Simon Ungar, Mark Blades and Christopher Spencer.

Department of Psychology, University of Sheffield, Sheffield S10 2TP.

E-mail: s.ungar@sheffield.ac.uk

Abstract: Mental rotation tasks have been used to probe the mental imagery of both sighted and visually impaired people. People who have been blind since birth display a response pattern which is qualitatively similar to that of sighted people but tend to respond more slowly or with a higher error rate. It has been suggested that visually impaired people code the stimulus and its (or their own) motion in a different way from sighted people - in particular, congenitally blind people may ignore the external reference framework provided by the stimulus and surrounding objects, and instead use body-centred or movement-based coding systems. What has not been considered before is the relationship between different strategies for tactually exploring the stimulus and the response pattern of congenitally blind participants. Congenitally blind and partially sighted children were tested for their ability to learn and recall a layout of tactile symbols. Children explored layouts of one, three or five shapes which they then attempted to reproduce. On half the trials there was a short pause between exploring and reproducing the layouts. In an aligned condition children reproduced the array from the same position at which they had explored it; in a rotated condition children were asked to move 90˚ round the table between exploring and reproducing the layout. Both congenitally blind and partially sighted children were less accurate in the rotated condition than in the aligned condition. Five distinct strategies used by the children in learning the layout were identified. These strategies interacted with both visual status and age. We suggest that the use of strategies, rather than visual status or chronological age, accounts for differences in performance between children.

A variety of tasks has been devised to explore the nature of mental imagery in sighted people (Kerr and Neisser, 1983; Kosslyn, 1975; Kosslyn, Ball and Rieser, 1978; Paivio, 1986; Shepard and Metzler, 1971). These generally indicate that representations of perceptible objects and events are picture-like or are at least in some way analogous to visual perception. For example, producing mental images of words facilitates their recall (Paivio, 1986), and images can be mentally scanned and rotated just as such events would occur in perception (Kosslyn, 1975; Kosslyn et al., 1978; Shepard and Metzler, 1971).

In a series of studies which have become paradigmatic in mental imagery research, Shepard and Metzler (1971) showed participants pictures of two three dimensional objects in different alignments and then asked them to determine if the two pictures depicted the same object, or if one was the mirror image of the other. Reaction times to recognize same shape pairs were a linear function of the degree of rotation of one object relative to the other, suggesting that participants mentally rotated an image of one of the stimuli until it came into congruence with the other. These and similar studies have been carried out with sighted participants but, as Paivio (1986) points out, imagery can be derived from all the sensory modalities and therefore even congenitally blind people could, in principle, form mental images of objects based on their intact sensory modalities (especially audition and touch).

Marmor and Zaback (1976) and Carpenter and Eisenberg (1978) presented congenitally blind and blindfolded sighted participants with variants of Shepard and Metzler's (1971) mental rotation task. Both studies found the same relation between reaction time and angular disparity of the target shape as in the original experiment. In both studies, visually impaired and sighted participants reported mentally rotating, moving or twisting the target shape until it was aligned with the standard shape. However, Marmor and Zaback also reported faster overall reaction times and lower error rates for the sighted participants. This suggests that, while congenitally blind people do form mental images which are functionally equivalent to those of the sighted, visual images may be easier to operate upon than haptic images. In the Carpenter and Eisenberg study, on the other hand, no differences were found between congenitally blind and sighted participants in errors or reaction times. The authors suggest that this discrepancy may have been due to the congenitally blind participants' previous experience with tactile graphic material.

Millar (1976) presented totally blind and blindfolded sighted children (aged 7 to 11 years) with a number of simple mental rotation tasks. The visually impaired children performed worse than the sighted children overall. From an analysis of the error scores Millar concluded that none of the children used external reference systems to solve the task resulting in low performance overall. The totally blind children could only accurately code rotations to orientations which were orthogonal to their own body co-ordinates. Millar suggests that young visually impaired children should be presented with tasks which encourage them to use more complex and flexible systems of reference.

Millar (1981; 1982; 1988) suggested that the differences in latencies or errors on spatial tasks between congenitally blind and sighted groups may be due to different strategies for coding spatial information. She argued that strategy preferences arise from differences in sensory experience; visual experience prompts people to adopt external frames of reference when coding spatial information while the lack of visual experience prompts the use of self-referent or movement-based coding strategies. However, Millar proposed that these differences in strategies are not mandatory but 'optional'. Just as sighted people may adopt self-referent strategies in the appropriate conditions (e.g. when determining whether it is the train in which one is sitting or the train on the next track that is moving), so visually impaired people have the potential to adopt external coding strategies. Tactile external coding systems may be functionally equivalent to those of the sighted, such that these forms of coding (visual external vs. tactile external) are optional forms of coding in Millar's sense (i.e. they are qualitatively different but functionally interchangeable).

