@SRI-KL.ARPA Subject: LISP machine benchmarks A thought for the day on the AI computer benchmark controversy. We need a single, simple measure for machine quality in order to decide which machine to buy. It must be simple and general because these are typically intended to be used as general purpose AI research machines where we cannot closely define and confine the application. We already have one single, simple measure called price. If there is no *simple* alternative number based on performance, others (i.e. those funding the effort) will use price as the only available measure, and we will have to continually struggle against it using secondary arguments and personal opinion. It should be possible to create a simple benchmark measure. It will - of necessity - be highly abstracted, necessarily crude. This has been done for conventional computer systems: the acronym MIPs is now fairly common, for good or ill. Yes, there are additional measures, but they are used in addition to simple ones like MIPs. We need good, extensive benchmarks for these machines: they will point out the performance bugs that are unique to particular designs. After we do the benchmarks, however, we need to boil it down to some simple number we can use for general purpose comparason to place in opposition to price. ------------------------------ Date: 19 Oct 84 10:32 PDT From: Schoppers.pa@XEROX.ARPA Subject: The Future of the English Auxiliary In response to Ken Kahn's question on language evolution, my own theory is that the invasion of a language by foreign cultures, or vice versa, has a lot to do with how simple a language becomes: cross-cultural speakers tend to use only as much as absolutely necessary for them to consider themselves understood. The English spoken in some communities, eg "Where they goin'?" (missing an auxiliary), "Why he be leavin'?" (levelling the auxiliary), "He ain't goin' nowhere" (ignoring double negatives), etc may well be indicative of our future grammar. On the other hand, "Hey yous" for plural "you" (in Australia), and "y'all" (here), are pointing towards disambiguation. Well, there does have to be a limit to the simplification, lest we "new-speak double-plus ungood". Then again, "ain't" can mean any one of "am not", "aren't", "isn't", "haven't", "hasn't" --- effectively replacing both the primary English auxiliaries (to be, to have) in all their conjugations! United States "English", being the lingo of the melting pot, will probably change faster than most. Marcel Schoppers Schoppers@XEROX ------------------------------ Date: Fri 19 Oct 84 15:23:26-MDT From: Stan Shebs Subject: Cases & Evolution of Natural Language Has anybody at all researched the origins of language? Not an expert on the subject, but I do know that the languages of aboriginal tribes are extraordinarily complicated, as languages go. But they probably don't give us much clue to what the earliest of languages were like. If you believe that the earliest of languages arose along with human intelligence, then you can suppose that the most primitive languages had a separate "word" for each concept to be expressed. Such concepts might include what would correspond to entire sentences in a modern language. Thus the most primitive languages would be completely non-orthogonal. When intelligence developed to a point where the necessary vocabulary was just too complex to handle the wide range of expressible concepts, then perhaps some individuals would start grouping primitive sounds together in different ways (the famous chimpanzee and gorilla language experiments suggest that other primates already have this ability), resulting in the birth of syntactic rules. Obvious question: can all known languages be derived as some combination of arbitrarily bizarre syntactic/semantic rules? (I would guess so, based on results for mathematical languages) Word cases can then be explained as one of the last concepts to be factored out of words. In the most ancient Indo-European languages, for instance, prepositions are relatively infrequent, although the notions of subject, object, verb, and so forth have already been separated into separate words. Perhaps in the future, singular and plural numbers will be separated out also (anyone for "dog es" instead of "dogs"?). stan shebs ------------------------------ Date: 19 Oct 1984 15:17-PDT (Friday) From: Rick Briggs Subject: Sastric Sanskrit Firstly, the language is NOT artificial. There is a LITERATURE which is written in this language. It is different from toy artificial languages like Fitch's in that for three thousand years scientists communicated and wrote texts in this language. There are thus two aspects which are interesting and relevent; one is that research such as I have been describing was carried out in its peculiar context, the other is that a natural language can function as an unambiguous, inference-generating language without sacrificing simplicity or stylistic beauty. The advantage of case is that (assuming it is a good case system) you have a closed set with which a correspondance can be made with a closed set of semantic cases, whereas prepositions can be combined in a multitude of ways and classifying prepositions is not easy. Secondly, the fact that prepositions are not attached to the word allows a possibility for ambiguity "a boat on the river near the tree" could be "a boat on the (river near the tree)" or "a boat (on the river) near the tree". Attaching affixes directly to words allows you (potentially) to express such a sentence without ambiguity. The Sastric approach is to allow one to express a sentence as a series of "facts", each agreeing with "activity". Prepositions would not allow this. If one hears "John was killed", some questions come to mind: who did it, how, why. These are actually the semantic cases agent, instrument, and semantic ablative (apaadaanakaaraka). Instead of "on" and "near" one would say "there is a proximity, having as its substratum an instance of boatness... etc." in Sastric Sanskrit. The real question is "How good a case system is it?". Mapping syntactic case to semantic is much easier than mapping prepositions since a direct correspondance is found automatically if you have a good case system, whereas prepositions do not lend themselves to easy classification. Again, Sanskrit is NOT long-winded, it is the english translation which is, since their vocabulary and methodology was more exact than that of English. "Caitra cooks rice in a pot" is not represented ambiguously. Since it is not specified whether the rice is boiled, steamed, or fried the correct representation should include the fact that the means of softening the rice is unspecified, and the language does have the ability to mark slots as unspecified (anabhihite). Actually, cooking is broken down even further (if-needed) and since rice is cooked by boiling in India, that fact would be explicitly stated. The question is how deep a level of detail is desired, Sanskrit maintains: as far as is necessary but "The notion 'action' cannot be applied to the solitary point reached by extreme subdivision", i.e. only to the point of semantic primitives. Sentences with ambiguity like "the man lives on the Nile" in Sastric is made up of the denotative meaning (the man actually lives on the river) and the implied meaning (the man lives on the bank of the Nile). The latter is the default meaning unless it is actually specified otherwise. There is a very complex theory of implication in the literature, but sentences with implied meanings are discouraged because: "when purport (taatparya) is present, any word may signify any meaning", thus the Sastric system where implied meanings are made explicit. I do not agree that languages need to tolerate ambiguity, in fact that is my main point. One can take a sentence like "Daddy ball" and express it as an imperative of "there is a desire of the speaker for an unspecified activity involving the ball and Daddy." By specifying what exactly is known and what is unknown, one can represent a vague mental notion as precisely as is possible. But do we really need to allow such utterances? Would something humanistic be lost if children simply were more explicit? Children in this culture are encouraged to talk this way by adults engaging in "baby talk". All this points to the fact that the language you speak has a tremendous influence on the your mental make-up. If a language more specific than english was spoken, our thoughts would be more clear and ambiguity would not be needed. I conclude with another example: Classical Sanskrit--> raama: araNye baaNena baalinam jaghaana (Rama killed Baalin in the forest with an arrow) ---> raamakartRkaa araNyaadhikaraNikaa baaNakaraNikaa praaNaviyogaanukuulaa parokSHaatiitakaalikii baalinkarmakaa bhaavanaa (There is an activity relating to the past beyond the speaker's ken, which is favourable to the separation of life, which has the agency of Rama, which has the forest as locus, Baalin as object, and which has the arrow as the implement. Note that each word represents a semantic case with its