AIMag07-02-007

7/29/2019 AIMag07-02-007

1/10

RESEARCH IN PROGRES S

Recent and Current Artificial IntelligenceResearch in the Department ofComputer Science, State University ofNew York at Buffalo

Shoshana L. Hardt & William J. Rapaport (Editors)Depmfmenf of Compufer Science,Sfafe University of New York at Buffalo,Buffalo, New York 74260

The Department of Computer Science at the State Uni-versity of New York (SUNY) at Buffalo, one of the old-est computer science departments in the United States togrant a Ph.D, has been active ly engaged in AI researchsince its establishment in 1965. With half of our full-timefaculty members (Shoshana L. Hardt, Wil liam J. Rapa-port, Stuart C. Shapiro, Sargur N. Srihari, and DeborahWalte rs), more than 40 graduate students, and several as-sociated faculty members invo lved in research programs inthe various branches of AI, the department is engaged in avariety of applied and theoretical AI research efforts. Thedifferent research groups use a variety of AI methodolo-gies and tools in a variety of different fields, including nat-ural language understanding and computational lingu is-tics, expert systems, computer vision and pattern recog-nition, knowledge representation, reasoning, and cognitivescience . The departmental laboratories include VAX es,LISP machines, a SPERRY 7000/40, SUN workstations, aGrinnell GMR 274 graphics system, an Imaging Technolo-gies IP512 image-processing system, and an Adage 3010graphics system.

Reasoning About the Temporal Structureof Narrative TextsA major part of what it means to understand a story isto know how the different events and states described inthe text relate to one another temporally. Although muchof this information is provided simp ly by the order of pre-sentation within the text, a number of other factors playsignificant roles in the production of an adequate tempo-ral model. Among these factors are changes in tense, theuse of the perfect tense, the distinction between progres-sive and nonprogressive, time adverbials, world knowledge,and the inherent aspectual properties of different class es ofpredicates. We also believe it is necessary to use some rep-resentation of the present moment within the story, thatis; a narrative now point. This narrative now point ismoved forward in time as the story progresses in time, andit functions as the temporal deict ic center to which all theevents and situations that occur in the story are related.The major focus of our research is on how the various fac-tors mentioned here interact with and affect this narrative

Natural-Language UnderstandingAnd Computational Lingu isticsOur natural-language understanding projects are focusedaround representing and reasoning about spatial and tem-poral information and belief and knowledge reports, as wellas a knowledge-engineering approach to natural language.

AbstractThis art ic le contains reports from the var ious research groupsin the SUNY Buffalo Department of Computer Science, VisionGroup, and Graduate Group in Cognit ive Science It is or-ganized b y the di fferent research topics. Howeve r, i t shouldbe noted that the individual projects might also be organizedaround the methodologies and tools used in the research, and,of course, m any of the projects fal l under more than one cate-gory.

THE AI MAGAZINE Summer, 1986 91

7/29/2019 AIMag07-02-007

2/10

now point. The system is being implemented in the SNePSSemantic Network Processing System, (Shapiro, 1979) andis a project of the SNePS Research Group and a part ofa research project on deix is in narrative being conductedby the Graduate Group in Cognitive Science. (For furtherinformation, see Almeida & Shapiro 1983.)Partic ipants: Michael J. Almeida and Stuart C. Shapiro.A Knowledge-Based Approachto Natural Language UnderstandingA significant feature of any natural language i s that i t canserve as its own meta-language. One can use a naturallanguage to talk about the language itself as well as togive instruction in the use and understanding of the lan-guage. Because human beings are able to use their nat-ural language to talk about that natural language itsel f;we have been investigating methods of knowledge repre-sentation and natural language understanding that wouldenable an AI system to do likew ise. We have implementeda language-understanding system in the role of an educablecognitive agent whose task domain includes language un-derstanding and whose discourse domain includes knowl-edge of its own language.This system has just one (initially primitive) language,which becomes increasingly more sophisticated as the sys-tem accepts instruction expressed in its evolving language.Such a system must start with some language facili ty, andwe have strived to make this initial kernel language assmal l and as independent of theory as possible . With anunbiased kernel language, teacher-users should ideally beable to bootstrap into the language of their choice. Thesystem is implemented in the SNePS Semantic NetworkProcessing System (Shapiro, 1979) and is a project of theSNePS Research Group. (The final report of this project isNeal, 1985; for further information, see Neal, 1981; Shapiro& Neal, 1982; and Neal & Shap iro 1984, 1985, and forth-coming.)Partic ipants: Jeannette G. Neal1 and Stuart C. Shapiro.Logical Foundations for Belie f RepresentationOur research consists of the design and implementation ofa logically and psycholog ically adequate computer systemcapable of representing and reasoning about the cognitiveattitudes of intelligent agents. The agents include users,other AI systems, and the system itsel f; the cognitive at-titudes include beliefs, knowledge, goals, and desires. Thesystem wil l be able to represent nested attitudes; it wil l besensit ive to the intensionality and indexicalit y of attitudes,in particular, to the phenomenon of quasi -indexicality, afeature at the core of self-referen tial beliefs; and it wil l beable to expand and refine its beliefs by interacting withusers in ordinary conversational situations. The system islCurrent affiliation: Calspan Corporation, Buffalo, NY.