There is evidence to suggest that visually impaired children and adults do not spontaneously adopt external coding strategies. In a series of studies, Millar (1974; 1975; 1976; 1979; 1981) examined the spatial coding of visually impaired children. In Millar's (1979) study visually impaired children were required to reproduce an arrangement of four objects which were placed at the corners of either a square or a diamond shaped base. The original array was always present, and the children were required to reproduce it by moving all the objects onto a second identical base in one of two conditions. In the first condition, both bases were placed directly in front of the children with the second base further away. In the second condition, the bases were placed on either side of the children's mid-line. High errors by the totally congenitally blind group in the cross mid-line condition, but not in the vertical condition, suggested that that group were using a self-referent coding strategy. Qualitative analysis of the errors added support to this interpretation, showing that, in the cross mid-line condition, the totally congenitally blind group often reproduced the array as a mirror image - i.e. objects closer to the participants' mid-line in the original array were placed close to the mid-line in the reproduced array, while those further from the mid-line were placed further away.

In another study, Hollins and Kelley (1988) asked early blinded and blindfolded sighted adults to learn the positions of five objects presented one by one on a circular table (91 cm diameter). Participants were then required either to aim a pointer at the locations of all the objects or to replace each one in its original position. These responses were made either from the position at which the objects had been examined or from a position 90˚ round the table. The visually impaired group were less accurate than the blindfolded group when making pointer responses from a novel position, but there were no differences between groups when participants pointed to the objects' locations from the position from which they had been explored. Nor were there any differences between groups when participants replaced objects from either the original or the novel position. However, an analysis of the participants' patterns of errors suggested that the visually impaired group tended to "foreshorten" the locations of objects on the table; distant objects were imagined as being closer to the position at which the participant had originally explored the layout. Therefore, Hollins and Kelley suggested that the visually impaired participants felt the distances of objects from their own position which they used as a cue when learning the locations of objects. This cue was subsequently available to them to guide them in the replacement condition (where foreshortening was not observed) but was not available in the pointing condition when participants were asked not to reach out onto the table top.

The evidence from these studies indicates that congenitally totally blind adults and children tend to use self-referent or movement-based strategies when learning the locations of objects in small-scale space. This would have important practical significance for teachers of visually impaired people who use tactile graphic materials such as diagrams and maps. An understanding of these materials requires a full apprehension of the spatial relations between all their parts and such an apprehension is best achieved if all the parts of the graphic are related to each other in space and to the frame of the display (Winn, 1991). One important factor which has not been considered in the studies of mental rotation but has been considered in the literature on tactile graphics (e.g. Berlá, 1972; Berlá, 1973; Berlá, 1981; Berlá, 1982; Berlá and Butterfield, 1977; Berlá, Butterfield and Murr, 1976), is the type of tactile strategies used by participants to explore experimental displays.

The experiment reported below was designed as a test of the ability of visually impaired children to reproduce a tactile array under aligned and under rotated conditions, and as a basis for considering the relative effectiveness of the various tactile exploration strategies spontaneously used by the children. Rather than restrict the children to specific exploratory movements (cf. Millar 1981) we allowed the children to generate their own strategies for exploring the layout. We believed that the participants' strategies would be an important factor in the stability of their memory for the array under rotated conditions.

Winn (1991) identified two relationships which define the location of any part of a tactile display. Firstly, the position of each part in terms of the edge of the display (i.e. top, bottom, left, right). Secondly, the position of a part in relation to all other parts of the display. One would expect that a successful strategy for exploring an array would take account of both these relationships. With reference to the two types of relationships defining the locations of parts of the array (Winn, 1991), we hypothesised that an effective exploration strategy would take account of both the relationship of each element to the other elements of the array and of the relationship of each element to the borders of the array. A strategy involving only the first type of relationship (elements to each other) would be moderately effective, while one involving only the second type (elements to the borders of the display) would provide only rudimentary (top-bottom-left-right) information - especially as, in the present experiment, the border of the area was circular. Therefore the participants' performance was analysed with reference to their strategies and whether the strategies were related to the participants' degree of visual impairment.