instantiation, (eg., raama-kartRkaa having as agent Rama), with the verb "kill" (jaghaana) being represented as an activity which is favourable (anukuulaa) to the separation (viyoga) of praana (life). Thus the sentence is a list of assertions with no possibility of ambiguity. Notice that Sanskrit expresses the notion in 42 syllables (7 words) and English takes 75 syllables (43 words). This ratio is fairly indicative of the general case. Rick Briggs ------------------------------ Date: 19 Oct 1984 15:41 EDT (Fri) From: "Daniel S. Weld" Subject: Seminar - Transformational Grammar and AI [Forwarded from the MIT bboard by SASW@MIT-MC.] Transformational Grammar and Artificial Intelligence: A View from the Bridge Robert Berwick It has frequently been suggested that modern linguistic theory is irreconcilably at odds with a ``computational'' view of human linguistic abilities. In part this is so because grammars were thought to consist of large numbers of explicit rules. This talk reviews recent developments in linguistic theory showing that, in fact, current models of grammar are quite compatible with a range of AI-based computational models. These newer theories avoid the use of explicit phrase structure rules and fit quite well with such lexically-based models as ``word expert'' parsing. Wednesday October 24 4:00pm 8th floor playroom ------------------------------ Date: 19 Oct 84 15:35 PDT From: Dietterich.pa@XEROX.ARPA Reply-to: DIETTERICH@SUMEX-AIM.ARPA Subject: PHD Oral: Theory-Driven Data Interpretation [Forwarded from the Stanford bboard by Laws@SRI-AI.] PHD ORAL: TOM DIETTERICH DEPARTMENT OF COMPUTER SCIENCE 2:30PM OCTOBER 25 SKILLING AUDITORIUM CONSTRAINT PROPAGATION TECHNIQUES FOR THEORY-DRIVEN DATA INTERPRETATION This talk defines the task of THEORY-DRIVEN DATA INTERPRETATION (TDDI) and investigates the adequacy of constraint propagation techniques for performing it. Data interpretation is the process of applying a given theory T (possibly a partial theory) to interpret observed facts F and infer a set of initial conditions C such that from C and T one can infer F. Most existing data interpretation programs do not employ an explicit theory T, but rather use some algorithm that embodies T. Theory-driven data interpretation involves performing data interpretation by working from an explicit theory. The method of local propagation of constraints is investigated as a possible technique for implementing TDDI. A model task--forming theories of the file system commands of the UNIX operating system--is chosen for an empirical test of constraint propagation techniques. In the UNIX task, the "theories" take the form of programs, and theory-driven data interpretation involves "reverse execution" of these programs. To test the applicability of constraint propagation techniques, a system named EG has been constructed for the "reverse execution" of computer programs. The UNIX task was analyzed to develop an evaluation suite of data interpretation problems, and these problems have been processed by EG. The results of this empircal evaluation demonstrate that constraint propagation techniques are adequate for the UNIX task, but only if the representation for theories is augmented to include invariant facts about the programs. In general, constraint propagation is adequate for TDDI only if the theories satisfy certain conditions: local invertibility, lack of constraint loops, and tractable inference over propagated values. ------------------------------ End of AIList Digest ******************** 24-Oct-84 12:01:22-PDT,17822;000000000001 Mail-From: LAWS created at 24-Oct-84 11:57:33 Date: Wed 24 Oct 1984 11:47-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #144 To: AIList@SRI-AI AIList Digest Wednesday, 24 Oct 1984 Volume 2 : Issue 144 Today's Topics: Courses - Decision Systems & Introductory AI, Journals - Annotated AI Journal List, Automatic Programming - Query, AI Tools - TI Lisp Machines & TEK AI Machine, Administrivia - Reformatting AIList Digest for UNIX, Humor - Request for Worst Algorithms, Seminars - Metaphor & Learning in Expert Systems & Representing Programs for Understanding ---------------------------------------------------------------------- Date: Tue 23 Oct 84 13:33:06-PDT From: Samuel Holtzman Subject: Responses to Decision Systems course. Several individuals have requested further information on the course in decision systems I teach at Stanford (advertised in AILIST a few weeks ago). Some of the messages I received came from non-ARPANET sites, and I have had trouble replying electronically. I would appreciate getting a message from anyone who has requested information from me and has not yet received it. Please include a US (paper) mail address for my reply. Thanks, Sam Holtzman (HOLTZMAN@SUMEX or P.O. Box 5405, Stanford, CA 94305) ------------------------------ Date: 22 Oct 1984 22:45:40 EDT From: Lockheed Advanced Software Laboratory@USC-ISI.ARPA Subject: Request for information A local community college is considering adding an introductory course in AI to its curriculum. Evening courses would be of benefit to a large community of technical people interested in the subject. The question is what will be the benefit to first and second year students. If anyone knows of any lower division AI courses taught anywhere, could you please drop me a line over the net. Also, course descriptions on introductory AI classes, either lower or upper division, would be appreciated. Comments on the usefulness or practicality of such a course at this level are also welcome. Thank You, Michael A. Moran Lockheed Advanced Software Laboratory address: HARTUNG@USC-ISI ------------------------------ Date: Tue, 23 Oct 84 11:34 CDT From: Joseph_Hollingsworth Subject: annotated ai journal list I am interested in creating an annotated list of the AI related journals list that was published in AIList V1 N43. I feel that this annotated list would be beneficial for those persons who do not have easy access to the journals mentioned in the previously published list, but who feel that some of them may apply to their work. I solicit information about each journal in the following form, (which I will compile and release to the AIList if there is enough interest shown). 1) Journal Name 2) Subjective opinion of the type of articles that frequently appear in that journal (short paragraph or so). 3) Keywords and phrases that characterize the articles/journal, (don't let formalized keyword lists hinder your imagination). 4) The type of scientist, engineer, technician, etc. that the journal would benefit. 5) Address of journal for subscription correspondence, (include price too, if possible). Please send this information to Joe Hollingsworth at jeh%ti-eg@csnet-relay (if you are on the ARPANET) jeh@ti-eg (if you are on the CSNET; I am on the CSNET) The following is the aforementioned list of journals: AI Magazine AISB Newsletter Annual Review in Automatic Programming Artificial Intelligence Artificial Intelligence Report Behavioral and Brain Sciences Brain and Cognition Brain and Language Cognition Cognition and Brain Theory Cognitive Pshchology Cognitive Science Communications of the ACM Computational Linguistics Computational Linguistics and Computer Languages Computer Vision, Graphics, and Image Processing Computing Reviews Human Intelligence IEEE Computer IEEE Transactions on Pattern Analysis and Machine Intelligence Intelligence International Journal of Man Machine Studies Journal of the ACM Journal of the Assocation for the Study of Perception New Generation Computing Pattern Recognition Robotics Age Robotics Today SIGART Newsletter Speech Technology ------------------------------ Date: 23 October 1984 22:28-EDT From: Herb Lin Subject: help needed on automatic programming information I need some information on automatic programming. 1. How complex a problem can current automatic programming systems handle? The preferred metric would be complexity as measured by the number of lines of code that a good human programmer would use to solve the same problem. 2. How complex a problem will future automatic programming systems be able to handle? Same metric, please. Of course, who can predict the future? More precisely, what do the most optimistic estimates predict, and for what time scale? 