RtLfl

0

N

DEPARTMENTofPSYCHOLOGY

Carnegie-Mellon University

COMPREHENSION PROCESSES

IN MECHANICAL REASONING

FINAL REPORT

PATRICIA A. CARPENTER

MARCEL ADAM JUST

CARNE•GIE MELLON UNIVERSITY

MAY 1JM

This research was sponsored by Cognitive Science Progranm 7- CEof the Office of Navat Research. under Contract No. N00014-MK-0584. fLECTEConttract Authority Identification No. N11702,014. Approved for public %A JUL 2 5 1989 I'

release: distribution unlimited. Reproduction in whole or in part ispemitted for any purpose of the United States GoverntneaL• iU

E .,

SECURITY CLASSIFICATION OF THIS PAGE

REPORT DOCUMENTATION PAGEIa. REPORT SECURITY CLASSIFICATION lb.-RESTRICTIVE MARKINGS

Unclassified2a. SECURITY CLASSIFICATION AUTHORITY 3. DISTRIBUTION/ AVAILABILITY OF REPORT

2b. OECLASSIFICATION I DOWNGRADING SCHEDULE Approved for public releaseDistribution unlimited

4. PERFORMING ORGANIZATION REPORT NUMBER(S) S. MONITORING ORGANIZATION REPORT NUMBER(S)

ONR89-1%6a. NAME OF PERFORMING ORGANIZATION 6b. OFFICE SYMBOL 7a. NAME OF MONITORING ORGANIZATION

Department of Psychology (If applicable) Cognitive Science '.rogramCarnegie Mellon University Office of Naval Research (Code 1142 PT)

G6 A001ESS (City, State. and ZIPCode) 7b. ADDRESS (City, State. and ZIP Code)

800 N. Quincy StreetPittsburgh, PA 15213 Arlington, VA 22217-5000

Sa. NAME OF FUNDING /SPONSORING Sb. OFFICE SYMBOL 9. PROCUREMENT INSTRUMENT IDENTIFICATION NUMBERORGANIZATION (If applicable)

N00014-85-K-O584

8c. ADDRESS (City, State, and ZIPCode) 10. SOURCE OF FUNDING NUMBERSPROGRAM PROJECT TASK WORK UNITELEMENT NO. NO. NO. ACCESSION NO61153N RR04206 RR04206-OC NR702-014

11. TITLE (include S@cul•ly Clasiitceaton)

ONR FINAL REPORT: Comprehension Processes in Mechanical Reasoning

f2. PERSONAL AUTHOR(S)Carpenter, Patricia A. and Just, Marcel Adam

13a. TYPE OF REPORT I13b. Tl~teE COVgREO 0.DT FRDA Y..fnhOy s PAGE COUNTFINAL IFROM 6/1./85 rO S/3/85 89/05/31 ' 1,

16, SUPPLEMENTARY NOTATION

7.COSATI COOES I&.SUSJECT TEAMS (Continu, on revqeo d ntcomary and identify by block numflbvr)PIELD IGROUP SUO.GROUP

___________ ................. chanical ability, individual differences,comprehension of texts and diagrans. ('. *'w

ikAOST "CT (Continue on teovtre it neceua.ri and identify by block numbet)

Several lines of research investigated how people reason about mechanical devices.One avenue explored the use of diagrams in conjunction with texts to understand a particularmachine. Another project investigated the psychological processes that distinguish peoplevho score high or low in a psychometric test of mechanical ability, A third project examinedthe visual scanning and decision processes that are used to evaluate a kinematic display ofa machine in motion. A fourth project was a simulation todel of how a person might generatethe kinematic imagery to represent a machine in motion. A fifth project examined thecognitive processes in a visually-based test of pure reasoning (Raven Progressive Matrices).A sixth project examined the use of kinematic computer displays in understanding a complexdevice. These projects together provide an overview of the psythological processes usedin mechanical comprehension, as well as indicating why some people are better at fechanicalcomprehension than others. ,-,

20 $ISTAIGUTIO/NIAVAILASIUTY O0 ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION8IUNCLASSIFEI•.NLIMITrD• 0 SAME AS ROY. OTic USE tI Unclassified

2a. NAME OF RESPONSIBLE IIVIDUAL i 22b. TELEPHONE (Include Area Code) 22C. OF•lCE SYMOLDr. Susan Chinman 1_ (202) 696-.4318 ! ONR 1142CS

CD FORM 1473. eB. D m Redition may bQ 'd until 41hauited. srCUI'RY CLASSIIC,..'ION 1 THIS -PAGEAll Other aditt"Ii sir obtaitle.

