Home

Just (1976) Eye fixations and cognitive processes

1. This model can be easily summarized During the initial search phase the subject encodes one segment on a figure She then switches her fixation to the other figure and searches for and encodes the corresponding 458 JUST AND CARPENTER segment The second stage transformation and comparison consists of iterative applications of two rules 1 Compare the two orientations Are they less than 25 apart a No Transform the currently fixated segment by 50 in the direction of the other figure Go to Rule 2 b Yes Go on to the confirmation stage 2 Retrieve the representation of the corresponding segment of the other figure and switch fixation to the other figure Go back to Rule 1 The model assumes a very close relationship between eye fixations and mental operations during the transformation process The rotation rule is always applied to the arm that is being fixated Applying a rotation rule to the representation of one arm may cause the representation of the other arm to be pushed down in the short term memory stack When the representation of that other arm is being retrieved to the top of the stack the arm is fixated anew Rule 2 According to this model the number of switches during transformation should increase monotonically with the angular disparity but the increase should be in the form of a particular step function There should be one switch between 0 and 25 two switches between 25 and 75 three switches between
2. 1c the short arm in the left figure is closed while the short arm in the right figure is open No pair of segments corresponds with respect to both length and openness so subjects must select a pair on the basis of length or openness In all seven Different trials involving stimulus pair 1c that we analyzed the initial selection was based on the feature of length In Different trials involving the objects depicted in Figs 1a and b the two open arms have the same length but differ in the way they are joined to the center In two of the three analyzed trials involving these objects the transformation was between open arms In the third case it was between arms that were similarly joined to the center The confirmation process is extremely important in the Different trials since it leads to the discovery that the intersegment relations are not the same in the two figures and hence that the figures are different In fact one of the most prominent features of the Different scan paths is the large amount of confirmation behavior that they contain In the 10 analyzed Different trials the confirmation process consumed an average of 4195 msec or 49 of the total duration The prolongation of the confirmation process is not the only reason for the very long response latencies for Different trials On some trials after going through a complete search transform and unsuccessfully confirm sequence subjects make a second attempt at searching trans
3. The eye fixation data also suggest a precise model of the transformation process We propose that rotations are executed and monitored in discrete steps of approxi mately 50 The estimate of the 50 step size is based on the result that there is one additional switch during the transformation stage for each additional increment of 50 in angular disparity as shown in Fig 10 A transformation may consist of applying a rotation rule that alters the representation of the orientation of a segment by 50 For example an 2 The 50 steps indicated by our data are suggestively close to 45 which has more intuitive appeal EYE FIXATIONS AND COGNITION 457 nz et gs x lt rs 5 o O L o ns 4 uZ 3 0 9 og ah 2p 22 W o O Estimated M2 IF o Observed a D2 1 jci cei meee a Es e za 0 40 80 120 160 ANGULAR DISPARITY degrees Fic 10 Mean number of observed and estimated switches during the transformation and comparison stage in Same trials as a function of angular disparity open arm represented as OPEN r 6 might be transformed into OPEN r 6 50 It is assumed that the representations of the two segments are rotated towards each other by applying the 50 rotation rules first to one segment and then to the other until they are within 25 of each other This form of representation and transformation does not impose any great computational burden in contrast to a truly an
4. often estimated to consume between 50 to 800 msec each The goals of this paper are to demonstrate that the locus duration and sequence of the eye fixations can be closely tied to the activity of the central processor and to exploit this relation in investigating the fine structure of the processor s activity in a number of cognitive tasks The primary proposal is that the eye fixates the referent of the symbol currently being processed if the referent is in view That is the fixation may reflect what is at the top of the stack If several symbols are The order of authors is arbitrary This paper represents a collaborative effort We are grateful to David Klahr Roger Shepard and Herbert Simon for their comments on earlier drafts of this paper We also thank Roger Shepard for providing copies of the stimulus figures used in the rotation experiment Finally we thank Lee Gregg and Chuck Faddis who have been responsible for much of the development of the on line tracking system in the Carnegie Mellon laboratory The project was supported in part by Research Grant NIE G 74 0016 from the National Institute of Education U S Department of Health Education and Welfare and Grant MH 07722 from the National Institute of Mental Heaith Requests for reprints should be sent to Marcel Adam Just Psychology Department Carnegie Mellon University Pittsburgh PA 15213 441 Copyright 1976 by Academic Press Inc All rights of reproduction
5. 75 and 125 and so on Figure 10 shows that the number of switches predicted by the model corresponds very closely to the observed number of switches The increase in switches is similar to the pattern obtained for the duration of the transformation process shown in Fig 7b The data in Fig 7b suggest that an upper bound on the duration of each step of the transform and compare process is about 800 msec In general the model of the transformation stage gives a good account of the data The confirmation process Being able to rotate two segments into similar orientations during the transformation stage does not guarantee that the two figures are the same Therefore the third stage confirmation determines whether segments other than the transformed ones correspond to each other The scan paths indicated at least two methods for confirming such correspondence One method applies the same sequence of rotation rules used in the transformation stage to another pair of segments If this second rotation is successful then the two figures are the same This method used on about half the trials produced scan paths similar to those in the transformation stage except that the switches were between a pair of corresponding segments other than the initially transformed pair A second confirmation method encodes the relation between the central joint and an arm of each figure and determines whether that relation is the same in both figures This method may
6. AND COGNITION 477 task there cannot be eye fixations switching back and forth between the two Rs since only one of them is externally present while the other is the long term representation of a normal R Nevertheless itis reasonable to speculate that the sequence of internal switches of attention in the Cooper and Shepard study is related to the external sequence of fixation switches observed in our study of rotation Thus the eye fixations observed in the current rotation experiment may reflect more general mental processes that also occur in the absence of eye fixations This view is supported by the performance in a purely tactile mental rotation task performed by blindfolded subjects The response latency curve as a function of orientation in deciding whether a wooden letter is mirror image or normal is similar to the curve found by Cooper and Shepard for visually presented letters Carpenter amp Eisenberg Note 1 The similarities in the total latency functions for the three types of tasks the visual and tactile studies with alphanumeric characters and the current rotation task with abstract figures suggest that at least some aspects of the rotation process are similar across all three tasks In this view eye fixations may be simply a convenient method for externalizing internal processes that are also used in nonvisual tasks If this speculation is correct the model we have proposed may apply to rotation tasks that involve mentally ge
7. a natural segment But in general the appropriate unit of analysis depends on the accompanying theory The current research demonstrates that analyses based on the gaze are compatible with models of cognitive processing The locus of the fixation The most general assumption of the current EYE FIXATIONS AND COGNITION 471 research is that the locus of the eye fixation can indicate what symbol is currently being processed Converging lines of evidence from very diverse tasks support this general assumption and also allow us to refine our theoretical consideration about the relationship between eye fixations and mental processes In tasks where the behavioral units are fairly large and open to conscious introspection the pattern of eye fixations correlates well with subjects verbal reports For example Winikoff 1967 see also Newell amp Simon 1972 found a high correlation in cryptarithmetic tasks where numbers are substituted for letters to solve a problem like DONALD GERALD ROBERT In general Winikoff s subject tended to look at the letter whose value he was computing or trying to recall as inferred from his concomitant verbal protocol Similarly eye fixations correlate with verbal protocols when subjects are choosing among several alternatives such as cars that differ in make year and condition Russo amp Rosen 1975 These studies provide evidence that the locus of the eye fixation corre sponds to the information being
8. arm will be called closed and the four central cubes constitute the central joint The locus of the eye spot was scored according to the locus of its centroid with respect to the three segments The simplest way to describe our scoring procedure is to apply it to a few representative scan paths Figure 4 shows a scan path for a Same trial with 0 disparity After the initial fixation on the center of the left figure the subject fixated corresponding closed arms at the upper part of each figure Then the open arms at the bottom of each figure were fixated To make the analysis of the scan paths precise we constructed rules for classifying instances of search transformation and comparison and confirmation The most prominent property of the scan paths was that the subject would repeatedly look back and forth between corresponding segments of the two figures We identified the repeated fixation of corresponding segments with the transformation and comparison process When the same pair of segments was involved in two transformation TABLE FIG 4 Locus AND DURATION OF THE FIXATIONS Duration Fixation Figure Location msec l Left Central joint 351 2 Closed arm 150 3 Right Closed arm 200 4 Open arm 200 5 Left Open arm 250 EYE FIXATIONS AND COGNITION 449 Fic 5 The figure indicates the sequence of fixations on a correct Same trial in which the disparity was 80 The subject s total res
9. forming and confirming a different pair of segments Occasionally a lengthy search stage involved an examination of all the possible ways of pairing the segments and that kind of search led directly to a response of Different without any transformation Thus the durations of all three stages increased during Different trials but the duration of confirmation increased the most One scan path that exemplifies the processing on Different trials is shown in Fig 9 Fixations 1 to 4 reflect the initial search for corresponding segments consuming 1436 msec Then there is a transformation and com EYE FIXATIONS AND COGNITION 455 Fic 9 The figure indicates the sequence of fixations on a correct Different trial The subject s total response latency was 5868 msec of which 6 had no visible eye spot See Table Fig 9 for the locus and duration of the fixations parison of the short arms of each figure in fixations 5 6 and 7 consuming 919 msec Fixations 8 to 15 reflect the confirmation process consuming 3175 msec We presume that it is during confirmation that the subject determined that the relation between the arm and central joints was different in the two figures In this trial the bulk of the processing time was consumed by the confirmation stage The Processing Model of the Rotation Task The internal representation We propose that the processor operates on one segment of the figure at a time and that the representa
