Retro- and prospection for mental time travel: emergence of episodic remembering and mental rotation in 5- to 8-year old children.

We investigate the common development of children's ability to "look back in time" (retrospection, episodic remembering) and to "look into the future" (prospection). Experiment 1 with 59 children 5 to 8.5 years old showed mental rotation, as a measure of prospection, explaining specific variance of free recall, as a measure of episodic remembering (retrospection) when controlled for cued recall. Experiment 2 with 31 children from 5 to 6.5 years measured episodic remembering with recall of visually experienced events (seeing which picture was placed inside a box) when controlling for recall of indirectly conveyed events (being informed about the pictures placed inside the box by showing the pictures on a monitor). Quite unexpectedly rotators were markedly worse on indirect items than non-rotators. We speculate that with the ability to rotate children switch from knowledge retrieval to episodic remembering, which maintains success for experienced events but has detrimental effects for indirect information.

Introduction

We naturally speak of “looking back on past events” or “looking forward to the future.” Clearly, we cannot literally look back or forward in time (Martin, 2001). We cannot see the future or the past in the same way as we see an event unfolding in front of us or behind us. The best we can do to capture past or future experiences of an event is to re-experience the event (retrospection), or imagine experiencing a future event (prospection).1

The ability to retrospect in this sense has become a central feature for episodic memory with Tulving’s (1985) introduction of “autonoetic consciousness”, namely the awareness that remembering consists of “calling back into consciousness a seemingly lost state that is then ‘immediately recognized as something formerly experienced’ (Ebbinghaus, 1885, p. 1).” Philosophers spoke of “experiential memory,” at least since Locke (Owens, 1996). Although the different terms all capture the phenomenon adequately, and “retrospection” does so in nice juxtaposition to “prospection,” we prefer “episodic remembering” as our standard term. The choice of “remembering”, rather than “memory” is to emphasise that it is more than just retrieval of knowledge about a past episode. It is a re-experiencing of that episode. This distinction is also brought out by the notion of “mental time travel” (MTT: Suddendorf & Corballis, 1997; Wheeler, Stuss, & Tulving, 1997): one has to not just retrieve information about the past or think about the likely future. It requires projecting oneself as an experiencing agent into the past or future. This distinction is also akin to the difference between having a theory of mind (Churchland, 1984; Gopnik & Astington, 1988) as opposed to simulating one’s own (or other people’s) mental processes (Goldman, 2006; Gordon, 1986; Heal, 1986).

There is the strong and widespread, but not uncontested, intuition that these three abilities, theory of mind, episodic remembering, and prospection belong together. They may be uniquely human abilities (Gilbert & Wilson, 2007; Roberts, 2002; Roberts et al., 2008; Suddendorf & Busby, 2003; Suddendorf & Corballis, 1997, 2007; Wheeler et al., 1997). There is now also fast growing empirical evidence from different quarters that episodic remembering (often investigated under the label or as part of autobiographical memory), prospection, and theory of mind share a developmental schedule, and common neural substrate as shown in coactivation patterns and in common deficits in clinical cases.

Brain imaging

Spreng, Mar, and Kim (2008) meta-analyzed, among other areas, all available brain imaging studies on theory of mind (n = 30, as a random selection of 50), autobiographical memory (n = 19), and prospection (n = 6). Direct overlap was observed in the medial temporal lobe (left parahypocampal gyrus: BA 36), medial parietal regions (precuneus, posterior cingulate, bilaterally: BA 31), left temporo-parietal junction (BA 39, touching on BA 19), medial prefrontal cortex (frontal pole: BA 10). Convergence within same Brodmann areas were also observed in right TPJ, left ventrolateral prefrontal cortex (BA 47), medial prefrontal cortex, and rostral anterior cingulate (BA 32), and lateral temporal lobe (BA 21, 22) especially left. All these regions tend to also be activated by navigation problems and default processing (areas that tend to be more strongly activated in the absence than in the presence of external stimulation). Two theories have been proposed as to the common denominator underlying these common activations. Hassabis and Maguire (2007) suggested that all these tasks require scene-construction and Buckner and Carroll (2007) that they all require projection of self into different time points or spatial locations.