3. In 30 years (if anyone is brave enough to look that far ahead), what will automatic programming be able to do? Please provide citable sources if possible. Many thanks. ------------------------------ Date: 22 Oct 1984 12:07:39-PDT From: William Spears Subject: TI Lisp machines The AI group at the Naval Surface Weapons Center is interested in the new TI Lisp Machine. Does anyone have any detailed information about it? Thanks. "Always a Cosmic Cyclist" William Spears Code N35 Naval Surface Weapons Center Dahlgren, VA 22448 ------------------------------ Date: 22 Oct 84 08:10:32 EDT From: Robert.Thibadeau@CMU-RI-VI Subject: TEK AI Machine [Forwarded from the CMU bboard by Laws@SRI-AI.] I have good product literature on the Tektronix 4404 Artificial Intelligence System (the workbook for their people). This appears to be a reasonable system which supports Franz Lisp, Prolog, and Smalltalk-80. It uses a 68010 with floating point hardware and comes standard with a 1024^2 bit map, 20mb disk, floppy, centronics 16 bit port, RS232, 3-button mouse, ethernet interface, 1 mbyte RAM, and a Unix OS. The RAM upgrades at least 1 more mbyte and you can have a larger disk and streaming tape. The major thing is that the price (retail without negotiation) is $14,950 complete. It is apparently real, but I don't know this system first hand. The product description is all I have. ------------------------------ Date: Sat, 20 Oct 84 23:10:53 edt From: Douglas Stumberger Subject: reformatting AILIST digest for UNIX For those of you on Berkeley UNIX installations, there is a program available which does the slight modifications to ailist digest necessary so it is in the correct format for a "mail -f ...". This allows using the UNIX mail system functionality to maintain your ailist digest files. For a copy of the program, net to: douglas stumberger csnet: des@bostonu ------------------------------ Date: Mon 22 Oct 84 10:30:00-PDT From: Jean-Luc Bonnetain Subject: worst algorithms as programming jokes [Forwarded from the Stanford bboard by Laws@SRI-AI.] After reading the recent complaint(s) about those people who slow down the system with their silly programs to sort a 150-element list, and talking with a friend, I came up with the following dumb idea : A lot of emphasis is understandably put on good, efficient algorithms, but couldn't we learn also from bad, terrible algorithms ? I have heard that Dan Friedman at Indiana collects elegant LISP programs that he calls LISP poems. To turn things upside down, how about LISP jokes (more generally, programming jokes) ? I'm pretty sure most if not all of programmers have some day (night) burst into laughter when encountering an algorithm that is particularly dumb, and funny for the same reason. I don't know whether anyone ever collected badgorithms (sorry, that was the worst name I could find), so I suggest that you bright guys send me your favorite entries. To qualify as a badgorithm, the following conditions should be met: (if you don't like them, send me your suggestions for a better definition) 1. It *is* an algorithm in the sense described by Knuth Vol 1. 2. It *does* solve the problem it addresses. Entering the Knuth-Bendix algorithm as a badgorithm for binary addition is illegal (though I admit it is somewhat funny). 3. Though it solves the problem, it must do so in an essentially clumsy way. Adding loops to slow down the algorithm is cheating. In some sense a badgorithm should totally miss the right structure to approach the problem. 4. The hopeless off-the-track-ness of a badgorithm should be humorous for someone knowledgeable with the problem addressed. We are not interested in alborithms, right ? Just being the second or third best algorithm for a problem is not enough to qualify (think of the "common sense" algorithm for finding a word in a text as opposed to the Boyer-Moore algorithm, or of the numerous ways to evaluate a polynomial as opposed to Horner's rule; there is nothing to laugh at in those cases). There is nothing funny in just being a O(n^(3/(pi^3)-1/e)) algorithm, I think. 5. It should be described in a simple, clear way. Remember that the best jokes are the shortest ones. I'm sure there are enough badgorithms for well-known problems (classical list manipulation, graph theory, arithmetic, cryptography, sorting, searching, etc). Please don't enter algorithms to solve NP problems unless you have good reasons to think they are interesting in our sense. If anyone out there is willing to send me an entry, please send the following: * a simple description of the problem (the name is enough if it's a well-known problem). * a verbal description of the badgorithm if possible. * a programmed version of the badgorithm (in LISP preferably). this is not necessary if your verbal description makes it clears enough how to write such a program, but still it would be nice. * a description of a good algorithm for the same problem in case most people are not expected to be familiar with one. Comparing this to the badgorithm should help us in seeing what's wrong with the latter, and I would say that this could have good educational value. To start things, let me enter my favorite badgorithm (I call it "stupid-sort"): * the problem is to sort a list, according to some "order" predicate. * well, that's easy. just generate all permutations of the list, and then check whether they are "order"ed. would you bet that someone in CS105 does actually use this one ? [I once had to debug an early version of the BMD nonparametric package. It found the min and max of a vector by sorting the elements ... (Presumably most users would also request the median and other sort-related statistics.) For a particularly slow sort routine see the Hacker's Dictionary definition of JOCK, quoted in Jon Bentley's April Programming Pearls in CACM. -- KIL] I understand perfectly that some people/organizations do not wish to have their names associated with badgorithms, but please don't refrain from entering something because of that. I swear that if you request it there will be no trace of the origin of the entry if I ever compile a list of them for personal or public use (you know, "name withheld by request" is the usual trick). jean-luc ------------------------------ Date: 17 Oct 1984 16:25-EDT From: Andrew Haas at BBNG.ARPA Subject: Seminar - Metaphor [Forwarded from the MIT bboard by SASW@MIT-MC.] Next week's BBN AI seminar is on Thursday, October 25th at 10:30 AM in the 3rd floor large conference room. Bipin Indurkhya of the University of Massachusetts at Amherst will speak on "A Computational Theory of Metaphor Comprehension and Analogical Reasoning". Abstract follows. Though the pervasiveness and importance of metaphors in natural languages is widely recognised, not much attention has been given to them in the fields of Artificial Intelligence and Computational Linguistics. Broadly speaking, a metaphor can be characterized as application of terms belonging to source domain in describing target domain. A large class of such metaphors are based on structural analogy between the two domains. A computational model of metaphor comprehension was proposed by Carbonell which required an explicit representation of a mapping which maps terms of the source domain to the terms of the target domain. In our research we address ourselves to the question of how one can characterize this mapping in terms of the knowledge of the source and the target domains. In order to answer this question, we start from Gentner's theory of Structure-Mapping. We show limitations of Gentner's theory and propose a theory of Constrained Semantic Transference [CST] that allows part of the structure of the source domain to be transferred to the target domain coherently. We will then introduce two recursive operators, called Augmentation and Positing Symbols, that make it possible to create new structure in the target domain constrained by the structure of the source domain. We will show how CST captures several cognitive properties of metaphors and then discuss its limitations with regard to computability and finite representability. If time permits, we will use CST as a basis to develop a theory of Approximate Semantic Transference which can be used to develop computational models of the cognitive processes involved in metaphor comprehension, metaphor generation, and analogical reasoning. ------------------------------ Date: Tue 23 Oct 84 10:45:51-PDT From: Paula Edmisten Subject: Seminar - Learning in Expert Systems [Forwarded from the Stanford SIGLUNCH distribution by Laws@SRI-AI.] DATE: Friday, October 26, 1984 LOCATION: Chemistry Gazebo, between Physical and Organic Chemistry TIME: 12:05 SPEAKER: Li-Min Fu Electrical Engineering ABSTRACT: LEARNING OBJECT-LEVEL AND META-LEVEL KNOWLEDGE IN EXPERT SYSTEMS A high performance expert system can be built by exploiting machine learning techniques. A learning method has been developed that is capable of acquiring new diagnostic knowledge, in the form of rules, from a case library. The rules are designed to be used in a MYCIN-like diagnostic system in which there is uncertainty about data as well as about the strength of inference and in which the rules chain together to infer complex hypotheses. These features greatly complicate the learning problem. In machine learning, two issues that can't be overlooked are efficiency and noise. A subprogram, called "Condenser," is designed to remove irrelevant features during learning and improve the efficiency. It works well when the number of features used to characterize training instances is large. One way of removing noise associated with a learned rule is seeking a state with minimal prediction error. Another subprogram has been developed to learn meta-rules which guide the invocation of object-level rules and thus enhance the performance of the expert system using the object-level rules. By embodying all the ideas