IRList Digest Wednesday, 8 October 1986 Volume 2 : Issue 50 Today's Topics: Query - Indexing Legal Depositions Discussion - Soundex (ref. IRList #44) COGSCI - Object Recognition as Visual Classification COGSCI - Generic Tasks in Expert Systems, Pattern Knowledge and Search COGSCI - General Logic COGSCI - Planning Simultaneous Actions in Temporally Rich Worlds CSLI - Figural Representation Discussion Group News addresses are ARPANET: fox%vt@csnet-relay.arpa BITNET: foxea@vtvax3.bitnet CSNET: fox@vt UUCPNET: seismo!vtisr1!irlistrq ---------------------------------------------------------------------- Date: Sat, 20 Sep 86 00:54:22 edt From: csnet_gateway (LAWS@SRI-STRIPE.ARPA) To: bboard.maintainer@a.cs.cmu.edu Subject: Text abstraction. References? [Forwarded from the CMU BBoard by Laws@SRI-STRIPE.] A friend of mine is interested in automated text abstraction. In particular, he is would like to write a system that does as much as possible toward abstraction and subject indexing of legal depositions. Does anyone know anything about this subject, or know of any good references? Dean [Note: Several IRList subscribers have worked or are working in this are, so I hope they will send information to IRList which can be filtered back to CMU. Carole Hafner is the first who comes to mind, and I remember Harry Wu being interested. - Ed] ------------------------------ Date: Fri, 19 Sep 86 12:35:05 edt From: rdj (R. Daniel Johnson) Subject: IRLIST V2#44 Ed, In V2#44, Stephen Page requested information on what is called the "Soundex Code" which is a moderately effective and inexpensive means of matching words (e.g. names) which "sound" alike. I used such matching in a film booking system some years ago to match teacher names in schools. ... The technique is simple and goes something like this: drop all vowels from a word to match, do a many-to-one mapping of the consonants, and do a search on the resulting 4-character code. I think I still have a listing of the routine in HP Systems Programming Language (ALGOL-like.) Perhaps I should offer it to Stephen Page? (It wasn't clear whether he wanted to use it or find literature about its use.) -d P.S. on Soundex Aha... Soundex is documented in Knuth's Searching and Sorting. ------------------------------ Date: Mon, 22 Sep 86 18:39:49 edt From: csnet_gateway (DEJONG%OZ.AI.MIT.EDU@MC.LCS.MIT.EDU) Subject: Cognitive Science Calendar Date: Monday, 22 September 1986 11:09-EDT subject: Center for Biological Information Processing Wednesday, 24 September 12:00pm Room: E25-401 Object Recognition as Visual Classification Aaron Bobick MIT - Department of Brain and Cognitive Sciences One of the most commonly proposed goals of a perceptual system is that of object recognition. Yet, to date, we have been unable to define precisely what the task of recognition really entails. In this talk I will provide a definition of recognition in terms of classifying objects into natural categories, and I will argue that natural categories can be defined in terms of processes that create objects and the interaction between objects, their environment and the observer. Given this definition of object recognition we can decompose the task into the three components of representation, speculation and evaluation and consider the knowledge or constraints that must be provided to the observer if he is th be able to classify objects successfully. Both an idealized world and the natural world will be considered. For the natural world, a procedure based upon the uncertainty of categorical inferences will be presented along with some results of an implementation that "naturally" categorizes leaves. Finally, the questions of whether recognition should be considered as a learning problem and exactly what can or should be learned will be addressed, i.e., some difficult questions will be deftly hedged. ------------------------------ Date: Tue, 30 Sep 86 20:39:32 edt From: csnet_gateway (DEJONG%OZ.AI.MIT.EDU@MC.LCS.MIT.EDU) Subject: Cognitive Science Calendar ++++++++++++++++++++++++++++++++++++++++++ Date: Friday, 26 September 1986 11:09-EDT Subject: Generic Tasks in Knowledge-Based Reasoning Wednesday, 1 October 3:00pm Room: NE43-512A Generic Tasks in Knowledge-Based Reasoning: Caracterizing and Designing Expert Systems at the "Right" Level of Abstraction B. Chandrasekaran Laboratory for Artificial Intelligence Research Department of Computer and Information Science Ohio State University Columbus, Ohio 43210 We outline the elements of a framework for expert system design that we have been developing in our research group over the last several years. This framework is based on the claim that complex knowledge-based reasoning tasks can often be decomposed into a number of generic tasks each with associated types of knowledge and family of control regimes. At different stages in reasoning, the system will typically engage in one of the tasks, depending upon the knowledge available and the state of problem solving. The advantages of this point of view are manifold: (i) Since typically the generic tasks are at a much higher level of abstraction than those associated with first generation expert system languages, knowledge can be acquired and represented directly at the level appropriate to the information processing task. (ii) Since each of the generic task has an appropriate control regime, problem solving behavior may be more perspicuously encoded. (ii) Because of a richer generic vocabulary in terms of which knowledge and control are represented, explanation of problem solving behavior is also more perspicuous. We briefly describe six generic task that we have found very useful in our work on knowledge-based reasoning: classification, state abstraction, knowledge-directed retrieval, object synthesis by plan selection and refinement, hypothesis matching, and assembly of compound hypotheses for abduction. ++++++++++++++++++++++++++++++++++++++++++ Date: Friday, 26 September 1986 11:09-EDT Subject: LCS Distinguished Lecturer Series Thursday, 2 October 3:30pm Room: 34-101 Pattern