Effect of semantic clustering on the memory of spatial locations

Journal of Experimental Psychology: Learning, Memory, and Cognition 1986, Vol 12, No 2, 182-189 Copyright 1986 by the American Psychological Association, Inc 0278-7393/86/W0.75 Effect of Semantic Clustering on the Memory of Spatial Locations Stephen C Hirtle and Michael F Mascolo State University of New York at Albany The effect of altering the labels attached to points was examined in three experiments The first experiment measured the extent of clustering that occurs based on the labels alone This experiment also established norms for the remainder of the study In the second and third experiments, subjects were required to learn the locations of points The points were labeled in such a way as to suggest certain spatial clusterings It was shown that subjects cluster points with regard to the labels attached to the points and these clusters may be based solely on the labels attached to the points Furthermore, an alteration of the learning sequence to induce an alternate clustering showed no noticeable effect One dominant aspect within the field of cognitive psychology in recent years has been discussion on the analysis of mental representations in long-term memory (e.g., Anderson, 1978; Kosslyn & Pomerantz, 1977; Mehler, Walker, & Garrett, 1982; Podgorny & Shepard, 1978) As a context for studying mental representations, the encoding and processing of spatial information provides a well-learned and complex domain Thus, the study of cognitive maps has received the attention of developmental psychologists (e.g., Liben, Patterson, & Newcombe, 1981), environmental psychologists (e.g., Evans, 1980), and cognitive psychologists {e.g., Stevens & Coupe, 1978) Interesting in its own right, an understanding of the mental calculations involved in spatial tasks will also provide insights into the organization of complex, interlocking data structures and the processes that act upon that organization actually is In each case the superordinate relationship biased subjects into making incorrect judgments Maki (1981) and Wilton (1979) demonstrated a clustering effect by measuring the time required to verify directional statements Judgments made on across-cluster pairs were uniformly fast and free from congruity and distance effects when compared to within-cluster pairs Using artificial stimuli, McNamara (in press) found that both cluster membership, as determined by boundary markers, and Euclidean distance were important predeterminers of performance on three distinct tasks: spatial priming, directional judgments, and distance estimation In each of the studies above, the stimulus set consisted of predefined clusters such as cities within states or artificial regions with marked boundaries Two recent studies offer evidence that clustering extends to poorly differentiated areas Our lab (Hirtle & Jonides, 1985) provided evidence of hierarchies in the cognitive maps of a central city area without predefined boundaries Subjects in the study first completed a memorization task of city landmarks By noting the order in which the landmarks were recalled, we were able to infer a hierarchical tree for each subject The same subjects participated in several spatial tasks, such as distance estimation and distance classification, using the same city landmarks In each case, the subjects showed a clustering bias, rating within-cluster pairs as closer together than equally distant between-cluster pairs Recent work on spatial representations provides evidence that cognitive structures are critical in one's memory of space (Allen, 1981; Hirtle, 1985; Stevens & Coupe, 1978) Experimental evidence suggests that specific locations serve as reference points for the internal organization of space (Allen, Siegel, & Rosinski, 1978; Evans, Marrero, & Butler, 1981) and that space is segmented into clusters (Hirtle & Jonides, 1985; McNamara, in press) The influence of reference points is strong enough to result in asymmetrical judgments (Holyoak & Man, 1982; Sadalla, Burroughs, & Staplin, 1980), and the influence of segmentation, or clusters, is strong enough to reverse directional judgments (Stevens & Coupe, 1978) Segmentation occurs for both well-defined hierarchies and poorly differentiated areas Stevens and Coupe (1978) observed systematic errors in directional judgments between cities in different states or countries For example, Reno was judged to be northeast of San Diego, not northwest as it actually is, and Seattle was judged to be southwest of Montreal, not northwest as it Allen (1981) also found that clustering occurred in an area with no predefined boundaries In his study, adults, fifth graders, and second graders learned an unknown walk by slide presentation Subjects then made a comparative distance judgment For all age groups, the distances across clusters were judged as longer than the corresponding distances within a cluster, where clusters were defined by changes in the landscape Segmentation is not the only dominant factor that results in biased distance judgments Performance in a spatial task can also be biased by perceptions about landmarks, knowledge of names or images, and other nonspatial characteristics For example, spatial information is not encoded independently of verbal and visual information (Pezdek & Evans, 1979), color coding improves spatial performance (Evans, Fellow, Zorn, & Doty, 1980), travel time is better than objective distance as a predictor ofsubjectivedistance(MacEachren, 1980), and aroute containing high-frequency names is estimated as longer than a route con- The ideas for this study grew out of a discussion with John Jonides, Janellen Huttenlocher, and Steven Shevell a number of years ago Their inspiration is appreciated We would also like to thank Maihee Lee and Lisette Fantuazzi for their assistance in data collection This research was supported by NIMH Grant MH39912 to Stephen C Hirtle Correspondence concerning this article should be addressed to Stephen C Hirtle, Department of Psychology, State University of New York at Albany, 1400 Washington Avenue, Albany, New York 12222 182 