10. information to localize the error on a particular trial to a particular stage However the eye fixations do provide clues about the reasons for some of the errors There was a total of nine errors on the Same trials all on angular disparities greater than 120 On five of the nine trials the subject attempted to transform noncorresponding segments That is the initial search process selected two segments that were in fact not corresponding The subsequent transformation and confirmation stages failed to detect this error The scan path in Fig 8 demonstrates this type of error in which the subject erroneously selected the open arm on the left and EYE FIXATIONS AND COGNITION 453 Fig 8 The figure indicates the sequence of fixations on an incorrect Same trial in which the disparity was 140 The subject s total response latency was 8567 msec of which 13 had no visible eye spot See Table Fig 8 for the locus and duration of the fixations closed arm on the right as corresponding then looked back and forth between them in fixations 2 to 8 and 10 to 12 and attempted confirmation in fixations 13 to 21 In the remaining four error trials subjects did successfully complete the initial search process and subsequent fixations alternated between corresponding segments of the two figures This suggests that the source TABLE FIG 8 Locus AND DURATION OF THE FIXATIONS Duration Fixation Figure Location msec 1 Lef
11. look at the schematic car while answering More interestingly when the picture was removed prior to the question subjects still tended to look where the appropriate object had been located Such fixations apparently play a place keeping organizational role rather than an encoding role The symbols in the short term memory may be indexed to particular spatial locations This formulation is reminiscent of the method of loci cf Bower 1970 and spatial interference effects in retrieval Byrne 1974 When the time comes to retrieve or operate on a symbol the eye may fixate the location from which the symbol was originally encoded It may be this mechanism that produces fixations on the referent of the symbol at the top of the stack assuming that the referent stays in the same location Duration of gaze In the tasks we investigated the time spent gazing at a figure reflected both the time to encode that figure as well as the time to operate on the encoded symbol Tachistoscopic recognition studies indicate that familiar figures like alphanumeric characters or even words can be internalized within a very short exposure duration as low as a few tens of milliseconds Yet in these cognitive tasks people gaze at very simple and familiar figures for much longer often for hundreds of milliseconds For example in the sentence verification task subjects looked at a star or a plus for 700 to 1200 msec depending upon the relation between the sentenc
12. mental operations examined gaze duration in a Sternberg memory scanning task Gould 1973 In Gould s experiment subjects had a memory set of one two or three letters and 12 probe letters were dis tributed along the perimeter of an imaginary clock face that corresponded to the display Only one of the 12 probe letters was a member of the memory set The subject s task was to scan around the clock face starting at 12 o clock and proceeding clockwise until he found the positive probe The amount of the time spent fixating each negative probe and the positive probe increased linearly with the memory set size at a rate of about 50 msec per item This is compatible with the explanation that each probe item was serially compared to each of the memory set items and as the memory set size increased the probe had to be compared to more items in memory The probe was fixated while the comparison operations occurred The importance of this finding is that the parameter of 50 msec per item inferred from the duration of the gaze is very close to the time of 38 msec per item inferred from reaction time studies with only a single probe Sternberg 1969 Gould s results provide an important validation for the eye fixation methodology The eye fixation measures yield results that are completely consistent with Sternberg s careful reaction time studies The only eye fixation research that reports a lack of correlation between fixation duration a
13. of sentences and pictures In L Gregg Ed Cognition in learning and memory New York Wiley 1972 Clark H H amp Chase W G On the process of comparing sentences against pictures Cognitive Psychology 1972 3 472 517 Cooper R M The control of eye fixation by the meaning of spoken language Cognitive Psychology 1974 6 84 107 Cooper L A amp Shepard R N Chronometric studies of the rotation of mental images In W G Chase Ed Visual information processing New York Academic Press 1973 Dearborn W The Psychology of Reading Columbia University Contributuions to Philosophy and Psychology XIV New York The Science Press 1906 Ditchburn R W Eye movements and visual perception Oxford England Clarendon 1973 Fairbank B A Jr Experiments on the temporal aspects of number perception Unpublished doctoral dissertation University of Arizona 1969 Gaarder K R Eye movements vision and behavior Washington Hemisphere Publishing 1975 Gould J Eye movements during visual search and memory search Journal of Experimental Psychology 1973 98 184 195 Groen G amp Parkman J A chronometric analysis of simple addition Psychological Review 1972 79 329 343 Henmon V A C The time of perception as a measure of differences in sensation Arch Philos Psychol Sci Method 1906 8 5 75 Hochberg J Components of literacy Speculations and exploratory research In H Levin amp J P Willia
14. phrase always referred to the location of a plus and to consider it to mean The plus is North or The plus isn t North The picture contained a plus at one of the four compass directions and a star at the other three Any one of these characters as well as the sentence was displayed only when the subject directly fixated it When an affirmative sentence was true or a negative sentence was false the plus was at the place specified by the directional term in the sentence In the false affirmative and true negative cases the plus could have been at any one of the three remaining locations This design was adopted to discourage subjects from recoding negatives like Isn t North into corresponding affirmatives like Is South The analysis however is concerned only with the cases where the plus was located on the same axis as the directional term in the sentence The sentence centered on the video monitor was 5 6 mm high and 45 mm wide 50 mm for negative sentences The plus and stars were 5 mm by 5 mm and they were at a distance of 75 mm from the center of the screen The subject s viewing distance was 64 cm on average however the distance was adjusted for each subject to keep the excursion of the eye spot constant On average the display subtended about 14 of visual angle For scoring purposes the viewing field was divided into an imaginary three by three grid such that the sentence was located in the center square while t
15. result in a scan from the center to the arm of one figure and then a similar scan of the other figure see Fig 5 for an example A combination of these two methods might EYE FIXATIONS AND COGNITION 459 explain why the confirmation duration increases with angular disparity but with a slower rate of increase than for the transformation duration shown in Fig 7 Either method of confirmation could determine the response of Same or Different Discussion The eye fixation data lead to a detailed model of the processing in the Shepard Metzler task but there are questions about the generalizability of the model Without examining a broad range of experimental situations there is no way of knowing which aspects of the model are invariants of the human processing system and which aspects are task induced Consider the proposed 50 rotation steps It is possible that the 50 steps are fundamental and invariant over tasks Alternatively people may be able to tune the size of the rotation step to the particular grain and range of orientation differences they are faced with in an experiment This is a clear empirical question of whether the rotation operation adapts itself to the task environment Similarly one can consider whether the representation of the figures is the same in all rotation tasks The representations proposed in the current model are highly schematic but they do contain sufficient information to perform the task The representat
16. the outcome of the computations These tasks are well structured in that the subjects goals are clear to them and to the experimenter Such tasks are more amenable to a precise processing analysis than tasks that require EYE FIXATIONS AND COGNITION 443 Fic 1 a A pair of Same figures with 0 disparity b a pair of Same figures with 180 disparity c a pair of Different figures with 120 disparity subjects simply to read or scan a display without any specified purpose or response Moreover the tasks to be analyzed are all speeded tasks in which the subject is asked to work accurately but quickly The total response latencies produced under these conditions can be divided into processing stages on the basis of the locus and sequence of fixations The purpose of analyzing several tasks is to abstract the general charac teristics of the central processor as they are revealed by eye fixation behavior Generally research programs and resulting papers revolve 444 JUST AND CARPENTER around a particular task such as mental rotation sentence comprehension or memory scanning attempting to discover or characterize the opera tions used in that particular task The goals here are slightly different While one goal is to learn about the fine structure of the processes used in each task an equally important goal is to examine the relation between eye fixations and cognitive processes PROCESSING ROTATED FIGURE
17. the larger group does not predict the gaze duration on the larger group EYE FIXATIONS AND COGNITION 467 Gaze on Larger group OF aa Ve e 170 e eee SEERNES Reece EE Ge 23 4 5 6 NUMBER OF DOTS IN LARGER GROUP DURATION msec Fic 16 Mean duration of gaze on the larger group of dots as a function of the number of dots in that group On some trials subjects fixated one group of dots and then switched their fixation to the other group The pattern of such switches between the two groups of dots was similar to the latency data The number of switches increased with the number of dots in the smaller group and decreased as the difference between the two groups increased However the mean number of such switches per trial was only 0 3 indicating that subjects were sometimes able to perform the task by using their peripheral vision The proposed model The results are consistent with a counting model cf Parkman 1971 Groen amp Parkman 1972 adapted to the dot inequality task The process might start by counting one or two dots in each group and checking to see if either group had been exhausted If one group had been exhausted it would be designated the smaller one If neither had been exhausted then one or two more dots might be counted in each group and again there would be a check to see if either group had been exhausted This process would continue until one of the groups the smaller one would be exhausted
18. the min effect should decrease The total latencies in Fig 17a show this trend When the split is small namely one then latencies increase by an average of 43 msec with each increment in the min the smaller group When the split is larger two or three the min effect is reduced to 16 msec Finally with a split of four there is no min effect A similar analysis of the Buckley and Gillman 1974 dot com parison data based on a larger range of mins and splits further supports this conclusion Their data also show a monotonic decrease in the min effect as the split increases Fig 17b For splits of one through seven the min effects are 66 57 49 34 25 16 and 8 msec respectively Also there is a main effect of the split such that the latency generally decreases as the split increases The important point as far as the two process explanation is concerned is that when the split is larger the category judgment mechanism may be used more often and so the min effect decreases Other types of explanations account for the quantitative comparison task in terms of a quasi logarithmic analogue representation of quantities such that small digits like and 2 are relatively far apart on the internal scale while larger digits like 8 and 9 are closer together Buckley amp Gillman 1974 Moyer amp Landauer 1967 Shepard Kilpatric amp Cun ningham 1975 These explanations account for the min and split effects by assuming that the fa
