nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (04/26/88)
NL-KR Digest (4/25/88 18:49:50) Volume 4 Number 45 Today's Topics: How Language Structures its Concepts Talk on knowledge compilation A Theory of Justified Reformulations Encoding "Here" and "There" in the Visual Field From CSLI Calendar, April 21, 3:25 Theoretical and Computational Issues in Lexical Semantics (TCILS) Seminar - AI Revolving -- Alan Bawden (MIT) A Parallel Model of Sentence Processing Submissions: NL-KR@CS.ROCHESTER.EDU Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU ---------------------------------------------------------------------- Date: Thu, 14 Apr 88 14:11 EDT From: Dori Wells <DWELLS@G.BBN.COM> Subject: Lang. & Cognition Seminar BBN Science Development Program Language & Cognition Seminar Series HOW LANGUAGE STRUCTURES ITS CONCEPTS: THE ROLE OF GRAMMAR Leonard Talmy Program in Cognitive Science University of California, Berkeley BBN Laboratories Inc. 10 Moulton Street Large Conference Room, 2nd Floor 10:30 a.m., Wednesday, April 20, 1988 Abstract: A fundamental design feature of language is that it has two subsystems, the open-class (lexical) and the closed-class (grammatical). These subsystems perform complementary functions. In a sentence, the open-class forms together contribute most of the *content* of the total meaning expressed, while the closed-class forms together determine the majority of its *structure*. Further, across the spectrum of languages, all closed-class forms are under great semantic constraint: they specify only certain concepts and categories of concepts, but not others. These grammatical specifications, taken together, appear to constitute the fundamental conceptual structuring system of language. I explore the particular concepts and categories of concepts that grammatical forms specify, the properties that these have in common and that distinguish them from lexical specifications, the functions served by this organization in language, and the relations of this organization to the structuring systems of other cognitive domains such as visual perception and reasoning. The greater issue, toward which this study ultimately aims, is the general character of conceptual structure in human cognition. ------------------------------ Date: Thu, 14 Apr 88 14:32 EDT From: Dori Wells <DWELLS@G.BBN.COM> Subject: Lang. & Cognition Seminar BBN Science Development Program Language & Cognition Seminar Series PALENQUE: AN INTERACTIVE DISCOVERY-BASED LEARNING EXPERIENCE FOR CHILDREN Kathleen Wilson Bank Street College BBN Laboratories Inc. 10 Moulton Street Large Conference Room, 2nd Floor 2:00 p.m., Thursday, April 21, 1988 Abstract: Educational technology designers have recently been exploring the potential uses of new interactive video technologies and, in particular, experimenting with a variety of structures and metaphors for the information on a disk. Palenque is a DVI (Digital Video Interactive) pilot application that uses both spatial and thematic structures to provide students with a surrogate travel experience through Palenque, an ancient Mayan site. It is designed to be used with the second season of the Voyage of the Mimi project, an interdisciplinary curriculum that includes broadcast TV shows, software and classroom activities for grades 4 through 8. The talk will include a description of the Mimi project (including a tape of a TV excerpt), a discussion of design and pedagogical principles underlying Palenque, and a description of its use in classrooms. ------------------------------ Date: Thu, 14 Apr 88 20:56 EDT From: Chris Tong <ctong@lightning.rutgers.edu> Subject: Talk on knowledge compilation The following thesis proposal defense will be held at 10am, Mar. 29, in Hill Center, room 423, Busch Campus, Rutgers University, New Brunswick, NJ., and will be chaired by Chris Tong. CONSTRAINT INCORPORATION USING CONSTRAINED REFORMULATION Wesley Braudaway wes@aramis.rutgers.edu ABSTRACT. The goal of this research is to develop knowledge compilation techniques to produce a problem-solving system from a declarative solution description. It has been shown that a Generate-and-Test problem-solver can be compiled from a declarative language that represents solutions as instances of a (hierarchically organized) solution frame; the generator systematically constructs instances of the solution frame, until one is found that meets all the tests. However, this Generate-and-Test architecture is computationally infeasible as a problem-solver for all but trivial problems. Optimization techniques must be used to improve the efficiency of the resulting problem-solving system. Test Incorporation is one such optimization technique that moves testers, which test the satisfaction of the problem constraints, back into the generator sequence to provide early pruning. This proposal defines a special kind of test incorporation called Constraint Incorporation. This technique modifies the generators so they enumerate only those generator values that satisfy the problem constraints defined by the tests. Because of this complete incorporation, the tests defining the incorporated constraints can be removed from the Generate-and-Test architecture. This