2

INTRODUCTION

This is the final report for ONR contract N00014-85K-0584 between the Office ofNaval Research WPersonnel and Training program) and Carnegie-Mellon University. PatriciaA. Carpenter and Marcel Adam Just were the principle investigators. This contract coverswork from May 31. 1985 through June 1. 1988.

The purpose of the research is to build psychological models of the technical thinkingthat constitutes understanding mechanical systems. These processes occur when a personreads about a mechanical device to understand its operation in preparation for operating,assembling, or repairing it. The fundamental cognitive processes involve comprehendinglanguage and graphic information to construct a representation of the mechanical andphysical properties of the device. One focus of the research are the processes incomprehending texts that describe mechanical devices and the diagrams that oftenaccompany them.

A second goal of this research is to analyze the differences between people who aregood at spatial and mechanical reasoning from those who are not. Some of the researchexamines the performance of individuals of varying levels of ability as they try tounderstand mechanical devices of varying complexity. Differences among individuals in theirability to reason and retain spatial information is of obvious practical and scientificsignificance, An important facet of completeness is the ability to account not only fortypical processes. but also to provide a systematic account of the variation amongindividuals,

The research approach is to develop fine.grained analyses of the reasoning and visual-perceptual processes in spatial problem solving. The project utilizes data-intensivemethodologies, such as eye fixations and verbal protocols, that us to monitor the cognitiveprocesses as they occur. Thus, these investigations seek to analyze the nicro-structure ofthe coniprehension of technical information.

The following sections briefly summarize the research associated with the project andprovide references to more complete published descriptions of the work.

L Indiuidual differences in mechanical and spatial knowledge

A cognitive analysis of a test of mechanical ability provides a fine-grainedcharacterization of the individual differences that traditional psychometric tests characterizeasi a unitary entity. The cognitivo analysis specifies the differences in knowledge amongindividuals and the types of strategies used by different Individuals. These analyses permitus to account for the nature of the errors that Individuals make when solving mechanicalproblems. Based on this analysis, test problems can be constructed that will elicit specifictypes of errors from individuals with different types of knowledge. We have used thisapproach to analyze the types of knowledge underlying mechanical skill and the processes inspatial skill.

To study differences in mechanical ability, we began our research lHagarty. Just. &Morrison. 19881 using problems of the type found in a widely used psychometric test ofmechanical ability, the Bennett Test of Mechanical Comprehension Test MBennett, 1909). Atypical item is shown in Pigure 1. In this problem. the subject must decide which of two

3

pulleys will require more force to lift a weight. (The test instructions state that the pulleysare weightless and frictionless). To solve the problem based on a correct understanding, asubject must know how the forces balance in the two pulley systems. If the system is inequilibrium, the force is equal throughout the rope and the sum of the upward forces atany point in the system is equal to the sum of the downward forces. If the person usingpulley system B exerts a unit force on the pull rope. there will be a force equal to twounits acting on the movable pulley. We will refer to the amount of force required to lift aweight with a pulley system as the effort. The ratio of the weight to be lifted to theeffort is the mechanical advantage of the system. In this case, pulley system B has lessmechanical advantage and requires more force than pulley system A.

Figure 1 - pulley system

People who understand mechanical devices can infer the principles of operation of anunfamiliar device from their knowledge of the device's components and their mechanicalinteractions. A general characterization of how individuals solve the test items is asfollows. Individuals decide which attributes of pulley systems are relevant to reducing theeffort required to lift a weight, They compare the depicted pulley systems by applyingrules that relate these attributes to the effort that must be exerted. When several differentrules are applicable in a given situation. preferences among these rules determine which ruleis used to generate an answer to the problem. Based on subjects' retrospective protocolsand response patterns, it was possible to identify rules that accounted for the performanceof subjects of different levels of mechanical ability Iflegarty. Just. & Morrison. 1988). Therules are explicitly stated in a simulation model which demonstrates the sufficiency of therules by producing the kinds of responses observed in the subjects.

Three abilities are proposed as the sources of Individual differences in performance: II)ability to -orrectly identify which attributes of a system are relevant to its mechanicalfunction. 121 ability to use rules consistently. and 13) ability to quantitatively combineinformation about two or more relevant attributes.