being implemented in the SNePS Semantic Network Pro-cessing System (Shapiro, 1979) using an augmented tran-sition network grammar for parsing and generation. Thisresearch is a project of the SNePS Research Group and ialso part of a research project on deix is in narrative beingconducted by the universitys Graduate Group in Cogni-tive Science.The research is sponsored by a grant from the NationalScience Foundation and a SUN Y Research Foundation Research Development Award. (For further information, seRapaport & Shapiro, 1984; Rapaport; 1984.)Partic ipants: William J. Rapaport and Janyce M. Wiebe.Dynamic Computation of Spatia l ReferenceFrames in Understanding Narrative TextThe actual meaning of a spatial description of an object(figure) relative to a background (ground) is deter-mined only when the reference is interpreted in a spec ificorientational system around the ground. In the expres-sion, an unaware boy playing in front of a truck rollingbackward down the hill, the real spatial relationship between the boy and the truck hinges on the front-back ref-erence frame set around the truck. In natural languages,spatial references are usually made without explicitly in-dicating how the reference frame is to be established, andthe burden of its determination is left to the hearers othe readers infe rences. Unlike the case of spoken languagewhere the hearer often need look only at the referent todisambiguate a spatial description , in the case of reading atext the data for computing reference frames is containedsole ly in the text and the readers background knowledge.We are developing heuris tics that can be used by computer story understander for dynam ically comput-ing the reference frames for sl :ial descript ions. The systems understanding of a spati; description wil l be demon-strated by drawing a picture of the situation. For this, wewil l use the graphic knowledge re,>rescnta tion techniquesbeing developed in the project on representation of visua lknowledge, described later in this article.This research is a project of the SNePS ResearchGroup and is also part of a research project on deix is innarrative being conducted by the Graduate Group in Cog-nitive Science.Partic ipants:Shapiro. Albert Hanyong Yuhan and Stuart C

Expert SystemsResearch in expert systems includes projects in music, image understanding, message processing , maintenance, anddiagnosis.An Expert System for Harmonizationof Chorales in the Style of J. S. BachWe are designing an expert system called CHORAL foharmonizing four-voice chorales in the style of Johann

92 THE AI MAGAZINE Summer, 1986

7/29/2019 AIMag07-02-007

3/10

Sebastian Bach. Using the Bach Chorales as the highstandard, as well as traditional music treatises and per-sonal intuitions, we have found approximately 270 rules,expressed in a form of first-order predicate calculus, fordescribing the musical knowledge to harmonize a givenchorale melody. Our rules represent knowledge from mul-tiple viewpoints of the chorale, such as the chord skele -ton, individual melodic lines, and Schenkerian voice lead-ing within the descant and bass. The program generateschorales from left to right, using a generate-and-test tech-nique with intelligent backtracking, until a solution sat is-fying all the constrain ts is found. A substantial numberof heuristics are used for biasing the search toward musi-cal solutions. Results of acceptable competence have beenobtained. To provide the execution efficiency that appearsto be mandatory for tonal music generation, we have de-signed BSL, a new logic programming language that com-piles into C, to implement the CHORAL system, and wehave investigated the foundations of BSL.This research is supported by a grant from the Na-

tional Science Foundation. (For further information, seeEbcioglu, 1985.)Participants: Kemal Ebcioglu and John Myhill.Rule-Based Expert Systemsfor Image UnderstandingResearch is current ly being done to develop rule-based ex-pert systems for the visua l domain. The systems beinginvestigated consis t of the following components: (1) aknowledge base that is made of multiple levels of produc-tion rules which embody the knowledge of the differentcharacteris tics of document images and (2) an inferenceengine that uses this knowledge base to perform detailedanalysis of the image data in order to arrive at consis-tent interpretations of the identities of the various logicalblocksi in the image.

The control system in the inference engine applies thedifferent leve ls of rules progressive ly on the image data inorder to identify the blocks . The lowest level of rules areknowledge ru les, which examine the intrinsic properties ofthe blocks. Control rules guide the search and act as thefocus-of-attention mechanism. Strategy rules determinewhether at any point in time a consistent interpretationof the image has been achieved. A system applicable topostal images is being implemented in Prolog. Anothersystem, applicable to computed tomography images: is inLISP and C.

This research is supported by the U. S. Postal Serviceand by a SUNY Research Foundation Research Develop-ment Award (For further information, see Srihari et al,1985, Kumar & Srihari, 1985.)

Participants: Debashish Niyogi, Rakesh Kumar,3 andSargur N. Srihari.High-Speed Interpretationof III-Formed MessagesWe have automated the interpretation of daily reportsfrom ships for the U.S. Coast Guard. Although very heav-ily knowledge-based, the system that emerged from theproject can be viewed as a nontraditional expert system,because it employs new methods to handle control anddata flow. The process of interpretation of daily reportsfrom ships requires considerable amounts of task-specificknowledge, a large degree of flex ibili ty in processing, andan efficient error-recovery mechanism. This is becausethese reports are ill formed and can contain enormousamounts of noise. Building a computer program that canhandle this interpretation task requires designing and im-plementing powerful focusing and understanding mecha-nisms which can operate under severe real-time demandswith over 75% reliability.

From the start; this project presented us with the in-teresting double challenge of coordinating the theoreticaland the applica tive aspects of the work with the goal ofdesign completion and implementing the system in twelvemonths. In addition, we had to make sure that the finalsystem (product) is a maintainable piece of software-a low-priority task in a university research environment.At the moment, we are investigating further the generaldynamics of variable-depth processing and of knowledge-based focusing of attention in natural language parsing.This project is sponsored by the U.S. Coast Guard.(For further information, see Hardt et al., 1985; Hardt &Rosenberg, 1985, 1986.)

Participants: Jay Rosenberg and Shoshana L. Hardt.Device Modeling for a VersatileMaintenance Expert SystemWe are developing a versat ile maintenance expert system(VME S) for troubleshooting a wide variety of electronicdevices. The system diagnoses a malfunctioning devicebased on structura l and functional descriptions of the de-vice: which have been widely used by other fault diagnosisresearchers, as a solution to the difficulties of empiricalrule-based diagnosis systems in knowledge acquisition, di-agnosis capabili ty, and system generalization. The diagno-sis efficiency and effectiveness, as well as the system gen-erality , are being investigated. We find the device model:that is, the structura l and functional representation of thedevice, to be vita l to the performance of the system. Thesystem is implemented in the SNePS Semantic NetworkProcessing System (Shapiro, 1979). The structure of the

2SUNY at Buffalo Department of Mathematics.3Current affiliation: Department of Computer and Information Sci-ewe, University of Massachusetts at Amherst

THE AI MAGAZINE Summer. 1986 93

7/29/2019 AIMag07-02-007

4/10

device is currently represented as instantiation rules, re- Computer Vision and Pattern Recognitionsulting in a compact representation and better system ver-satil ity. Functions are implemented as Lisp functions thatallow device simulation during diagnosis. Graphic displaysare used to communicate with the user (see Representationof Visual Knowledge below). We are also investigating theapplicability of SNePS Belief Revision (SNeBR) for faultisolation (see SNeBR: A Belief-Revision Package).