19. were displayed on a standard video monitor Eye fixations were monitored with a corneal reflectance eye tracking system that was under computer control This system beams a small light onto the left cornea captures 446 JUST AND CARPENTER z 16 000F t F v be H e v E 14 000r H s 5 12000 L F 2 E D E 10 000 H F g r F p t 8000 H ee F f 6 6000 a e N ad w 4000p e e X e i i e lol e e pees te 7 e a Cs es es e pode dg dp N E A E AE EE A E E E E E E O 40 80 120 I60 O 40 80 120 i60 o 40 80 120 160 ANGULAR DISPARITY degrees Fic 2 Mean response latency for Same trials as a function of angular disparity for the three subjects the reflection of the light with a television camera and inputs the video signal to a digitizer The digitizer determines the position of the eye spot relative to the vertical and horizontal synchronization pulses of the video signal The position of the eye spot in the video frame is output as a pair of rectangular coordinates which is then transferred to the room interface of a Honeywell DDP 116 computer The coordinates can be evaluated immediately by the program in order to make the stimulus presentation contingent on the location of the eye spot For example to initiate a trial the subject was required to fixate a fixa tion point and simultaneously push a ready button When the button was pushed the program did not start the
20. which people verify whether a sentence is true or false of an accompanying picture Reaction time 460 JUST AND CARPENTER ZGK BXVQ Fic 11 a Schematic diagram of the visual display in the sentence verification task when the eye spot denoted by black spot is on the sentence not to scale b visual display within the same trial when the eye spot is in the North location studies of sentence verification show that people make more errors and take longer to respond when verifying a negative sentence The extra processing time for a negative lies between 300 and 1200 msec depending on the linguistic structure of the negative sentence Carpenter amp Just 1975 The processing stages involved in verification include reading the sentence and internally representing it looking at the picture and representing it and comparing the two representations Carpenter amp Just 1975 Chase amp Clark 1972 Clark amp Chase 1972 Trabasso Rollins amp Shaughnessy 1971 An eye fixation analysis may indicate how the processing time is allocated among the various processing stages More over the analysis may indicate which stage of processing consumes the extra time due to negation Elsewhere we have developed a processing model of sentence veri fication that suggests that elements in the sentence representation are compared sequentially to elements encoded from the picture Carpenter amp Just 1975 Mismatches between elements resul
21. COGNITIVE PSYCHOLOGY 8 441 480 1976 Eye Fixations and Cognitive Processes MARCEL ADAM JUST AND PATRICIA A CARPENTER Carnegie Mellon University This paper presents a theoretical account of the sequence and duration of eye fixation during a number of simple cognitive tasks such as mental rotation sentence verification and quantitative comparison In each case the eye fixation behavior is linked to a processing model for the task by assuming that the eye fixates the referent of the symbol being operated on A widely accepted view of the human information processing system is that most of the symbol manipulation takes place in a central processor sometimes referred to as the active memory Neisser 1967 working memory Newell amp Simon 1963 operational memory Posner 1967 or the immediate processor Newell 1973 This paper is concerned with the rapid mental operations of the central processor and how they are reflected by the pattern and duration of eye fixations during a task involving visual input We will examine the basic operators parameters and control structure of the central processor as it performs such tasks as the com parison of rotated figures Shepard amp Metzler 1971 mental arithmetic Parkman 1971 sentence verification Carpenter amp Just 1975 and memory scanning Sternberg 1969 These tasks generally take less than 5 or 10 sec to complete and can be decomposed into very rapid mental operations
22. If the subjects were answering the question Which group contains more dots they would simply indicate the group that had not been exhausted The number of counts or iterations in this process would be proportional to the number of dots in the smaller group If the gaze duration is proportional to the number of increments then it follows that duration of gaze on the smaller group should increase with the number of dots in the smaller group as it does This is called the min effect since latencies increase with the size of the smaller or minimum group Futhermore the duration of gaze on the larger group should be independent of the number of dots in the larger group which it is One further prediction of this model is that the duration of gaze on the larger group should increase with the size of the smaller group This prediction follows from the proposal that the dots in both groups are counted only until one group the smaller one is exhausted This prediction is confirmed with gaze durations on the larger group increasing 468 JUST AND CARPENTER a Data from Current Experiment 1000 o F i split 1 900 D yo split 2 800 E split 3 split 4 TOOP split 5 a ar ap 10001 Data from Buckley and Gillman 900p 800 700F TOTAL RESPONSE LATENCY msec 600 500F L JL 12345 67 8 NUMBER OF DOTS IN SMALLER SET Fic 17 a Mean response latency as a function of the number of
23. NI 2 4000 8 O o 2000 l N Sit a se on cm 0 40 80 120 160 ANGULAR DISPARITY degrees Fic 6 Mean response latency as a function of angular disparity for all correct Same trials and for 100 correct Same trials in which eye fixations were scored We identified the third process confirmation as a short sequence of fixations between corresponding parts of the two figures other than the transformed segments Confirmation could appear as a scan from the central joint to an arm on one figure then a similar scan on the other figure Figure 5 shows an example of the confirmation process where the fixations proceed from the central joint to the open arm on the left figure fixations 9 and 10 and then a similar scan is executed on the right fixations 11 and 12 In the scan path in Fig 4 the last two fixations on the open arms fixations 4 and 5 also exemplify confirmation While confirmation generally followed transformation some confirmation occasionally occurred between episodes of transformation Any fixation or sequence of fixations that did not conform to the definition of search transformation or confirmation were classified as other To see how well the model fits the eye fixation data the scan paths were scored for 100 of the 171 correct Same trials Seventy one trials were not scored because of apparatus failure in which the optical system failed to capture an eye spot that was visible at least 85 of the tim
24. S Eye fixations are intimately involved with our ability to visually encode spatially distributed information It is possible that eye fixations can also indicate how visual information is internally manipulated This question can be explored in the task domain of mental rotation in which people compare two figures in order to determine whether or not they depict the same three dimensional object Shepard amp Metzler 1971 In these studies subjects were timed while they decided whether two figures were views of the same object Figs la or b or views of different objects The two objects in the Different trials Fig 1c differed by a reflection as well as by rotation The main independent variable was the angular disparity between the two views of the same object that is the amount of physical rotation necessary to align the two figures into congruence The response latencies for the Same trials increased linearly with the degree of angular disparity Shepard and Metzler attributed this increase in response time to a process of mental rotation The slope of the response times as a function of the angular disparity was postulated to reflect the rate of mental rotation There are several key questions about the processes underlying per formance in this task that are not easily answered by the response latency studies We proposed that the following questions about the microstruc ture of the processes could be addressed by an eye f
25. ack and forth between the left and right figure For example at 0 disparity subjects initially fixated the left figure then looked over at the right hand figure then looked back at the left and frequently looked back at the right hand figure for a second time for a total of three switches between the two figures The mean number of such switches between figures increased with angular disparity as shown in Fig 3 The next step of the analysis was designed to determine exactly what subjects were looking at and how the pattern of their fixations might reveal the microstructure of the underlying cognitive operations To classify the locus of the eye fixation we divided each figure into three main segments the arm whose third face of the end cube was visible open 1 Metzler and Shepard 1974 report some preliminary observations on the eye movements of two subjects performing the mental rotation task their subjects also looked back and forth between the two figures 448 JUST AND CARPENTER Fic 4 The figure indicates the sequence of fixations on a correct Same trial in which the disparity was 0 The subject s total response latency was 1296 msec of which 11 had no visible eye spot See Table Fig 4 for the locus and duration of the fixations the arm whose third face of the end cube was not visible closed and a central joint For example in Fig la the upper arm will be called the open arm while the lower
26. alogue holistic representation of the entire figure rotated by a parallel compu tation of the position of all its points This model of the transformation stage is most easily explained by working through an example say when two Same figures have an angular disparity of 80 Suppose that the subject has encoded a particular segment of the left hand figure The first switch occurs when she fixates and encodes the corresponding segment of the right figure Then the orientations of the two segments are compared The orientations differ by more than 25 therefore she rotates her representation of the right hand figure by 50 counterclockwise After this transformation she retrieves the representation of the segment on the left In doing this retrieval she switches her fixation to that figure After it is retrieved she compares the two orientations They would still be more than 25 apart Therefore she transforms the representation of the orientation of the left hand figure by 50 clockwise Then she retrieves the representation of the segment on the right in order to compare the two again In doing this retrieval she switches fixation over to the right figure After it is retrieved she compares the two orientations At last after three switches and two applications of the rotation rule the two segments are represented at fairly similar orientations within 25 of each other The subject would then continue on to the confirmation process
27. amp J P Williams Eds Basic studies on reading New York Basic books 1970 Loftus G R Eye fixations and recognition memory for pictures Cognitive Psychology 1972 3 525 551 Loftus G R A framework for a theory of picture recognition In R A Monty amp J W Senders Eds Eye movements and psychological processes Hillsdale NJ Lawrence Erlbaum 1976 in press Mackworth N The line of sight approach to children s reading and comprehension In S Wanat H Singer amp M Kling Eds Extracting Meaning from Written Language Newark DE International Reading Association 1974 Mackworth N H amp Bruner J S How adults and children search and recognize pictures Human Development 1970 13 149 177 Mackworth N H amp Morandi A J The gaze selects informative details within pictures Perception amp Psychophysics 1967 2 547 552 Mehler J Bever T G amp Carey P What we look at when we read Perception amp Psychophysics 1967 2 213 218 Metzler J amp Shepard R Transformational studies of the internal representation of three dimensional objects In R Solso Ed Theories in cognitive psychology The Loyola Symposium Potomac Md Lawrence Erlbaum 1974 Meyer D E On the representation and retrieval of stored semantic information Cognitive Psychology 1970 1 242 300 Moyer R S Comparing objects in memory Evidence suggesting an internal psychophysics Perception amp Psych
