Temporal organization of pattern structure

Transcription

1 Memory & Cognition 1979,7 (3), Temporal organization of pattern strctre STEPHEN K. REED and JAMES L. BROWN Case Western Reserve University, Cleveland, Ohio Two pattern reprodction experiments examined the relations among the figral goodness of a pattern, the organization of two parts within the pattern, and the interpart interval (ISI), which ranged from 40 to 200 msec. Ifthe parts contained connected line segments, performance was slightly better (3%-5% gain in accracy) at a 40-msec ISI than at a 200 msec ISI. If the parts contained nconnected line segments, reprodction accracy of the first part declined sharply between 40 and 200 msec. These reslts were interpreted by assming that the parts were perceived as a single whole pattern at a 40-msec ISI bt as two separate patterns at a 200 msec ISI. One srprising finding, the lack of an interaction between figral goodness and ISI, was explained in terms of a response bias in favor of figrally good patterns. A secondary maniplation revealed that a part was more accrately reprodced in a good figre context than in a poor figre context bt was most accrately reprodced when it appeared alone. The strctral organization of patterns, emphasized by Gestalt psychology, contines to attract the attention of psychologists. Experiments on pattern strctre are smmarized by Garner (1974), and the relation of strctral theories to other theories of pattern recognition is discssed by Reed (1978). The present series of experiments was motivated by Palmer's (1977) research on how the organization of a pattern into component parts affects the synthesis of the parts. Sbjects in Palmer's experiment saw two parts, each consisting of three straight lines connecting dots in a 3 by 3 dot matrix. The two dot rnatrices were adjacent and the task was to synthesize the two parts into a whole pattern. Palmer fond that abot 1.5 sec were needed to synthesize the parts if they had a high level of organization, bt abot 4.0 sec were needed if the parts had a medim or low level of organization. The high level of organization was the most natral way of organizing the patterns according to both sbjects' ratings and Palrner's theoretical predictions. One way of extending the paradigm sed by Palmer (1977) wold be to ask sbjects to drawa pattern from two parts that occrred seqentially rather than simltaneosly. Both Haber (I978) and Hochberg (1978) have recently emphasized the importance of integrating sccessive glances in forming a description of a pattern. Haber arges that since the resoltion of fine detail is only possible in the center of the visal field, several glances are necessary to achieve a clear representation of the visal scene. Also, since mch of perception occrs while the head and body are in motion, the continos transformations of the visal field have to be integrated into a coherent percept. Reqests for reprints shold be sent to Stephen K. Reed, Department of Psychology, Case Western Reserve University, Cleveland, Ohio We appreciate the comments of Grover C. Gilmore on an earlier draft of the manscript. Hochberg has arged that the problem of how to combine sccessive sensory inpt is central to the stdy of vision, hearing, and toch. In the case of visal perception, Hochberg shows how the integration of sccessive glimpses is necessary both for the active viewing of stationary objects and the perception of motion pictres. The crrent experiments investigate the temporal organization of pattern strctre by sing a modification of the seqential part techniqe designed by McFarland (1965). McFarland sccessively presented either the three sides or the three angles of a triangle. Each side was presented for lo msec, and the sbject's task was to jdge whether the parts appeared as simltaneos, overlapping, or sccessive. The percent "simltaneos" jdgments declined to 0% as the interpart interval (ISI) increased to 100 msec, and the percent "sccessive" jdgments increased to almost 100% as the ISI increased to 200 msec. Similar reslts were fond for presenting the three angles, except that the parts appeared joined more often when the sides were presented rather than angles. The stimli sed in the present experiments were similar to the line patterns sed by Palmer (1977). The two parts were seqentially presented in a tachistoscope, separated by dark intervals ranging from 40 to 200 msec. The task reqired that the sbjects draw the entire sixsegment pattern and mark the three segments that occrred in the first exposre. Or expectations were that sbjects wold perceive the pattern less as a whole and more as two separate parts as the ISI increased. The primary prpose of or stdy was to investigate whether varios strctral characteristics of patterns wold interact with the temporal separation of the parts. We distingish between two kinds of strctral characteristics. The first refers to the figral goodness of the whole pattern and is the major independent variable in the experiments reviewed by Garner (1974). Copyright 1979 Psychonomic Society, Inc X/79/ $01.15/0