results in a significant increase of problem-solving efficiency over test incorporation when the test cannot be partitioned into subtests that affect a single generator. These cases seem to occur when a mismatch exists between the language used to represent (and construct) solutions and the language used to define the problem constraints. To incorporate these constraints, the representations of solutions and problem constraints should be shifted (i.e., reformulated) so as to bridge the gap between them. One method for bridging the gap is to search the space of solution and problem representations until incorporation is enabled. However, because of the difficulties encountered (e.g., the space is large and difficult to generate), an alternative method is proposed that will constrain the reformulation process. This method incorporates constraints by compiling an abstract solution description into a problem-solver. By using an abstract solution description, the system does not commit prematurely to a detailed and biased representation of the solution description. The problem constraints are refined into procedural specifications and merged to form a partial specification of the problem-solver. The problem-solver is partial in that it only generates those solution details mentioned in the constraints. In this way, the compiler is focusing on just those details of the solution language that are relevant to incorporating the constraints. The partial problem-solver is then extended into a complete one by adding generators for the remaining details. Any such extension is guaranteed to have successfully incorporated all the constraints. This method has been applied to a house floorplanning domain, using extensive paper traces. It is currently being implemented, and will be applied to a second domain. ------------------------------ Date: Tue, 12 Apr 88 13:04 EDT From: Mary E. Spollen <SPOLS%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU> Subject: A Theory of Justified Reformulations A THEORY OF JUSTIFIED REFORMULATIONS Devika Subramanian Department of Computer Science Stanford University Tuesday, April 19, 1988 Refreshments...4:00 p.m. Lecture...4:15 p.m. NE43-512A ABSTRACT Present day systems, intelligent or otherwise, are limited by the conceptualizations of the world given to them by their designers. This research explores issues in the construction of adaptive systems that can incrementally reformulate their conceptualizations to achieve computational efficiency or descriptional adequacy. In this talk, a special case of the reformulation problem is presented: we reconceptualize a knowledge base in terms of new abstract objects and relations in order to make the computation of a given class of queries more efficient. Automatic reformulation will not be possible unless a reformulator can justify a shift in conceptualization. We present a new class of meta-theoretical justifications for a reformulation, called irrelevance explanations. A logical irrelevance explanation proves that certain distinctions made in the formulation are not necessary for the computation of a given class of problems. A computational irrelevance explanation proves that some distinctions are not useful with respect to a given problem solver for a given class of problems. Inefficient formulations make irrelevant distinctions and the irrelevance principle logically minimizes a formulation by removing all facts and distinctions in it that are not needed for the specified goals. The automation of the irrelevance principle is demonstrated with the generation of abstractions from first principles. We also describe the implementation of an irrelevance reformulator and outline experimental results that confirm our theory. Host: Prof. Gerald Jay Sussman ------------------------------ Date: Mon, 18 Apr 88 09:35 EDT From: William J. Rapaport <rapaport@cs.Buffalo.EDU> Subject: Encoding "Here" and "There" in the Visual Field STATE UNIVERSITY OF NEW YORK AT BUFFALO The Steering Committee of the GRADUATE STUDIES AND RESEARCH INITIATIVE IN COGNITIVE AND LINGUISTIC SCIENCES PRESENTS ZENON PYLYSHYN Center for Cognitive Science University of Western Ontario ENCODING "HERE" AND "THERE" IN THE VISUAL FIELD: A Sketch of the FINST Indexing Hypothesis and Its Implications I introduce a distinction between encoding the location of a feature within some frame of reference, and individuating or indexing a feature so later processes can refer to and access it. A resource-limited indexing mechanism called a FINST is posited for this purpose. FINSTs have the property that they index features in a way that is (in most cases) transparent to their retinal location, and hence "point to" scene locations. The basic assumption is that no operations upon sets of features can occur unless all the features are first FINSTed. A number of implications of this hypothesis will be explored in this talk, including its relevance to phenomena such as the spatial stability of visual percepts, the ability to track several independently moving targets in parallel, the ability to detect a class of spatial relations requiring the use of "visual routines", and various mental imagery phenomena. I will also discuss one