28. amp Winston 1972 478 JUST AND CARPENTER Atkinson R C amp Juola J F Factors influencing speed and accuracy of word recognition In S Kornblum Ed Attention and performance IV New York Academic Press 1973 Bower G H Analysis of a mnemonic device American Scientist 1970 58 496 510 Buckley P B amp Gillman C B Comparisons of digits and dot patterns Journal of Experimental Psychology 1974 103 1131 1136 Buswell G T Fundamental reading habits a study of their development Chicago Chicago University Press 1922 Buswell G T How people look at pictures Chicago University of Chicago Press 1935 Buswell G T How adults read Supplementary Educational Monographs No 45 1937 Byrne B Item concreteness vs spatial organization as predictors of visual imagery Memory amp Cognition 1974 2 53 59 Carpenter P A amp Just M A Semantic control of eye movements during picture scanning in a sentence picture verification task Perception amp Psychophysics 1972 12 61 64 Carpenter P A amp Just M A Sentence comprehension A psycholinguistic processing model of verification Psychological Review 1975 82 45 73 Carpenter P A amp Just M A Linguistic influences on picture scanning In J Senders amp R Monty Eds Eye movements and psychological processes Hillsdale NJ Lawrence Erlbaum 1976 in press Chase W G amp Clark H H Mental operations in the comparison
29. an response latency for the true affirmative TA false affirmative FA false negative FN and true negative TN conditions Response latencies are plotted as a function of the hypothesized number of comparison operations for each condition where K is the number of comparisons for the true affirmative condition was timed from the onset of the display until his response terminated the trial Each of the 12 subjects had 15 practice trials and two blocks of 48 test trials Results and discussion As Figure 12 shows the total response times in the four information conditions did increase linearly from true affirma tive to false affirmative to false negative to true negative In fact a Straight line accounts for 98 6 of the variance among the four means The residual 1 4 of the variance is not significant F 2 33 lt 1 Thus the pattern of total latencies for the current task resembles the latency pattern found in other experiments cf Carpenter amp Just 1975 These analyses concern only those trials in which the subject gave a correct response The frequency of incorrect responses was very low as indicated in Fig 12 The important advantage of the current methodology is that the location and duration of the gaze allow us to break down the total response time into finer components For this analysis we divided the gazes into four categories the initial gaze on the sentence subsequent gazes on the sentence after having looke
30. ation condition as Fig 13 shows The function plotted in Fig 12 is simply the sum of the functions in Fig 13 The nonlinearity of the initial gaze on the sentence is obviously small relative to the dominant linear trend and so its effect is not apparent in the total response latency These results indicate how the total processing time in sentence veri fication is distributed among various stages The duration of the initial gaze on the sentence suggests that the time needed to read and represent the sentence is 700 msec at most This 700 msec enters primarily into the intercept of the total response time What accounts for the difference between the response time for the fastest condition the 1400 msec for the true affirmative and the slowest condition the 1900 msec for the true negative This 500 msec is consumed by the operations that compare the sentence and picture to determine their relation In fact these com parison operations are reflected in the duration of the gaze on the location specified by the directional term This analysis can tell us why negative sentences take longer to process than affirmatives The total response time was 346 msec longer for negatives than for affirmatives This can be partitioned into several components The largest component is the comparison time reflected in the duration of gaze at the picture which was 267 msec longer for negatives Secondly negative sentences took 57 msec longer to read And thir
31. between accounted search comparison mation stages for Total 0 0 0 1 0 0 9 0 6 0 2 2 7 20 0 4 1 1 0 9 0 8 0 1 3 3 40 0 2 1 7 1 2 0 9 0 2 4 2 60 0 3 1 7 1 5 0 7 0 4 4 6 80 1 0 3 0 1 5 0 4 0 3 6 2 100 0 5 2 7 1 5 0 8 0 5 6 0 120 1 1 2 9 2 1 0 6 1 2 7 9 140 2 2 3 6 1 8 0 7 0 7 9 0 160 1 3 4 0 2 2 0 8 0 9 9 2 180 1 6 5 7 2 3 0 8 1 8 12 2 another involves a switch to the other figure The number of such switches remains fairly constant across angular disparities As might be expected the average number of fixations increased with angular disparity from six fixations at 0 to 31 at 180 Also the average duration of a fixation increased from 200 msec at 0 to 320 msec at 180 Incorrect Same trials Error trials have often been ignored by chron ometric models of cognitive processes because it is difficult to attribute errors to a particular stage of processing exceptions are the work on the speed accuracy trade off cf Wickelgren Note 5 and the work on multiple processes in word recognition cf Atkinson amp Juola 1973 An incorrect response in the rotation task could result from an error during any one of the stages of searching transforming and comparing con firming or in executing the final motor response An example of a trans formation error would be to rotate a segment about the wrong axis and incorrectly conclude that two Same figures represent different objects The total response latency alone provides insufficient
32. buted over two testing sessions separated by at least one day The subjects received 60 practice trials before the experiment began The three paid subjects were right handed females of college age with 20 20 corrected vision Five other subjects were eliminated because they made more than 15 errors during the 60 practice trials The locus of the eye spot relative to the 10 cubes that made up each figure was scored on each frame of the videotape namely once every 16 7 msec When the eye spot was EYE FIXATIONS AND COGNITION 447 NUMBER OF SWITCHES Po fe ef eS 40 80 120 160 ANGULAR DISPARITY degrees GF Fic 3 Mean number of switches for Same trials as a function of angular disparity located on the same cubes in a sequence of successive video frames for at least 100 msec the frames were aggregated into a single fixation Latency results The mean response latencies for correct Same trials shown in Fig 2 increased monotonically with increasing angular disparity All three subjects showed a linear increase between 0 and 100 but the curves were positively accelerated beyond 100 The subjects here had considerably less practice than Metzler and Shepard s 1974 subjects Nevertheless the mean latencies from 0 to 100 disparity have a pattern similar to that obtained by Metzler and Shepard 1974 Eye fixation results One striking feature of the eye fixation behavior was that subjects systematically looked b
33. cond condition is that the task goals be specified for the subject Asking subjects simply to look at a picture or read some prose permits them to adopt their own definitions of what processing is required and this again makes it difficult to infer the relationship between eye fixations and underlying mental processes And of course speeded tasks discourage extraneous processing and the concomitant extraneous fixations Some of the rules that govern fixations are general scanning strategies while other rules are highly specific to the processing in the task being performed Eye fixations will reveal the mental processes in a particular task only if the task structure minimizes the use of general scanning strategies An example of this structuring is evident in the study of how people looked at pictures after reading sentences involving affirmative quantifiers e g A small proportion of the dots are red or negative quantifiers e g Few of the dots are red Carpenter amp Just 1972 The pictures always had a small subset of dots at the top and a large subset at the bottom Thus the subject knew to look at the top or at the bottom depending on whether he wanted to determine the color of the small subset or the large one This task structure eliminated the need first to search for the desired subset and then to encode its color The relation between eye fixations and mental operations is even clearer when the role of peripheral information is cont
34. d away gazes on the location specified by the directional term in the sentence and finally gazes in any other locations Thus the durations of all four types of gazes add up to the total response time The important question was whether these durations varied systemat ically as a function of the four information conditions The initial gaze on the sentence should reflect the time to read and represent the sentence As Fig 13 shows the duration of the initial gaze was 57 msec longer for negatives than for affirmatives F 1 33 14 93 p lt 01 This result indicates that the negative sentences take about 57 msec longer to read and represent than the affirmatives After having looked away from the sentence subjects occasionally refixated it later in 3 These latency results are similar to those of Krueger 1973 for the comparable conditions EYE FIXATIONS AND COGNITION 463 Initial gaze on the sentence 800 o e 600 Subsequent gazes A on the sentence g 200p Ee emm o P l L N o Da S Gazes on location named by directional term o 1200F p gt 2 1000 Boot li 1 pt Gazes on all other locations 200F OES 4 ll f ni Si K K i K 2 K 3 TA FA FN TN Number of comparisons Fic 13 Average duration spent gazing on various locations of the display for the true affirmative TA false affirmative FA false negative FN and true negative TN conditions The
35. dly subsequent gazes on the sentence were an insignificant 20 msec longer for negative sentences Thus the bulk of the additional processing time for negatives is consumed by the operations that compare the information from the sentence to the picture The results show that there is a systematic correspondence between the mental operations and eye fixations in a sentence verification task Under well controlled conditions the sequence of gazes on the external display corresponds to the sequence of mental operations in the processor Moreover the duration of the gaze is proportional to the duration of the underlying operations 4Somewhat different scanning strategies are used when the presentation is not gaze contingent and the entire display is visible Carpenter amp Just 1976 Under those conditions subjects can occasionally perform an entire trial while fixating on only the sentence Moreover they sometimes detect the plus and then fixate it even when it is not in the location mentioned in the sentence EYE FIXATIONS AND COGNITION 465 Fic 14 Schematic diagram of the visual display in the dot quantitative comparison task not to scale The dashed lines which indicate the boundaries between the four sectors did not appear in the display QUANTITATIVE COMPARISON A quantitative comparison requires an order judgment e g Which is larger or Which is brighter or Which is longer of two or more objects along a common under
36. dots in the smaller group for various splits Data from the current experiment b Mean response latency as a function of the number of dots in the smaller group for various splits The graph is based on cell means estimated from a graph of the latencies for comparing random configurations of dots Gillman amp Buckley Note 3 The aggregated data appear in Buckley and Gillman 1974 monotonically from 160 msec when the smaller group contains one dot to 296 msec when the smaller group contains five dots The proposed counting model requires supplementation to account for a persistent finding from this and previous research namely that compar isons are faster when the absolute difference or split between the two groups is larger cf Henmon 1906 Johnson 1939 for the data on line length comparisons Buckley amp Gillman 1974 Fairbank 1969 Moyer amp Landauer 1967 Parkman 1971 Sekuler Rubin amp Armstrong 1971 for data on digit comparisons and Buckley amp Gillman 1974 for data on dot comparisons The split effect is present in both the total latencies see Fig 17a and in the gaze durations on both the smaller and larger groups of dots We attribute the split effect to the presence of a second mechanism that can sometimes make the quantitative comparison by categorizing each of the two groups of dots as a small group or a large group Groups of one two or three dots may be classified as small while groups o