First. individuals who score high on tests of mechanical ability typically know whichparts of a mechanical system are relevant to Its mechanical functioning and which areirrelevant. Low-scoring subjects are often misled into thinking that an irrelevant propertyof a mechanical system. such as the height of a pulley system is relevant to its mechanicaladvantage. For example. in problems in which the depicted pulley systems differed onheight. high-scorlng subjects had a higher proportion of correct responses 1.90) than low-scoring subjects 1.44L,

Second. high-scoring subjects were more consistent in their rule use. Thus. high-scoring subjects typically used the same rule to answer problems of a given type Iproblemsin which the pulley systems differed on the same attributes) while low-scorhig subjects useddifferent rules on different problems of the same type. We interpreted this result to meanthat high-scoring subjects have a clear set of preferences among rules which are applicablein a given situation while low-scoring subjects do noL

Third. high-scoring subjects were more likely than lowscoring subjects to quantitativelycombine information about two relevant attributes in a single rule. In problems thatrequired the quantitative combination of two relevant attributes Iwelght and mechanicaladvantagel subjects used one of three strategies to solve the problems. One strategy was to

Figure 1: A typical, pulley problem.

B

With which pulley systei

does the man have to piwith more force to liftthe weight?

ABIf no difference.mark C.

ITTS GRAIlDTIC TAR

U,~ .ic r

0

Distribt 1o____Avfh1hbilitY e~O

Avail and/orDist SeOC1&)

iillr

4

compute the effort directly by computing a ratio of the weight to some attribute, such asthe mechanical advantage or the number of pulleys. The second strategy was to use aprinciple whereby differences in mechanical advantage are considered to compensate fordifferences in weight. The third was to use only one of the applicable rules to solve theproblem. The first strategy was used by the highest scoring subjects while the thirdstrategy was used by the lowest scoring subjects.

Figure 2 presents a schematic description of the range of individual differences thatwe found. According to the description presented in the figure, low-scoring subjects arecharacterized as using rules based on visible attributes of pulley systems. These rules arequalitative, the attributes on which they are based can be either relevant or irrelevant.Also. low scoring subjects have no clear preferences among their rules so that theirresponses are inconsistent. High-scoring subjects. on the other hand, prefer rules based onattributes that are highly correlated with mechanical advantage. Also their rules arequantitative and take configural properties of the system into account.

Figure 2 - Schematic representation of the progression

In order to specify mechanisms that can underlie performance on the problems andthat can account for the individual differences identified in Experiment 1. we developed asimulation model. The model simulates the performance of one high-scoring and one low-scoring protocol subject. It simulates the response choices that the subjects gave to theproblems in Experiment 1. as well as stating the rationale for each choice. The simulationmodel Is written in the Soar production system language ILaird. Newell, & Rosenbloom.19871. As in other production systems. Soar's procedural knowledge Is contained inproductions, some of which. in this case. are intended to correspond to the rules subjectsuse in sonling the pulley ptblems. This research Illustrates how a cognitive analystsprovides an understanding of the knowledge that underlies mechanical ability. Consequently.it provides a better foundation for how one might acquire that knowledge. as well as howone could test for Its effects.

Individual differs, ccs in iporiul skill, In earliet work. we examined individualdifferences in spatial skills, such as those tapped by p/ychometric tests IJust & Carpenter.19851 of spatial rotation and transformation. Two such tests are the Cube comparisonstask and the Shepard and Metzler 11971) meatal rotation task. Strategic differences Insuch spatiol tasks can be explained in terms of different cognitive coordinate systems thatsubjects adopt. The strategy of mental rotation that occurs in many recent experimentsuses a coordinate system defined by the standard axes of our visual world It. e.. horizontal.vertical. and depth axesi. Several other possible coordinate systems land hence otherstrategies for solving the problems that occur in psychometric tests of spatial ability areexamined in this article. One alternative strategy uses a coordinate system defined by thedemands of each test item. resulting in mental rotation around arbitrary. task-defined axes.Another strategy uses a coordinate system defined exclusively by the objects. producingrepresentations that are invariant with the objects' orientation. A theory of the mentalrotation of individuals of low and high spatial ability solving problems taken frompsychometric twsts Is instantiated as two related computer simulation models whoseperformance corresponds to the response latencies. eye-lxation patterns, and retrospsctlvestrategy reports of the two ability groups,

In another series of studies. we examined two approaches to the analysis of spatial

Figure 8: Schematic representation of the progression from low to highability.

Ht. -quantitative rules based on system configuration.High Ability- preference for rules based on determining attributes.

. mechanical advantage computed.

. ratio strategy used,

- relevant attributes identified.

compensation strategy used.

* rules based on relevant attributes preferred.

- qualitative rules used consistently.

. qualitative rules based on visible components.

Low Ability • all attributes considered relevant,• no preferences among applicable rules.