183 SEMANTIC CLUSTERING laining low-frequency names (Sadalla, Staplin, & Burroughs, 1979) Thus, semantic information is seen to be important in two related ways First, it can assist in encoding spatial information when first acquiring spatial knowledge Second, semantic information can assist in later accessing spatial information The ease of access, in turn, alters the memory of the physical layout This is not to say that spatial attributes are ignored but rather that cognitive maps consist of both spatial and nonspatial attributes The present study extends previous work in several ways First, we examined the role of semantic labels on the formation of clusters Experiment established semantic categories of common landmarks through a sorting task From these categories, maps were constructed for the remainder of the study Experiment questioned if altering the semantic labels attached to locations would alter the memory of the space We hypothesized that distance estimates are biased by clustering and that clustering depends not only on the relative location of landmarks but also on the labels affixed to the locations Finally, we examined the role of experience in encoding spatial relations Experiment varied the clusters that subjects experienced in learning spatial relationships Because Experiments and study the acquisition of spatial knowledge, a map-learning task was employed using artificial maps Experiment To demonstrate that semantic labels bias memory for spatial locations, one must first establish natural clusterings that occur on the basis of names alone Subjects in this experiment sorted index cards according to the perceived likelihood that they would be located near each other in a typical city On completion of the task, the experimenter recorded the members of each resulting category onto a response sheet and then reshuffled the cards Subjects performed the task a total of three times Results The data for all three trials were entered into a 39 X 39 matrix, where the rows and columns represented the names of the 39 locations The cells of the matrix represented the number of trials in which a subject categorized a row location in the same group as the column location Thus, the scores contained in the matrix acted as an index of association between the names A hierarchical cluster analysis (Johnson, 1967) was performed to determine strong clusters of five or more items Both complete and single linkage methods of clustering were used Both methods resulted in similar groupings at the upper level Starting with the single linkage method, there were only two strong clusters of more than five items We defined a strong cluster as one with a minimum strength of 60 or greater, where the strength could range from to 123 These two clusters were {court house, town hall, police station, firehouse, post office, and bank} and {playground, park, pond, golf course, dock, beach, bathhouse, and pool} These two clusters were also dominant in the complete linkage method, as was the cluster {library, college, senior high, junior college, elementary school} We chose to be conservative and only consider clusters that appeared in both methods We reduced these clusters to five items each by eliminating one member of any orthographically similar pairs, such as park and pond or firehouse and court house The resulting clusters consisted of a government cluster {court house, town hall, ponce station, post office, and bank} and a recreation cluster {playground, pond, golf course, dock, and beach} These clusters formed the basis of the remaining experiments Method Subjects Forty undergraduates attending the State University of New York at Albany participated in the present study in order to fulfill a course requirement The task required hour of participation Materials Thirty-nine names of city landmarks were chosen on the basis of the following criteria: (a) The choice of landmarks was limited to landmarks expected to be found in a typical city, (b) landmarks were not permitted to be exemplars of other landmarks on the list (for example, if restaurant was used, then hamburger stand would not be allowed), and (c) landmarks were not permitted to be variants of other landmarks on the list (for example, if grocery store was used, then supermarket would not be allowed) Furthermore, it was decided not to include names ending in the word store, such as, grocery store, or shoe store, due to the confusability of names Each landmark name was typed in large typeface (6 cpi) and centered in the middle of a in X in index card The experimenters recorded the data through use of paper and pencil Procedure Each subject was individually seated in a room with the set of index cards positioned on a table in front of the subject The experimenter read the following instructions: Sort the cards into piles containing the names of places which you think you would find located near each other in a typical city For example, if you thought that you might find an "ice cream parlor" near a "restaurant," then you would put these items in the same pile If you think that it would be unlikely to find an "auto parts store" next to a "restaurant," then you would sort these items into separate piles The piles may be as large or as small as you like, or may contain as few as one item Experiment Experiment focuses on the ways in which semantic labels affect the acquisition of spatial information from maps In order to measure labeling effects independent of spatial characteristics, a map of 10 points was constructed using the labels from two conceptually distinct sets, as determined by an independent group of subjects in Experiment through the sorting task Learning in Experiment took the form of placing points on a computer screen in relationship to two other anchors The use of anchors was to encourage the learning of relative locations, rather than absolute locations That is, without anchors, buildings can be placed relative only to the screen By providing a few randomly chosen anchors, learning of the interrelations between buildings would be encouraged Method General design Experiment consisted of three