37. e Memory amp Cognition 1975 3 267 276 Sekuler R Rubin E amp Armstrong R Processing numerical information Journal of Experimental Psychology 1971 90 75 80 Shepard R N Kilpatric D W amp Cunningham J P The internal representation of numbers Cognitive Psychology 1975 7 82 138 Shepard R amp Metzler J Mental rotation of three dimensional objects Science 1971 171 701 703 Simon H A amp Barenfeld M Information processing analysis of perceptual processes in problem solving Psychological Review 1969 76 473 483 Sternberg S The discovery of processing stages Extensions of Donder s method In W G Koster Ed Attention and performance II Acta Psychologia 1969 30 276 315 Tichomirov G K amp Poznyanskaya E D An investigation of visual search as a means of analyzing heuristics Voprosy Psikhologii 1966 12 39 53 Tinker M A Recent studies of eye movements in reading Psychological Bulletin 1958 55 215 230 Trabasso T Rollins H amp Shaughnessy E Storage and verification stages in processing concepts Cognitive Psychology 1971 2 239 289 Tversky B Eye fixations in prediction of recognition and recall Memory amp Cognition 1974 2 275 278 Volkmann F C Vision during voluntary saccadic eye movements Journal of the Optical Society of America 1962 52 571 578 Williams L G The effect of target specification on objects fixated during visual searc
38. e The mean response latencies from the 100 trial sample are very similar to the data for all 171 correct Same trials as shown in Fig 6 so the sample appears to be representative The analysis of the scan paths makes it possible to examine how the total processing time shown in Fig 6 is distributed across search trans formation and comparison and confirmation stages as a function of angular disparity As Fig 7a shows the time spent initially searching the figures increased with angular disparity from about 300 msec at 0 to about 1600 msec at 180 The bulk of the processing time was spent in trans EYE FIXATIONS AND COGNITION 451 2000 a SEARCH 4000 b TRANSFORMATION AND COMPARISON 3000 m so e 2000 a s to0o00F e z E F fae a T 2000 CONFIRMATION ie e A 1000 oi e e roook d OTHER S P NE AOS A S 1000 SACCADES e eo E AE ET yy yf o 40 80 120 160 ANGULAR DISPARITY degrees Fic 7 Mean duration of various processing stages in Same trials as a function of angular disparity formation and comparison as shown in Fig 7b The duration of this stage increased markedly with increasing angular disparity from about 500 msec at 0 to 3800 msec at 180 The average time spent in the third stage confirmation increased from 450 msec at 0 to 2300 msec at 180 as shown in Fig 7c Thus for a typical trial say at 80 disparity 21 of the time was cons
39. e and the figure Clearly the duration of the gaze includes not only encoding time but also the time for subsequent operations on the encoded symbol There are a number of reasons why a subject might continue to fixate a figure after the relevant information has been encoded If the processor is busy operating on the most recently encoded information there is no reason for it to direct the eye to seek other information So the eye may remain stationary simply because it is not instucted to move An alternative view of the persistence of the gaze is that the processor might actively instruct the eye not to move during the processing of the most EYE FIXATIONS AND COGNITION 473 recently encoded information The reason for avoiding new fixations might be that a saccade automatically initiates an encoding activity cf Loftus in press that could interrupt the ongoing processing Perhaps the reason that people often gaze upwards or close their eyes altogether while computing the answer to a demanding question is that they are avoiding extraneous encoding operations that could interrupt processing Thus the persistence of the gaze could be due to the absence of an instruction to move the eye or the presence of an instruction not to move the eye In either case the gaze duration on a particular figure provides a measure of the time spent processing the corresponding symbol One of the most elegant studies of the relationship between gaze duration and
40. f four five or six may be classified as arge but the boundary may be variable across trials and subjects If one group of dots belongs EYE FIXATIONS AND COGNITION 469 to the small category and the other to the large category then the one that belongs to the large category is larger The category judgment mechanism may be much quicker than the counting mechanism but it would not work when the split is small since in those cases the two groups would tend to belong to the same category Pairs with large splits splits of three four or five could be processed with the quick category membership judgment much more often than pairs with small splits one or two The mean response latencies for any pair would be a mixture of the trials where the fast category membership judgment is used and trials where the counting mechanism is used As the split increases the number of fast trials contributing to the mean should increase and so on average the mean latencies should decrease The two process explanation is supported by an interaction between the min effect and the split effect observed in this experiment and others When the split is small the counting mechanism is more likely to be used resulting in a strong min effect That is the response latencies increase with the size of the smaller group When the split between the two groups is larger the category judgment mechanism should be used more often and so
41. gaze consisting of any number of consecutive fixations on the same part of the stimulus For example in the sentence picture verification task any number of consecutive fixations on a plus were aggregated into a single gaze While each gaze was associated with a particular stage of processing the models had little to say about the distribution of fixations within a gaze Fixations may be aggregated into still larger units comprised of clusters of adjacent fixations as well as some intervening fixations on other parts of the display For example such large aggregation units in problem solving research cf Winikoff 1967 Newell amp Simon 1972 are com patible with models that describe mental operations that take on the order of a few seconds each The models of problem solving are based to some extent on verbal protocols which tend to describe only large scale operations The large scale aggregation of eye fixations was therefore compatible with the time frame of the verbal report and the emerging psychological theory Unaggregated fixations may serve as the unit of analysis either for theoretical reasons or simply as a default option For example Gaarder s 1975 theoretical framework based to some extent on evoked potential research treated an individual fixation as a unit of encoding and pro cessing In other cases where there is no available theory to specify a unit of analysis the individual fixation may be adopted since it is
42. h Perception amp Psychophysics 1966 1 315 318 Winikoff A Eye movements as an aid to protocol analysis of problem solving behavior Unpublished doctoral dissertation Carnegie Mellon University Pittsburgh PA 1967 Woodworth R S Experimental psychology New York Henry Holt 1938 Yarbus A L Eye movements and vision New York Plenum Press 1967 REFERENCE NOTES 1 Carpenter P A amp Eisenberg P Mental rotation of tactile stimuli Complex Informa tion Processing Paper No 313 Pittsburgh PA Carnegie Mellon University 1976 2 de Groot A amp Jongman R Perception and memory in chess An experimental analysis of the master s professional eye RITP Memorandum No 024 University of Amsterdam Holland 1973 3 Gilman C B amp Buckley P B Numeric comparison Paper presented at the Fourteenth Annual Psychonomic Society Meeting St Louis Mo November 2 1973 McConkie G W amp Rayner K Identifying the span of the effective stimulus in reading Reading and Learning Series Research Report No 3 Cornell University Ithaca NY 1974 Wickelgren W A Speed accuracy tradeoff and information processing dynamics Unpublished manuscript University of Oregon Eugene Oregon 1974 a a Accepted February 17 1976
43. he stars and plus were in the middle top middle bottom middle left or middle right squares Any single fixation or sequence of fixations on one of these squares was scored as a gaze on that location During a trial the digitizer determined the locus of the eye spot every 16 msec Sixteen milliseconds after the eye spot was first detected in a square the stimulus material for that Square appeared on the screen As soon as the eye spot moved from that square the stimulus was replaced by a place holder The place holder for each star and plus was a dollar sign The place holder for the sentence location was composed of a string of three random letters to replace the copula s or Isn t and four random letters to replace the directional term The place holders remained the same on all trials The rapidity of the replacement within 16 msec after the initial fixation of a square made it relatively unobtrusive The place holders in the periphery provided markers where the subject could look to get information but the subject could not know what was there until he actually looked Half a second after the subject fixated a target in the center of the display field and pressed a ready button the sentence appeared at the central fixation place The subject 462 JUST AND CARPENTER S 3 2000 3 1800 o 5 1600 v 1400 x a v S 1200 Jio 2 5 X rey K K i K 2 K 3 TA FA FN TN Number of Comparisons Fic 12 Me
44. in any form reserved 442 JUST AND CARPENTER processed in a particular sequence then their referents should be fixated in the same sequence and the duration of fixation on each referent may be related to the duration that the corresponding symbol is operated on The obvious advantage of monitoring eye fixations is that the behavior within any particular trial can potentially be decomposed into various stages whose durations can be directly measured By contrast a single response latency cannot be interpreted or decomposed without reference to laten cies in other conditions Another reason that eye fixations provide an appropriate measure in cognitive tasks is that the rapidity of the fixation behavior matches the rapidity of the processor The fixation behavior can be sampled at high densities per unit time say once every 200 msec and so the durations of individual processing stages and hence changes in the duration can be measured directly The relation between duration of processes and sampling rate can be elucidated with an analogy to time lapse photographs of slow or rapid processes To study the behavior of glaciers it is sufficient to take a photograph once every few weeks but to study the blossoming of a flower it might be necessary to take photographs every hour Similarly to study the rapid mental operations of the central processor it is desirable to monitor its behavior many times per trial so as to separate the behavior into
45. ions might be more complex in tasks that demand that more information be encoded from the figures Just as eye fixation analyses led to a precise model for the Shepard Metzler task this methodology should also distinguish the invariant from the transient processes and so lead to a general theory of mental rotation The current model proposes that rotation in this task occurs in steps of approximately 50 It is possible that within each 50 step there are intermediate stages corresponding to intermediate orientations But even with 50 steps a 150 rotation involves intermediate steps corresponding to 50 and 100 rotations Thus to some extent Metzler and Shepard s 1974 proposal of an analogue process is compatible with the current proposal In summary the scan paths enabled us to separate the processing into search transformation and confirmation stages and to measure the duration of each stage Switches in fixation during the transformation stage indicated that the rotation was monitored in steps of approximately 50 This analysis was applicable not only to the correct Same trials but also provided evidence on error trials and Different trials The research shows how eye fixations can reveal the sequence of mental operations during the internal manipulation of spatial information COMPARING SENTENCES WITH PICTURES One linguistic comprehension task that lends itself well to an eye fixation analysis is sentence verification in
