IRList Digest Friday, 9 May 1986 Volume 2 : Issue 23 Today's Topics: Query - Interest in Vector Processors and Document Similarities Announcement - Job in AI and Education COGSCI - Probability, Maximum Entropy, and Expert Systems CSLI - Models in Semantics; Prolog and Geometry CSLI - Structures in Written Language; Functional-Typological Syntax; On Visual Communication CSLI - German-English Transfer of f-structures ---------------------------------------------------------------------- Date: Mon, 5 May 86 13:38:15 edt From: ltw@eecs.umich.CSNET Subject: Ed, FYI Date: Mon, 28 Apr 86 16:30:44+0100 From: user Subject: (forwarded) request for information Resent-From: Lou Salkind Resent-To: supercomputer@nyu.ARPA Resent-Date: Wed, 30 Apr 86 11:45:10 EDT I am working on a non-numerical problem (document similarity computations in information retrieval) with a VP200 vector processor. Do you know of other people working on non-numerical problems with vector processors. I would like to get in contact with other researchers in this field and I thought you might have some addresses. Thank you in advance Juerg Grossmann c/o Institute of Informatics University of Zurich Winterthurerstrasse 190 CH-8057 Zurich ...!cernvax!unizh!circe!gross ------------------------------ Date: Thu, 8 May 86 00:53:50 edt From: RICHER@SUMEX-AIM.ARPA To: richer@sumex-aim.arpa Subject: JOB ANNOUNCEMENT SIMON FRASER UNIVERSITY A major new initiative has resulted in the following positions for which applications are invited. o Knowledge Engineer o Post-doctoral Fellow o Research Scientist The Instructional Psychology Research Group is seeking qualified researchers trained in AI techniques. Applicants will join our staff in work aimed at transfering basic AI technology to educational applications and conducting specific research in: o Expert Systems o Planning Systems o Knowledge Understanding and Representation o Intelligent CAI The successful applicant will have an advanced degree in computer science and demonstrated experience in at least one of the aforementioned areas. Experience with Symbolics hardware, LISP, PROLOG and other AI languages is desirable. Preference will be given to applicants eligible for employment in Canada at the time of application. For further information contact Dr. Philip H. Winne Director, Instructional Psychology Research Group Faculty of Education Simon Fraser University Burnaby, British Columbia V5A 1S6 (604) 291-3395 Questions may be directed to the above address, or: Wolfgang_Rothen%SFU.Mailnet@MIT-Multics.ARPA ------------------------------ Date: Mon, 5 May 86 13:37:53 edt From: DEJONG%OZ.AI.MIT.EDU@mc.lcs.mit.edu Subject: Cognitive Science Calendar [Extract - Ed] From- Lisa F. Melcher Monday, 5 May 4:00pm Room: NE43-453 AUTOMATIC INDUCTION OF PROBABILISTIC EXPERT SYSTEMS Peter Cheeseman NASA Ames Many have realized that expert systems that make decisions under uncertainty must represent this uncertainty and manipulate it correctly. This cannot be done in general by "symbolic" (i.e. non-numeric) methods or by sprinkling numbers over logical inference, as advocated by many authors in AI. Probability has been proved to be the only consistent inference scheme if uncertainty is represented by a real number. Probabilistic inference requires assessing the effect of ALL the relevent evidence on the hypothesis of interest though ALL the possible chains of inference (rather than establishing a single path from axioms to theorem, as in logic). However, some methods used in probabilistic inference in AI (e.g. Prospector) impose strong constraints on the structure of the information (e.g. conditional independence) or require large amounts of information. The solution to this problem is to use Maximum Entropy to spread the uncertainty over the set of possibilities as evenly as possible consistent with the known information. A computationally efficient method for performing the maximum entropy calculation will be presented as well as a method for extracting the necessary probabilistic information directly from data. The result is a complete probabilistic expert system without using an expert. Sponsored by TOC, Laboratory for Computer Science Ronald Rivest, Host ------------------------------ Date: Thu, 24 Apr 86 01:34:20 est From: EMMA@su-csli.ARPA Subject: Calendar, April 24, No. 13 [Extracted and edited - Ed] THIS WEEK'S SEMINAR (April 24, 1986) Uses and Abuses of Models in Semantics Jon Barwise and John Etchemendy Barwise@su-csli and Etchemendy@su-csli The use of set-theoretic models as a way to study the semantics of both natural and computer languages is a powerful and important technique. However, it is also fraught with pitfalls for those who do not understand the nature of modeling. In this talk we hope to show how a proper understanding of the representation relationship implicit in modeling can help one exploit the power while avoiding the pitfalls. ... The talk will presuppose some familiarity with the techniques under discussion. PIXELS AND PREDICATES Prolog and Geometry Randolph Franklin, UC at Berkeley wrf@degas.berkeley.edu 1:00 p.m., Tuesday, April 29, CSLI trailers The Prolog language is a useful tool for geometric and graphics implementations because its primitives, such as unification, match the requirements of many geometric algorithms. We have implemented several problems in Prolog including a subset of the Graphics Kernal Standard, convex hull finding, planar graph traversal, recognizing groupings of objects, and boolean combinations of polygons using multiple precision rational numbers. Certain paradigms, or standard forms, of geometric programming in Prolog are becoming evident. They include applying a function to every element of a set, executing a procedure so long as a certain geometric pattern exists, and using unification to propagate a transitive function. Certain strengths and weaknesses of Prolog for these applications are now apparent. ------------------------------ Date: Thu, 1 May 86 18:41:21 edt From: EMMA@su-csli.ARPA Subject: Calendar, May 1, No. 1 [Extracted and edited - Ed] THIS WEEK'S COLLOQUIUM Structures in Written Language Geoff Nunberg (Nunberg@csli) 4:15, Thursday, May 1, Redwood G-19 Just about all contemporary research on linguistic structure has been based exclusively on observations about the spoken language; the written language, when it is talked about at all, is generally taken to be derivative of speech, and without any independent theoretical interest. When we consider the written language in its own terms, however, it turns out to have a number of distinctive features and structures. In particular, it contains a number of explicitly delimited ``text categories,'' such as are indicated by the common punctuation marks and related graphical features, which