Knowledge and Search: The Suprem Architecture Hans J. Berliner Senior Research Scientist Computer Science Department Cartnegie Mellon University We describe a problem solving architecture that combines rapid search through a state space coupled with highly sophisticated evaluation of states. Speed of searching is made possible by special purpose hardware tailored to producing state-change operators for the domain. Evaluation of states is done by more general machines that can extract small patterns in parallel and use these to produce much better evaluations than are possible in non-pattern oriented methods. In this connection, we present a hierarchy of knowledge. This method has been applied to the development of Hitech, a chess machine that is now ranked within the top 200 chess players in the US. While Hitech's chess understanding is still far short of that of the very top players in the world, its ability to calculate accurately, and look at one million times as many possibilities in unit time as the best humans can, make up for this to a large degree. We discuss the issues involved in creating such architectures and how they can be improved. Performance data on Hitech will also be presented. Host: Michael L. Dertouzos ------------------------------ Date: Tue, 30 Sep 86 20:44:01 edt From: csnet_gateway (DEJONG%OZ.AI.MIT.EDU@MC.LCS.MIT.EDU) Subject: Cognitive Science Calendar Date: Tuesday, 30 September 1986 11:00-EDT From: Lisa F. Melcher Thursday, 2 October 2:00pm Room: NE43-512a " GENERAL LOGIC " Gordon Plotkin Department of Computer Science University of Edinburgh, Scotland A good many logics have been proposed for use in Computer Science. Implementing them involves repeating a great deal of work. We propose a general account of logics as regards both their syntax and inference rules. As immediate target we envision a system to which one inputs a logic obtaining a simple proof-checker. The ideas build on work in logic of Paulson, Martin-Lof and Schroeder-Heister and in the typed lambda-calculus of Huet and Coquand and Meyer and Reinhold. The slogan is: Judgements are Types. For example the judgement that a proposition is true is identified with its type of proofs; general and hypothetical judgements are identified with dependent product types. This gives one account of Natural Deduction. It would be interesting to extend the work to consider (two-sided) sequent calculi for classical and modal logics. Sponsored by TOC, Laboratory for Computer Science Albert Meyer, Host ------------------------------ Date: Thu, 2 Oct 86 18:59:31 edt From: csnet_gateway (DEJONG%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU) Subject: Cognitive Science Calendar Date: Thursday, 2 October 1986 13:30-EDT From: Brad Goodman Monday, 6 October 10:30am Room: 3rd floor large conference room, BBN Labs, 10 Moulton Street, Cambridge BBN Laboratories Science Development Program AI/Education Seminar Planning Simultaneous Actions in Temporally Rich Worlds Professor James Allen University of Rochester (james@rochester) This talk describes work done with Richard Pelavin over the last few years. We have developed a formal logic of action that allows us to represent knowledge and reason about the interactions between events that occur simultaneously or overlap in time. This includes interactions between two (or more) actions that a single agent might perform simultaneously, as well as interactions between an agent's actions and events occuring in the external world. The logic is built upon an interval-based temporal logic extended with modal operators similar to temporal necessity and a counterfactual operator. Using this formalism, we can represent a wide range of possible ways in which actions may interact. ------------------------------ Date: Sat, 4 Oct 86 00:58:19 edt From: csnet_gateway (EMMA@CSLI.STANFORD.EDU) Subject: Late Newsletter Entry Date: 2 Oct 86 17:43 PDT From: dlevy.pa@Xerox.COM Emma: Could you send this out through the usual channels as a late newsletter entry? Thanks, David Reading and Discussion Group on Figural Representation Organizers: David Levy, Geoff Nunberg First meeting: Thursday, October 9 at 10 AM, Ventura Hall We are forming a reading and discussion group to explore the nature of figural (roughly speaking, visual) representation. Systems of figural representation include writing systems, systems of musical notation, screen "icons," bar graphs, architectural renderings, maps, and so forth. This topic lies at the intersection of various concerns relevant to a number of us at CSLI, at Xerox PARC, and at SRI -- theoretical concerns about the nature of language and representation and their manifestation in the building of systems and the design of visual notations for formal languages. There is currently no well-motivated framework for discussing such material, no map on which to locate important terms such as "document," "text," "icon," and "format." But there is clearly a coherent subject matter here waiting to be explored. Topics we want to look at in early meetings include: 1. Properties of the figural. 2. Figural representation and representation in general. 3. The typology of figural systems. 4. Writing as a figural representation system; distinctive properties of written language. 5. The technological basis for figural representation (from writing to print to the computer). Initially, we plan to organize the discussion around readings drawn from the literatures of a number of disciplines, among them linguistics, psychology, literary theory, art criticism, AI, anthropology and history. We expect to meet once a week (or once every two weeks) at Ventura Hall (CSLI), starting Thursday morning, October 9, at 10AM. Please note that we consider this to be a working group, not a general public forum or a TINLunch. At our first meeting, we will be discussing a short paper, "Visible Language," which outlines some of the areas we will be concerned with. Copies are available at the Ventura Hall desk. ------------------------------ END OF IRList Digest ********************