46. ixation study 1 How does the subject know which parts of the figure are to be rotated into each other Before rotating one figure into another the subject must decide which parts potentially correspond to each other Eye fixations may indicate how this initial decision about correspondence is made 2 How does the subject know how far to rotate one of the objects One possibility would be that the subject makes some estimate of the angular disparity and then performs a ballistic rotation i e with the target orientation predetermined Alternatively the rotation process may be monitored at various points along the way The eye fixations may show whether the process is monitored 3 Once the required rotation has been performed how does the subject know whether the two figures represent the same object or not The eye fixation behavior may reveal the comparison process that determines whether the two figures match or not after rotation Our objective was to identify component processes in this task by ana EYE FIXATIONS AND COGNITION 445 lyzing the scan paths and by observing how they changed with angular disparity In a pilot eye fixation experiment subjects compared two figures with different orientations in the picture plane the plane perpendicular to the subjects line of sight The results suggested that there were three stages in the processing that will be called 1 search 2 transformation and comparison and 3 confirmatio
47. lying dimension The comparative judgment re quires that the two objects be represented and their representations be compared In order to obtain more detailed evidence about the processes in this task an experiment was devised in which subjects eye fixations were monitored while they decided which of two groups of dots was larger The response latencies for selecting the larger of two groups of dots strongly resemble the latencies for digit comparisons Buckley amp Gillman 1974 so this task may produce results generalizable to digit comparisons Furthermore prior data summarized by Klahr 1973 have shown that the time to determine how many dots there are in a group increases monotonically from about 500 msec for one dot to 2200 msec for nine dots These results suggest that larger groups of dots might be fixated longer if they are to be quantified The hypothesis was that the duration of fixation on each of the groups of dots might tell us how the two groups of dots were represented and processed during a quantitative comparison task Method Subjects eye fixations were monitored as they compared the sizes of two groups of dots Each group contained from one to six dots so there were 15 possible pairs of unequal groups If the word more appeared on the left side of the display as shown in Fig 14 subjects indicated whether the upper or the lower group contained more dots by pressing an upper or lower response button If the word wa
48. ms Eds Basic studies on reading New York Basic Books 1970 Huey E The Psychology and pedagogy of reading The Macmillan Co 1908 Reprinted Cambridge MA The M I T Press 1968 Javal E Essaie sur la physiologie de la lecture Annales d oculistique 1878 79 97 Johnson D M Confidence and speed in the two category judgment Archives of Psychology 1939 241 1 52 Just M A Comprehending quantified sentences The relation between sentence picture and semantic memory verification Cognitive Psychology 1974 6 216 236 EYE FIXATIONS AND COGNITION 479 Just M A amp Carpenter P A Comprehension of negation with quantification Journal of Verbal Learning and Verbal Behavior 1971 10 244 253 Kahneman D Attention and effort Englewood Cliffs NJ Prentice Hall 1973 Kahneman D amp Lass N Eye position in tasks of association and memory Unpublished manuscript Hebrew University Jerusalem 1971 Klahr D Quantification processes In W G Chase Ed Visual information processing New York Academic Press 1973 Kolers P A Three stages of reading In H Levin amp J Williams Eds Basic studies on reading New York Basic Books 1970 Krueger L E Sentence picture comparison A test of additivity of processing time for feature matching and negation coding stages Journal of Experimenal Psychology 1972 95 275 284 Levin H amp Kaplan E Grammatical structure and reading In H Levin
49. n In the first stage there is a search for segments of the two figures that superficially correspond to each other for example two segments at the end of the figures that both have three visible faces The function of the search process is to select segments of the two figures that can potentially be transformed one into the other During the next stage transformation and comparison the two corresponding segments are rotated into each other A transform and compare operation is applied stepwise to the representations of the two segments Each step of the transformation may correspond to a rotation such that at the end of the transformation the segment is represented at a new orientation Each step of the transformation is followed by a comparison to determine whether the two orientations are now congruent This stepwise trans form and compare process continues until the necessary number of transformations has been made to make the internal representations of the two segments sufficiently congruent in orientation The third stage con firmation involves a check of whether the rotation that brought the two segments into congruence will also bring other portions of the two figures into congruence Processes roughly similar to search transformation and confirmation have been suggested by Metzler and Shepard and their subjects introspective reports supported the suggestions 1974 pp 169 178 The eye fixation data make it possible to separa
50. nd performance concerns memory for pictures Loftus 1972 Loftus found that during learning the number of fixations not their total duration was the best predictor of subsequent recognition memory However Tversky 1974 has recently found a positive corre lation between the duration of individual eye fixations and later memory Tversky suggests that the critical variable in relating fixation duration to picture memory might be the kind of features being encoded in the learning phase The present research does not attempt to account for eye fixations in picture scanning and recognition for relevant work on this topic see Buswell 1935 Mackworth amp Brunner 1970 Mackworth amp Morandi 1967 Noton amp Stark 1971 Potter amp Levy 1969 Our concern has been with ongoing computation rather than search processes in long term 474 JUST AND CARPENTER memory so the picture recognition issue would take our discussion too far afield However the present theoretical framework suggests that the resolution of this problem requires a model of what is encoded during the initial learning and what tests are made during the subsequent recognition phase As yet the data relating fixation duration to recognition memory are insufficient to construct a complete model of the mental processes in picture memorization and retrieval While the duration of the gaze is closely related to the duration of cognitive processes the two durations are not neces
51. nerated stimuli If processing models based on eye fixation studies are to be generalized to nonvisual tasks then the factors that influence only visual encoding must be identified For example picture scanning processes might be affected by perceptual saliency Williams 1966 and there may be no parallel in semantic memory retrieval Conversely semantic memory re trieval may be affected by factors such as semantic distance cf Rips Shoben amp Smith 1973 which has no parallel factor in picture encoding processes If these modality specific processes can be isolated then eye fixa tions may provide a way to investigate the fundamental operations that occur in the central processor Operations whose durations lie between 50 and 800 msec seem especially susceptible to this approach as shown by the current work on rotation sentence verification and quantitative com parison For these rapid operations there is a very close link between the symbol that is being processed and the locus sequence and duration of eye fixations because of the eyes tendency to fixate the referent of the symbol that is tat the top of the stack REFERENCES Alpern M Movements of the eyes In H Davson Ed The eye New York Academic Press 1962 Vol 3 Alpern M Effector mechanisms in vision In J W Kling amp L A Riggs Woodworth and Schlosberg s experimental psychology Volume I Sensation and perception New York Holt Rinehart
52. ns in reading cf Buswell 1922 1937 Hochberg 1970 Kolers 1970 Levin amp Kaplan 1970 Mack worth 1974 McConkie amp Rayner Note 4 Mehler Bever amp Carey 1967 Tinker 1958 there is no convergence on a theory of reading The difficulty is that there is no single treading process because we read differently in different situations For example a newspaper article is read differently from a legal contract and the same contract is read differently depending on whether one is looking for typographical errors or buying a house In order to develop models of reading it will probably be necessary to study performance in a number of well understood task environments so as to determine the influence of the environments on the reading process Generalization of the models Certain kinds of operations in the central processor appear to function similarly irrespective of the source of encoding of the operated on symbol be it a visual display tactile input semantic memory retrieval or whatever The invariant operations would presumably be very basic ones such as comparing two symbols for identity retrieving the next symbol in an ordered list or incrementing an internal counter In those cases in which the operations are invariant conclusions gained from the eye fixation methodology may generalize to processing of symbols in non visual domains One example from recent psycholinguistic research demonstrates how sentences that
53. ophysics 1973 13 180 184 Moyer R S amp Landauer T K Time required for judgments of numerical inequality Nature London 1967 215 1519 1520 Neisser U Cognitive psychology New York Appleton Century Crofts 1967 Newell A Production systems Models of control structures In W G Chase Ed Visual information processing New York Academic Press 1973 Newell A amp Simon H A Computers in psychology In R D Luce R R Bush amp E Galanter Eds Handbook of Mathematical Psychology New York Wiley 1963 Vol 1 Newell A amp Simon H A Human problem solving Englewood Cliffs NJ Prentice Hall 1972 Noton D amp Stark L Eye movements and visual perception Scientific American 1971 224 34 43 Parkman J A Temporal aspects of digit and letter inequality judgments Journal of Experimental Psychology 1971 91 191 205 Posner M I Short term memory systems in human information processing In A F Sanders Ed Attention and performance Acta Psychologica 1967 27 267 284 480 JUST AND CARPENTER Potter M C amp Levy E I Recognition memory for a rapid sequence of pictures Journal of Experimental Psychology 1969 81 10 15 Rips L J Shoben E J amp Smith E E Semantic distance and the verification of semantic relations Journal of Verbal Learning and Verbal Behavior 1973 12 1 20 Russo J amp Rosen L An eye fixation analysis of multialternative choic
54. orner of the left field then the search for the corresponding segment begins in the upper right hand of the right field If there is no segment in the upper right then the segment nearest the upper right is examined The duration of this search process increases with angular disparity for two reasons First with increasing disparity the corresponding segments are in successively more dissimilar locations At 0 disparity corresponding segments have identical locations in their respective fields However as the disparity increases from 0 absolute location is a successively poorer cue for finding corresponding segments and the heuristic must be supplemented by an active search The second reason for the increase is that at larger disparities the probability of selecting and attempting to transform noncorresponding segments increases and this incorrect transformation is counted as part of initial search Figure 8 shows an example of the search process selecting non corresponding segments that are both at the top of their respective fields In this trial the incorrect search led to an error On other trials the incorrect selection of a pair of segments was detected after some trans formation had been attempted Thus the eye fixations allow us to trace the initial search for corresponding features and to determine the reason for the increase in the duration of the search process with angular disparity The transformation and comparison process