5

skills: psychometrics and information processing (Carpenter & Just. 1986). The centralquestion was what processes could be responsible for the general correlation observed amongperformance on spatial tests. the general spatial factor? This chapter suggests that bothquantitative and qualitative differences constitute high spatial ability. In general. high,patial subjects are better at generating, maintaining, and coordinating information duringspatial transformations. In the tasks we have studied, high spatial subjects were less likelyto re-execute a spatial transformation or regenerate a spatial representation. They appearedto find it easier to encode spatial information and might do so more accurately or in moredetail than low spatial subjects. While more research is needed on this topic, it may bethat high spatial subjects have a better spatial "vocabulary". that is. that they have chunksthat allow for more efficient encoding and construction and more accurate retrieval ofspatial information. High spatial subjerf.ts also have more facility with spatialtransformations. When there are alternative ways of solving rotation problems, the highspatial subjects are more likely to use the more complex trajectories. They may also befaster at performing basic spatial operations. but this facility could reflect the more efficientrepresentation that they use. Finally, our definition of spatial ability in terms ofinformation processing, unlike the traditional psychometric one, distinguishes betweensubjects who use spatial transformations and those who use non-spatial processes. even ifthe two groups have roughly similar speed and accuracy on a psychometric test.

In another series of studies, we have investigated individual differences in a morecomplex type of visual problem solving (Carpenter & Just. 1989). We analyzed thecognitive processes in a widely-used non-verbal test of analytic intelligence, the RavenProgressive Matriees Test MRaven. 1962). The analysis determines what processes arecommon to all subjects and all items on the test. and what processes differentiate betweenhigher-scoring and lower.scoring subjects. The analysis is based on detailed performancecharacteristics such as verbal protocols and eye fixation patterns. The theoretical model isexpressed as a pair of computer simulation models that perform like the median or bestcollege students in the sample. The processes that distinguish among Individuals areprimarily the ability to induce abstract relations and the ability to dynamically manage alarge set of problem-solhing goals. The processing characteristic that is common to allsubjects is an incremental, re-iterative strategy for encoding and inducing the regularitiespiece by piece. Additional experiments and discussions relate the theoretical account of theprocessing in the Raven test to performance in other tests of intellectual ability.

I1, Learning nechtanical information from texts and diagrams

Diagrams accompanied by text have been a common means of recording andconveying qcientific and technical information since the 15th Century. Illustrated technicalbooks originated in engineers' notebooks and manuals of technical processe. These booksrelied hwivily on graphics and when they included text. it served to explain the pictures.The invention of the printing press in the I-6th century made these illustrated booksavailable to a large audience. Some histodians have suggested that their availability mayhave been a major cause of the large technological advances between the 16th and 18thcenturies IFerguson. 19771. In recent years. there has been an analogous advance in thecapabilities of graphic technologies, as well as their availability. Graphics innovations. suchas animation software. computer aided drawing. and plotting programs. have made thetechniques of graphic communication available to an ever growing community of users.These innovations have made clear the need for a theory of communication that wouldspecify which media are suited to conveying different types of information, where and whengraphics should be included, and the extent to which information in graphics and text

6

should overlap (Bertin, 1983). But such prescriptions must be grounded in a theory of theprocesses in understanding texts and diagrams.

The chapter of Hegarty. Carpenter and Just (in press) describes the beginnings ofsuch a theory, focusing on how readers understand technical texts and diagrams, particularlydiagrams that have a close correspondence to their concrete referents. Of course, theprocesses in text comprehension have been the focus of considerable research in the last 15years OJust & Carpenter. 1987: Pearson. Barr, Kamil. & Mosenthal. 1984: van Dijk &Kintsch. 1983). In this chapter, we build on what is known about text processing todescribe how the process changes when diagrams accompany the text. Our discussionfocuses on how text comprehension is influenced by the diagram. how the diagram itself isprocessed, and how information from the two sources is integrated. But a psychologicalanalysis that considered only the properties of the text and diagram would miss asignificant component of the story. The processes in understanding a text accompaniud bydiagrams also depends on the reader's profile of cognitive aptitudes. so that the theorymust take into account the differences among individuals.

One of the important results to come from this research is that low mechanical abilityindividuals are "text driven.' They are dependent on the text to draw their attention toparticular apects of the diagram and they use to text to help interpret the diagram, evenwhen the diagram is highly representational. rwa't-h-i'a abstract. In contrast, subject ofhigh mechanical ability can use either source for information about the structure andfunction of the device. As our earlier research revealed, an important aspect of highmechanical ability is knowledge about the relevance of various features of mechanicaldevices, which can guide the inspection and interpretation of diagrams.