Twenty nine children attending schools for the visually impaired participated in the study. They were divided into two age groups: 18 children of mean age 10:2 (range 8:2 to 12:3) and 11 children of mean age 6:9 (range 5:6 to 7:6), and two sight groups: 14 children were totally blind from birth and 15 had some residual vision (perception of light or better).

A circular box and lid of diameter 34 centimetres was used to display the stimuli (see Figure 1). Both in the floor of the box and on the lid, a circular area of 30 centimetres was delimited with a ring of thick cardboard. Twelve circular pieces of paper could be fastened into the box, on each of which a different layout was drawn. Two sets of five clay shapes (square, circle, triangle, cross and star) could be firmly fixed into the box according to the drawn layout; these were all approximately two centimetres in diameter and were painted green to be visible to children with residual vision.

Children were tested individually. The task was introduced with three practice sessions. In these, two of the shapes were fastened into the box, the children were asked to explore inside the box making sure they knew where the shapes were. When the children were confident that they had memorised the positions of the shapes, the lid of the box was placed on it. The children were then given two identical shapes and asked to reproduce, on the lid of the box, the layout they had explored. The importance of absolute accuracy was emphasised and children were reminded to place each shape exactly above the corresponding one below. After each practice trial, the lid was raised so that the children could compare the original layout with their estimation. This feedback was only allowed after the practice trials.

The experimental trials followed the same procedure as the practice trials but three variables were manipulated in a balanced fashion across trials:

Testing took place over four sessions of approximately twenty minutes each, with one week between sessions. Children were presented with six layouts at each session. In each session, children received two trials with one shape layouts, two trials with three shape layouts and two trials with five shape layouts presented in the order 135135. In three trials of each session, reconstruction was carried out with no delay and in the other three trials reconstruction took place after one minute. The No Delay and Delay trials were given alternately so that in each session participants reconstructed one, three and five shape layouts with No Delay and a corresponding set of layouts with Delay. Sessions one and two were carried out under the aligned condition while sessions three and four were carried out under the rotated condition.

After each trial, a sheet of transparent plastic marked with a grid of 1 cm squares was placed over the lid of the box and the x and y co-ordinates of the mid point of each shape in the child's reproduction were recorded. A complete videotaped record was obtained of the visually impaired children performing the task to allow analysis of the strategies used by the children performing the task. No feedback was given to the children during the experimental trials.

For each child on each trial an error score was calculated as the distance of the child's estimated positioning(s) from the position(s) of the shape(s) in the original layout; the mean error score was calculated for each of the three- and five-shape layouts. The error scores were therefore examined in a 2 (age group) x 2 (visual status) x 2 (alignment) x 3 (number of shapes) x 2 (delay) analysis of variance, with repeated measures on the last three factors (see Table 1).

A significant effect of alignment showed that the children were less accurate in the rotated condition than in the aligned condition (mean error: Aligned = 4.50, Rotated = 7.62; F (1,25) = 49.47, p < 0.001). An effect of number of shapes showed that accuracy decreased as the number of shapes increased (Mean Error: one shape = 4.72, three shapes = 6.12, five shapes = 7.35; F (2,50) = 43.21, p < 0.001). Pair-wise Tukey tests indicated that these differences were significant at the 0.01 level. There were no other significant effects.

Analysis of videotaped recordings of the visually impaired children performing the three- and the five-shape tasks revealed five distinct strategies. These were labelled and defined as follows:

Rating was carried out by one of the authors and an independent rater assessed a subset of 40 trials. Inter-rater reliability was .85. Children were classified according to the strategy they used on more than half of their trials. Table 2 shows the proportion of children in each age and sight group using each strategy type and the mean error scores for the children falling into each category. The distribution of strategies across age and sight groups differed slightly between three- and five-shape layouts. The distribution of strategy use was similar for male and female participants.

Error scores from the three-shape and the five-shape layouts were subjected to separate 5 (strategy) x 2 (alignment) x 2 (delay) analyses of variance (see Table 3). For the three shape layouts, an effect of strategy was found (mean error: 1=4.25, 2=5.88, 3=7.67, 4=7.96, 5=5.88; F(4,24) = 4.1, p < 0.05). Tukey paired comparison tests indicated that children using strategy 1 were significantly more accurate than those using strategy 3 and than those using strategy 4 (both at the 0.05 level).

With the five shape layouts, Tukey paired comparisons on the significant effect of strategy (mean error: strategy 1=4.14, 2=5.74, 3=9.91, 4=8.68, 5=7.22; F(4,24) = 9.6, p < 0.001) indicated that children using strategy 1 were significantly more accurate than children using strategies 3 and 4 and that children using strategy 2 were significantly more accurate than those using strategy 4 (all at the 0.01 level). In both analyses, a significant effect of alignment reflected the finding reported above (see Table 4). There were no other significant results.