of the main reasons for postulating FINSTs: the possibility that such indexes might be used to bind per- ceived locations to arguments in motor commands, thereby serving as a step towards perceptual-motor coordination. Monday, April 25, 1988 4:00 P.M. 110 Knox, Amherst Campus ^^^^^^^^ There will also be an informal evening discussion at Judy Duchan's home, 130 Jewett Parkway, at 8:00 P.M. Call Bill Rapaport (Dept. of Computer Science, 636-3193 or 3180) for further information. ------------------------------ Date: Wed, 20 Apr 88 20:25 EDT From: Emma Pease <emma@russell.stanford.edu> Subject: From CSLI Calendar, April 21, 3:25 [Excerpted ...] NEXT WEEK'S CSLI TINLUNCH Reading: "Constraints, Meaning and Information" by Ian Pratt, (then at) Princeton University Discussion led by Keith Devlin (devlin@csli.stanford.edu) April 28 The paper claims to establish a fatal flaw in situation semantics as developed in "Situations and Attitudes" by Barwise and Perry, arguing that the meaning of a declarative sentence, whatever it is, cannot be a constraint. As a mathematician trying to help build a theory around the ideas developed in S&A, this claim, needless to say, bothers me. My own reading of the paper leads me to believe that Pratt bases his argument on two basic misreadings of S&A. But maybe I am misreading him. I am hoping that the linguists and philosophers at CSLI (but not Koll Construction) will be able to help me out and reassure me that all is not built on sand. -------------- NEXT WEEK'S CSLI SEMINAR Types and Tokens in Linguistics Sylvain Bromberger (sylvain@csli.stanford.edu) April 28 This paper takes as its point of departure three widely -- and rightly -- accepted truisms: (1) that linguistic theorizing rests on information about types, that is, word types, phrase types, sentence types, and the like; (2) that linguistics is an empirical science and that the information about types on which it rests is empirical information, that is, information obtained by attending with ones senses to something -- normally tokens (utterances); (3) that the facts that linguistics seeks to uncover -- for instance, facts about the lexicon or the range of sentences in a given language -- follow, in part at least, from facts about people's mental makeup. It (the paper) seeks to reconcile (1) with (2) and with (3). Few, if any, practitioners feel the need for such a reconciliation, but wide-eyed philosophers like me who think (rightly) that types are abstract entities, that is, nonspatial, nontemporal, unobservable, causally impotent entities, have trouble seeing how information about such entities can be obtained by attending to spatial, temporal, observable entities, though they cannot deny that it can; and they have even greater trouble seeing how features of minds (some misguided souls would say brains) can have repercussions in the realm of abstract entities. The reconciliation proposed in the paper is based on two conjectures. First, that tokens of a type (for instance all the utterances of `cat', or all the utterances of `Mary sank a ship') form what I call a "quasi-natural kind," a grouping like that formed by all the samples of a chemical substance (for instance, all the samples of mercury). Second, that tokens of different types form what I call "categories," a grouping like that formed by the samples of different chemical substances (mercury, water, gold, sulfuric acid, etc.). The possibility of inferring facts about types from facts about tokens follows from these two conjectures like day follows night. And so does the conclusion that linguistics is grounded on mental realities. The conjectures, if true, also reveal that the subject matter of linguistics, like the subject matter of any natural science, is defined by configurations of questions as well as by the makeup of the world. The paper is an essay in the philosophy of science as it applies to linguistics. ------------------------------ Date: Wed, 20 Apr 88 20:42 EDT From: James Pustejovsky <jamesp@brandeis.csnet> Subject: Theoretical and Computational Issues in Lexical Semantics (TCILS) THEORETICAL AND COMPUTATIONAL ISSUES IN LEXICAL SEMANTICS (TCILS) A Workshop with Support from AAAI Computer Science Department Brandeis University April 21-24, 1988 FINAL SCHEDULE THURSDAY EVENING, April 21, 8:00 - 12:00, Welcoming Reception. ``A Cambridge House'', Cambridge. (617)-491-6300. FRIDAY, April 22. Brandeis University, Usdan Student Center. 8:45-9:15 Coffee and Bagels 9:15-9:30 Opening Remarks, Pustejovsky SESSION 1 9:30-10:00 Jackendoff ``X'-Semantics'' 10:15-10:45 Talmy ``The Lexicalization of Aspect and Result'' 11:00-11:30 Pustejovsky ``Structured Representations for the Lexicon'' 11:45-2:00 LUNCH. Cold Buffet. SESSION 2 2:00-2:30 Grimshaw ``On the Representation of Two Different Nominal'' 2:45-3:15 Williams ``Nominals and Binding'' 3:30-4:00 Ingria ``Adjectives, Nominals, and the Status of Arguments'' 4:15-5:30 SPIRITS Evening FREE SATURDAY, April 23. Brandeis University, Usdan Student Center. 