One series of studies has examined the way high and low mechanical ability subjectsexamine texts and their accompanying diagrams (Hegarty & Just. 1989), The g"al hasbeen to develop a model of the types of proceses engendered by the two media, Byanalyzing the patterns of eye fixations while readers are leamoing about a device, we havefoulnd characteristic distinctions that reflect the difference* betweven high and low mechanicalability oubjects iHegurty & Just. 1989: Hegarty. 1988). In the experiments, subjects arepresented with a page of text lapproximately 80 words) aad a diagram describing amechanical devices, typically pulley systems of va.r-ing complexity. The Initial researchexamined how long and how well subjects examined the diagram as a function of variousclwracteriutics of the text.

In general. high ability subjects construct a good representation of the pulley systemafter reading the first sentence and then inspecting the diagram. After the first sentenceIwhen both highs and low ability subjects tend to scan the devo-el, the highs gaze alonglines of action, rather than just single components. The high ability subject make fewermid-text inspections than the lows. and in general. use the diagram to verify that the text

* is consistent.

By contrast. low ability subjects are "text driven* In one experiment. we found thatif the text was disorganized (but still accurate) the high ability subjects could and wouldrompensate iby rereading the text and inspecting the diagraml: the low ability subjectscould not compensate. The quality of the text was manipulated by rearranoing some of theinternal sentences, so that they still had internal coherence Ithat is. they were notambiguous), but they were reanranged so that they sentences referring to the samecomponents were not clustered together. The high ability subjects reread the text moreoften for the disorganized text and inspected the diagram slightly more often. By contrs.

7

the low ability subjects could riot compensate: in fact. they inspected the diagram even lessoften for the disorganized texts. Correspondingly. the low ability subjects had particularlypoor scores on a subsequent comprehension test.

Low ability subjects construct their representation as they read the text. Indeed,other studies IHegarty & Just. 1989) have shown that for coherent texts, the more oftenthe text refers to information in the diagram. the more often the low ability subjectsexamine the diagram. The experiment varied the amount of structural detail given in thetext fit was always also available in the diagram). Low ability subjects spend more time onthe diagram when the text has more information about the diagram. This reflects their"text drivenness." By contrast. high ability subjects can utilize the diagram to encode ororganize information about the device. even if the text is very concise or even disorganized.

In some. these studies illuminate basic differences in how individuals of varyingmechanical ability learn new information from texts and diagram. The differences mirrorthe knowledge differences we examined in our analysis of individual differences, described inSection 1.

1M1. Perception of diagrams

A final project under the current proposal examined how people examine animateddisplays to determine if the device was realistic. When a person inspects a machine that isoperating. perhaps to check that it is working correetly. he collects information about themotions of the machine's components, and evaluates this information using his mechanicalknowledge. Several studies examined how people collect such information, and the knowledgeabout machines that they use to evaluate this information. and its rotation to mechanicalability IFallside. 1988). The importance of this project from a scientific vantage point isthat it 6s the first clear demonstration of stimulus sampling during decision nmking.Stimulus sampling, the idea that e0idenre is accumulated in a probabilistie nmnner, hasbeen a major theory In several donains, including learning. decision making, and patternrerognition. The current reaearch externa#Ized the sampling process by recording howpeople inspect a machino to decide if it is mallunctioning.

In the experimental paradigm, subjects were prented with dynamic displaysIcomputer animationsl of a pulley system. and they were asked to judge whethe or not areal pulley system could actually work like the animation, In the experiment, the ropesand wtight were animated so that they move up and down: also, the pulleys could moveclorkwUe or counterclockwise. Finally. a computer simulation, called PULLMAN. wasdeveloped: It uses a production system architectute and it is able to account for the patternof eye fixations of high mechanical ability subjects inspecting the pueys. as well as theirerrors a response times for different pulley syttems.

Three experiments investigate how people san dynamic displays of a pulley system.to determine whether it could be realistic-. The purpose of the expertimnts is to discoverwhat psychologkal processes are inv•lved in making this decision. The first experimentexaminm the sequenc in which pulley components are scanned and the decision processes"that are used in compering interacting components to det-laine whether their motions aremutually consistent. The experinmet examines the eye-fixations of a group of subjects whoperfomed the task. The eyef"*lteon and reation time data are used to develop asimulation modal that perfortrs the task and exhibits similar characteristics to the humansubjects. The second experiment uses the framework devoloped within the model to examine

8

how the spatial layout of the pulley system components affects the nature of the scan andthe decision processes. It also examines the role of individual differences in mechanicalability in performing the task. The third experiment examines performance on the taskwhen the unrealistic pulley animations are quantitatively incorrect rather than qualitativelyincorrect.