developed in this work, an expert program called JAUNDICE is built, which can diagnose the likely cause and mechanisms of a patient with jaundice. Experiments with JAUNDICE show the developed theory and method of learning are effective in a complex and noisy environment where data may be inconsistent, incomplete, and erroneous. Paula ------------------------------ Date: Tue, 23 Oct 84 00:08:10 cdt From: rajive@ut-sally.ARPA (Rajive Bagrodia) Subject: Seminar - Representing Programs for Understanding [Forwarded from the UTexas-20 bboard by Laws@SRI-AI.] Graduate Brown Bag Seminar: Representing Programs For Understanding by Aaron Temin noon Friday Oct. 26 PAI 3.36 Automatic help systems would be much easier to generate than they are now if the same code used to create the executable version of a program could be used as the major database for the help system. The desirable properties of such a program representation will be discussed. An overview of MIRROR, our implementation of those properties, will be presented with an explanation of why MIRROR works. It will also be argued that functional program representations are inadequate for the task. If you are interested in receiving mail notifications of graduate brown bag seminars in addition to the bboard notices, please send a note to briggs@ut-sally ------------------------------ End of AIList Digest ******************** 27-Oct-84 22:12:24-PDT,16286;000000000000 Mail-From: LAWS created at 27-Oct-84 22:06:38 Date: Sat 27 Oct 1984 21:56-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #145 To: AIList@SRI-AI AIList Digest Saturday, 27 Oct 1984 Volume 2 : Issue 145 Today's Topics: Administrivia - Usenet Disconnection, AI Languages - Buzzwords, Expert Systems - Logic-Based Expert Systems & Critique, Humor - Expert Systems & Recursive Riddle & Computational Complexity, Algorithms - Bad Algorithms as Programming Jokes, Seminars - Nonmonotonic Inference & Mathematical Language, Symposium - Expert Systems in the Government ---------------------------------------------------------------------- Date: Sat 27 Oct 84 21:36:47-PDT From: Ken Laws Subject: Usenet Disconnection The SRI-UNIX host that has been the AIList gateway between Arpanet and Usenet has been undergoing system changes. This broke the connection about a week ago, and I do not know how soon communication will be restored. Meanwhile the discussion continues asynchronously in the two networks. -- Ken Laws ------------------------------ Date: Mon 22 Oct 84 11:18:59-MDT From: Stan Shebs Subject: Re: buzzwords for different language types My favorite buzzwords are "low-level" for C, Pascal, and Ada, and "high-level" for Lisp :-) But seriously, one can adopt a very abstract (i.e. applicative/functional) programming style or a very imperative (C-like) style when using Lisp. On the other hand, adopting an applicative style in C is difficult (yes, I've tried!). So Lisp is certainly more versatile. Also, Lisp's direct representation of programs as data facilitates the construction of embedded languages and the writing of program-analysing programs, both important activities in the construction of AI systems. On the other hand, both of these are time-consuming, if not difficult to do in C or Pascal. Incidentally, these remarks largely apply to Prolog also (although Prolog doesn't make it easy to do "low-level" programming). stan shebs ------------------------------ Date: Thu 25 Oct 84 20:59:56-CDT From: Charles Petrie Subject: Logic-based Expert Systems Regarding expert system tools: would anyone like to offer some reasoned opinions regarding the suitability of logic-based systems for such? I have no strong definition of "logic-based" to offer, but I have in mind as prime examples MRS from Stanford and DUCK from SST which provide interfaces to LISP, forward and back chaining, and various extra-logical functions to make life easier for the system builder. I am interested in large systems (1000+ rules desirable) and the control and performance problems and solutions that people have found. Can such systems be built successfully? What techniques to constrain search have been tried and worked/failed? Any references? Charles Petrie ------------------------------ Date: Sun, 21 Oct 84 20:28:24 pdt From: weeks%ucbpopuli.CC@Berkeley (Harry Weeks) Subject: Expert system critique. An article appears in the current (November/December) issue of ``The Sciences'' (New York Academy of Sciences) by Hubert and Stuart Dreyfus of Berkeley. The article ``Mindless Machines'' asserts that `computers don't think like experts, and never will,' invoking, in part, Plato's ``Euthyphro'' (Euthyphro is a theologian queried by Socrates as to the true nature of piety) as an allegory. The basic assertion is that so-called expert systems reason purely from rules, whereas human experts intuit from rules using the vast experience of special cases. They cite this `intuition' as being an insurmountable barrier to building intelligent machines. Harry Weeks (Weeks@UCBpopuli) ------------------------------ Date: Fri 26 Oct 84 06:46:39-CDT From: Werner Uhrig Subject: is there an Expert-System like that ?? (-: [ cartoon in from InfoWorld, Nov 5, 84, page 7] ( 2 ladies having tea in the 'parlor', chatting. with a somewhat perplexed expression, one stares at a small dirt-pile on the carpet, while the obvious hostess explains with a smug grin:) "I thought he was a vacuum cleaner salesman. He came in, sprinkled dirt on the carpet and then tried to sell me a software program that would show me how to clean it up. " ------------------------------ From: gibson@unc.UUCP (Bill Gibson) Subject: Recursive Riddle [Forwarded from Usenet by SASW@MIT-MC.] How many comedians does it take to tell a Light Bulb Joke ? Two - one to say, "How many comedians does it take to tell a Light Bulb Joke? Two - one to say, "How many comedians does it take to tell a Light Bulb Joke? Two - one to say, "How many comedians does it take to tell a Light Bulb Joke? Two - one to say, "How many comedians does it take to tell a Light Bulb Joke? ... and one to ask nonsense riddles." ... and one to ask nonsense riddles." and one to ask nonsense riddles." and one to ask nonsense riddles." and one to ask nonsense riddles." and one to ask nonsense riddles. - from the parallel process of - Bill Gibson ------------------------------ Date: Wed 24 Oct 84 19:13:16-PDT From: Jean-Luc Bonnetain Subject: minor correction on my msg on "badgorithms" Afte reading again the message, I *do* find interesting and unusual an O(n^(3/(pi^3) - 1/e)) algorithm. I'd be real glad to see, and maybe even touch, one. ------------------------------ Date: Thu, 25 Oct 84 07:38 EDT From: MJackson.Wbst@XEROX.ARPA Subject: Re: worst algorithms as programming jokes A very interesting idea, but "badgorithm" as a label should have been strangled at birth. How about "algospasm"? Mark ------------------------------ Date: Thu, 25 Oct 84 08:14:31 cdt From: "Duncan A. Buell" Subject: Bad Algorithms Jean-Luc Bonnetain suggests worst algorithms (badgorithms) as programming jokes. In a similar vein, with interests in winning the Cold War by shipping some of these to the Soviet Union, what is the slowest possible way to sort a list of N items? The only requirement should be one (this problem may not be well-defined yet, but I'm sure people could produce subproblems that were) to the effect that repetition of a state or sequence of states should not take place, and that the method actually at some future date sort the list. As an example of how to think about this, consider generating the permutations of N things, then comparing the existing list against each permutation. How slowly, then, can we generate the permutations of N things? We could isolate one element, generate permutations of N-1 things, and then insert the isolated element in N different places. Ignoring the symmetry of the situation, we could isolate a second element and continue (is this cheating on the rule?). And generating permutations of N-1 things? ------------------------------ Date: 25 Oct 84 09:58 PDT From: Kahn.pa@XEROX.ARPA Subject: Re: Badgorithms in AIList Digest V2 #144 The examples of badgorithms that come to mind (including the sorting by selecting an ordered permutation and find min and max by sorting, or for that matter defining the last element of a list as CAR of the reverse of the list or empty intersection by computing the entire intersection and then seeing if its empty) all have in common that they are making use of existing constructs that do what is desired and much more. I think that these are very reasonable PROGRAMS even if they normally correspond to bad ALGORITHMS. The point is that various projects in program transformation (especially partial evaluation) take as input such programs and automatically transform them into programs that correspond to very reasonable algorithms. Also, true fans of logic programming who believe that an algorithm = logic + control use sort as ordered permutation as their classic example. They add control anontations that cause the permutation activity to be coroutined with