55. periment replicated the major latency results that have been previously reported for this task Buckley amp Gillman 1974 The mean latencies ranged from 700 to 1100 msec Trials with incorrect responses were rare 2 8 and were not considered in any of the analyses The response latencies will be discussed in more detail after an analysis of the eye fixation results The first analysis concerns the duration of gaze on the smaller group If subjects were computing the number of dots in the group one might expect that the more dots there were in the group the longer people would spend looking at it As expected the gaze duration on the smaller group increased by about 26 msec for each additional dot and a linear model accounts for 95 8 of the variance among the means shown in Fig 15 The slope of 26 msec is within the range of subitizing rates cited by Klahr 1973 although it is at the low end Thus it is plausible that the subjects compute the number of dots in the smaller group But what about fixation on the larger group If subjects determine the quantity of dots in the larger group then gaze duration on the larger group should also increase with the number of dots in that group However Fig 16 shows that the duration of gaze on the larger group is independent of the number of dots there Thus the two groups of dots are fixated differently The size of the smaller group predicts the gaze duration on the smaller group but the size of
56. ponse latency was 3574 msec of which 9 had no visible eye spot See Table Fig 5 for the locus and duration of the fixations episodes separated by extraneous fixations their durations were com bined Extraneous fixations were classified as other The trans formation and comparison process is evident in fixations 5 to 8 of the scan path shown in Fig 5 where the figures have an 80 disparity In fixations 5 to 8 the subject looked back and forth between the closed arms of the two figures for a total of 1185 msec We identified the search process with the initial portion of the scan path that preceded the first instance of transformation Applying these rules to the scan path in Fig 5 fixations 1 to 4 would be attributed to search for a total of 818 msec In Figure 4 where the angular disparity is much smaller the duration of the search process 351 msec fixation 1 is much shorter TABLE FIG 5 Locus AND DURATION OF THE FIXATIONS Duration Fixation Figure Location msec l Left Central joint 200 2 Central joint 301 3 Right Open arm 167 4 Central joint 150 5 Closed arm 167 6 Left Closed arm 200 7 Closed arm 317 8 Right Closed arm 501 9 Left Central joint 250 10 Open arm 200 11 Right Central joint 484 12 Open arm 317 450 JUST AND CARPENTER i Segl in 10000 i o I7l correct Same trials bd e 00 trial sample E o 8000 gt Q 2 R amp 6000 8 ee EE
57. processed in tasks where subjects can verbalize what they are processing Some aspects of problem solving involve operations too rapid for verbal protocols but the eye fixations still reveal what symbols the subjects are processing A good example are the few scan paths that have been recorded of chess masters scanning a board position for 5 sec de Groot amp Jongman Note 2 Tichomirov amp Posnyanskaya 1966 The locus of eye fixations is accounted for by assuming that the master scans between pairs of pieces that are related by attack or defense Simon amp Barenfeld 1969 Again these data support the assumption that the locus of the eye fixations reflects what is being internally processed Since eye fixations are sensitive to the structure of the internal repre sentation being constructed or operated upon they provide a valuable methodology for examining how linguistic material is interpreted One research strategy is to present a linguistic stimulus followed by a picture and examine how the internal representation of the prior sentence alters the way the picture is scanned in a verification task For example this methodology has been used to examine the processing of affirmative and implicitly negative sentences Carpenter amp Just 1972 The affirmative sentences e g A small proportion of the dots are red and the implicitly negative sentences e g Few of the dots are red have the same truth value However linguistic and p
58. refer to information from different sources like pictures vs semantic memory may be processed similarly Just 1974 timed subjects while they verified quantified sentences like Some of the red figures are round with respect to a picture that included red and round figures The overall pattern of latencies was similar to the pattern obtained when the sentences refer to concepts in semantic memory e g Some men are doctors Meyer 1970 In fact even though the relevant informa tion was encoded from a picture in one case and retrieved from semantic memory in the other both sets of data could be explained in terms of the same operations Just 1974 Obviously the initial encoding stages involve different processes but in this and certain other cases cf Carpenter amp Just 1975 the information seems to be manipulated similarly once it is past the encoding stages This suggests that processing models of these subsequent stages derived from eye fixation studies may generalize to nonvisual domains Internal rotation processes may also be somewhat independent of the visual modality When subjects are deciding whether a visually presented rotated R is normal or a mirror image Cooper amp Shepard 1973 the response latencies resemble those for the Shepard and Metzler task in certain respects The resemblance led Cooper and Shepard to argue that the processes in the two tasks were similar In the Cooper and Shepard EYE FIXATIONS
59. rolled The extreme case of this is the computerized tunnel vision in the sentence verification task in which there is no peripheral information so the duration of gaze at any locus cannot reflect encoding of information from another locus These features of the task structure minimize the role of general scanning strategies and thereby make the design more sensitive to the cognitive processes of interest In all of these tasks the eye scan is very much goal directed in fact directed by the information present at the top of the stack There are two possible sources of such information namely the task structure and information computed during the trial Both sources influenced fixations in the sentence verification task where the instructions to fixate the sentence determined the first fixation but the locus of the second fixation was determined by information computed during the trial After the sentence e g Plus isn t North was fixated the directional term in the sentence determined the locus of the next fixation in this case North Since both the task structure and the ongoing processing can determine the locus of fixation both factors must be taken into account in developing a complete processing model One domain of eye fixation research that has been hampered by the 476 JUST AND CARPENTER absence of task analyses is the area of reading While there have been many promising empirical studies of eye fixatio
60. rovides an upper bound on the estimate In any case the difference between gaze durations in different conditions may provide a good estimate of the duration of the cognitive process by which they differ To the extent that absolute gaze duration does not provide a precise estimate of processing time a subtractive technique can be used For example we were able to compare the gaze duration in sentence picture verification for affirmative and negative sentences The general point here is that most of the analytic power of mental chronometry cf Sternberg 1969 can be applied to gaze durations as well as to total response latencies The conjoint chronometric analysis of gaze durations and response latencies can often yield a very fine grained model of cognitive operations Task conditions that optimize the use of eye fixations The locus of fixation is not always synonymous with the direction of attention Subjects can be instructed to fixate one referent while attending elsewhere The possibility of such disassociation makes it important to specify the EYE FIXATIONS AND COGNITION 475 conditions under which eye fixations are an accurate reflection of what is being processed One of the most important conditions is that the task require that information from the visual environment be encoded and processed If the visual display is not relevant there are no mapping rules between what is being fixated and what is being internally processed A se
61. rther apart two quantities are located on the internal logarithmic scale the faster is the quantitative comparison process The advantage of these alternative explanations is that they are parsimonious and they seem readily applicable to continuous dimensions such as sizes of animals Moyer 1973 However these approaches cannot easily account for the finding that the gaze duration on both groups of dots 470 JUST AND CARPENTER was proportional to the size of the smaller group By contrast a counting model is easily compatible with this aspect of the data The duration of gaze on the sectors other than the larger and smaller group of dots did not vary from condition to condition and showed little evidence of a min effect or a split effect The mean processing time in this task 793 msec was distributed as follows On average 371 msec were spent gazing at the more or less 178 msec gazing at the larger group of dots 165 at the smaller group and 79 msec at the sector between the two groups The eye fixation data also showed how this distribution of processing time was affected by the sizes of the smaller and larger groups The results indicated that the two groups of dots are fixated in a manner consistent with an upward counting process GENERAL OVERVIEW The unit of analysis The appropriate unit of analysis in relating eye fixations to cognitive processes depends on the theory motivating the analysis The unit used in this paper is the
62. s less they judged which group contained fewer dots A total of 60 stimuli was formed by orthogonally combining the two words more and less with the 15 pairs of groups and the responses designating either the upper or the lower group Each subject had four blocks of 60 stimuli presented in a random order A trial started 500 msec after the subject fixated a point at the locus of the word and pressed a ready button The computer generated display was presented on a video monitor at a distance of 53 466 JUST AND CARPENTER Gaze on Smaller group 230p s oO 2 210b E A z 190F E I7OF e 150f 5 2 130 atte he fate 1 1 23 4 5 NUMBER OF DOTS IN SMALLER GROUP Fic 15 Mean duration of gaze on the smaller group of dots as a function of the number of dots in that group to 68 cm The word more or less 2 8 cm wide appeared 13 cm to the left of the dot display The dots formed two vertical lines one above the other separated by a vertical distance of at least 5 cm Each group of dots was 0 5 to 6 cm long depending on the number of dots in the group On average the center to center distance between the two groups of dots was 8 of visual angle For scoring purposes the screen was divided into the four imaginary sectors indicated by the dashed lines in Fig 14 The analysis was concerned primarily with the distribution of the gaze across the four sectors Results The response latencies showed that this ex
63. sarily identical The duration of the gaze may overestimate processing time because other factors enter into the duration of a fixation and consequently into the duration of a gaze These other factors include the time to plan the next fixation and the additional time to detect the stimulus because of the raised threshold after a saccade Both these factors have a demonstrated effect in some circumstances but these circumstances are very different from the ones in the cognitive tasks we have examined For example the planning of a fixation takes between 180 and 250 msec as estimated from the fixation response latency but in these studies the time of onset and the locus of the stimulus were unpredictable Alpern 1972 In our tasks the time to plan a fixation may be very much shorter because the viewing field is stable and the subject himself decides when and where to look for information Similarly the finding that the detection of threshold level lights is impaired before during and for some time after a saccade Volkman 1962 may have little implication for tasks like the current ones where the stimuli are all clearly suprathreshold In sum while the factors other than cognitive processes may contribute to the duration of fixations and gazes current data are insufficient to estimate the magnitude of their contribution At best the gaze duration may provide a rough estimate of the absolute duration of a stage of processing or at least it p