2 206 REED AND BROWN The second refers to the figral goodness or organization of the two parts within the whole and is the major independent variable investigated by Palmer (1977). Figre 4 shows this distinction. The pattern in the first row was rated high in goodness and the pattern in the second row was rated low in goodness. The patterns in the left colmn have a high level of organization since they are divided into natral parts. The patterns in the right colmn have a low level of organization since they are divided into nnatral parts. The experiments were designed to test the hypothesis that the figral goodness of the whole pattern wold inflence performance more at short ISIs, and the organization of the parts wold inflence performance more at long ISIs. EXPERIMENT 1 Previos research has demonstrated that it is easier to encode and reprodce patterns scoring high on figral goodness than patterns scoring low on figral goodness (see Garner, 1974). Experiment I examined the effect of figral goodness on a reprodction task as the ISI was varied. If the pattern is perceived as a whole, the figral goodness of the whole shold inflence performance. If the pattern is perceived as two separate parts, performance shold be inflenced more by the figral goodness and organization of the parts within the pattern. A second isse explored in Experiment 1 was the effect of context on perception. If high- and low-goodness pairs share a common part, it is possible to stdy whether the same part is reprodced more accrately when it occrs in a high-goodness (H) pattern than when it occrs in a low-goodness (L) pattern. Method Stimli. Two scaling sessions were eondeted to scale the figral properties of the stimli sed in Experiment 1. Sbjects in both stdies were ndergradates at Case Western Reserve University who were tested in a single grop session. The first stdy asked 32 sbjects to jdge the figral goodness of 54 six-segment patterns, eonstreted by conneeting dots in a 3 by 3 matrix. The patterns eonsisted of 27 pairs that shared a common three-segment part. The patterns were drawn in an attempt to make one member of a pair a figrally good pattern and the other member a figrally bad pattern. Following the procedre sed by Garner and Clement (1963), sbjects were instrcted to rate each pattern on a 7-point scale ranging from 1 for the best (well strctred) patterns to 7 for the poorest (poorly stretred) patterns. The patterns were photocopied to form test booklets, and sbjects were allowed to work at their own pace. After eompleting the first part of the task, sbjects divided eaeh pattern into two mtally exclsive three-segment parts, chosen to be the "best" or the "most natral" parts within the pattern. Their divisions were sed to seleet patterns for Experiment 2. The average figral goodness ratings were sed to seleet the 12 pairs that best satisfied the criterion that one member of a pair have a low rating and one member have a high rating. Figre 1 shows the 12 pairs, divided into two three-segment parts to illstrate that each pair of patterns shared a common part. The 24 patterns were sed in Experiment 1..,, LN,, L} 1H I L, [-.--; ll. ',, 4H 7H 4L 7L 10 H 10 L Jj I J J,./. I!..J 2H 2L /7/7 ",,_', L_J 5H 5L 3H 3l,' '! l7 '--., 4 L, * I /,, " BH BL 9H 9L, ", I ', /t\.l >C I --.,. ("-; r/ l'1 J 6H -, v '--. I v. IIH IIL 12H 12L Figre 1. Patterns sed in Experiment 1. The prpose of the second scaling session was to scale the ftgral goodness of each part and measre the organization of the parts within the whole. Ten ndergradates received test booklets containing the 36 parts shown in Figre 1. They were instreted to rate each of the three-segment parts on a 7-point scale of figral goodness. The 36 parts were presented in a random order and were not shown in the context of the whole pattern. The seeond half of the task sed the techniqe of Palmer (1977) to measre the organization of the parts within the whole. Each of the 24 patterns was divided into the two parts shown in Figre 1. Each item consisted of three dot matrices with the whole pattern in the left matrix and the two parts in the other two matrices. Sbjects were asked to jdge how obvios the parts were within the pattern. Ratings were made on an l l-point scale ranging from 0 for a very bad organization to 10 for a very good organization. Table 1 shows the average ratings obtained in the two scaling sessions. The average figral goodness ratings of H patterns was 2.0, and the average figral goodness ratings of the L patterns was 5.1. The H and L patterns did not differ significantly in the figral goodness ratings assigned to their parts or in the organization measre of how well the two parts fit within the whole pattern. The average rating of the parts that differed in eaeh pair was 3.3 for H patterns and 3.7 for L patterns [t(22) =.83). The average organization rating was 6.3 for H patterns and 5.9 for L patterns [t(22) =.40]. Procedre. Eight sbjeets participated in the experiment. All were ndergradates at Case Western Reserve who received corse credit. Eaeh experiment consisted of three 50-min sessions given on different (sally consective) days, Sbjects were told that they wold be shown two three-segment parts separated by abrief interval. Their task was to draw a sixsegment pattern, marking the three lines that occrred in the first exposre. The answer sheet contained 3 by 3 dot matrices, and sbjects drew the patterns by eonnecting dots. They were reqired to always draw six lines and mark three lines, gessing when ncertain. A Gerbrands three-ehannel tachistoseope was sed to present the 24 patterns shown in Figre 1. Eaeh trial consisted of a fixation field presented for 500 msec, a 500-msee dark interval, abrief presentation of the first part, a dark interval of 40, 80, 120,160, or 200 msee dration, and abrief presentation ofthe second part followed by a dark interval. A part was presented for either 5 msec (for sbjects) or 10msee (for sbjects) in order to determine whether slight differences in exposre dration wold inflence performance. The fixation field was 6L