phases In the maplearning phase, subjects learned one of two maps through a series of reconstructions of the map In the distance-estimation phase, subjects used their knowledge of the map to provide distance estimates, from memory, between pairs of locations In the comparative-judgment phase, subjects used their knowledge of the map to judge, from memory, which of two locations was closer to a third location The experiment required a total of hours' participation, hour per day over a 2-day period The 184 STEPHEN C HJRTLE AND MICHAEL F MASCOLO first hour was devoted to the map-learning task, and the second hour was devoted to the distance-estimation and comparative-judgment tasks Subjects Twenty-one undergraduates attending the State University of New York at Albany participated in order to fulfill a course requirement Six additional subjects participated in pilot versions of the map-learning phase i s order to provide criteria for successful learning of the maps.1 On the basis of criteria obtained from pilot subjects, of the 21 subjects was eliminated due to a failure to successfully learn the map One other subject eliminated himself due to scheduling difficulties, leaving a total of 19 subjects available for analysis Stimuli and apparatus An Apple He microcomputer controlled the presentation of stimuli and data collection All stimuli were presented on a 25 X 20 cm CRT screen positioned in a sound-proof chamber Sutgects sat approximately 50 cm from the screen, with either a numerical keypad or a response box placed between the subject and the screen, depending on the task The Apple computer itself was located in another room For the map-learning phase and the distance-estimation phase, subjects entered responses using an Apple numerical keypad For the comparative judgment task, subjects entered responses using a two-key response box The centers of the keys were positioned cm apart Two hypothetical maps depicting 10 locations were created The first map is shown in Figure The second map differed from the first only with respect to the labels, not the locations In both maps aa asterisk was used in place of the number to indicate the exact location of the landmark The locations were constructed to include eight embedded critical triads, each containing two identical lengths Thus, each critical triad formed an isoceles triangle For example, length [1, 3] - length [1, 10], and length [2, 6] - length [2, 8], Labels were assigned to the locations to suggest specific clusterings Half the subjects were given labels that suggest clustering locations {1, 3, 4, 5, 6} and clustering locations {2, 7f 8, 9, 10}, whereas for the remaining subjects, the labels of location I and were reversed, suggesting clusters {1, 7, 8, 9, 10} and (2, 3, 4, 5, 6} The labels were determined from the semantic classification task in Experiment For the distance-estimation task, the stimuli consisted of a total of 72 pairs of locations This set consisted of 20 practice pairs, 32 critical pairs (16 distinct critical pairs presented twice), and 20 abstractor pairs* A critical pair consisted of the apex and one base point of the isoceles triangle formed by a critical triad, for example, [1, 3], The abstractor and practice pairs were randomly chosen from all other pairs For the comparative-judgment task, the stimuli consisted of 100 triads of locations, of which 16 were critical triads (8 distinct triads presented twice) and 84 were distractor triads For any given triad, the three names were arranged on the computer screen in the form of a triangle The referent for the Dock GoR Course Tfcwn Hall Playground Beach 10 Potics Station Pond Court House Bank Figure Map A presented in Experiment 2- (Map B differed by the reversal of the labels of Points and 2.) triad formed the apex of the triangle Two additional anchor locations were positioned at each of the two lower corners of the screen Subjects were to judge which of the two anchors was closer to the referent in the original map Procedure The experiment required participation for two 1-hour sessions on two separate days not separated by more than days The first session was devoted to map learning Half of the subjects learned Map A, and the other half learned Map B For the first day, the subjects spent hour learning the map locations through a series of brief study periods and reconstructions The study periods were kept short so that learning would take place primarily in response to the feedback during the placement of targets In this way the learning strategy used by a subject was more tightly controlled than would be possible under a free-study method The sequence of events for the map-learning task was as follows The subject was individually seated in a sound-proof chamber in front of a CRT screen and a numerical keypad The experimenter read the instructions for the map-learning segment to the subject, after which the map then appeared on the CRT screen for 30 s The subject was instructed to memorize the map After 30 s, the map disappeared The name of a target location, chosen randomly, appeared at the bottom of the screen outlined in white The subject was required to place the target location in its correct position on the map To assist the subject in placing the point, two additional locations, again chosen randomly, were depicted in their correct locations and remained on the screen while the subject attempted to place the target location The subject used the numerical keypad to move the point around the screen until the point was located in the position believed to be correct On placing the point, the subject pressed a key to enter the response into the computer At this point, a number sign {#) flashed on the screen depicting the correct location of the target point In addition, the computer presented an error score, which provided a measure of the proximity of the subject's placement of the point to the actual location of the point on the map in terms of screen units The subject was encouraged to use both pieces of information (the actual location and the error score) as an aid in learning the map After the subject placed the