are either wholly absent in the spoken language, or at best are present there only implicitly. In the course of uncovering the principles that underlie the use of text categories like the text-sentence, paragraph, and parenthetical (i.e., a string delimited by parentheses), we have to provide three levels of grammatical description: a semantics, which sets out the rules of interpretation associated with text categories by associating each type with a certain type of informational unit; a syntax, which sets out the dependencies that hold among category-types; and a graphology, which gives the rules that determine how instances of text categories will be graphically presented. Each of these components is a good deal more complex and less obvious than one might suppose on the basis of a recollection of what the didactic grammars have to say about the written language; what emerges, in fact, is that most of the rules that determine how text delimiters are used are not learned through explicit instruction, and are no more accessible to casual reflection than are the rules of grammar of the spoken language. NEXT WEEK'S TINLUNCH (5/8/86) Definiteness and Referentiality Vol. 1, Ch. 11 of Syntax: A Functional-Typological Introduction by Talmy Givon Discussion led by Mark Johnson (Johnson@csli) The relationship between syntactic structure and meaning is one of the most interesting lines of research being undertaken here at CSLI. One of the questions being addressed in this work concerns the way that grammatical or syntactic properties of an utterance interact with its semantics, i.e., what it means. Givon and others claim that discourse notions of topicality and definiteness interact strongly with grammatical processes such as agreement---and moreover, that there is no clear dividing line between grammar and discourse; one cannot understand agreement or anaphora viewing them as purely grammatical processes. Linguists here at CSLI are tentatively moving toward this position, for example Bresnan and Mchombo (1986) make explicit use of a theory of ``discourse functions'' to explain the distributional properties of Object Marking in Chichewa, so a discussion of what it would mean to have an ``integrated'' theory of language is quite timely. Givon's treatment of definiteness and referentiality explicitly rejects earlier philosphical treatments as being ``too restrictive to render a full account of the facts of human language.'' He starts by listing some observations on the interactions between definiteness and a variety of other linguistic phenomena (e.g. modality) and goes on to propose a model based on a ``Universe of Discourse'' and the notion of ``referential intent.'' After examining examples of how referentiality is coded in various languages and how it interacts with various other syntactic and semantic phenomena, he finishes by discussing degrees of definiteness and referentially, and introduces the notion of communicative importance. This chapter raised several interesting questions. For example, what are the key properties of referentiality and definiteness, and how would one go about building a theory that expresses them? What are Givon's insights into this matter, and how could these be reconstructed within a formal theory such as DRS theory or Situation Semantics? NEXT WEEK'S SEMINAR (5/8/86) On Visual Communication David Levy, Xerox Palo Alto Research Center (Dlevy.pa@xerox) Lately there has been much talk around CSLI about representation as a concept transcending and unifying work being done in different research groups and domains. Various points have emerged and recurred in recent presentations and discussions: the distinction between the representing state of affairs (A) and the state of affairs represented (B); examples of the dangers inherent in conflating them; forms of structural correspondence between aspects (objects, properties, and relations) of A and aspects of B; the partiality of representation (the fact that only certain aspects of A correspond to aspects of B, and that only certain aspects of B correspond to aspects of A); the priority of B over A; and so on. The use of computers is largely mediated by representations. Many of these are transparent to us: We talk of ``typing an A'' when we actually press a key, causing a character code (a character representation) to be generated from which an actual character is rendered. We talk of ``viewing'' data structures, when in fact we do nothing of the sort, since data structures ``inside'' machines are inherently non-visual, much as are mental states ``inside'' heads; rather, we view *visual representations* of data structures. In many contexts the transparency of representations (leading to the conflation of A and B) is tremendously useful and powerful. The term ``direct manipulation'' denotes a style of user interface design in which the user is led (or encouraged) to conflate the visual objects on the screen (e.g. icons) with the things they represent (e.g. printers), and to conflate the representation of these visual objects with the visual objects themselves. But there seem to be times when our facility for seeing through representations is a hindrance rather than a help, as Barwise and Etchemendy argued recently for the case of model theory. As a theoretician and observer of certain classes of computer systems, and, equally importantly, as a *designer* of them, I believe that we need an understanding of representation (and of the sorts of issues described in the first paragraph) to help us build truly rational systems. In this talk I will focus on the problem of developing an analysis of visual representation. I will use examples from the surface of computer screens (e.g. windows, scroll bars, and icons) to illustrate the importance of distinctions such as visual vs. non-visual entities, representing vs. represented entities, and (active) processes vs. (static) representation relations. ------------------------------ Date: Thu, 8 May 86 00:53:56 edt From: EMMA@SU-CSLI.ARPA Subject: Calendar, May 8, No. 15 [Extracted - Ed] NEXT WEEK'S COLLOQUIUM (May 15, 1986) Transfer of f-structures Across Natural Languages Tom Reutter, Weidner Communications Corp., Chicago A recursive algorithm for mapping functional structure from a source natural language into a target natural language is presented and its implementation in the programming language CPROLOG is discussed. The transfer algorithm is guided by a symmetrical bilingual lexicon. It was prototypically implemented for German-English as part of a transfer-oriented machine translation system at the University of Stuttgart (Germany). Special emphasis is placed on asymmetiral transfer, e.g., mapping of f-structures with different semantic valencies, unequal NUM and SPEC attributes, etc.