64. se components add up to the total response times shown in Fig 12 the trial The durations of such subsequent gazes on the sentence were similar for all four information conditions as Fig 13 shows The directional term in the sentence can be viewed as an instruction for where to direct the next fixation irrespective of whether the sentence was affirmative or negative In fact the location specified by the directional term was the locus of the second gaze on 92 of the trials Subjects tended to fixate this location only once during a trial The time spent gazing at this location increased linearly with the number of hypothesized comparison operations as Fig 13 shows The straight line accounts for 98 1 of the variance among the four means The residual 1 9 is not significant F 2 33 lt 1 The slope 135 msec per operation may be inter preted as an estimate of the time to compare an element from the sentence representation to one from the picture representation These results suggest that after reading and representing the sentence the subject fixated 464 JUST AND CARPENTER the location specified by the directional term encoded the figure that was there and continued fixating there while performing the comparison operation Occasionally subjects would gaze at a location other than the sentence or the location specified by the directional term The frequency and duration of these other gazes did not vary as a function of inform
65. stages The trace of the stages may provide a specification of their respective durations and the sequence in which they occur Eye fixation studies have their historical roots in cognitive research dealing with reading Almost 100 years ago in 1878 Javal cited by Mackworth 1974 observed young children s eyes during reading and contrary to the then popular conception of a continuous sweep across a line of print he discovered that the eye made a series of discrete pauses separated by jumps While some research pursued the role of eye fixations in reading cf Buswell 1922 Dearborn 1906 Huey 1908 Woodworth 1938 much of the subsequent psychological research focused on the jumps saccades rather than the pauses cf Alpern 1962 Ditchburn 1973 Yarbus 1967 for overviews and the behaviors that were investi gated were oculo motor rather than cognitive Recently there has been renewed research interest in the pauses themselves and how they relate to underlying cognitive processes cf Tichomirov amp Posnyanskaya 1966 Winikoff 1967 The current paper will examine eye fixations in several situations and account for the locus sequence and duration of eye fixations in terms of their relationship to underlying cognitive processes The tasks to be examined all require that the subject must encode some information from a visual display do some mental computations on that information and then produce a response that is contingent on
66. sychological evidence suggests that the two sentences have different internal representations Just amp Carpenter 1971 The affirmative sentence is represented as an affirmation that the small subset has some property in this case redness We predicted that after reading the affirmative that refers to the small subset e g A small proportion of the dots are red people should tend to fixate the small subset By contrast an implicit negative is represented as a negation 472 JUST AND CARPENTER of some property of the large subset in this case redness It was predicted that after reading an implicit negation about the large subset e g Few of the dots are red people would tend to fixate the large subset As predicted subjects looked at the location in the picture specified by the underlying representation of the sentence The locus of the eye fixation is sensitive to the internal representation even when subjects are not consciously aware of the nature of the linguistic stimulus or of their pattern of eye fixation While people are listening to spoken questions or passsages they tend to fixate the pictorial referent of words that occur in the text Cooper 1974 Kahneman amp Lass 1971 cited by Kahneman 1973 For example in the Kahneman and Lass study people were shown a schematic drawing of four objects such as a car person tree and airplane asked a question like What makes of cars can you name Subjects tended to
67. t Central joint 334 2 Open arm 134 3 Right Closed arm 200 4 Closed arm 200 5 Left Open arm 468 6 Open arm 317 7 Right Closed arm 200 8 Closed arm 334 9 Left Central joint 334 10 Open arm 117 11 Right Closed arm 401 12 Left Closed arm 150 13 Right Central joint 150 14 Left Central joint 418 15 Right Closed arm 251 16 Central joint 200 17 Central joint 568 18 Left Open arm 768 19 Right Central joint 450 20 Central joint 902 21 Central joint 534 454 JUST AND CARPENTER of the error must have occurred in some subsequent stage such as the transformation confirmation or response execution Different trials The response latencies for Different trials were long an average of 4 sec longer than Same trials and variable The angular disparity between two figures is not really well defined for a Different trial since the two figures cannot be physically rotated into congruence The total response latencies alone give no indication of how processing time was distributed across the three stages However the pattern of eye fixations allows us to follow the sequence of processing stages and to determine which stages consume the extra 4 sec of processing The scan paths indicate that the initial search process in Different trials starts out similarly to Same trials However in a Different trial the segments selected by the search stage cannot be in complete cor respondence For example in the Different pair shown in Fig
68. t in additional com parisons thereby consuming additional processing time The model postulates that because of the form of the internal representation and the EYE FIXATIONS AND COGNITION 461 number and nature of the mismatches the number of comparison opera tions increases linearly from the case of true affirmative sentences to false affirmatives to false negatives to true negatives In fact the verification latencies in a number of studies have been found to increase linearly corresponding to the increasing number of postulated comparisons The current experiment examined which parts of the display were fixated longer during the conditions with longer response latencies An important innovation in the current methodology was that the display was made contingent on the locus of the gaze The only part of the display either the sentence or the picture that was visible to the subject was the part at the locus of the gaze as depicted in the schematic diagram in Fig 11 This gaze contingent display creates a functional tunnel vision in the subject by eliminating all peripheral information relevant to the true false decision The subject could not encode new information unless he looked at the relevant position in the display Method The sentences in the experiment were either affirmative like Is North or negative isn t North and involved one of the four directions North South East or West The subject was told the
69. te the performance on each trial into the three stages and specify the nature of the processing within each stage Method The experiment was a Same Different task in which the subject was timed and her eye fixations recorded while she decided whether two figures depicted the same object or two objects that were mirror images of each other The stimuli were three drawings shown on the left hand side in Fig 1 as well as their mirror images for a total of six basic figures In the Same trials the left hand figure could be rotated clockwise 180 or less in the picture plane to bring it into complete congruence with the right hand figure The amount of rotation necessary to bring the two figures into congruence varied from 0 to 180 in steps of 20 for a total of 10 possible angular disparities To construct a Different pair the right hand figure of a Same pair was replaced by its mirror image isomer The mirror image figure was constructed by reflecting the original figure through a plane in three dimensional space see Metzler amp Shepard 1974 There was a Same and a Different pair for each of the six basic figures at each of the 10 angular disparities for a total of 120 pairs of stimulus figures The two figures were displayed side by side with the left hand figure randomly assigned to one of three orientations The center to center distance between the figures was 15 5 cm and each figure was between 10 and 10 5 cm wide The stimulus pairs
70. tion of TABLE FIG 9 Locus AND DURATION OF THE FIXATIONS Duration Fixation Figure Location msec 1 Left Central joint 234 Qs Central joint 518 3 Right Closed arm 367 4 Left Central joint 317 5 Closed arm 184 6 Right Open arm 434 F Left Closed arm 301 8 Central joint 25i 9 Closed arm 217 10 Right Open arm 635 11 Central joint 518 12 Central joint 234 13 Left Closed arm 585 14 Central joint 167 15 Right Central joint 568 456 JUST AND CARPENTER the segment is schematic The representation must include information about the segment s absolute orientation in space as well as some defining feature such as its length or whether it is a closed or open arm This informa tion can be efficiently represented as the vector formed by the major axis of the segment Moreover if the vector has its initial point at the origin of the reference frame then the segment can be represented by the spherical coordinates of the end point of the vector For example an open arm might be represented OPEN r 6 where r is the length of the segment and 0 and define the orientation of the segment The initial search process The scan paths indicate that the search for corresponding segments uses a simple heuristic Once a segment of one figure has been identified then the search for the corresponding segment starts in the corresponding location of the other field For example if the long arm is in the upper right hand c
71. trial unless the coordinates of the eye spot were very close to the fixation point This contingent aspect of the presentation assured that the apparatus was calibrated at the beginning of each trial The gaze contingent programming also allowed more sophisticated stimulus presentations in experiments to be described later As well as monitoring the fixations on line the system also produced a videotape record of the eye spot superimposed on the stimulus field Since the amount of deflection of the eye spot varied with the curvature of the cornea the defiection was normalized by setting the viewing distance individually for each subject The viewing distance was always between 53 and 68 cm so that each figure subtended about 10 of visual angle and the center to center distance between the two figures subtended about 15 Subjects initiated a trial by fixating a point in the middle of the left hand side of the screen and pushing a ready button Before each trial the eye spot was calibrated with respect to this fixation point The fixation point disappeared after calibration and half a second later the stimulus appeared The subject responded Same or Different by pressing one of two microswitches with the index and third finger of her dominant hand The stimulus presentation and timing of the response were monitored by the computer Head movements were minimized by using a bite bar The 120 stimuli were presented in a random order and distri
72. umed by initial search 39 by transformation and comparison and 26 by confirmation The remaining 14 was distributed between saccades about 10 and remaining other fixations about 4 that did not fit any of the three categories The durations in Fig 7 panels a through e add up to the total time shown in Fig 6 Just as the total reponse latency can be decomposed so can the switches in a fixation between the two figures be ascribed to each of three processing stages As Fig 3 showed subjects repeatedly looked back and forth between the two figures and the number of such switches between the figures increased with angular disparity As Table 1 shows the number of switches associated with the search stage remained quite low usually one or less at all disparities Most of the switches occurred during the transform and compare process during which the number of switches increased monotonically with the angular disparity The switching data from this stage will play a key role in the development of the model Finally the switches during confirmation increased with angular disparity but not as much as for transformation The classification procedure also categorizes the switches that occur if the transition from one stage to 452 JUST AND CARPENTER TABLE 1 DISTRIBUTION OF SWITCHES IN 100 TRIAL SAMPLE Mean number of switches during Transfor Switches Angular mation Transition not disparity Initial and Confir

Just (1976) Eye fixations and cognitive processes

Contents

Download Pdf Manuals

Related Search

Related Contents