the order selection. I'm looking forward to the day when one can write programs that if interpreted naively correspond to badgorithms and yet are either tranformed automatically or interpreted cleverly enough so that they run like a bat out of hell. ------------------------------ Date: 24 Oct 1984 10:35-EDT From: MVILAIN at BBNG.ARPA Subject: Seminar - Nonmonotonic Inference [Forwarded from the MIT bboard by SASW@MIT-MC.] "A Non-Monotonic Inference System" James W. Goodwin, University of Linkoping. BBN Laboratories, 10 Moulton St, Cambridge. Third floor conference room, 10:30 AM. Tuesday October 30th. We present a theory and implementation of incomplete non-monotonic reasoning. The theory is inspired by the success of inference systems based on dependency nets and reason maintenance. The process of inference is conceived as a monotonic accumulation of constraints on belief sets. The "current database" is just the set of constraints accumulated so far; the current beliefs are then required to be a set which satisfies all the constraints in the current database, and contains no beliefs which are not forced by those constraints. Constraints may also be thought of as reasons, or as dependencies, or (best) simply as individual inference steps. This approach allows an inference to depend on aspects of the current state of the reasoning process. In particular, an inference may support P on the condition that Q is not in the current belief set. This sense of non-monotonicity is conveniently computable (by reason maintenance), so the undecidability of Non-monotonic Logic I and its relatives is avoided. This makes possible a theory of reasoning which is applicable to real agents, such as computers, which are compelled to arrive at some conclusion despite inadequate time and inadequate information. It supports a precise idea of "reasoned control of reasoning" and an additive representation for control knowledge (something like McCarthy's Advice Taker idea). ------------------------------ Date: 26 Oct 84 15:47:53 EDT From: Ruth.Davis@CMU-RI-ISL1 Subject: Seminar - Mathematical Language [Forwarded from the CMU bboard by Laws@SRI-AI.] Date: Monday, October 29 Title: PRL: Practical Formal Mathematics Speaker: Joe Bates, Cornell University Time: 1:30 pm Location: 4605 WEH PRL: Practical Formal Mathematics Joseph Bates Cornell University PRL is a family of development environments which are designed to support the construction, validation, execution, and communication of large bodies of mathematics text (eg, books on graph algorithms or group theory). The design of these systems draws on work in many areas, from philosophy to Lisp hackery. Tuesday, Constable will speak on certain issues in the choice of PRL's mathematical language. I will present, in detail, the most significant aspects of the current system architecture, and will suggest directions for future work. ------------------------------ Date: 26 Oct 1984 9:27:12 EDT (Friday) From: Marshall Abrams Subject: Symposium - Expert Systems in the Government I am helping to organize a Symposium on Expert Systems in the Federal Government. In addition to papers, I am looking for people to serve on the program committee and the conference committee, and to serve as reviewers and session chairmen. The openings on the conference committee include local arrangements, publicity, and tutorials. Please contact me or the program chairman (or both by net-mail) with questions and suggestions. The call for papers follows. Call for Papers Expert Systems in Government Conference October 23-25, 1985 THE CONFERENCE objective is to allow the developers and implementers of expert systems in goverenment agencies to exchange information and ideas first hand for the purpose of improving the quality of existing and future expert systems in the government sector. Artificial Intelligence (AI) has recently been maturing so rapidly that interest in each of its various facets, e.g., robotics, vision, natural language, supercomputing, and expert systems, has acquired an increasing following and cadre of practitioners. PAPERS are solicited which discuss the subject of the conference. Original research, analysis and approaches for defining expert systems issues and problems such as those identified in the anticipated session topics, methodological approaches for analyzing the scope and nature of expert system issues, and potential solutions are of particular interest. Completed papers are to be no longer than 20 pages including graphics and are due 1 May 1985. Four copies of papers are to be sent to: Dr. Kamal Karna, Program Chairman MITRE Corporation W852 1820 Dolley Madison Boulevard McLean, Virginia 22102 Phone (703) 883-5866 ARPANET: Karna @ Mitre Notification of acceptance and manuscript preparation instructions will be provided by 20 May 1985. THE CONFERENCE is sponsored by the IEEE Computer Society and The MITRE Corporation in cooperation with The Association for Computing Machinery, The American Association for Artificial Intelligence and The American Institute of Aeronautics and Astronautics National Capital Section. This conference will offer high quality technical exchange and published proceedings. It will be held at Tyson's Westpark Hotel, Tysons Corner, McLean, VA, suburban Washington, D.C. TOPICS OF INTEREST The topics of interest include the expert systems in the following applications domains (but are not limited to): 1. Professional: Accounting, Consulting, Engineering, Finance, Instruction, Law, Marketing, Management, Medicine Systems, Intelligent DBMS 2. Office Automation: Text Understanding, Intelligent 3. Command & Control: Intelligence Analysis, Planning, Targeting, Communications, Air Traffic Control 4. Exploration: Space, Prospecting, Mineral, Oil Archeology 5. Weapon Systems: Adaptive Control, Electronic Warfare, Star Wars, Target Identification 6. System Engineering: Requirements, Preliminary Design, Critical Design, Testing, and QA 7. Equipment: Design Monitoring, Control, Diagnosis, Maintenance, Repair, Instruction 8. Project Management: Planning, Scheduling, Control 9. Flexible Automation: Factory and Plan Automation 10. Software: Automatic Programming, Specifications, Design, Production, Maintenance and Verification and Validation 11. Architecture: Single, Multiple, Distributed Problem Solving Tools 12. Imagery: Photo Interpretation, Mapping, etc. 13. Education: Concept Formation, Tutoring, Testing, Diagnosis, Learning 14. Entertainment and Intelligent Games, Investment and Expert Advice Giving: Finances, Retirement, Purchasing, Shopping, Intelligent Information Retrieval ------------------------------ End of AIList Digest ******************** 27-Oct-84 22:22:24-PDT,16425;000000000000 Mail-From: LAWS created at 27-Oct-84 22:19:14 Date: Sat 27 Oct 1984 22:10-PDT From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #146 To: AIList@SRI-AI AIList Digest Sunday, 28 Oct 1984 Volume 2 : Issue 146 Today's Topics: Report - CSLI Description, Linguistics - Indic Interlingua & Evolution & Shastric Sanscrit, Seminars - Knowledge and Common Knowledge & Gestalt Tutorial & AI and Real Life ---------------------------------------------------------------------- Date: Wed 24 Oct 84 18:33:02-PDT From: Dikran Karagueuzian Subject: Institute Description - CSLI [Excerpted from the CSLI Newsletter by Laws@SRI-AI.] NEW CSLI REPORT Report No. 16, ``The Center for the Study of Language and Information,'' has just been published. It describes the Center and its research programs. An online copy of this report can be found in the directory in the file ``Report-No-16.Online.'' In addition to this report, the directory contains other valuable information about the Center and Turing. To obtain a printed version of Report No. 16, write to Dikran Karagueuzian, CSLI, Ventura Hall, Stanford 94305 or send net mail to Dikran at Turing. ------------------------------ Date: Sun, 21 Oct 84 20:06:59 pdt From: weeks%ucbpopuli.CC@Berkeley (Harry Weeks) Subject: Indic interlingua. If I recall correctly, the continuing colloquy on Sastric Sanskrit was motivated by the desire for a natural interlingua for machine trans- lation. Pardon my ignorance, but I do not see the efficacy of trans- lating a language first into something like Sastric Sanskrit with its concomitant declensional, conjugational and euphonic complexity, then from there into the target language. Are not less complex (and less verbose) formalisms more appropriate, not being weighted with aesthe- tic amenities and cultural biases? If Sastric Sanskrit is otherwise being offered as a paradigm for such a formalism, a more detailed in- sight into its grammar is needed. Another facet of the colloquy is its focus on ambiguity in the rela- tionship of semantic elements (viz. words) in sentences. There is also the problem of determining unambiguously the meaning of a word, when in natural languages words often have more than one meaning de- pending on context. Is Sastric Sanskrit unique in its vocabulary as well as its grammar that each word has but one precisely circumscribed meaning, and how eclectic and deep