As the children were given unlimited time to explore the layouts, the time taken to explore the layouts was analyses in order to rule out the possibility that children using different strategies had different amounts of exposure to the layout. The time taken to explore was the time (in seconds) from when the children first touched or looked at each layout to the moment when they declared that they had memorised it. The three and five shape layouts were subjected to a separate 5(strategy) x 2(alignment) x 2 (delay) analyses of variance with repeated measures on the last factor. No significant differences were found for strategy or delay in either analysis. Alignment was significant with the three shape layouts (mean exploration time: aligned = 35.4, rotated = 39.5; F(1,25) = 4.46, p < 0.05) but not with the five shape layouts.

The effects of alignment and number of shapes were qualified by the strategies used by the children. Examination of Table 2 shows that visually impaired children who used a methodical tactile strategy (Edge & Relative or Relative) were more accurate than those who simply pointed at the shapes or those who used vision alone. Furthermore, while the scores of the Edge & Relative and Relative strategy users differed little from three shapes to five, the children relying on vision were more severely affected by the number of shapes. This may be due to a superiority of the relational strategies over vision in this task, or it may be that the children using those strategies were older and had more experience of exploring tactile materials. The latter possibility could be tested by training young children to use higher level strategies. The possibility that more effective strategies were simply associated with lengthier, more careful exploration of the layouts was excluded, as the analysis of times taken to explore the layouts yielded no differences between the different strategy categories.

The results are consistent with previous studies which found that people with visual impairments tend to perform poorly on tasks involving tactile mental rotation (Hollins and Kelley, 1988; Marmor and Zaback, 1976; Millar, 1976; 1981). Similarly, in the present study overall error scores were high under the rotation condition. It has been inferred from such results that people with visual impairments (in particular those who are congenitally totally blind) tend not to use external frames of reference for coding tactile information (Millar 1979; 1981). The present study also considered the exploration strategies spontaneously used by visually impaired children to learn the tactile layout. Children who physically explored the internal and external relationships in the layout performed better than children who ignored these relationships. This adds further weight to Millar's (1976; 1988) claim that visually impaired children from the earliest ages should be encouraged through various tactile tasks to gain experience of complex spatial relationships in small-scale space.

The results support the claim by Winn (1991) that an effective strategy for learning graphic information must take account of the relation between elements of the graphic and also situate the array within an external framework. Children who examined the relationship of the symbols with both the sides of the display and with each other were extremely accurate and performed well across Number of Shapes and Alignment (see Tables 3 and 4). Those who concentrated only on the interrelationships of shapes were highly accurate across Number of Shapes although the lack of external referents marred performance in the Rotated condition. Those children who considered only the relationship of symbols to the edges of the display performed relatively poorly as the round display provided only very rudimentary information about the positions of symbols. It is of course impossible to establish which features of the layout were attended to by participants who used vision, but it is clear that at least the relationships between shapes and the orientation of the array within the box were simultaneously available to them.

The findings of the present study confirm that visually impaired children can learn a tactile array but that the quality of the derived representation depends in part upon the particular strategy used (cf. Berlá and Butterfield, 1977). We found that those children using more sophisticated strategies were less affected by the number of shapes and the rotation of the array. The reason for this is probably that those strategies provided children with a more extensive reference framework, both within the layout of shapes and external to it. This would allow these children to use external coding strategies (such as referring each shape to the others, finding a pattern in the shapes or referring the layout to the top, bottom and sides of the box) which are likely to be less perturbed by rotation (Millar, 1988). It is possible that training children to use an edge and relational strategy would encourage them to notice and make use of cues which are external to their own bodies and activities. Future research could examine this possibility experimentally.

These findings also have important practical implications for the education of visually impaired children. Various tactile graphical media, such as pictures, diagrams and maps, form an ever increasing part in the curriculum for visually impaired children. However there are no guidelines on how to teach these children to use tactile graphical media. The findings of the present study, along with other work (Berlá, 1973; Berlá, 1981; Berlá, 1982; Berlá and Butterfield, 1977; Ungar, Blades and Spencer, in press) suggests that visually impaired children should be actively encouraged to relate elements of the graphic to each other and to the frame of the display. This is likely to be particularly important for wayfinding in the environment with a tactile map, which often requires the construction of an extensive and flexible reference framework (Ungar, Blades & Spencer, 1994; Ungar, Blades & Spencer, in press).