This work is sponsored by the Rome Air Develop-ment Center (RADC) and the Air Force Office of ScientificResearch through the Northeastern Art ific ial IntelligenceConsortium (AIC). (For further information, see Shapiro,Srihari et al ., 1985, 1986.)

In these areas, our research includes projects in letter andword recognition, document image understanding, and vi-sual processing.A Cognitive Model of LetterLearning and RecognitionMost cognitive science research in letter recognition isbased on the use of feature models. In this research, adifferent approach has been tried. At acquisition time,letters are subjected to a form of object-oriented quadtreeanalys is. A ll acquired letters are integrated into the samequadtree, and object knowledge and density values of thegraphic representation are added to this structure. Atrecognition time, quadtree analysis and densities of thepresent letter are compared to the stored density quadtree.The algorithm wil l then return the name of the letter thatthe present letter is most similar to. Experiments in recog-nizing new distortions have so far been quite encouraging.(For further information, see Geller , 1985.)Partic ipant: James Geller.

Partic ipants: Stuart C. Shapiro, Sargur N. Srihari, Ming-Ruey Taie, James Geller, and Scott S. Campbell.Spatial Structure and Function in Diagnosis.We are developing a model-based expert system for neuro-logical consultation: NEUREX. A key element in buildinga model-based expert system i s the representation of spa-tial structure and the function associated with the compo-nents of that structure. Spatial structure, synonymouswith three-dimensional physical structure, captures thephysical characte ristics of the components and their in-terconnections. The representation of three-dimensionalspatial structu re can be divided into two categories: ana-logica l and propositional representations. An analogicalrepresentation is a detailed geometrical description, suchas the exact shape and position of the spatial entity rep-resented. A propositional representation represents enti-ties and spatial relations between the entities topologically.Because much of the relevant structure of the nerve sys -tem and of digital circuits is networklike, semantic networkstructures implemented in the SNePS Semantic NetworkProcessing System (Shapiro, 1979) will form analogical,as well as propositional, representations of these domains.Because most methods proposed for such systems are usu-ally based on a structural representation from one of thetwo categories, they lack generality and their capabil ityis limited. The method of representation should acceptboth analogical and propositional information concerningthe structure, support function association , provide for di-agnostic reasoning, and allow for graphics and languageinterfaces. We are investigating the features and benefitsof this dual representation.

A Computational Theory of Word RecognitionThis project centers on the development of a computa-tional theory of word recognition by adapting componentsof psycho logical theories to the machine reading problem.When explanations of human reading are compared to thestate of the art in comparable algorithms, a large dis-crepancy is apparent. For example, it has been knownsince the nineteenth century that word recognition is nota character-by-character identification process followed bydictionary lookup. However, this is the way most algo-rithms are designed. As an alternative to this strategy, aword-recognition algorithm with a mixture of holis tic pro-cessing and feature analysis is being investiga ted.The holistic cue of word outline shape that has beensuggested as an aid to word recognition in humans was thesubject of a recent series of computational experiments todetermine its usefulness to a reading algorithm. The re-sults of these experiments provide convincing evidence forthe use of such a holistic cue in a word-recognition al-gorithm. Because on the average only a smal l numberof words have the same shape, the decision space of anyfurther processing is much smaller than it would other-wise be. The reexamination of psycho logical results at-tributed to word outline shape is also suggested becauseof the predictive abili ty of the new definitions as well ascomputational results which show that much more visua linformation than just outline shape is available to humanreaders. Future work in this project includes the develop-ment of a complete word-recogn ition algorithm in whichthe feature analysis strategy i s determined from the subsetof the dictionary specified by the shape of an input word.Work is under way on the modeling of a dictionary by a

This research is supported by a SUNY Research Foun-dation Research Development Award. (For further infor-mation, see: Xiang, Sriha ri, et al., 1984, 1985; Xiang &Srihari, 1985; Xiang, Chutkow, et al ., 1986.)Part ic ipants: Zhigang Xiang, Sargur N. Srihari, StuartC. Shapiro, and Jerry G. Chutkow.44SUNY at Buffalo Department of Neurology


7/29/2019 AIMag07-02-007

5/10

graph that condenses prefixes and suffixes and the devel-opment of a hypothesize-and-test recognition strategy thatuses this data structure. This approach has analogies totheories of human word recognition based on the impor-tance of the first and the last characters in a word and theselect ive extraction of features from a word image.This research is supported by the U. S. Postal Service.(For further information, see Srihari & Hull, 1982; Hull,1985.)Participants: Jonathan J. Hull and Sargur N. Srihari.Document Image UnderstandingA document image is an optically scanned and digitizedrepresentation of, say, the title page of a journal article ,a newspaper page, the face of a stamped and addressedenvelope, and so on. Document image understanding in-volves deriving a high-level description that retains thespatial structure of the document; assigns labels to var-ious components such as text, figures, titles, addresses,and so on; and allows extraction of relationships such asreading order. We are investigating the development andcoordination of the visual, spatial, and linguistic processesnecessary for this task. They include visual processes todetermine objects from the background by edge detecting,edge grouping, and texture analyzing, spatial processes tolabel regions using knowledge of how a typica l document isstructured, and determining the font and identity of char-acters. Reading words of text i s being investigated as aprocess that involves visual, spatial, as well as linguisticknowledge.The interpretation of images of postal m ail pieces isthe domain of this investigation . Our efforts have includedthe development of various operators for visua l data pro-cessing and image segmentation. The invocation of theseroutines and the interpretation of the information they re-turn is determined by a control structure that uses a vari-ant of relaxation combined with a rule-based methodol-ogy. This approach is designed to capitalize on the spatialrelationsh ips between blocks of image data (for example,postage is to the right and above the destination addressin, say, 89% of all cases) and to propagate the influence ofclassifications in a controlled fashion.This research is supported by the U.S. Postal Service.(For further information, see Srihari et al., 1985.)Participants: Sargur N. Srihari, Jonathan J. Hull: PaulPalumbo, and Ching-Huei Wang.Select ion and Use of Image Featuresin Early Visual ProcessingOne of the important issues in vision i s determining whichfeatures of an image should be processed in the early, par-allel stages of visual processing. Another way of saying thisis that we need to determine what types of image represen-tations are most useful for preattentive processing. This