is this vocabulary? Certainly the professed unequivocality of the syntax is an aid to determining mean- ings of the words whose interrelationship is thus well defined, but it would seem preferable not to rely on context or on clumsy defining clauses in an interlingua. As an aside on ambiguity being requisite for a literature in a lan- guage, I might profer two opinions. A great writer is often charac- terized by his ability to mold sentences which have an uncommon flui- dity and expressivity -- would an unambiguous language allow such freedom? Great poetry invokes thoughts and emotions which defy written expression through the use of rhythm and juxtaposition of disparate images through words set in defiance of strict grammatical precepts. Further, the beauty of prose or poetry lies in good part in the use of ambiguity. Especially in poetry, distilling many emotions into a compact construction is facilitated by ambiguity, either semantic or phonetic. The beauty of poetry is a very different one from the beauty of logic or mathematics. Harry Weeks (Weeks@UCBpopuli) ------------------------------ Date: Mon, 22 Oct 84 10:06 EDT From: Tim Finin Subject: language evolution Marcel Schoppers (AIList Digest V2 #143) seems to suggest that certain dialects (e.g. those which include "Why he be leavin'?" and "He ain't goin' nowhere") are the result of forces which SIMPLIFY the grammar of a language: ".. my own theory is that the invasion of a language by foreign cultures, or vice versa, has a lot to do with how simple a language becomes: cross-cultural speakers tend to use only as much as absolutely necessary for them to consider themselves understood." The analyses that I have seen show that such dialects are just as complex, linguistically, as the standard dialect. They are just complex in different ways. As I understand it, simplified PIDGIN languages quickly evolve into complex CREOLE languages - all it takes is one generation of native speakers. Tim ------------------------------ Date: Wed 24 Oct 84 23:23:56-PDT From: Bill Poser Subject: linguistics I would like to respond to several linguistic questions discussed recently. First, in response to Rick Briggs re-assertion that Shastric Sanskrit is a natural language, his claim that there was a literature written in it and that it was in use for over three thousand years is simply irrelevant. The same could perfectly well be true of an artificial language. There is literature written in Esperanto, an artificial language which is also used for scientific communication. It is perfectly possible that Esperanto will remain with us for thousands of years. But we all know that it is an artificial language. What makes it artificial is that it was consciously designed by a human being-it did not evolve naturally. This leads to the question of whether Shastric Sanskrit is a natural language. It looks like it isn`t. Rather, it is an artificial language based on Sanskrit that was used for very limited purposes by scholars. I challenge Rick Briggs to present evidence that (a) it was in use for anything like 3000 years; (b) that anyone ever spoke it; (c) that even in written form it was used extensively at any period; (d) that it was not always restricted to scholars just as mathematical language is today. There has also been some speculation about the historical development of languages. One idea presented is that languages evolve from morphologically complex to morphologically simple. This is just not true. It happens to be true of a number of the Indo-European languages with which non-linguists are most familiar, but it is not true in general. Second, someone claimed that the languages of "aboriginal people" (I assume he means "technologically primitive") are complex and badly organized, and that languages evolve as people become technologically more advanced. This was a popular idea in the early nineteenth century but was long ago discarded. We know of no systematic differences whatever between the languages spoken by primitive people and those spoken by technologically advanced people. There is no evidence that language evolves in any particular direction. Finally, Briggs mistakenly characterizes linguists as prescriptivists. That is quite false. In fact, the prescriptivists are mainly English and Literature people or non-academics like William Safire. Linguistics is non-prescriptive by definition since we are interested in describing what occurs in natural language and characterizing the possible natural languages. Finally (here comes a minor FLAME), why don't you guys read some serious Linguistics books or ask a linguist instead of posting ignorant speculation about linguistic issues? Some of us do Linguistics for a living and there is extensive technical literature on many of these questions. If I want to, say, know about algorithms I don't sit around guessing. I look it up in a book on algorithms or ask a computer scientist.  ------------------------------ Date: Thu, 25 Oct 1984 00:09 PDT From: KIPARSKY@SU-CSLI.ARPA Subject: Even "shastric" Sanskrit is ambiguous Take the example "Caitra is cooking rice in a pot". It is ambiguous in both Sanskrit and English as to whether it is the rice that is in the pot, or Caitra himself. Clearly the "shastric" paraphrase "There is an activity subsisting in a pot..." doesn't resolve this ambiguity. That can only be done by distinguishing between subject- and object- oriented locatives (which, incidentally, some natural languages do). The reason why the Sanskrit logicians' paraphrases don't make that distinction is that they follow Panini in treating locatives, like all other karakas, simply as arguments of the verb. In general, shastric paraphrases, though certainly very explicit and interesting, are by no means an "unambiguous language". What they make explicit about the meanings of Sanskrit sentences is limited by the interpretations assigned to those sentences by the rules of Panini's grammar. This grammar introduces only such semantic categories as are needed to account for the distribution of Sanskrit grammatical formatives. So shastric paraphrases wind up leaving some of the ambiguities of the corresponding ordinary Sanskrit sentences unresolved. This sentence and its shastric paraphrase are ambiguous in other ways as well, namely with regard to aspect ("cooks" or "is cooking"), and definiteness ("the pot" or "a pot"). These categories don't play a role here though they do in other areas of Sanskrit. E.g. the generic/progressive distinction is important in derived nouns, where English in turn ignores it: Sanskrit has two words for "driver", depending on whether the activity is habitual/professional or not; a shastric paraphrase might make the distinction explicit for such nouns. The prevalence of this logicians' system of paraphrasing should not be exaggerated, by the way. There is no evidence of it having been around for anything like 3000 years(!), and it is not, to my knowledge, used in any "literature" other than technical works on philosophy. ------------------------------ Date: Thu, 25 Oct 84 10:28 EST From: Kurt Godden Subject: reply to schoppers@xerox 'United States "English", being the lingo of the melting pot, will probably change faster than most.' The historical linguists tell us that in fact when groups of speakers physically move and establish a new language group, as has happened here in the US, that the 'new' language dialect actually changes more slowly than the original language group, in this case British English. As simple evidence, witness the fact of the diverse English dialects in the British Isles versus the far more homogeneous regional dialects in the US. There is also textual evidence from poetry (rhythm, etc) showing that present day American English has preserved the patterns of Middle English and early Modern English whereas present day British English has changed. -Godden@gmr [Note that he availability of national radio and television broadcasts in this century may be altering the evolution of modern dialects. -- KIL] ------------------------------ Date: 25 Oct 1984 15:54-EDT From: AHAAS at BBNG.ARPA Subject: Seminar - Knowledge and Common Knowledge [Forwarded from the MIT bboard by SASW@MIT-MC.] There will be an AI seminar at 10:30 AM Friday November 2, in the 3rd floor large conference room. Abstract follows: Knowledge and Common Knowledge In Distributed Environments Yoram Moses, Stanford University Knowledge plays a fundamental role in distributed environments. An individual in a distributed environment, be it a person, a robot, or a processor in a network, depends on his knowledge to drive his decisions and actions. When individuals' actions have an effect on one another, it is often necessary that their actions be coordinated. This coordination is acheived by a combination of having a predetermined common plan of some kind, and communicating to expand and refine it. The states of knowledge that are relevant or necessary in order to allow the individuals to successfully carry out their individual plans vary greatly according to the nature of the