Peoples' mechanical abilities were investigated in a series of experiments that usedpulley systems as the target machines. Pulley systems were chosen because they haveinteresting properties and they are easily adapted to visual-processing tasks. One of theinteresting properties of pulley systems is that they are dynamic: weights are lifted, ropesmove. and pulleys cotate. Any mental model of a pulley system probably includes arepresentation of some of the mutually constraining motions of interacting components.Another interesting property of pulley systems is that they appear to provide something fornothing, in the sense that they make it possible to lift heavy weights with ease. withoutany external sources of force. This property. mechanical advantage, is also associated withvisually perceptible cues such as the different relative velocities of a system's components.In general. pulley systems are amenable to visual inspection tasks because the interestingstructures and motions of their components are visually available,

A single experimental paradigm wian used in each of the three studies described inthis paper. Subjects were presented with dynamic displays (computer animations) of a pulleysystemn and they were asked to judge whether or not a real pulley system could actuallywork like the animation. Subjects were also asked to localize the difference between a realpulley system and the animation. if they judged that a real pulley system could not worklike the animation. Some of the displays were realistic, but in others. one or twocomponents of the system behaved inconsistently with respect to their neighbors. Foreu.mple. consider the simple pulley system. which consist. of a pulley attached to theceiling. a weight on the ground. and a rope that is attached to the weight, The rope fromthe weight paise over the pulley froni left to right. and its free end is between the pulleyand the ground. In a tealistic display of this system. when the free end is pulled. thesegment of rope between the free end and the pulley travels downwards. the pulley rotatesclockwise. and the left hand rope segment and the *etight travel uwards. In an unrealisticdisplay, the pulley might be shown rotating counte-clockwlse. a direction intconsistet withthe motiwn of the two rope segmetts and the welght.

The subjects' use of a stochastic sampling process to inpet the pulley systemIoimpwient* implies that their inspection has some interesthng and counter-Intuitiveptoperties. First. the sampling process does not select every romponent in a display. sosmne components may never be oloted at before the subject responds. Second. some othercomponents are slected more than once. The results of the first esperiment reveal some-of the niajor char•cteristics of the procesms subjects used to determine wheathr a pulleytystem was behaving realistically. These characteristics we, used as the main buildingblnks i the coinstrution of a computer simulation model which perorms the sam task.

Pullman is a production system model that simulates how people judg dynamic pulleydisplays. Pullman examines the components of a pulley system and decides whether thesysten is behaving realisticy or not by checking the consistency of its compoentaWmotikvs. It was developed from the performance data obtained from the subjects inExperiment I. The first pant of this section describes the model and how it works. and thesecon part compare its performanc, with the performance of the human sdubet.

Pullnan juuges a pulley systm by collecting evidence about the consisteaies and

9

inconsistencies of the motions of the system's components. It collects one piece of evidenceat a time, adds it to the existing evidence, and then assesses the accumulated evidence. Ifthe evidence for or against the existence of an inconsistency passes a threshold. thenPullman makes a response. Otherwise, another piece of evidence is collected and the cyclestarts over. Three processes are responsible for collecting evidence, assessing it, and makinga response: (1f a selection process, W2) an evaluation process, and 13) a response process.The flow of control between these processes and the operations underlying each one aresummarized in Figure im-process). The three processes and their operations are firstdescribed in general terms, and then in more detail.

10

PUBLICATIONS ASSOCIATED VvITH THE CONTRACT

Books and Dissertations

Fallside. D. C. (1988). Understanding Machines in Motion. Doctoral thesis. CarnegieMellon University. Department of Psychology, Pittsburgh. PA.

Hegarty. M. (1988). Comprehension of Diagrams accompanied by text. Doctoral thesis.Carnegie Mellon University. Department of Psychology. Pittsburgh. PA.

Just. M. A.. & Carpenter. P. A. (1987). The psychology of reading and languageconiprehlnsion. Newton. MA: Allyn and Bacon.

Articles

Beck. I. & Carpenter. P. A. U 9861. Understanding reading and improving instructionalpractice. Americanz Psychologist. 41. 1098-1105.

Dixon. P., & ,Just. M. A. (1986!. A chronometric analysis of strategy preparation in choicereactions. Memory 4 Cognition. 14. 488-500.