point in the location believed to be correct, the computer cleared the screen and presented another target location at the bottom the screen The procedure was repeated until the subject placed all 10 points on the map, completing the block At the completion of the block, the subject was provided with a cumulative error scots for that block The experimenter urged the subject to keep the error score as low as possible The subject spent the remainder of the hour reconstructing the map in this manner On the second day the subject was seated in front of the CRT screen and completed one block of the map-leaming task as just described After the subject had completed the block, the experimenter read the instructions for the distance-estimation task On each trial, the phrase, "From to " appeared on the screen, with the names of different map locations appearing in the Hanks The experimenter instructed the subject to assume that the width of the screen represented 100 units On the basis of this information, the subject estimated the distance between the two locations and entered the result into the computer by using the numerical keypad The subject made judgments on a total of 72 pairs, presented as six blocks of 12 trials The subject was permitted a brief, self-terminated rest period between blocks of trials On completion of the distance-estimation phase of the experiment, the subject rested for minutes before commencing the comparative judgment task For the comparative judgment task, the subject judged from memory which of two points was closer to a third The subject was individually seated in front of a CRT screen The experimenter asked the subject to An error score for each trial of the reconstruction was calculated Subjects were eliminated if the average error score for the last four blocks of reconstruction was greater than 10 screen units 185 SEMANTIC CLUSTERING lightly rest the index finger of each hand on the keys of the response box The experimenter read the instructions to the subject Each trial began with the presentation of a plus sign (+), which appeared in the middle of the CRT screen for 0.5 s The experimenter instructed the subject to use the plus sign as a fixation point The plus sign was replaced by the name of one of the locations; this location was designated the referent for the trial Two additional locations, or anchors, then appeared in the lower left and right corners of the screen The experimenter instructed the subject to judge, from memory, which of the two anchors was closer to the referent on the original map The subject was instructed to press the left key if the left anchor was judged as closer to the referent and to press the right key if the right anchor was judged as being closer The subject was told to make the responses as quickly but as accurately as possible Each subject made judgments on 100 trials, which were divided into 10 blocks of 10 trials As before, the subject was permitted to pause momentarily at the end of each block of trials VAB \ \ Results Analysis Map reconstruction, distance judgments, and distance classification were analyzed separately Within- and between-cluster pairs distance judgments were compared using an analysis of variance (ANOVA) test, and the classification data were analyzed with a chi-square test It was predicted that betweencluster pairs would be judged further apart than identically distant within-cluster pairs In the following sections, the term criticial triad will refer to a set of three locations, a critical point (either Point or 2), a location from the recreation cluster (either Points 7, 8,9, or 10), and a Location from the government cluster (either Points 3, 4, 5, or 6) In these critical triads the distance from the critical point to the location in the recreation cluster equals the distance from the critical point to the location in the government cluster Within a critical triad, both pairs of locations involving the critical point will be refered to as critical pairs; the distance between the points in a critical pair will be refered to as the critical distance As the critical triad forms an isoceles triangle, the critical distance is identical for both critical pairs Map reconstruction In analyzing the data obtained from the map-reconstruction phase, four maps representing the average placement of locations on the CRT screen were constructed, so that each map represented one of two days of the experiment and one of two groups of subjects The coordinates of the Day map were obtained by averaging the last four blocks of trials across subjects, whereas the coordinates of the Day map were obtained by averaging data from a single block of trials across subjects The major data of interest are the displacement of the critical points relative to the correct locations For each day and for each group, we measured the average displacement in terms of the X and Y coordinates On the first day, subjects learning Map A displaced Point on average by (.70,2.75) units, whereas subjects learning Map B displaced Point I on average by ( 10,0.00) units Subjects learning Map A displaced Point on average by (.30, 50) units, whereas subjects learning Map B displaced Point on average by (-.80, 2.25) units To measure the differential effect of the labels, we then calculated the vector from the averaged Map A location to the averaged Map B location, as shown in the hypothetical drawing in Figure In terms of polar coordinates,2 the resulting vectors were (2.78, -93.8°) for Point and (2.07, 122.2°) for Point These data are summarized in Table Thus, \ Actual Figure An illustration of the vector used to measure the shift from the Map A placement to the Map B placement of points in the map-reconstruction task there was a tendency for both groups to shift the critical points toward the center of their natural clusters The effect was even more pronounced on the second day On the second day, subjects learning Map A displaced Point on average by (-1.20, 4.25) units, whereas subjects learning Map B displaced Point on average by (1.90, 25) units Subjects learning Map A displaced Point on average by (.90, -.50) units, whereas subjects learning Map B displaced Point on average by (—1.50, 2.50) units The resulting vectors from the averaged Map A location to the averaged Map B location were (5.06, -52.2°) for Point and (3.84, 128.7°) for Point Thus, the effect was more pronounced on Day 2, with a shift that on average was 63% greater Together, these results suggest that subjects tended to place within-group critical points closer to their respective clusters, thus differentially pulling the points away from their actual locations toward their respective semantic clusters Although the map reconstruction data give an overall impression of a shift, many researchers have noted problems in the interpretation of map reconstruction (see Siegel, 1981, for a review) More sensitive measures include distance estimation and comparative judgment Thus, we examined the shifts with regard to these measures Distance estimation For each subject, eight difference scores, calculated from the estimates made on each of the eight critical triads, were computed in the following manner As subjects in Any vector can be expressed in terms of any of several coordinate systems The Cartesian coordinates (x, y) of a vector refer to the relative displacement along the x and y axes Polar coordinates (r, 0) provide an alternative representation, where r refers to the length of the vector and refers to angle By convention, 0° represents due east, 90° represents due north, -90° represents due south, and 180° represents due west In this manner, the two numbers, r and 6, represent independently the size of the shift and the direction of the shift 186 STEPHEN C HIRTLE AND MICHAEL E MASCOLO that the mean difference scores did not differ significantly among the critical triads The interaction between map and triad also failed to reach significance, Fil, 119) = 1.63, p > 10 Comparative judgments For the comparative-judgment task, we calculated the number of times that subjects classified withmciuster pairs as closer or farther away from equidistant betweencluster pairs for each critical triad Each subject made two judgments on each of eight critical triads for a total of 16 judgments per subject The data obtained from the comparative-judgment task, displayed in Table 2, were analyzed using four chi-square tests When the critical location was Point 1, subjects in the Map A group classified locations in the recreation cluster {7, 8, 9, 10} as closer 61 out of 80 times This result differs significantly from chance classifications, x2 (1, N = 80) = 21.01, p < 001 Subjects in the Map B group classified the same locations {7, 8, 9, 10} as closer to Point only 22 out of 72 times This result differs significantly from chance classification, x2 U» AT = 72) = 10,12, p < 01 The direction of this effect was opposite to that found for the Map A group Table Displacement of Critical Points in Experiment Map A Critical point Point Day Day Point Day Day MapB Shifty x y r 2.75 4.25 10 1.90 00 25 2.78 5.06 -98.3 -52.2 50 -.50 -.80 -1.50 2.25 2.50 2.07 2.84 122.2 128-7 X y 70 -1.20 30 90 Note Units are in terms of screes units each group estimated the distance between each critical pair twice, the mean of these pairs of estimates was computed For triads involving Critical Point 1, the mean judged distance from Point I to the point in the government cluster was subtracted from the mean judged distance from Point to the corresponding point in the recreation cluster For triads involving Critical Point 2, the mean judged distance from Point to the location in the recreation cluster was subtracted from the mean judged distance from Point to the corresponding location in the government cluster These difference scores are presented in the equation below For critical-triad classifications where the critical location was Point 2, the Map A group classified locations in the government cluster {2,3,4,5} as closer 67 out of 80 times This result differs significantly from chance, x2 (1, N = 80) = 35.1 \,p < 001 The Map B group classified the same locations {3, 4, 5, 6} as closer only 14 out of 72 times This result differs significantly from chance classifications, x (1, N = 72) » 25.65, p < 001, and again the direction of the effect is in the opposite direction relative to that found for the Map A group Thus, for any given critical triad, subjects judged within-cluster critical pairs as closer than equidistant between-cluster pairs DUJ = dist[\J] - d i s t f t j ] D2ij = dist [2J] - dist [2, i], f o r / = {3, 4, 5, 6} a n d ; - {7,8,9, 10} The difference scores were constructed so that positive scores indicate that Critical Point I is clustered with {7, 8, 9, 10} and Critical Point is clustered with {3, 4, 5, 6} Thus, a shift of points towards their respective clusters would be reflected as positive difference scores for the Map A group and negative difference scores for the Map B group For each group of subjects, a mean difference score was calculated for each of the eight critical triads The difference scores were entered into a Triad X Map mixed-design ANOVA The mean difference score produced by Map A subjects was 10.8, whereas the mean difference score produced by Map B subjects was -13,5 The main effect of map was significant, F(l, 17) = 22.8, p < 001, indicating that subjects made a differential shift, moving points toward their natural clusters There was no significant difference between triads, ^ , 1 ) = ! 