dependence of their plans on the actions of others. This work introduces a hierarchy of states of knowledge that a system may be in. We discuss the role of communication in ``improving'' the system's state of knowledge of a given fact according to this hierarchy. The strongest notion of knowledge that a group can have is Common Knowledge. This notion is inherent in agreements and coordinated simultaneous actions. We show that common knowledge is not attainable in practical systems, and present a variety of relaxations of common knowledge that are attainable in many cases of interest. The reationship between these issues and communication and action in a distributed environment is made clear through a number of well known puzzles. This talk should be of interest for people interested in distributed algorithms, communication protocols, concurrency control and AI. This work is joint with Joe Halpern of IBM San Jose. ------------------------------ Date: Fri, 26 Oct 84 15:56:32 pdt From: chertok@ucbcogsci (Paula Chertok) Subject: Seminar - Gestalt Tutorial BERKELEY COGNITIVE SCIENCE PROGRAM Fall 1984 Cognitive Science Seminar -- IDS 237A SPEAKER: Steven E. Palmer, Psychology Department and Cognitive Science Program, UC Berkeley TITLE: ``Gestalt Then and Now: A Tutorial Review'' TIME: Tuesday, October 30, 11 - 12:30 PLACE: 240 Bechtel Engineering Center DISCUSSION: 12:30 - 2 in 200 Building T-4 ABSTRACT: I will present an overview of the nature and importance of the Gestalt approach to per- ception and cognition with an emphasis on its relation to modern work in cognitive science. First I will discuss the nature of the contribution made by Gestalt psycholo- gists in the historical context in which they worked. Then I will trace their influ- ence on some current work in cognitive sci- ence: textural segmentation (Julesz, Beck & Rosenfeld), Pragnanz (Leeuwenberg, Palmer), soap-bubble systems (Marr & Poggio, Attneave, Hinton), and global precedence (Navon, Broadbent, Ginsberg). Beginning with this talk, the Cognitive Science Seminar will periodically present tutorials as a service to its interdisciplinary audience. Each tutorial will review the ideas in some research area for workers outside ------------------------------ Date: Thu, 25 Oct 84 15:17:33 EDT From: "Martin R. Lyons" <991@NJIT-EIES.MAILNET> Subject: Seminar - AI and Real Life ARTIFICIAL INTELLIGENCE AND REAL LIFE "Artificial Intelligence and Real Life", a talk by Paul Levinson of The New School for Social Research, will be one of several topics discussed as part of the Second Colloquium on Philospohy and Technology. The event is co-sposored by the Media Studies Program of the New School for Social Research and the Philosophy & Technology Studies Center at the Polytechnic Institute of New York. The talk will be held at the New School's 66 W. 12th St. Building, NYC, Monday November 12th, at 8pm, and the general public is invited. Admission is free. I am passing this info on for Paul Levinson, the aforementioned speaker. He can be reached directly at this site as: Lev%NJIT-EIES.Mailnet@MIT-MULTICS.ARPA or @MIT-MULTICS.ARPA:Lev@NJIT-EIES.Mailnet Please do not address inquiries to me, as all the info I have is above. MAILNET: Marty@NJIT-EIES.Mailnet ARPA: Marty%NJIT-EIES.Mailnet@MIT-MULTICS.ARPA USPS: Marty Lyons, CCCC/EIES @ New Jersey Institute of Technology, 323 High St., Newark, NJ 07102 (201) 596-2932 "You're in the fast lane....so go fast." ------------------------------ End of AIList Digest ******************** 30-Oct-84 22:20:11-PST,16234;000000000001 Mail-From: LAWS created at 30-Oct-84 22:15:52 Date: Tue 30 Oct 1984 22:04-PST From: AIList Moderator Kenneth Laws Reply-to: AIList@SRI-AI US-Mail: SRI Int., 333 Ravenswood Ave., Menlo Park, CA 94025 Phone: (415) 859-6467 Subject: AIList Digest V2 #147 To: AIList@SRI-AI AIList Digest Wednesday, 31 Oct 1984 Volume 2 : Issue 147 Today's Topics: LISP - Function-itis & Comparison with C, Algorithms - Pessimal Algorithms & Real Programmers, Seminars - Robot Navigation & Accessibility of Analogies & Student Models ---------------------------------------------------------------------- Date: Sun 28 Oct 84 17:40:10-PST From: Shawn Amirsardary Subject: Lisp Function-itis [Forwarded from the Stanford bboard by Laws@SRI-AI.] Lisp with its very elegant syntax suffers from acute function-itis. When adhering to the traditional lispish style of using very few setq and god forbid even fewer progs, you end up with about a million and a half functions that get called from usually only one place. Of course LET and lambda help, but not that much. My question is does anybody know of a good method for ordering and perhaps even naming the little suckers? In pascal you have to define procedures before you use them, but the lack of such restrictions in lisp means that functions are all over the place. What is the cure? --Shawn ------------------------------ Date: Tue, 30 Oct 84 21:42:58 -0200 From: eyal@wisdom (Eyal mozes) Subject: Re: different language types > But seriously, one can adopt a very abstract (i.e. applicative/functional) > programming style or a very imperative (C-like) style when using Lisp. > On the other hand, adopting an applicative style in C is difficult (yes, > I've tried!). So Lisp is certainly more versatile. Really!! I've never yet seen an "imperative" non-trivial LISP program which is not impossible to read, full of bugs which nobody knows how to correct, and horribly time-consuming (most of UNIX's VAXIMA is a good example of what I mean). Writing "imperative" style in LISP is a programming equivalent of "badgorithms". You can be as abstract as you want to be in C or Pascal. I don't think there is anything for which you can't come up with a good program in C, if your writing style is good (and if it isn't, no language will help). Of course, there are some activities, especially in some areas of AI, which are made much easier by the functional style of LISP, and its representation of programs as data - but even this wouldn't be true for *all* AI systems. But in terms of versatility, I don't think there can be much question about the big advantage of C, Pascal, and languages of this type. ------------------------------ Date: 29 Oct 84 14:41 PST From: JonL.pa@XEROX.ARPA Subject: Pessimal Algorithms ("Badgorithms"?) The following "badgorithm" comes from a practical joke, perpretated many years ago; although it is not a naturally occuring "badgorithm", it does have a humorous side. In high school, I worked part-time for a university computing center ** programming on an IBM 650 (don't ask how many years ago!). [One really shouldn't pooh-pooh the 650 -- it was the world's first List processing computer! explication follows]. It's main memory consisted of 2000 10-digit decimal words, stored on a rotating magnetic drum; there were 40 rotational channels on the drum, with each channel holding 50 words (one complete revolution). Since the various instructions took a variable amount of time, it would be most unwise to sequence the instructions by merely incrementing each instruction address by 1, for an instruction which took more time than that which would elapse between two successive words in a "channel" would thus be blocked for one full drum rotation time. An instruction consisted of an operator, an operand address, and a "next instruction address" (i.e., a CDR link in Lisp parlance); thus one could assign the sequenceing of instructions to be "optimal" in that the successor of an instruction at word offset A (mod 50) would be A+n (mod 50) where n is the time in, fiftieths of a drum rotation, required for the instruction exection of the operator stored at A. The IBM 704/709 series had a machine language assembler called SAP, for "Symbolic Assembly Program"; the 650 had SOAP, for "Symbolic Optimal Assembly Program". One would speak of "Soaping" a program, meaning to assemble a symbolic deck of cards into a self-loading machine-language deck. My cohorts and I dubbed the obvious pessimizing version of SOAP as SUDS, for "Symbolic Un-optimal Disassembly System" (it would assign the "next instruction" to a word offset just 1 short of optimal, and thus would slow down the resultant object code by up to a factor of 50). As a gag, we SUDS'd the SOAP deck, and left it for others to use. Imagine the consternation when a program that normally took 10 minutes to assemble suddenly began taking over an hour! Of course, we were quickly found out, and SUDS was relagated to the circular hacks file. -- Jon L White -- ------------------------------ Date: 22 Oct 84 16:18:41 EDT From: Michael.Jones@CMU-CS-SPICE Subject: Real Programmers [Excerpted from the CMU bboard by Laws@SRI-AI.] [I regret having to truncate this, but the original was too long to distribute on AIList. I have decide to proceed with the following because it fits in with other recent AIList messages. -- KIL] A recent article devoted to the *macho* side of programming made the bald and unvarnished statement: Real Programmers write in Fortran. [...] I feel duty-bound to describe, as best I can through the generation