is important because the efficiency of visua l computationsdepends on the representation used; many representationswil l be complete, but few wil l be ideal ly matched to therequired computations.Another important issue is to determine how the fea-tures are used (where features might be relations). An im-portant distinct ion here is the difference between the usesof features in the early preattentive stages versus the usesin stages of higher processing . It is possible for differentstages to use different features or for the same features tohave different uses at different stages. One of the thrustsof our research i s to investigate novel uses of features atthe preattentive level. For example, the presence of a par-ticula r feature can be used to alter the processing of anassociated region of the image. Thus, it is possib le to haveearly processing that is both se lective and parallel.

These issues are addressed by searching for featuresthat capture the structure and uniformity of nature usingthe following four criteria : (I) features should be perceptu-ally valid for humans (as determined through psychophys-ica l experimentation); (2) feature sets should be geometri-cal ly complete; (3) features should have low probab ility ofaccidenta l occurrence and should allow useful inductive in-ferences to be made from simple assumptions, such as therepresentativeness of both viewpoint and position; and (4)features should be loca lly computable in paralle l

This research is supported by a grant from the Na-tional Science Foundation. (For further information, seeWalters & Weiss tein, 1982a, 198213; Walte rs, Biedermann,& Weiss tein, 1983; Walters 1984, 1985; and forthcoming.)The Vision GroupIt is becoming increasingly important for vision researchersin diverse fields to interact, and the Vision Group at SUNYBuffalo was formed to facilitate that interaction Currentmembership includes 25 faculty and 25 students from 10departments (computer science , elect rical and computerengineering, industrial engineering, geography, psychology,biophysics, physiology, biochemistry, philosophy, and me-dia studies). The group organizes a colloquium series andprovides centralized information about activ ities both oncampus and in the local area that are of interest to visionresearchers.Contact: Deborah Walte rs.

Knowledge RepresentationAlthough most of the projects discussed so far have knowl-edge representation as a major component, the projectsmentioned in this section are direc tly concerned with it.Representation of Visual KnowledgeWe are investigating methods of representing visua l knowl-edge in a knowledge representation system integrated with


7/29/2019 AIMag07-02-007

6/10

traditional conceptual and propositiona l knowledge. Vi -sual knowledge is knowledge about how to display ob-jects, attributes of objects, and relations between objectson a graphics terminal screen. At the most basic level,the visual form of an object is a Lisp function that drawsthe object on the screen when evaluated. However, visua lknowledge can also be distributed among nodes in the tra-ditional hierarchies: The knowledge of how to display aparticula r hammer can be stored at the leve l of the clas sof hammers; the knowledge of how to display a person canbe distributed among the nodes for heads, arms, hands,and so on; locations of parts are relative to their who les.Attributes of objects can be seen as functionals that mod-ify graphic-form functions. We are using visual knowledgein our development of a versat ile expert maintenance sys -tem for digital circuits (see Device Modeling for a Ver-satile Maintenance Expert System above) implemented inthe SNePS Semantic Network Processing System (Shapiro,1979). Interaction with the user is carried out through thegraphic images, including requesting and obtaining data,display ing dynamic traces of the reasoning, and showingthe final conclusions.This research is a project of the SNePS ResearchGroup, and is sponsored by RADC and the Air Force Of-fice of Scientifi c Research through the Northeastern AICand by the Southeastern Center for Electrical EngineeringEducation. (For further information, see Shapiro, Srihar i,et al., 1985, 1986.)Participants: James Geller, Stuart C. Shapiro, Sargur N.Srihari, and Ming-Ruey Taie.Knowledge Organization Schemesfor Psychia tric DiagnosisCognitive modeling of the expertise exhibited by a psy-chiatrist during a diagnostic session provides us with anopportunity to confront some of the central research prob-lems in the field of AI. In this project, we are developingan on-line, computerized assistant that can aid a clinicianperforming psych iatric diagnoses. In particular, we areinvestigating (1) the advantages of a computer assistantover a complete computer diagnostic system, (2) the pa-rameters of a domain that indicate its suitability for gain-ing support from a computer assistant , (3) a functionalspecification for a computer assistant, and (4) the con-straints that this specification places on how to organizethe knowledge base around processing structures. Cur-rently, the project centers on the implementation of thepsychiatric diagnosis system, Diagnosis and Understand-ing of Natural Experiences (DUNE), which specializes inaffective and anxiety-related disorders . DUNE is based onthe idea that successful diagnos is in a complicated domaininvolves the parallel pursuit of multiple hypotheses. Eachhypothesis is implemented as a cluster of processors, andthe system involves elaborated processor communicationand self-evaluat ion schemes.

Participant: Shoshana L. Hardt.Intensional Semantics forPropositional Semantic NetworksA semantic network is a representation in which each con-cept (including relations between concepts) is representedby a spec ific node, and nodes are linked to each other by asmall set of arcs. The SN ePS Semantic Network Process-ing System contains an inference package that supportsforward, backward, and bidirectional inference using ruleswhich are represented in the SNePS network. It also pro-vides for interactive graphics for data entry and explana-tion, image analysis, and natural language interfaces.SNePS is one of the few semantic network processingsystems that is fully intensional in the sense that nodesonly represent intensiona l entities. Prev ious attempts bysome researchers to provide a semantic interpretation forthe system have relied on the semi-in tensiona l formal-ism of possible world semantics. In this project, we areproviding a semantic interpretation using a fully inten-sional theory based on the philosophical theories of AlexiusMeinong.This research is a project of the SNePS ResearchGroup and is supported by a grant from the National Sc i-ence Foundation. (For further information, see Rapaport,1978, 1981, 1985a; Shapiro, 1979; Maida & Shapiro 1982;and Shapiro & Rapaport, 1985.)Participants: William J. Rapaport and Stuart C. Shapiro.