89, p > 05, indicating Experiment As a next step, we tried to manipulate the formation of clusters by altering the learning experience We reasoned that if a subject were only allowed to experience, say, the town hall with items from the recreation cluster, then that subject might be induced to cluster the town hall with that cluster, rather than with the government cluster as occurred in Experiment To test this hypothesis, we altered the map-learning phase so that points were presented in the context of specific clusters Specifically, the anchors in the map-reconstruction task were drawn from one of two sets, either {town hall, dock, golf course, playground, beach} or {pond, court house, police station, bank, post office}, depending on the point to be located Table Classification of Shortest Distance in Experiment Point Point i Ouster Map A MapB Map A MapB Recreation Government 61 (76.2) 19(23.8) 22 (30.6) 50 (69.4) 13(16.2) 67 (83.8) 80 72 80 58 (80.6) 14(19.4) 72 Total Note The numbers reflect the frequency with which a given point is classified as being closer to the recreation or government cluster Percentages are in parentheses 187 SEMANTIC CLUSTERING Method Procedure The same three-phase procedure was used as in Experiment The only difference was in the map-learning phase The set of locations was split into two discrete sets, each consisting of an incongruous critical point and a congruous set of distractors The two sets were {town hall, dock, golf course, playground, beach} and {pond, court house, police station, bank, post office} When locating each of the points during the map-learning phase, the two distractors that were presented on the screen were chosen so that all three points were from the same set Thus, subjects never experienced the town hall with the government cluster or the pond with the recreation cluster As before, the experiment required a total of hours' participation, hour per day over a 2-day period The first hour was devoted to the maplearning task, and the second hour consisted of one map-learning trial, the distance-estimation task, and the comparative-judgment tasks Subjects Twenty-five undergraduates attending the State University of New York at Albany participated in order to fulfill a course requirement On the basis of criteria use in Experiment 2, of the 25 subjects were eliminated due to a failure to successfully learn the map, leaving a total of 23 subjects available for analysis Results If the learning experience influenced either the perception or memory of locations, then we would expect the critical points to be displaced towards the implied clusters That is, the town hall should be placed close to the recreation cluster, whereas the pond should be placed close to the government cluster Expressing these relationships in terms of locations, for the Map A group we would expect to find Point shifted towards the government cluster, whereas Point should be shifted towards the recreation cluster The reverse relationship should hold true for the Map B group Note that these predictions are the exact opposite of the predictions in Experiment Map reconstruction The displacement vectors were calculated for the two critical points for each map group and for each day In contrast to the predictions above, the actual displacements of the critical points were very similar to those found in Experiment The displacement coordinates are shown in Table Looking at the displacement of each point by each group, there is some indication that the Map A group displaced points less, whereas the Map B group displaced points more, in comparison to Experiment However, the vector from the averaged Map A location to the averaged Map B location was virtually identical to those found in Experiment On Day 1, the shift vectors were (2.21, -21.2°) for Point and (3.67, 92.5°) for Point On Day 2, the shift vectors were (5.39, -30.4°) for Point I and (4.23, 179.6°) for Point Distance estimation The distance estimation data show a similar relationship Difference scores were computed as in the previous experiment These were entered into a Triad X Map mixed-design ANOVA The mean difference score for Map A subjects was 7.9, whereas the mean difference score produced by Map B subjects was - , F{\, 20) = 22.2, p< 001, indicating that subjects made a differential shift, moving points toward their semantic clusters Although the means show a slightly diminished effect from the previous experiment, there is no evidence of the formation of alternative clusters To be sure, an Experiment X Triad X Map ANOVA was performed in order to compare the results of Experiment and Experiment Replicating the in- Table Displacement of Critical Points in Experiment Map A Critical point Point Day Day Point Day Day X ShiftAB MapB y 00 82 -.83 2.50 88 50 X y 02 2.06 3.82 -.23 -.10 72 3.57 42 -3.73 45 r 2.21 5.39 -21.2 -30.4 3.67 4.23 92.5 179.6 Note Units are in terms of screen units dividual analyses, we found that the main effect of map was significant, F\\, 37) = 45.25, p < 001, whereas the effect of experiment was not significant, F\l, 37) < 1.00, p < 10 One minor difference appeared in the analysis of Experiment In contrast to Experiment 2, the triad main effect was also significant, ^ , 140) = 7.01, p < 01, with the mean difference scores ranging from -2.7 to 12.0 across the eight triads To examine this effect closer, we compared the difference scores with the actual distance Not suprisingly, we observed that relatively close triads such as {1, 3,10} resulted in small difference scores, whereas relatively far triads such as {2, 6, 9} resulted in large difference scores As in Experiment 2, the interaction failed to reach significance, F(l, 140) = 1.53, p < 10 Comparative judgments Finally, the comparative-judgment task showed a similar, although slightly weaker, relationship The data are shown in Table When the critical location was Point 1, subjects in the Map A group classified locations in the recreation cluster as closer 55.2% of the time, x20> N- 96) = 1.04, ns Subjects in the Map B group classified the same locations as closer 36.8% of the time, x , N ^ 96) = 6.58, p < 05 For critical-triad classifications where the critical location was Point 2, the Map A group classified locations in the recreation cluster as closer 22.7% of the time, %2(U N = 88) = 35.11,p < 001 The Map B group classified the same locations as closer 67.0% of the time, x , AT = 88) = 10.22, p < 01 Thus, three of the four tests of significance showed a marked bias, indicating a virtually identical pattern of results to Experiment It appears that the alteration in the learning strategy had little effect on the formation of clusters General E>iscussion The three experiments taken together provide evidence that the semantic labels attached to clusters can produce mental clusters and that these clusters can