Hega, , ".. Just. M. A., & Morrison, I. R. 119881. Menral models of mechanical systems:,inevtdual differences in qualitative and quantitative reasoning. Cognitive Psychology,20. 191-236,

MacDonald. M. C.. & Just. M. A. tin press). Changes in activation levels with negation.Journtal of Experlmen tao Psychology: Learning. MetMory. attd Cogntition.

Chapters

Carpenter. P. A.. & Just. M. A. 11989). The role of working memory In languagecomprehension. In D. Klahr & K. Itotovsky VFds.). Comolex injormatost promessing:The Impact of Herbert A. Stmon Hillsdale. NJ: Erlbaum.

Carpenter. P. A., & Just. M. A. 1198681 Spatial ability: An information processing approachto psychometrics. In R, Sternberg MEd.). Advances in the psychology of humiantotelllgene•e (Vol. 31. Hillsdale. NJ: Lawren"e Eribaum.

Carpenter. P. A.. & Just. M. A. 41986). Cognitive procmm in reading. In J. Orasanu4iWd.). Reading compreheniow. From resev ah to pacdtce Hllodl, NJ: LawrenceErlbum.

Hegarty. M.. Carpenter. P. A.. & Just. M. A. (in press). Dlagrams in the comprehensioof scientific text,. In I. Barr, M. KamIL P. Mowenthal & P. D. Pearson Eds.).Handbook of eatdig ros•earh (VoL WI. New York: Longm

Hqprty. M., & Just, M. A. (1989). Undetandilng machime from text and dialrans. InH. Mandi & J. f. Levin (Eds.). Knowledge acqui•i•on rmo text and pictUe. NorthHol•and-. Elevier Sckece Pabser. 11S.

11

Just. M. A.. & Carpenter, P. A. (1988). Reading and spatial cognition: Reflections fromeye fixations. In G. Luer. U. Lass. & J. Shallo-Hoffmann (Eds.) Eye movementresearch: Physiological and psychological aspects (Vol. 21. Lewiston, NY:C. J. Hogrefe.

Just. NI. A., & Carpenter. P. A. 11987). Eye fixations and visual cognition. In G. Luer &U. Lass (Eds.) Fourth European Conference on Eye Movements Wol 1). LewistonNY: C. J. Hogrefe.

Just. M. A.. & Carpenter. P. A. (1985). Cognitive coordinate systems: Accounts of mentalrotation and individual differences in spatial ability. Psychological Review. 92,137-172.

Technical Reports

Carpenter P. A.. & Just. M. A. (1989). What one intelligence test measures: Atheoretical account of the processing in the Raven Progressive Matrices Test.ITechnical Report). Pittsburgh, PA: Carnegie Mellon University, Department ofPsychology.

Additional References

Bennett. 0. K. 11969). Beatien Mechanical Comprehension test, New York: ThePsychological Corporation.

Bertin. J. 11983i. Seminology of graphs. W. J. Berg, Trans. Madison. Wl: University ofWisconsin Press.

Laird. J. E.. Newell. A.. & Rosenbloom. P. 8. (1987). Soar: An architecture for generalintelligence. Artificial Intelligence. 33. 14-4.

Ferguson. E. S. 11977). The mind's eye: Nonverbal thought In technology. Science. 197.827-836.

Pearson. P. D,. Barr. RK. Kamil. M. L.. & Mosenthal. P. (Mds.). 11984). Handbook ofreading research. New Yor LoWnan.

Raven. J. C. U1962. Advanced Progresskit Matriees. Set II, London: H. K. L~w1s & Co,Distributed in the USA by The Psychologcal Corporation. San Antonio. Tomas.

van Dijk, T. A.. & Ktntsch. W. (1983). Staeegirs of discourse comptehension. NewYork: Academic Press

Dr. Bruce Bloxom,Defense Manpower Data Center,550 Camino El Estero,Suite 200,Monterey, CA 93943-3231,Fairfax, VA 22030

Technical Document Center,AFHRL/LRS.TDC,Wright.Patterson ,AFBOH 45433.6503Dr. Sue Bogner.Army Research lnstitute,ATTN: PERI.SF,5001 Eisenhower Avenue,

Alexandria, VA 22333-5600Dr. Robert Breaux,Code 78,Naval Training Systems Center,Orlando, FL 32813-7100Dr. Raymond E. Christal,UES LAMP Science AdvisorAFHRLIMOEL,Brooks AFB, TX 78235LT COL William L. Derrlck.HO USAFIDPXA,The Pentagon. Rm. 5C360,Washington DC 20330Dr. Dattprasad DivgiCenter for Naval Analysis,4401 Ford Avenue,P.O. Bo 16268