ReasoningOur reasoning projects include commonsense reasoning,intuitive reasoning, and belief revision .Commonsense Reasoning aboutDiffusional ProcessesNaive physics is the body of knowledge that people haveabout the surrounding physical world. This knowledge isan important part of the commonsense knowledge that en-ables people to effectively deal with the world. The mainenterprise of studying naive physics as a part of AI is todiscover or invent the information-processing schemes thatcan enable computer programs to explain, describe, andpredict changes in the physical world. Because peoplehave decades of experience interac ting with their imme-diate physical surroundings, they inevitably develop ex-tremely rich knowledge structures that capture this con-tinuous experience. In an important sense, people are ex-perts in the domain of simple physical events, utiliz ingamazing ly large amounts of knowledge.Because most people find the dynamics of diffusiona lprocesses to be counterintuit ive, investigat ing the way peo-ple fail to correctly reason about situations involving theseprocesses might serve as a window on the intuitions peo-ple have about the physical world. From the viewpoint of


7/29/2019 AIMag07-02-007

7/10

the science of physics, there are at least three general anddistinc t perspectives from which the process of diffusioncan be described. These perspectives look at the process(1) from the level of the movements of individual parti-cles, (2) from the level of isolated flows of collect ions ofparticles, and (3 ) from the level of the system as a whole.These three perspectives correspond roughly to the threetheories developed in phys ics to capture the dynamics ofprocesses at different levels of description, namely, the ki-netic theory, fluid mechanics, and thermodynamics. Whenexamining the way people reason about diffusion prob-lems presented to them, it becomes apparent that theyuse knowledge structures which mix these clear ly definedleve ls of description. This seems to support the conjecturethat mental models which provide people with the rea-soning power for dealing with everyday physics problemsdeviate in a significant fashion from scien tific perspectives.The current state of the research project involves thedevelopment of theoretical concepts related to the natureof commonsense reasoning in general and naive physics inparticular. Among the research topics involved are knowl-edge organization schemes that can facilitate the reason-ing task, a vocabula ry for the representation of geometricalshapes, the differences in quality and quantity between ex-pert and novice knowledge, and effective knowledge reor-ganization schemes. Al l of the theoretical work is centeredaround the development of the computer program HighIntuition Yie lds Advanced Learning (H IYAL) that shouldlearn and reason about diffusional processes in complicatedgeometries.This project is sponsored by a grant from the NationalScience Foundation.Participant. Shoshana L. Hardt.Architectures for Intuitive ReasonersWe are investigating possible architectures for systemsthat reason about causal models of the world in generaland the physical world in particular. Our reasoners viewthe world as consisting of behaviors that can be generatedby typical devices. A device is an abstract model thatgenerates the same behavior in many situations, and thisobserved behavior can be decomposed into interacting de-vice behaviors. A database of typica l behaviors can bebuilt by modeling the devices and generating their behav-ior by simulation. Substant ial work has been done by otherresearchers on device simulation in the past five years.We view the knowledge in our reasoning systems asorganized around two computational metaphors: the com-piled knowledge that constitutes the database of behaviorsand the general princip les part. When the reasoner en-counters an observed behavior, it wil l attempt to decom-pose it to match parts with the database entries. If thesearch succeeds, the reasoner can then proceed to analyzeand explain the physical situation and make predictions.

However, the more interesting case is when the search foran atypica l behavior fails. This situation can arise if thedecomposition could have been done in more than one way.In this case, the reasoner will attempt alternative decom-positions guided by the general princip les part of theknowledge base. If all these attempts fail, then the rea-soner wil l start making educated guesses about the situ-ation at hand. This mechanism wil l also be guided by thegeneral principles part of the reasoners knowledge base.We are currently concerned with identifying and modelingtypica l behaviors and identifying and representing gen-eral principles.Partic ipants: Shoshana L. Hardt and Kulb ir S. Arora.SNeBR: A Belief-Revision Package.The SNePS Semantic Network Processing System (Shapiro,1979) has been extended to handle belief revision, an areaof AI research concerned with the study of the representa-tion of beliefs and belief dependence, the development ofmethods for selecting the subset of beliefs responsible forcontradictions , and the development of techniques to re-move some subset of beliefs from the original set of beliefs.(For an overview of the field, see Martins, forthcoming.)

SNeBR is an implementation in SNePS of an abstractbelief-revision system called the Multiple Belief Reasoner(MBR), which, in turn, is based on a relevance logic calledSWM (Shapiro & Wand, 1976; Martins, 1983; Martins &Shapiro, 1983, 1984). SWM deals with supported well-formed formulas (wffs) of the form: Alt, o, r , where A is awff representing a proposition, t is an origin tag indica tinghow A was obtained (for example, as a hypothesis or as aderived proposition), o is an origin set containing all andonly the hypotheses used to derive A, and r is a restrictionset containing information about contradictions known toinvolve the hypotheses in o. The origin tag, origin se t,and restriction set of a wff are computed when the wffis derived, and its restriction set can be updated whencontradictions are discovered.

In MBR a context (any set of hypotheses) determines abelief space, which i s the set of all the hypotheses definingthe context together with all propositions derived exclu -sive ly from them. The origin sets of the propositions inthe belief space defined by a given context are containedin that context. The only propositions that are retrievableat a given time are the ones belonging to the current be-lief space (whose context is the set of all hypotheses underconsideration at that time).