alter a subject's memory of spatial locations The strength of the effect was demonstrated in Experiment 2, whereas the impenetrability of the effect was demonstrated in Experiment The last experiment suggested that information being acquired is more important than the manner in which it is acquired Of course, we only tested a single learning strategy Thus, although we showed no relationship between learning and clustering, we not want to imply that future examination of learning strategies will not uncover a positive relationship 188 STEPHEN C HIRTLE AND MICHAEL F MASCOLO Table Classification of Shortest Distance in Experiment Point Point Cluster Map A MapB Map A MapB Recreation Government 53 (55.2) 43 (44.8) 20 (22.7) 68 (77.3) 35 (36.8) 60 (63.2) 59 (67.0) 29 (32.6) 96 88 96 88 Total Note The numbers reflect the frequency with which a given point is classified as being closer to the recreation or government cluster Percentages are in parentheses The results of Experiment 2, on the other hand, seem conclusive in their demonstration of the clustering effect One issue to address is whether the effect is due to perceptual or memorial processes To address this issue, we compared performance on the first and second day on the map-reconstruction task Whereas a mild shift of the critical points towards their natural clusters occurred on the first day of learning, the shift was amplified on the second Thus, we would argue that the clustering is primarily a memory phenomenon, rather than a perceptual one That is, it appears that verbal and spatial information are not encoded independently (Pezdek & Evans, 1979), but rather verbal information assists in the encoding and retreival of spatial information Semantically related locations are encoded together and thus remembered as being closer in space than semantically unrelated information Furthermore, the clustering effect occurred without explicit barriers, in contrast to many of the previous studies on clustering (e.g., Kosslyn, Pick, & Fariello, 1974; McNamara, in press; Stevens & Coupe, 1978) That is, barriers can be created semantically, as well as physically, in space We would further speculate that the minimum number of objects required to induce such clusters would be similar to that in verbal memory experiments (Ericsson, Chase, & Faloon, 1980; Miller, 1956) and that given a large enough set of objects that the clusters would become hierarchical in nature (Hirtle & Jonides, 1985) A possible constraint on the findings is the artificial nature of the stimuli in this study Of course, it is easier to reverse a pond and a town hall on a computer screen than it is to reverse a pond and a town hall in an actual spatial environment However, by using either a model town with surface-mounted video cameras (e.g., Walsh, Krauss, & Regnier, 1981) or a video disk (e.g., Hooper, 1981), one might be able to recreate a natural setting that can be altered It would be of interest to adopt the techniques used in the learning and test phases of this study to a naturalistic setting in this manner In doing so, one could independently vary the visual, semantic and spatial information, thus adding a third dimension to our analysis At present, this study adds converging evidence to the notion that clustering of spatial landmarks is critical in the formation of mental representations of space Through manipulation of the labels assigned to landmarks, we have induced semantic clusters on spatial locations in an artificial setting Earlier work in our lab demonstrated clustering in a natural setting (Hirtle & Jonides, 1985) without manipulation of an actual environment Taken together, it appears that clustering does occur in undifferentiated spatial areas and that the clusters can be formed on the basis of semantic information interacting with the spatial information References Allen, G L (1981) A developmental perspective on the effects of "subdividing" macrospatial experience Journal of Experimental Psychology: Human Learning and Memory, 7, 120-132 Allen, G L., Siegel, A., & Rosinski, R (1978) The role of perceptual context in structuring spatial knowledge Journal of Experimental Psychology: Human Learning and Memory, 4, 617-630 Anderson, J R (1978) Arguments concerning representations for mental imagery Psychological Review, 85 249-277 Ericsson, K A., Chase, W G., & Faloon, S (1980) Acquisition of a memory skill Science, 208, 1181-1182 Evans, G W (1980) Environmental cognition Psychological Bulletin, 88, 259-287 Evans, G W., Fellow, J., Zorn, M., &Doty, K (1980) Cognitive mapping and architecture Journal of Applied Psychology, 65, 474-478 Evans, G W., Marrero, D., & Butler, P (1981) Environmental learning and cognitive mapping Environment and Behavior, 13, 83-104 Hirtle, S C (1985) Cognitive structures in cognitive maps: Evidence of hierarchies in spatial representation Unpublished manuscript Hirtle, S C , & Jonides, J (1985) Evidence of hierarchies in cognitive maps Memory and Cognition, 13, 208-217 Holyoak, K J., & Mah, W A (1982) Cognitive reference points in judgments of symbolic magnitude Cognitive Psychology, 14, 328-352 Hooper, K (1981) The use of computer-controlled video disks in the study of spatial learning Behavior Research Methods & Instrumentation 13, 77-84 Johnson, S C (1967) Hierarchical clustering schemes Psychometrika, 32, 241-254 Kosslyn, S M., Pick, H L., & Fariello, G R (1977) Cognitive maps in children and men Child Development, 45, 707-716 Kosslyn, S M , & Pomerantz, J P (1977) Imagery, propositions, and the form of internal representations Cognitive Psychology, 9, 52-76 Liben, L., Patterson, A., & Newcombe, N (Eds.) (1981) Spatial representation and behavior across the life span New York: Academic Press MacEachren, A M (1980) Travel time as the basis of cognitive distance Professional Geographer 32, 30-36 Maki, R H (1981) Categorization and distance effects with spatial linear orders Journal of Experimental Psychology: Human Learning and Memory, 7, 15-32 McNamara, T P (in press) Mental representation of spatial relationships Cognitive Psychology Mehler, J., Walker, E C X, & Garrett, M (Eds.) (1982) Perspectives