3 TEMPORAL ORGANIZAnON 207 Table 1 Ratings of Figral Goodness for the Whole Pattern, the Two Parts, and the Organization of the Parts within the Whole Whole Pattern Common Part Different Part Organization Pattern H L H L H L H L Average Note-Figral goodness was jdged on a 7-point scale ranging from I for weil strctred pattems to 7for poorly strctred pattems. The organization 0 f parts was jdged on an li-point scale renging from 0 for very poor organization to I 0 for very good organization: a 3 by 3 dot matrix sbtending a visal angle of 55 min. The three-segment parts were presented within a dot matrix of the same size. The fixation field and the two parts were presented at a lminance of 4.4 ftl. Each of the three sessions began with six practice trials on patterns that differed from the patterns sed in the experiment. Each of the 24 patterns was presented twice at ISls of 40, 80, 120, 160, and 200 msec. Two presentations were sed in order that each part wold occr once in the first exposre and once in the second exposre. There were a total of 16 presentations at each interval for each of the three sessions. The ISI did not change dring a block of 16 trials. The order of presenting the blocks varied both across sessions and sbjects. Reslts The effect of exposre dration was nonsignificant (F< I), so the data from a11 eight sbjects were combined. Figre 2 shows the percent of lines correctly f- U 1LI BO 0 f- Z 1LI o 1LI a H2 -- _eh1 $>----<>11 50 '----..L-...l l.. l.-_---l INTERSTIMULUS INTERVAL (MsEC) Figre 2. Percent of lines drawn correctly from the first part (l) and second part (2) for high-goodness (H) and low-goodness (L) patterns in Experiment I. drawn at each of the five ISIs. The reslts are shown separately for the Hand L patterns and for the firstexposre and second-exposre parts. An analysis of variance revealed that ISI had a significant main effect [F(4,24) =7.73, p<.01] and interacted significantly with the order of presentation of the parts [F(4,24) =9.55, p<.01]. The order of presentation also had a significant main effect [F(l,6) =13.04, p<.05], with the second part being drawn more accrately than the first part. These findings are primarily the reslt of the sbstantial metacontrast effect illstrated in Figre 2. Accracy in reprodcing the lines from the first part was greatly affected by ISI, whereas accracy on the second part was relatively independent of ISI. The metacontrast mask-ng of the first part was greatest at an 80-msec ISI for H patterns and at a 120-msec ISI for L patterns. One somewhat srprising finding was that the overall level of performance at a 40 msec ISI was not sbstantially better than performance at a 200-msec ISI. For H patterns, sbjects correctly drew 91% of the lines at a 40-msec ISI, compared to 87% at a 200-msec ISI. For L patterns, sbjects correctly drew 80% of the lines at a 40-msec ISI and 77% at a 200-msec ISI. Althogh these reslts were in the anticipated direction, a greater difference was expected. The effect 01' figral goodness was significant [F(I,6) = 35.30, p<.01], and the better performance on H patterns was maintained over a11 ISIs. The scaling reslts from Experiment I revealed that althogh Hand L patterns differed in the goodness ratings assigned to the whole patterns, they did not differ in the goodness or organization of the parts. One might, therefore, expect a Goodness by ISI interaction, with thc differcnce between Hand L patterns being more evident at short ISIs when the patterns are more likely to be perceived as wholes rather than as separate parts. However, there was Iittle cvidence for sch an intcraction [F(4,24) = 1.44, p >.05J. Sbjects were

4 208 REED AND BROWN more accrate in drawing the H patterns at all ISI separations. Another interaction of interest concerns whether the common part was reprodced more accrately when it occrred in a H pattern than in a L pattern. The overall reslts revealed that 86% of the lines in H patterns were correctly drawn, compared to 77% ofthe lines in L patterns. This difference was the reslt of an 8% difference on the common parts and a 10% difference on the different parts. Since this interaction was not significant [F(l,6) = 2.57], it seems reasonable to conclde that there was a significant context effect-the same part was more accrately reprodced when it occrred in a H pattern.' In addition to drawing the six lines that composed the pattern, sbjects were asked to always mark the three lines that occrred in the first exposre. Figre 3 shows their ability to identify in which exposre the lines occrred. Only lines that were correctly drawn are inclded in these reslts. For example, if a sbject marked two of three correct lines from the first exposre and one of two correct lines from the second exposre, his percentages wold be 67% (pper crve) and 50% (lower crve). The reslts show that, at a 40-msec ISI, 70% of the correct lines from the first exposre were marked as having occrred in the first exposre. This percentage increases to 82% at an 80-msec ISI bt does not improve with Ionger ISIs. Discssion The significant effect of figral goodness on performance is consistent with many previos research findings (Garner, 1974). It is somewhat srprising, however, that figral goodness did not interact with ISI. Figrally good patterns were reprodced more accrately even when it was more likely that the patterns were perceived as two three-segment parts rather than as a six-segment whole. Neither of the two most common explanations of figral goodness wold seem capable of explaining the lack of interaction between figral goodness and ISI. One explanation proposes that figrally good patterns can be encoded more qickly then figrally poor patterns (see Bell & Handel, 1976). An alternative explanation proposes that figrally good patterns are easier to remember (see Pomerantz, 1977). Both factors cold have contribted to the advantage of figrally good patterns at a 40-msec ISI, when it is likely that patterns cold be perceived as a whole. However, since the average figral goodness of the parts did not differ for H patterns and L patterns, both explanations shold predict no difference between Hand L patterns when the patterns are encoded and reprodced from the two parts. Another possible explanation of the reslts, the visal synthesis explanation (Palmer, 1977), wold also have difficlty acconting for the advantage of H patterns over L patterns at a 200-msec ISI. According to this I- 60 z w w EXPERIMENT' EXPERIMENT " 6-_--6 EXPERIMENT 2 O'----_..l- -L.. ---I INTERSTIMULUS INTERVAL l..