gap, how a Real Programmer wrote code. I'll call him Mel, because that was his name. I first met Mel when I went to work for Royal McBee Computer Corp., a now-defunct subsidiary of the typewriter company. [...] Mel's job was to re-write the blackjack program for the RPC-4000. [...] The new computer had a one-plus-one addressing scheme, in which each machine instruction, in addition to the operation code and the address of the needed operand, had a second address that indicated where, on the revolving drum, the next instruction was located. In modern parlance, every single instruction was followed by a GO TO! [...] Since Mel knew the numerical value of every operation code, and assigned his own drum addresses, every instruction he wrote could also be considered a numerical constant. He could pick up an earlier "add" instruction, say, and multiply by it, if it had the right numeric value. His code was not easy for someone else to modify. I compared Mel's hand-optimized programs with the same code massaged by the optimizing assembler program, and Mel's always ran faster. That was because the "top-down" method of program design hadn't been invented yet, and Mel wouldn't have used it anyway. He wrote the innermost parts of his program loops first, so they would get first choice of the optimum address locations on the drum. The optimizing assembler wasn't smart enough to do it that way. Mel never wrote time-delay loops, either, even when the balky Flexowriter required a delay between output characters to work right. He just located instructions on the drum so each successive one was just *past* the read head when it was needed; the drum had to execute another complete revolution to find the next instruction. [...] Mel called the maximum time-delay locations the "most pessimum". [...] Perhaps my greatest shock came when I found an innocent loop that had no test in it. No test. *None*. Common sense said it had to be a closed loop, where the program would circle, forever, endlessly. Program control passed right through it, however, and safely out the other side. It took me two weeks to figure it out. The RPC-4000 computer had a really modern facility called an index register. It allowed the programmer to write a program loop that used an indexed instruction inside; each time through, the number in the index register was added to the address of that instruction, so it would refer to the next datum in a series. He had only to increment the index register each time through. Mel never used it. Instead, he would pull the instruction into a machine register, add one to its address, and store it back. He would then execute the modified instruction right from the register. The loop was written so this additional execution time was taken into account -- just as this instruction finished, the next one was right under the drum's read head, ready to go. But the loop had no test in it. The vital clue came when I noticed the index register bit, the bit that lay between the address and the operation code in the instruction word, was turned on-- yet Mel never used the index register, leaving it zero all the time. When the light went on it nearly blinded me. He had located the data he was working on near the top of memory -- the largest locations the instructions could address -- so, after the last datum was handled, incrementing the instruction address would make it overflow. The carry would add one to the operation code, changing it to the next one in the instruction set: a jump instruction. Sure enough, the next program instruction was in address location zero, and the program went happily on its way. I haven't kept in touch with Mel, so I don't know if he ever gave in to the flood of change that has washed over programming techniques since those long-gone days. I like to think he didn't. In any event, I was impressed enough that I quit looking for the offending test, telling the Big Boss I couldn't find it. [...] I didn't feel comfortable hacking up the code of a Real Programmer." -- Source: usenet: utastro!nather, May 21, 1983. [The Cray is so fast it can execute an infinite loop in three minutes? This machine might beat it! -- KIL] ------------------------------ Date: 28 Oct 1984 14:43 EST (Sun) From: "Daniel S. Weld" Subject: Seminar - Robot Navigation [Forwarded from the MIT bboard by SASW@MIT-MC.] Wednesday, October 31; 4:00pm; 8th Floor Playroom Navigation for Mobile Robots Rodney A. Brooks There are a large number of interesting questions in how to build a mobile robot capable of navigating through unknown surroundings in order to complete some desired task. Issues include obstacle avoidance using local observations, overall path planning, registration with a map and building a map from observations. There is a lot of ongoing and promising work on the first two of these problems. Less has been done on the last two. Registration work has been most succesful with detailed a priori maps in two domains: (1) indoors uncluttered areas with flat walls giving unambigous geometric clues, and (2) areas with reliably identifiable and accurately locatable landmarks visible over a large area. Re-registration with maps generated from a robot's own observations has mainly been successful in two modes: (1) incremental re-registration involving small motions from a known location, or (2) in an environment with active beacons providing reliably indentifiable and locatable landmarks. This talk focus on some of the issues in building a map from unreliable observations and in re-registering the robot to that map much later, again using unreliable observations. In particular we consider a new map represention, the requirements on the representations of the world produced by vision, the role of landmarks, and whether other sensors such as compasses or inertial navigation systems are needed. COMING SOON: Kent Pitman [Nov 7], Ryszard Michalski [Nov 14], Phil Agre [Nov 28] ------------------------------ Date: 29 Oct 1984 14:10-EST From: Brad Goodman Subject: Seminar - Accessibility of Analogies [Forwarded from the MIT bboard by SASW@MIT-MC.] "Mental Models of Electricity" Yvette Tenney, BBN Laboratories Hermann Hartel, University of Kiel, West Germany BBN Laboratories, 10 Moulton St, Cambridge. Third floor large conference room, 10:30 AM. Monday November 5th. The presentation will consist of two short talks that were part of a conference on Representations of Students' Knowledge in Electricity and the Improvement of Teaching, held in Ludwigsburg, Germany this fall. Talk 1: Yvette Tenney (in collaboration with Dedre Gentner) "What makes analogies accessible: Experiments on the water-flow analogy for electricity." In analogy, knowledge can be transferred from a known (base) domain to a target domain, provided the learner accesses the analogy. We used the water-electric current analogy to test the hypothesis that prior familiarity with the base domain (Experiment 1) and pre-training on the base domain (Experiment 2) increase the likelihood of noticing the analogy. Results showed that greater knowledge of the base domain did not increase accessibility, although it did increase the power of the analogy if detected. Talk 2: Hermann Hartel "The electric circuit as a system: A new approach." [...] ------------------------------ Date: Tue 30 Oct 84 09:28:59-PST From: Paula Edmisten Subject: Seminar - Student Models [Forwarded from the Stanford SIGLUNCH distribution by Laws@SRI-AI.] DATE: Friday, November 2, 1984 LOCATION: Chemistry Gazebo, between Physical and Organic Chemistry TIME: 12:05 SPEAKER: Derek Sleeman School of Education & HPP ABSTRACT: The PIXIE Project: The Inference and Use of Student (user) Models For a decade or more the importance of having accurate student models to guide Intelligent Tutoring Systems (ITSs) has been stressed. I will give an overview of the several types of models which have been inferred and will talk in some detail about a system which infers overlay models and Pixie which uses process-orientated models. Currently, all these techniques effectively determine whether the current user's behaviour falls within a previously defined model-space. The focus of some current work is to see whether these techniques can be extended to be more data-sensitive. (Analogous issues arise when an ITS or ES is attempting to reason with an incomplete database.) Issues which arise in the use of models to control (remedial) dialogues will be addressed. The seminar will conclude with an overview of the fieldwork shortly to be undertaken. PIXIE now runs on a PC (in LISP) and several of these machines will be used to "diagnose" the difficulties which high school students have with Algebra and maybe Arithmetic. It is envisaged that PIXIE will be used to screen several classes, and that the class teachers will remediate students on the basis of the diagnostic information provided by PIXIE. These sessions will then be analyzed to determine how "real" teachers remediate; remedial subsystem(s) for PIXIE will then be implemented. Paula ------------------------------ End of AIList Digest ********************