8:45 Coffee and Bagels SESSION 3 9:15-9:45 Wilks ``Dictionary Texts as Tractable Resources'' 10:00-10:30 Calzolari ``Deriving Lexical and World Knowledge'' 10:45-11:00 Coffee and Pastries SESSION 4 11:00-11:30 Hobbs ``Commonsense Knowledge and Lexical Semantics'' 11:45-12:15 Sowa ``Lexical Inference vs. Commonsense Inference'' 12:30-2:00 LUNCH. Cold Buffet. SESSION 5 2:00-2:30 Zaenen ``Thematic Roles between Syntax and Semantics'' 2:45-3:15 Rapoport ``Lexical Subordination'' 3:30-3:45 Coffee and Cookies SESSION 6 3:45-4:15 Palmer ``The Status of Verb Representations in Pundit'' 4:30-5:00 Fawcett ``A Way to Model Probabilities in Relations'' 5:15-6:00 SPIRITS Evening FREE SUNDAY, April 24, Brandeis University, Usdan Student Center. 8:45 Coffee and Bagels SESSION 7 9:15-9:45 Kegl ``The Interface between Lexical Semantics and Syntax'' 10:00-10:30 Tenny ``Aspectual Interface Hypothesis'' 10:45-11:00 Coffee and Pastries SESSION 8 11:00-11:30 Sondheimer ``How to Realize a Concept'' 11:45-12:15 Nirenburg ``How Lexical Semantics Can Best Contribute'' 12:30-1:00 Closing Remarks Speaker List with Complete Titles Nicoletta Calzolari ``Deriving Lexical and World Knowledge from On-line Dictionaries'' Robin Fawcett ``A way to model probabilities in relations between main verbs, participants roles and the semantic features of things'' Jane Grimshaw ``On the Representation of Two Different Kinds of Nominals'' Jerry Hobbs ``Commonsense Knowledge and Lexical Semantics'' Robert Ingria ``Adjectives, Nominals, and the Status of Arguments'' (Response) Ray Jackendoff ``X'-Semantics'' Judy Kegl ``The Interface between Lexical Semantics and Syntax'' Sergei Nirenburg ``How Lexical Semantics Can Best Contribute to Natural Language Application'' (Response) Martha Palmer ``The Status of Verb Representations in Pundit'' James Pustejovsky ``Structured Semantic Representations for the Lexicon'' (Response) T.R. Rapoport ``Lexical Subordination'' Norm Sondheimer ``How to Realize a Concept: Lexical Selection and the Conceptual Network in Text Generation'' John Sowa ``Lexical Inference vs. Commonsense Inference'' Leonard Talmy ``The Lexicalization of Aspect and Result Typologically Parallels that of Motion'' Carol Tenny ``Aspectual Interface Hypothesis and Lexical Decomposition'' Yorick Wilks ``Dictionary texts as tractable resources for computational semantics'' Edwin Williams ``Nominals and Binding'' (Response) Annie Zaenen ``Thematic Roles between Syntax and Semantics'' ------------------------------ Date: Mon, 25 Apr 88 12:06 EDT From: Peter de Jong <DEJONG%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU> Subject: Seminar - AI Revolving -- Alan Bawden (MIT) Thursday, 28 April 4:00pm Room: NE43- 8th floor Playroom The Artificial Intelligence Lab Revolving Seminar Series Thinking About State Alan Bawden (Alan@ai.mit.edu) MIT AI Lab It is generally agreed that the unrestricted use of state can make a program hard to understand, hard to compile, and hard to execute, and that these difficulties increase in the presence of parallel hardware. This problem has led some to suggest that constructs that allow state should be banished from programming languages. But state is also a very useful phenomenon: some tasks are extremely difficult to accomplish without it, and sometimes the most perspicuous expression of an algorithm is one that makes use of state. Instead of outlawing state, we should be trying to understand it, so that we can make better use of it. In this talk I will propose a way of modeling systems in which the phenomenon of state occurs. Using this model I will characterize those systems in which some components of a system perceive other components as having state. I will propose that systems that exhibit state-like behavior are those systems that must rely on their own nonlocal structure in order to function correctly, and I will make this notion of nonlocal structure precise. This characterization offers some new insights into why state seems to cause the problems that it does. I will suggest how these insights might lead us towards better ways of thinking about state, and make our programming languages more expressive when we program with state. ------------------------------ Date: Thu, 14 Apr 88 17:52 EDT From: KASH@OZ.AI.MIT.EDU Subject: A Parallel Model of Sentence Processing MIT Center for Cognitive Science Parsing Seminar A PARALLEL MODEL OF SENTENCE PROCESSING Robin Clark Department of Philosophy Carnegie Mellon University In this talk, I will describe the Constrained Parallel Parser (CPP) currently under development at CMU. The model uses grammatical constraints (e.g., Case theory and thematic theory) to constrain the hypotheses considered by the processor while analyzing a string. As a result of the interaction between grammatical constraints and constraints on memory used by the processor, the CPP is subject to garden path effects. Furthermore, the number of hypotheses considered by the CPP while processing will vary depending on the interaction between lexical ambiguity and the constraints; the model therefore predicts that the relative complexity of processing will vary during the parse. I will relate garden path effects and relative complexity to some of the relevant psycholinguistic literature. Wednesday, April 20, 2:00 p.m. Eighth Floor Playroom Building NE43 MIT ------------------------------ End of NL-KR Digest *******************