-_---j 100 (MSECI 200 Figre 3. Percent of correctly drawn lines indicated as occrring in the first part, explanation, sbjects wold synthesize the two parts to form the whole pattern, before attempting to reprodce the pattern. Palmer has shown that the difficlty of synthesizing a pattern is determined by whether it is divided into its most natral parts, bt this variable was controlled across Hand L patterns. The visal synthesis hypothesis cold accont for the present reslts if the figral goodness of the whole pattern determines the ease of synthesis in addition to the organization of the parts. Recent data by Thompson and Klatzky (1978) offer spport for this interpretation. Irreglar forms were very difficlt to synthesize when compared with more familiar forms, sch as triangles, trapezoids, and parallelograms. The H forms shown in Figre I are more familiar and may have been easier to synthesize. There are other reasons, however, that the visal synthesis explanation is nattractive. Visal synthesis is a very difficlt mental task, particlarly when a pattern is not divided into its most natral parts (palmer, 1977). Thompson and Klatzky (1978), in fact, qestioned whether even the more familiar forms were completely synthesized, since sbjects' ability to verify whether a part was present in a pattern was faster if the part matched one of the parts presented in the synthesis task. There are two aspects of the present paradigm that might frther discorage sbjects from attempting the kind of mental synthesis stdied by Palmer and by Thompson and Klatzky. First, the synthesis wold have to combine two parts that were recently presented bt no longer present. In contrast, the task sed by Palmer and by Thompsonand Klatzky reqired the synthesis of parts that were present. Second, sbjects in Experiment 2 were not reqired to

5 TEMPORAL ORGANIZATION 209 synthesize the parts bt were allowed to reprodce the pattern sing whatever method they chose. If synthesis is a difficlt mental task, and if its difficlty is increased by having to image the parts, there shold be little incentive for sbjects to synthesize the whole pattern before reprodcing it. The interpretation we favor proposes a response bias for familiar or figrally good forms. The response-bias hypothesis states that sbjects attempt to constrct a farniliar or figrally good pattern when they are ncertain ab ot which lines were presented. This interpretation is consistent with Bear's (1973) research on predicting the elements of a pattern. He hypothesized that if good figres are characterized by well organized elements, each element of a good pattern shold be strongly sggested or implied by the other elements. This elose relationship shold not exist for the elements of a poor figre. This differential predictability was tested by showing sbjects a for-dot sbpattern and asking them to fill in the missing fifth dot. The dots comprising a figrally good five-dot pattern were gene rally more predictable than the dots comprising a figrally poor pattern. It shold be noted that a bias toward drawing figrally good patterns shold exist at all ISIs and cold therefore accont for the data in Experiment 1. EXPERIMENT 2 The prpose of Experiment 2 was to investigate the organization of the parts within the pattern as an independent variable. The constraint that Hand L patterns share a common part in Experiment I preelded the systematic constrction of different part strctres for each pattern. Sbjects in a scaling task were asked to rate the organization of alternative parts based on the techniqe sed by Palmer (1977). Pairs of patterns differing in levels of organization were then sed in a reprodction task. Method Stimli. Two sealing sessions were needed to seleet the stimli for Experiment 2. The prpose of Experiment 2 was to examine the effeet of dividing a pattern into different eombinations of parts, having either a high or a low level of organization. Eaeh of the original 54 patterns was divided into three different two-part eom binations, with the two parts appearing to the right of the whole pattern. The three partitions of a pattern were eonstreted to represent a high, medim, or low level of organization following the proeedre sed by Palmer (1977). The high level of organization was based on the most freqently constreted partition of the pattern fond in Experiment 1. Sbjeets were asked to jdge how obvios or natral the parts were within the pattern, sing an ll-point seale ranging from 0 for a very bad organization to 10 for a very good organization. Thirteen ndergradates made a total of 162 jdgments, The second session asked a different grop of 26 ndergradates to rate the figral goodness of the threesegment parts ereated in the previos session. Sbjeets rated the 224 parts on a 7-point seale. The parts were presented in a random order and were not shown within the context of the whole pattern. The ratings were sed to seleet 18 pairs of patterns for se in Experiment 2. Each pair consisted of the same pattern broken down into either a high (natral parts) or a low (nnatral parts) level of organization. Nine pairs were figrally good patterns and nine pairs were figrally poor patterns, based on the ratings from Experiment 1. The criterion for selecting patterns was that the ratings on both figral goodness of the whole pattern and the organization of the parts be at the ex treme ends 0 f the scale. The stimli eonsisted of 18 pairs of patterns. Nine pairs were figrally good patterns (average rating =1.90) and nine pairs were figrally poor patterns (average rating = 5.28). Eaeh