A contradiction can be detected either because an as-sertion is derived that is the negation of an assertion al-ready in the network or because believed assertions in-validate a rule being used (where an assertion invalidatesa rule). In the former case the contradiction is notedwhen the new, contradictory assertion is about to be builtinto the network, because a uniqueness principle (Maida


7/29/2019 AIMag07-02-007

8/10

& Shapiro, 1982) guarantees that the contradictory as-sertions will share network structure. In the latter case(where an assertion invalida tes a rule), the contradictionis noted in the course of applying the rule. In this case,it might be that the contradicto ry assertions are in differ-ent belief spaces (only the new one being in the currentbelief space); if so: the restriction sets are updated to re-flect the contradicto ry sets of hypotheses, and nothing elsehappens. If the contradictory assertions are both in thecurrent belief space (which wil l be the case when one ofthem is a rule being used), then, besides updating the re-striction sets! the user wil l be asked to delete at least oneof the hypotheses underlying the contradiction from thecurrent context. Management of origin sets according toSWM guarantees that, as long as the current context wasoriginally not known to be contradicto ry removal of anyone of the hypotheses in the union of the o rigin sets of thecontradictory assertions from the current context will re-store the current context to the state of not being knownas inconsistent.This research is an ongoing project of the SN ePS Re-search Group. It has been supported by the National Sc i-ence Foundation and by RADC and the Ai r Force Officeof Scientifi c Research through the Northeastern AIC.Participants: Joao P. Martins5 and Stuart C. Shapiro.

Cognitive ScienceMany of the projec ts discussed above have a cognitivescience aspect to them. In this section, we discuss twoprojects that are explicitly within the multi-disciplinaryfield of cognitive science.Machine Understanding and Data Abstrac tionThis project consists of an investigation of the applica-bili ty of the notion of abstract data types and their im-plementations to resolve various philosophical controver-sies surrounding John Searles Chinese Room thoughtexperiment, Daniel Dennetts notion of the intentionalstance, and the problem of the existence and nature ofqualia. In particular, mental states and processes (forexample, understanding a language) are properly under-stood as abstractions (on a par with abstract data types)that both humans and computers implement. (For furtherinformation, see Rapaport, 1986a, 198610.)Participant: William J. Rapaport.Graduate Group in Cognitive ScienceCognitive science is an interdisciplinary effort intended toinvestigate the nature of the human mind. This effortrequires the theoretical approaches offered by computerjCurrent affiliation : Departamento de Engenharia Mecanica, Insti-tuto Superio r Tkcnico, Lisbon

science, linguistics, mathematics, philosophy, psychology,and a host of other fields related by a mutual interest inintelligent behavior.The Graduate Group in Cognitive Science was formedto facilitate cognitive science research at SUNY at Buf-falo. Its activ ities have focused on language-related issuesand knowledge representation. These two areas are impor-tant to the development of cognitive science and are wellrepresented at UB by the research interests of faculty andgraduate students in the group.Since its formal recognition in April 1981, the gradu-ate group has grown quickly . Currently, its membership ofover 150 faculty and graduate students is drawn from thedepartments of computer science ; psychology; lingu istics ;communicative disorders and sciences; philosophy; instruc-tion; communication; counseling and educational psycho l-ogy; educational organization, administration, and policystudies; the intensive English language institute; geogra-phy; and industrial engineering as well as other area col-leges and universities. The group sponsors lectures and in-formal discussions with visiting scholars; discussion groupsfocused on group members current research; an interdis-ciplinary, team-taught graduate course, Introduction toCognitive Science; and a cognitive science library.Deictic Centers in Narrative: An InterdisciplinaryCognitive Science Project. A research subgroup ofthe Graduate Group in Cognitive Science is developing amodel of a cognitive agents comprehension of narrativetext. (The term cognitive agent includes both normal andlanguage-impaired humans as well as nonhuman or arti-ficial agents.) Our model wil l be tested on a computersystem that wil l represent the agents beliefs about theobjects, relations, and events in narrative as a function ofthe form and content of the successive sentences encoun-tered. In particular, we wil l concentrate on the role ofspatial, temporal, and focal-character information for thecognitive agents comprehension.

We propose to test the hypothesis that the construc -tion and modification of a deictic center is of crucial im-portance for much comprehension of narrative. We seethe deictic center as the locus in conceptual space-time ofthe objects and events depicted or described by the sen-tences current ly being perceived. At any point in the nar-rative, the cognitive agents attention is focused on partic-ular characters (and other objects) standing in particula rspatial and temporal relations to each other. Moreover,the agent looks at the narrative from the perspective ofa particula r character, spatial location, or temporal loca-tion. Thus, the deictic center consists of a WHERE-point,a WHEN-point, and a WHO-point. In addition, refer-ences to characters beliefs, persona lities, and so on, arealso constrained by the deict ic center.We plan to develop a computer system that wil l reada narrative and answer questions about the deictic info r-


7/29/2019 AIMag07-02-007

9/10

mation in the text. To achieve this goal, we intend to carryout a group of projects that wil l allow us to discover thelinguistic devices in narrative texts, test their psycholog-ica l reality for normal and abnormal comprehenders, andanalyze psychological mechanisms which underlie them.Once we have the results of the individual projects, wewil l integrate them and work to build a unified theory andrepresentational system that incorporates the significantfindings. Finally, we will test the system for coherenceand accuracy in modeling a human reader and modify itas necessary.Partic ipants: In the department of Computer Science,William J. Rapaport and Stuart C. Shapiro, principalinvestigators, Michael Almeida, Janyce M. Wiebe, andAlber t Hanyong Yuhan; in the department of Psychol -ogy, Gail Bruder, Erwin Segal, principal investigators, andJoyce Daniels; in the Department of Communicative Dis -orders and Sciences, Judith Duchan, principal investigator,and Lynne Hewitt; in the department of Linguistics, DavidA. Zubin, principal investigator, Naicong Li, and SoteriaSvorou.

Northeastern Art ific ial Intelligence ConsortiumThe Northeastern AIC is an associat ion of AI researchersat nine universities-Syracuse University, SUNY at Buf-falo, the Unive rsity of Rochester, the Unive rsity of Mas-sachusetts at Amherst, Clarkson College of Technol-ogy, Colgate Un iversity, Rensselaer Polytechnic Institute,Rochester Institute of Technology, and the Ai r Force In-stitute of Technology. Initial sponsorship of AIC is beingprovided by RADC and the Air Force Office of ScientificResearch.