on mental representations Hillsdale, NJ: Erlbaum Miller, G A (1956) The magical number seven, plus or minus two: 189 SEMANTIC CLUSTERING Some limits on our capacity for processing information Psychological Review, 63, 81-96 Pezdek, K., & Evans, G W (1979) Visual and verbal memory for objects and their spatial locations Journal of Experimental Psychology: Human Learning and Memory, 5, 360-373 Podgorny, P., & Shepard, R N (1978) Functional representations common to visual perception and imagination Journal of Experimental Psychology: Human Perception and Performance, 4, 21-35 Sadalla, E K., Burroughs, W J., & Staplin, L J (1980) Reference points in spatial cognition Journal of Experimental Psychology: Human Learning and Memory, 5, 516-528 Sadalla, E K., Staplin, L J., & Burroughs, W J (1979) Retrieval processes in distance cognition Memory and Cognition, 7, 291-296 Siegei, A W (1981) The externalization of cognitive maps by children and adults: In search of ways to ask better questions In L Liben, A Patterson, & N Newcombe {Eds.), Spatial representation and behavior across the life span (pp 167-194) New York: Academic Press Stevens, A., & Coupe, P (1978) Distortions in judged spatial relations Cognitive Psychology 10, 422-437 Walsh, D., Krauss, I., & Regnier, V (1981) Spatial ability, environmental knowledge, and environmental use: The elderly In L Liben, A, Patterson, & N Newcombe (Eds.), Spatial representation and behavior across the life span (pp 32 \ -357) New York: Academic Press Wilton, R (1979) Knowledge of spatial relations: The specification of the information used in making inferences Quarterly Journal of Experimental Psychology, 31, 133-146 Received March 29, 1985 Revision received June 7, 1985 Instructions to Authors The editorial policy of the journal favors publication of mtegrative articles containing multiple experiments and extensive theoretical development However, articles reporting single experiments are also appropriate, so long as they report compelling data with a clear theoretical message Other appropriate forms of submission include brief comments on articles published in the journal and replies to such comments In addition, occasional literature reviews advancing our knowledge in the field wiH also be considered For further information on content, authors should refer to the editorial in the January 1985 issue of the journal (Vol 11, No 1, pp 1-2) For information on the other three JEP journals, authors should refer to editorials in those journals Authors should prepare manuscripts according to the Publication Manual of the American Psychological Association (3rd ed.) AH manuscripts must include an abstract of 100— 150 words typed on a separate sheet of paper Typing instructions (all copy must be double-spaced) and instructions on the preparation of tables, figures, references, metrics, and abstracts appear in the Manual Also, all manuscripts are subject to editing for sexist language APA policy prohibits an author from submitting the same manuscript for concurrent consideration by two or more journals APA policy also prohibits duplicate publication, that is, publication of a manuscript that has already been published in whole or in substantial part in another journal Also, authors of manuscripts submitted to APA journals are expected to have available their raw data throughout the editorial review process and for at least years after the date of publication Authors will be required to state in their initial submission letter or sign a statement that they have complied with APA ethical standards in the treatment of their sample, human or animal A copy of the APA Ethical Principles may be obtained from the APA Ethics Office, 1200 17th Street, N.W., Washington, DC 20036 Anonymous reviews are optional, and authors who wish anonymous reviews must specifically request them when submitting their manuscripts Each copy of a manuscript to be reviewed anonymously should include a separate title page with authors' names and affiliations, and these should not appear anywhere else on the manuscript Footnotes that identify the authors should be typed on a separate page Authors should make every effort to see that the manuscript itself contains no clues to their identities Manuscripts should be submitted in quadruplicate (the original and three photocopies), and all copies should be clear, readable, and on paper of good quality Authors should keep a copy of the manuscript to guard against loss Mail manuscripts and address all correspondence at all stages of handling to the Editor, Henry L Roediger III, Department of Psychological Sciences, Purdue University, West Lafayette, Indiana 47907 For the other JEP journals, authors should submit manuscripts (also in quadruplicate) to one of the editors at the following addresses: Journal of Experimental Psychology: General, Sam Glucksberg, Department of Psychology, Princeton University, Princeton, New Jersey 08544; Journal of Experimental Psychology: Human Perception and Performance, William Epstein, Department of Psychology, University of Wisconsin, W J, Brogden Psychology Building, 1202 West Johnson Street, Madison, Wisconsin 53706; Journal of Experimental Psychology: Animal Behavior Processes, Michael Domjan, Department of Psychology, University of Texas, Austin, Texas 78712, When one of the editors believes a manuscript is clearly more appropriate for an alternative journal of the American Psychological Association, the editor may redirect the manuscript with the approval of the author  ... closer to the recreation or government cluster Percentages are in parentheses The results of Experiment 2, on the other hand, seem conclusive in their demonstration of the clustering effect One issue... of each hand on the keys of the response box The experimenter read the instructions to the subject Each trial began with the presentation of a plus sign (+), which appeared in the middle of the. .. alter a subject's memory of spatial locations The strength of the effect was demonstrated in Experiment 2, whereas the impenetrability of the effect was demonstrated in Experiment The last experiment