pair eonsisted of the pattern divided into a high and a low level of organization. The two parts representing the high level of organization were the most freqent divisions made by the sbjeets in Experiment 1. The average organization rating was 9.13 for figrally good patterns and 8.72 for figrally poor patterns. The two parts representing the low level of organization had at least one line that was not connected to the other two lines in apart. The nmber of nconnected lines ranged from I to 6, with an average of 3.2 for figrally good patterns and 1.8 for figrally poor patterns. In contrast, all lines were conneeted in the parts eomprising the high level of organization. The average rating for the low-organization patterns was 2.71 for figrally good patterns and 3.47 for figrally poor patterns. Figre 4 shows an example of the for types of patterns." Procedre. The proeedre was similar to that sed in Experiment 1. Each trial consisted of a 500-msec presentation of the fixation field (the dot matrix), a 50Q-msec dark interval, a 7-msec presentation of the first part, a variable dark interval, a 7-msec presentation of the second part, and a dark interval. The ISI was 40, 120, or 200 msec. Sbjects were asked to always draw six line segments and mark the three that occrred first. Seven sbjeets participated in the three sessions of the experiment. The 36 stimli were presented twiee at each ISI with each of the two parts oecrring onee dring the first exposre and onee dring the second exposre. A session eonsisted of one block of 24 trials for each of the three ISIs. The order of the three blocks varied both across sbjects and sessions. Stimli ocerred in a random order nder the constraint that the same stimls eold not appear on sccessive trials. Reslts An analysis of variance revealed that figral good- r---"1' HIGH GOODNESS HIGH ORGANIZATION "(;: v. LOW GOODNESS HIGH ORGANIZATION HIGH GOODNESS LOW ORGANIZATION "",L l,.i Figre 4. Examples of patterns sed in Experiment 2. LOW GDODNESS LOW ORGANIZATION

6 ' 210 REED AND BROWN ness [F(I,6) = 22.90, p<.01], organization of parts [F(I,6) = 47.18, p <.01], order of presentation [F(I,6) =11.50, p<.05], and ISI [F(2,12) =10.10, p<.01] had significant effects on performance. Figral goodness did not interact with ISI (F< 1), spporting the reslts from Experiment 1 in which the better performance on figrally good patterns was maintained over all intervals of part separation. The reslts of the experiment are shown in Figre 5. Figre Sa shows sbjects' performance on patterns that were jdged to have a high level of organization. Althogh the level of performance is slightly lower, the effects of the experimental variables are very similar to the effects obtained in Experiment 1 (see Figre 2). There is a slight improvement in accracy when the parts are separated by 40 msec rather than 200 msec. Sbjects correctly reprodced 76% of the lines from figrally good patterns at a 40-msec separation, compared to 71% at a 200-msec separation. They were correct on 63% of the lines from figrally poor patterns at a 40-msec separation, compared to 60% at a 200-msec separation. The relative lack of effect of ISI on organizing the natral parts of a pattern changes dramatically when the pattern is poorly organized. Figre Sb shows that an ISI of 120 msec or greater prodces a large decrement in performance when sbjects attempt to reprodce the first part of a poorly organized pattern. A significant Organization by Interval interaction [F(2,12) = 10.68, p <.01] indicates that varying the ISI mainly inflences poorly organized patterns. The decline in performance with increasing ISI is most apparent for the first part, as revealed by the interaction between organization and order of presentation [F(I,6) = 6.21, p<.05]. There was no interaction between figral goodness and the organization of the parts (F < 1). A poorly organized pattern has similar effects on figrally good and figrally poor patterns, althogh performance is better on the former (see Figre Sb). One interesting finding in Experiment 2 reqires comparing Figres Sa and Sb at the 40-msec ISI. Althogh the organization of the pattern had a highly significant main effect on performance, it had little effect at a 40-msec ISI. For figrally good patterns, sbjects correctly drew 76% of the lines from well organized patterns and 74% of the lines from poorly organized patterns. For figrally poor patterns, the sbjects correctly drew 63% of the lines from well organized patterns and 61% of the lines from poorly organized patterns. These reslts cold be explained by the assmption that sbjects were perceiving the pattern as a whole at this ISI, so the part strctre of the pattern was nimportant. Figre 3 shows sbjects' ability to identify which line segments occrred in the first exposre. Temporal discrimination was poorer in Experiment 2 than it was in Experiment 1. Performance was elose to the 50% chance level at a 40-msec ISI bt improved with longer delays between the two parts. It is not entirely elear why temporal discrimination and reprodction accracy were at lower levels in Experiment 2. One possibility is that there were more patterns in Experiment 2, and the patterns were seen less often, making it more difficlt to learn the pattern set. Some of the patterns were different from those sed in Experiment 1 and may have been more difficlt. Another possibility is simply that there were individal differences among the two grops of sbjects w 60 0 z w 50 w Q. <, (0) _H L1 L o z tj 50 w Q. t b) \-- \ , _- " H2,,, L L-...L INTERSTIMULUS INTERVAL (MSECl 30 L-..L..-...l INTERSTIMULUS INTERVAL (MSEC) 200 Figre 5. Percent of lines drawn correctly from the first part (1) and second part (2) for high-goodness (H) and low-goodness (L) patterns in Experiment 2. (a) High organization. (b) Low organization.