The purpose of AIC is to enhance research and educa-tion in AI at the participating universities and at RADC.In particular , the participants are conducting joint and co-operative research, seminars, workshops, and conferences;providing cooperative educational programs in which stu-dents and RADC staff members can take courses at vari -ous participating universities; and expanding their AI re-search and instruction by recruiting additional AI-orientedfaculty and students, expanding AI course offerings, andenhancing their AI computing facilities.Partic ipants: Stuart C. Shapiro and Sargur N. Srihari,principal investigators.

ReferencesAlmeid a, Michael J , & Shapiro, Stuart C. (1983) Reasoning about

the tempo ral structure of narrative texts Cognitive Science Soci-et y 5Ebcioglu, Kema l (1985) An expert system for schenkerian synthe-

sis of chorales in the stule of J 5 Bach. 1984 Cornouter MusicConference: 135-142 I

Geller, James (1985) The teachab le Zetter recognizer IJCAI 9: 249-51

Hardt, Shoshana L , & Rosenberg, Jay (1985) The ERIK project:Final report and manuals Tech Rep. 85-08, Department of Com-puter Science, State University of New York at Buffalo

Hardt, Shoshana L , & Rosenberg, Jay (1986) Developing an expertship message interpreter: Theoretic al and practical conclusionsOptical Engineering (forthcoming)

Hardt, S. L , Rosenberg, J., Haefner, M , & Arora, K (1985) Thethree ERIK-A MVER progress reports Tech Rep. 85-07, De-partment of Computer Science, State University of New York atBuffalo

Hull , J J (1985) Word shape analysis in a knowledge-b ased systemfor reading text IEEE Artificial Intelligence Applications 2

Kumar, R., & Srihari, S. N (1985) A n expert system for the interpre-tation of cranial CT scan images. Expert Systems in GovernmentSymposium: 548-57.

Maida, Anthony S. & Shapiro, Stuart C (1982) Intensional conceptsin propositional semantic networks Cognitive Science 6: 291-330.Reprinted in R J Brachman & H J. Levesque (eds ), Readingsin knowledge representation (1985) Los Altos, Calif: MorganKaufmann.

Martins, Joie (1983) Reasoning in multiple belief spaces Tech Rep.203, Department of Computer Science, State University of NewYork at Buffalo

Martins, Joao (forthcoming) Belief revision In S C Shapiro (Ed ),Encyclopedia of artificial intelzigence. New York: John Wiley

Martins, Jo%o, & Shapiro, Stuart C. (1983) Reasoning in mult ip lebelief spaces IJCAI 8: 370-73

Martins, Jo%o, & Shapiro, Stuart C (1984) A model for belief revi-sion AAA I Non-Monotonic Reasoning Workshop: 241-94

Neal, Jeannette G (1981) A knowledge engineering approach to nat-ural language understanding Tech Rep 179, Department ofComputer Science, State University of New York at Buffalo

Neal, Jeannette G (1985) A knowledge based approach to naturallanguage understanding Tech Rep 85-06, Department of Com-puter Science, State University of New York at Buffalo

Neal, Jeannette G , & Shapiro, Stuart C. (1984) Know ledge BasedParsing Tech. Rep 213, Department of Computer Science, StateUniversity of New York at Buffalo.

Neal, Jeannette G , & Shapiro, Stuart C. (1985) Parsing as a form ofinference in a multiprocessing environment Intelligent Systemsand Machines Rochester, Mich : Oakland University

Neal, Jeannette G , & Shapiro, Stuart C (forthcoming) Knowledgebased parsing In L Bolt (Ed ), Natural language parsing sys-tems. Berlin: Springer-Verlag

Rapaport, William J (1978) Meinongian theories and a Russellianparadox No& 12: 153-80.

Rapaport, William J. (1981) How to make the world fit our language:An essay in Meinongian semantics Grazer Philosophische Stu-dien 14: l-21

Rapaport, William J (1984) Belief representation and quasi-indicators Tech Rep 215, Department of Computer Science,State University of New York at Buffalo.

Rapaport, William J (1985a) M eznong ian semantics for proposi-tional semantic networks Twenty-third Annual Meeting of theAssociation for Computational Linguistics: 43-48

Rapaport, William J (1985b) M ac me understanding and data ab-straction in Searles Chinese Room Cognitive Science Society 7:341-45

Rapaport, William J. (1986a) Searles experiments with thoughtPhilosophy of Science 53Rapaport, William J (1986b) Philosophy, artificial intelligence, andthe Chinese-Room argument Abacus 3(Summer)

Rapaport, William J , & Shapiro, Stuart C (1984) Quasi-indexicalreference in proposi tional semantic networks Tenth Interna-tional Conference on Computational Linguistics (COLING 84):65-70

Shapiro, Stuart C. (1979) The SNeP S semantic network process-ing system In N V. Findle r (Ed ), Associative networks: Therepresentation and use of knowled ge by computers. New York:Academic Press: 179-203


7/29/2019 AIMag07-02-007

10/10

Shapiro, Stuart C., & Neal, Jeannette G (1982) A knowledge engi-neering approach to natural language understanding. Computa-tional Linguistics 20: 136-144.

Shapiro, Stuart C , & Rapaport, William J (1985) S NePS con-sidered as a fully intensional propositional semantic networkTech. Rep 85-15, Department of Computer Science, State Uni-versity of New York at Buffalo Forthcoming in G McCalla &N Cercone (Eds ), Knowledge representation. Berlin: Springer-Verlag

Shapiro, Stuart C., Srihari, Sargur N., Taie, Ming-Ruey, & Geller,James (1985) Development of an intelligent maintenance assis-tant SIGART Newsletter, 92 (April 1985): 48-49

Shapiro, Stuart C., Srihari, Sargur N , Taie, Ming-Ruey, & Geller,James (1986) VME S: A network-based versatile maintenance ex-pert system. Applications of Artificial Intelligence to EngineeringProblems 1.