Alexandria. VA 22302-0268LT COL Bill ErcolIne,AFHRL/MOE,Brooks AFB, TX 78235Dr. P.A. Federico,Code 51,NPRDC,San Diego, CA 92152.6800Or. Myron Fischl.U.S. Army HeadquartersDAPE-MRR,The PentagonWashington DC 20310-0300Dr. Alfred R. Fregly,AFOSR/NL. Bldg. 410,Bolling AF8, DC 20332.6448Or. Sherrie GottAFHRL/MOMJ.Brooks AFB, TX 78235-5601Dr. Patrick R. Harrison.Computer Science Department.U.S. Naval Academy

Annapolis. MD 21402-5002Ms. Rebecca Hetter.Navy Personnel R&D CenterCode 63.San Diego, CA 92152.6800Mr. John F. HodgkissSP(N), Room 429A, ABSOId Admiralty Buildings.Spring

Gardens,London SWIA 28EUNITED KINGDOMMr. Paul L. Jones.Research Division.Chief of Naval Technical Training.

Building East.t,Naval Air Station MemphisMillington. TN 38054-5056Dr. Michael Kaplan,Oflice of Basic Research.U.S. Army Research Instltute.5001

Eisenhower Avenue,Alexandria, VA 22333.5600Or. Richard Kern.Army Research Institute.5001 Eisenhower AvenueAlexandria.

VA 22333Dr. Keith Kramer.HCl Lab. Code 5530.Naval Research Laboralory,4445 Overlook

Avenue.Wask .iglon, DC 20375.5000Dr Anit. .. ,ncaster.Accession Policy.OASDIMI&LIMP&FMIAP.Pentagon.WaShlngton. OC 203010,. Vth,Jeng Lee.Depattment of Computer Science.Code 52,Naval Postgraduate

School.Monterey. CA 93943Dr. Clessen J. MIlinOhfice ol Chief of Naval. Operations tOP 13 F)

Navv "•nnex. Room 2832.Washington, DC 20350

Do. Clarenr.e C. McCofrirck.HO. USMEPCOMPAEPCT.2500 Green Bay Road,North Chicago. IL '.•06C

Or. Mrlvln 0 k'antemerloNASA I'leadquarties.Code RCWashmnglon, DC 20546'hair, Depattmenl of Weapons and, Sysleme EngineetingU.S. Naval Academy,

Annapolis. MO 214C2

Df. ,qudth Orasanu.B&SIC Researev, OfticeArmy Retearch Institute.$00t Ftisnhowet Avenue.Atetandrtia, VA 22333

Of- RandOLah Pa, .Army Research Inbtllute.5001 Eisenhower Blvd..Aletandria.VA 22333

LCOR Frank C Pelho,Otticp of he Chile of. N&val Oprallons'OP.-11JWashington. OC 20350

Deot 0f Administralive Sciences, lode 14,Naval Postgraduate School.Monterey, CA 93943.#026

O. MalcelM Ree,AFHRLIPAOA.Btooks AFP. 1A ?0235Mt. Corn Shpsatel UXC2 Block 3,AdmitalltV kesearch Establishment,

Ml,-istrý oe Defence PrtisdownPortemouth HMnte P0$4AAU'tTIT) KINGDOMDr. Rann~all Shumahak.Naval 4eatre-h L.Waltloty.Code 5110AS45 Overltook

Av.nue, S.W,,Wa.tington. O, 27?65100MMt. Joe SilvetmmnCode CNavy P0r4 --41 R&U Center,S' Clego, CA 921524M00Or. H. Waillie Sinelko.Maohowqt Resepoth. and Adviny Services,

Smiohsonian tnowtittonAl0t Nort'l Pitt ileet, Suite t20.Alevendla., VA 223t4.i71tOt Aithard C: SoenSonf.Navy Persovnh, RAO Cente.San Diego. CA 925248600Or. Kutl Stvuct,AFHR..1VOA,Btof.tw AFPSan Antonio, TX ?823&.5601Or, John TangneyAFOSRANL, Bldg. 410,Boiling AFS, OC 20332.-448

SMajor D. 0, Tucker #OUC. Code MA. Room 4023.Washlogton, OC 203 0OD. Btian Wateir.MumRRO, 1100 S. Wafoto1on.Aoew"Iotita, VA 22314Or. Hilda WingNRC MHRI76.2101 Constitution mve.,Wallhlnglot. OC 20418Or. Matlin F, Wiskot,PERSEAREC49 Puitei St., Suite 455.Monetatey, CA 93940Mr. John H. Woat.Navy Personncl W Ceot•,San Dieo. CA P'254800