7 TEMPORAL ORGANIZATION 211 Discssion The lack of interaction between figral goodness and ISI replicates the reslts of Experiment 1. The slight advantage of the 40-msec ISI over the 200-msec ISI was confirmed for patterns having a high level of organization (Figre 5a). For figraily good patterns, reprodction accracy was 5% higher at a 40-msec ISI than at a 200-msec ISI. For figrally poor patterns, there was a 3% advantage at the 40-msec ISI. The reslts are almost identical to the 4% advantage for figra11y good patterns and the 3% advantage for figrally poor patterns fond in Experiment 1. Ths in ail for cases, the prob ability of reprodcing the lines of a pattern at a 40-msec ISI ranged from 3% to 5% higher than at a 200-msec ISI. The most salient finding in Experiment 2 was that the temporal separation of the parts dramatically affects performance when a pattern is poorly organized. Figre 5b shows that accracy in reprodcing the first part declines sharply with increasing ISI. It is likely that sbjects had great difficlty in remembering the first part after the second part was presented. The parts in the low-organization condition contained many nconnected lines, which shold be mare difficlt to remember. In contrast, a11 lines were connected in the parts of high-organization patterns, and most of the lines were connected in the parts of the patterns sed in Experiment 1 (see Figre 1). Ths, the nmber of nconnected lines wold seem to be an important variable in determining sbjects' ability to reprodce a pattern as a fnction of the temporal separation of its parts. GENERAL D1SCUSSION Or intention in this conclding section is to smmarize the principal findings, offer a tentative interpretation, and sggest experimental paradigms for testing or interpretation. Smmary of the Reslts Figral goodness. Figrally good patterns were reprodced more accrately than figrally poar patterns at ail ISIs. Better performance on figrally good patterns occrred even at intervals (sch as 200 msec) when sbjects perceived two separate parts rather than the whole pattern. This occrred even when the average organization of the parts did not differ between the two sets of patterns (Experiments 1, 1', and 2). Context. A three-segment part was reprodced more accrately when it occrred in a H pattern, than when it occrred in a L pattern (Experiments land 1'), However, performance was best when a part was presented alone (Experiment 1'). Organization of the parts. The organization of the parts did not inflence performance at a 40-msec ISI. However, reprodction accracy of the first part declined dramatica11y with increasing ISI for poorly organized parts. Highly organized parts were affected only slightly, bt consistently, with increasing ISI. Interstimls intervai. The most interesting comparison involved the extreme intervals since metacontrast masking occrred at the intermediate intervals. For the patterns sed in Experiment 1 and the patterns having weil organized parts in Experiment 2, there was a 30/0'5% improvement in performance at a 40-msec ISI over performance at a 200-msec ISI. For patterns having poorly organized parts, reprodction of the first part was either 21% (for figraily poor patterns) or 29% (for figraily good patterns) higher at the 40-msec ISI. A Tentative Explanation The present series of experiments was designed to test how several task parameters wold inflence performance on a seqential part task. Frther research is necessary to determine the psychological processes sed by sbjects to perform the task. We wold like, however, to offer what we feel is a plasible explanation and sggest some extensions of the paradigm to test the explanation. The ISIs were chosen in an attempt to create a task in which perception wold continosly change from perception of a whole pattern at a 40-msec ISI to the perception of two separate parts at a 200-msec ISI. Althogh the change may occr along a continm, metacontrast masking at msec makes the intermediate intervals less interesting, so we have mainly focsed or attention on the two extreme points, 40 msec and 200 msec. Several aspects of the data are consistent with the hypothesis that sbjects perceive the pattern as a whole at a 40-msec separation and as two separate parts at a 200-msec separation. The clearest evidence is provided by comparing the low- and high-organization conditions of Experiment 2. At a 40-msec ISI, it did not make any difference whether the pattern was divided into well organized or poorly organized parts. This finding is consistent with the proposal that the pattern was perceived as a whole. However, at a 200-msec ISI, patterns that were divided into weil organized parts were reprodced less weil than patterns that were divided into poorly arganized parts. These data prodced a highly significant interaction between organization and ISI. The reslts of Eriksen and CoIlins (1967) also spport the interpretation that a pattern becomes increasingly difficlt to perceive as a whole as the ISI increases from oto 100 msec. Their stimli consisted of pairs that were perceived as a random collection of dots when viewed alone or as a three-letter nonsense syllable when sperimposed. A niform decline in sbjects' ability to report the nonsense syilable with increasing temporal separation sggested an increasing tendency to perceive the two exposres as separate rather than sperimposed.