Shapiro, Stuart C , & Wand, Mitchell (1976) The relevance of rele-vance Tech Rep 46, Department of Computer Science, IndianaUniversity

Srihari, Sargur N., & Hull, Jonathan J (1982) Knowledge integrationin text recognition AAAI-82: 148-51

Srihari, Sargur N , Hull, Jonathan J , Palumbo , Paul W , Niyogi,Debashish, & Wang, Ching-Huei (1985) Address recognition tech-niques in mail sorting: Research dire&ions. Tech kep 85-09;Denartment of Cornouter Science, State University of New Yorkat Buffalo.

Walters, D. (1984) Local connections in line drawin g perception: Acomputational model Investigative Ophthalmology and VisualScience 25: 200.

Walters, D (1985) The use of natural constraints in ima ge segmenta-tion International Society of Optical Engineering, Applicationsof AI II. 548: 27-34.

Walters, D (forthcoming) A computer vision module b ased on psy-chophysical experiments. In E. C. Schwab & H C Nusbaum(Eds ), Theoretica l issues in the perception of speech an d visualform New York: Academic Press.

Walters, D , Biederman, I , & Weisstein, N (1983) The combinationof spatial frequency and orienta tion is not effortlessly perceivedInvestigative Ophthalmology and Visual Science 24: 124

Walters, D , & Weisstein, N (1982a) Perceived brightness is influ-enced by structure in line drawings Investigative Ophtha lmolog yand Visual Science 22: 124

Walters, D , & Weisstein, N (1982b) P erceived brightness is a func-tion of line length and perceived connectivity Bulletin of thePsychonomic Society (Sept. 1982): 130

Xiang, Zhigang, Chutkow, Jerry G , Shapiro, Stuart C , & Sri-hari, Sargur N (1985) Computerized neurological diagnosis: Aparadigm of modeling and reasoning Health Care Instrumenta-tion (Dee 1985)

Xiang, Zhigang, & Srihari, Sargur N. (1985) Spatial structure andfunction representation in diagnostic expert systems. Expert Sys-tems and Their Applications 5: 191-206

Xiang, Zhigang, Srihari, Sargur N : Shapiro, Stuart C , & Chutkow,Jerry G (1984) Anal ogica l and proposi tional representations ofstructure in neurological diagnosis Artificial Intelligence Appli-cations 1: 127-32

Xiang, Zhigang, Srihari, Sargur N , Shapiro, Stuart C , & Chutkow,Jerry G. (1985) A mod eling scheme for diagnosis Symposiumon Expert Systems in Government: 538-47

AAA I Membership Benefits:l Subscr iption to the AI Magazinel AAA I Membership Directoryl Reduced subscr iption rate to the AI Journall Reduced registration fee at IJCA I Conferencel Early announcement of AAAI-sponsored activi t iesl Affil iation with the principal AI asso ciation

AAAI-86 CONFERENCE RECXTRATIONSA final rcmintlcr that late rcgistratiorl for the AAA I-86 co~~fcrc~~cc11-15 August 19UO) must be rcccived in thAAAl offices by 11 July 1986. (On-site registration will Iavailaldc at. a highor price.) Late l?cdinical lrogr~m Registrntion fees arc: $lt)c).OO for rncr~ibcrs ($225 for wnmm-hers) or $90.00 for st.udcnt nmnbcrs ($125 for nonnwn~ber

studcnt,s). Late Tutorial Registration fees arc $225 petutorial for mcrnbcrs ($260 for nomwmbcrs) or $110 petutorjal for xtwtcnt nmnbers ($125 for noriincmber slu-urchasc orders, checks drnwn OUS banks, international money orders, travclcrs checksor lmnk transfers (in US funds) will ba acccptcd.For bank tmnsfcls, plcnsc ndtl an atlclit ional $18.00 tcover bsnk chnrgcs.

Tlwstlay, JO July: f1cndlinc for hotel rcscrvntions awptcd by the! Ihilatlelpf:ia AAAI-86 JIousing J3rwcau.E+itiq~~ 11 July: Deadline for Lete Rcgistrn tionslcchilic*a l ;wtl Tutorial 13wgran35.llastlay, 15 July: Confcrancc? Rcfmd DcacJliw (writ-

ten rcqwsts OJlty 710 lelcp holle culls.)Stintlay, 10 August: Conlmenccnlcu t, of on-site rcgistmtion, lt~il~~ ddldiia Civic Ccntcr.lWmicnl Proiyiun I.cTW fi ts: 13cncfits aSS0ciih.d witfitile twlunici~l program rc.@trat~iOJJ fee arc: ~dmi.ssioii tt.lw scicnw and engineering tmcfts awl ttic nia.in plenarysession ; 3 copy of tllc /l/111 1-86 Confe7wrce Proceedings;

a confcrcncc rcgist,mtiOn pncltct. aJld ~1 copy of the confwcwc issue of the Al Mapzinc; admission to the GxhibiIrogm~n; and ndiiiissioli to all receptions, except. t,tw galrcccption on wcdiwsday night, 13 August 1986.Iktosiul I~ogsam lhnejits: I3cucfits associated withthe tutorial rcgistra t~ion fee (each tutorid is a sclmmtcfw) inch&: atte~ldnncc id the tutorial(s); ii copy of tlltutorial syllabus; iL confcrcncc rcgist.ration packet; admission to tlic c!xJlibit program; iUId admission to dl co1lfemice: receptions, except the gida reception On wcdnesday13 August 1986.ll@is~~tion Inquiries: Pioase sciul 311 inquiries tAAAI-80, Amcri call Associatiotl for Art.ifi& l Intdligcncc:445 I~urgcss Drive, Menlo Rtrk, Cdifornia 94025. (415322-1118 or (415) 328-3123.--- --.-_--

100 THE AI MAGAZINE Summer , 1986

AIMag07-02-007

Documents

Transcript of AIMag07-02-007