8 212 REED AND BROWN If we assrne that the above explanation is correct, we can then ask if there is any advantage in perceiving the entire pattern rather than its separate parts. Experiments 1 and 2 sggest that when patterns are at least moderately weil organized, perceiving the whole pattern prodces a slight bt consistent advantage, casing a 30/0-5% improvement in performance. It is likely that perception of the entire pattern allows for more efficient groping and chnking of line segments, making it easier to remember the pattern. The reslts of Experiment 2 sggest that perceiving the separate parts prodces a dramatic decline in performance when the parts are poorly organized (Figre Sb). One distingishing characteristic of this particlar set of patterns is the nmber of nconnected lines. There was an average of 2.5 nconnected lines in the two parts, compared to 0 nconnected lines for the highorganization patterns. There was a sbstantial range in the average organization ratings assigned to the patterns sed in Experiment 1, bt the nmber of nconnected lines averaged only.3 for the two parts. The poor performance in reprodcing the poorly organized parts was cased by sbjects' difficlty in reconstrcting the first part. This finding is consistent with the reslts of Hines and Smith (1977), who demonstrated that an attended random shape cased sbstantial retroactive interference in identifying a random shape that occrred msec earlier. If sbjects did not have to identify the second shape, there was virtally no interference in identifying the first shape. Or reslts indicate that a second pattern makes it more difficlt to reprodce the first pattern (Experiment 1'), bt the magnitde of the effect is greatly inflenced by the strctre of the two parts (Experiment 2). Perhaps or most srprising finding was the lack of interaction between the figral goodness of the whole pattern and ISI. Sbjects were more accrate in reprodcing figrally good patterns even when perceived as parts. The reslt was srprising becase the strctre and organization of the parts did not differ between figrally good and figrally poor patterns. We considered several possible explanations of this finding when discssing the reslts of Experiment 1 and selected the one postlating a response bias in favor of figrally good patterns. The response-bias hypo thesis states that sbjects fill in missing featres by attempting to constrct a figrally good pattern. To smmarize or interpretation of the reslts, the effect of increasing the ISI from 40 msec to 200 msec is to case sbjects to perceive the pattern as a six-segment whole at one extreme (a 40 sec separation) and as two three-segment parts at the other extreme (200 msec). Following the application of Rmelhart's (1970) featre recognition model to metacontrast masking, we believe that featre recognition occrs simltaneosly from the two parts at short separations (40 msec or less), bt the featres from the first part are recognized before the second part is presented at longer separations (200 msec or more). At intermediate separations ( msec), sbjects are still in the process of forming a strctral description of the first part when the second part occrs. Recognizing the featres of the second part therefore interferes with forming adescription of the first part, casing metacontrast masking. Becase the line segments in a six-segment whole pattern can be more efficiently groped and chnked than the line segments in the two parts, reprodction accracy is higher at a 40 msec ISI than at a 200-msec ISI. If the organization of the pattern is divided into poorly organized parts, retroactive interference cased by the second part makes it extremely difficlt to remember the first part. When sbjects perceive two separate parts, they do not mentally synthesize the parts before attempting to reprodce the whole. They therefore constrct a six-segment pattern from the two parts bt are biased toward constrcting a figrally good pattern when ncertain abot some of the featres. Frther Research There are a nmber of possible extensions of the crrent paradigm that cold be sed to test or tentative explanation. The visal synthesis paradigm offers a methodology for examining whether a person has encoded the pattern as a whole or as two separate parts. Following the techniqe designed by Palmer (1977), sbjects cold be shown a six-segment test pattern after seeing the two parts. The test pattern wold be either similar or identical to the pattern formed from the parts and the task wold be to decide if the two patterns were identical. A con trol condition wold present a six-segment whole pattern followed by the test pattern. We wold expect that accracy and reaction times for the whole and part conditions wold be sirnilar at a 40-msec ISI, bt sbjects' performance wold decline at a 200-msec ISI becase sbjects wold not know what the whole pattern looked like. Another prornising paradigm wold se two line segments as the test pattern. This paradigm was designed by Cnningham (Note 1) to bild a tree strctre showing the relations among the line segments of a pattern. Cnningham sed the time needed to verify that two line segments were contained within a pattern as a measre of the "distance" between the two line segments. For example, line segment a wold be eqidistant from lines b and c if reaction times to pairs ab and ac were the same. Now divide the pattern into two parts sch that lines a and b belong to the same part, bt c belongs to a different part. If the two parts are perceived as a whole pattern, reaction times to pairs ab and ac shold remain the same. However, if the pattern were encoded as two separate parts, Thompson and Klatzky's (1978) reslts wold imply that the verification time for ab shold be less than for ac.

9 TEMPORAL ORGANIZATION 213 Systematic testing and the se of Cnningham's graph techniqes shold enablc one ot examine how the memory representation changes as a fnction of the ISI. One possible test of the response-bias interpretation, proposed to explain the lack of an interaction between figral goodness and ISI, wold be to delete a Une segment on some of the trials. Ths, sbjects wold expect and draw a six-segment pattern bt wold be shown only five lines on some trials. We wold expect on the basis of Bear's (1973) findings that sbjects wold correctly draw the rnissing segment more often when a figrally good pattern was presented than when a figrally poor pattern was presented. In conclsion, we have attempted to explore the temporal organization of line patterns by examining the relations among figral goodness, organization of the parts, and the ISI. We proposed a tentative explanation of the findings and sggested some extensions of the paradigm to test or explanation. REFERENCE NOTE I. Cnningham, J. P. Memory for simple visal patterns. Paper presented at the annal meeting of the Psyehonomie Society, Washington, D.C., November REFERENCES BEAR, G. Figral goodness and predictability of figral elements. Perception & Psychophysics, 1973, 13, BELL, H. H., & HANDEL, S. The role of pattern goodness in the reprodetion of baekward masked patterns. Jornal ofexperimental Psychology: Hman Perception and Performance, 1976, 2, ERIKSEN, C. W., & COLLINS, J. F. Some temporal charaeteristics of visal pattern perception. Jornal of Experimental Psychology, 1967,74, GARNER, W. R. The processing of information and strctre. Potomae: Erlbam, GARNER, W. R., & CLEMENT, D. E. Goodness of pattern and pattern ncertainty. Jornal of Verbal Learning and Verbal Behavior, 1963,2, HABER,R. N. Visal perception. In Annal review ofpsychology. Palo Alto, Calif: Annal Reviews, Ine., HINES, D., & SMITH, S. Reeognition of random shapes followed at varying delays by attended or nattended shapes, digits, and line grids. Jornal ofexperimental Psychology: Hman LearningandMemory, 1977,3, HOCHBERG, J. E. Perception (2nd ed.). Englewood Cliffs, N.J: Prentiee-Hall,1978. McFARLAND, J. H. Seqential part presentation: A method for stdying visal form pereeption. British Jornal ofpsychology, 1965,56, PALMER, S. E. Hierarehieal stretre in perceptal representation. Cognitive Psychology, 1977,9, POMERANTZ, J. R. Pattern goodness and speed of eneoding. Memory & Cognition, 1977,5, POMERANTZ, J. R., SAGER, L. C., & STOEVER, R. J. Pereeption of wholes and of their eomponent parts: Some configral speriority effeets. Jornal of Experimental Psychology: Hman Perception and Performance, 1977,3, REED, S. K. Sehemes and theories of pattern reeognition. In E. C. Carterette & M. P. Friedman (Eds.), Handbook of perception (Vol. 9). New York: Aeademic Press, RUMELHART, D. E. A mlti-cornponent theory of perception of briefly exposed stimls displays, Jornal of Mathematical Psychology, 1970,7, THOMPSON, A. L., & KLATZKY, R. L. Stdies of visal synthesis: Integration of fragments into forms. Jornal of Experimental Psychology: Hman Perception and Performance, 1978, 4, WILLIAMS, A., & WEISSTEIN, N. Line segments are pereeived better in a eoherent eontext than alone: An object-line effeet in visal pereeption. Memory & Cognition, 1978, 6, NOTES 1. Reeent research has indieated that, nder certain conditions, apart is reeognized better when it appears in the eontext of a pattern than when it appears by itself (pomerantz, Sager, & Stoever, 1977; Williams & Weisstein, 1978). This finding was tested in a slightly modified version of Experiment 1 sing a different grop of eight sbjeets. The procedre was identical except that only three ISIs were sed (40, 120, and 200 msee) and sbjeets had to reprodce only a single part on some trials (the fixation field replaced by the second part). The context prodced a signifieant main effeet [F(2,i4) =18.64, p <.01 J. When the part oeerred alone, 95% of the lines were drawn correctly, compared to 85% when it oeerred as part of a high-goodness whole and 78% when it oeerred as part of a low-goodness whole. A Newman-Kels analysis revealed that presenting apart alone differed significantly at the.01 level from presenting it in either eontext, and presenting it in a goodfigre eontext differed at the.05 level from presenting it in a poor-figre context. The latter finding eonfirms the facilitating effeet of a good-pattern eontext fond in Experiment 1. The effeet of context did not interact with ISI [F(4,28) =1.62], so the sperior performance in reprodeing apart that oeerred alone was not cased by metaeontrast masking when both parts oeerred. Sbjects' ability to ternporally discriminate the two parts, when both occrred, is shown in Figre 3 as Experiment 1'. 2. The average figral goodness rating of parts was 2.88 for high-organization patterns and 4.77 for low-organization patterns. In general, weil organized patterns tend to consist of figrally good parts and poorly organized patterns tend to consist of figrally bad parts. The crrent research does not attempt to independently evalate these two variables. (Revision accepted for pblication March 13, 1979.)