Brain injury

Patients without autonoetic consciousness in Tulving’s sense, i.e., amnesics with a special impairment of episodic remembering and autobiographical memory, have been reported to also have severe deficits of prospection, Patients K.C. (Tulving, 1985), R. (Stuss, 1991), M.L. (Levine et al., 1998), and D.B. (Klein, Loftus, & Kihlstrom, 2002). Loss of autonoetic consciousness does, however, not lead inevitably to an impairment in theory of mind (patients K.C. and M.L.: Rosenbaum, Stuss, Levine, & Tulving, 2007). This finding does not preclude theory of mind being necessary for autonoetic consciousness and episodic remembering, in particular, it may be crucial as a developmental requirement or linkage. In fact, there is growing evidence that theory of mind development around 4 and 5 years is linked to both, episodic remembering as well as prospection.

Development

There are studies showing a specific relationship between advances in theory of mind and free recall as a measure of episodic remembering in relation to cued recall (Tulving, 1985). Perner and Ruffman (1995) were able to show that between 3 and 5 years, children’s improvement on free recall correlates significantly with their understanding of how knowledge depends on experience. Even when cued recall and verbal intelligence were partialled out, correlations stayed above .30. Tasks used included children’s ability to explain why they know the contents of a box (How-do-you-know test: Wimmer, Hogrefe, & Perner, 1988a; Wimmer, Hogrefe, & Sodian, 1988b), to distinguish a lucky guess from proper knowledge (Miscione, Marvin, O’Brien, & Greenberg, 1978), and to understand which sense modality to use to find out about colour or weight of an object (O’Neill et al., 1992). These results were largely replicated by Naito (2003) on a Japanese sample. She also found a relationship between free recall and children’s ability to understand when they had learned a fact (Taylor, Esbensen, & Bennett, 1994).

Perner, Kloo, and Gornik (2007) used a different measure of episodic remembering. They contrasted recall of experienced events with recall of indirectly conveyed events. In the experience condition children put cards with drawings of simple objects into a box. In the indirect-information condition they were blindfolded and so could not see which cards they put inside. They were afterwards shown on a monitor the pictures that were on these cards (information about individual cards was thus indirectly conveyed). The reasoning was that children can have an episodic memory of putting a particular card inside the box only when they experienced putting that card into the box. Only with this experience can they later re-experience their action. When blindfolded, they cannot experience which card they had put inside. When later shown what was on these cards, they can only infer that they put that card inside the box. Hence there is no experience of putting that card inside, which they could re-experience. Free recall of experienced events correlated with performance on various theory of mind tasks (How-do-you-know, When-did-you-learn, and the modality-specificity test). Sprung (2008) and Sprung and Harris (2009) found that theory of mind abilities (especially introspective understanding) modulates children’s ability to report on their intrusive thoughts, which in case of traumatic experiences (e.g., Hurricane Katrina victims) provides a basis for outgrowing the trauma.

Rapidly growing evidence suggests that children’s prospective abilities develop at about the same age. Moore, Barresi, and Thompson (1998, Experiment 1) reported some correlations between 3- and 4-year-old children’s understanding of desire and belief and of the benefits of increased but delayed reward.

Bischof-Köhler (2000) investigated several abilities in a group of 3- to 4.5-year-olds with a large battery of tasks including understanding of duration, theory of mind (false belief and deception), planning (shopping, need for preparation), delay of gratification, and motivational conflict. These abilities all underwent a noticeable improvement from 3 to 4.5 years and correlated significantly even when children’s age was partialled out.

Atance and O’Neill (2005) investigated 3- to 4-year-olds’ ability to think about what to pack for a trip. About half of these children went beyond packing typical (scripted) or attractive things. They also thought of providing for possible eventualities (e.g., pack telephone to contact someone in case of an emergency). In another task children were given the beginning of a drawing, e.g., a straight line, and then were asked what they wanted to draw on the basis of this first element. There was a strong correlation r = .65 between children’s ability to anticipate events in the trip task and their choice of a drawing that was feasible with the given element (e.g., a sun but not a ladder starting from a circle) even when language ability was controlled.

Atance and Meltzoff (2005) tested for mental time travel and found that 3 year olds above chance chose a suitable item (e.g., winter coat) for going to a snowy place (as shown in a photograph) and they were able to explain their choice by reference to their future need created by that environment. By 4 and 5 years all children could do this. It remains unclear whether this shows children’s ability to project themselves into the future state of going to a wintery landscape. They might just know about the need for winter coats in a cold environment. Atance and Meltzoff (2006) overly satiated children on salty pretzels, which made them change their preference from pretzels to water. Even 5 year old children showed little sign of understanding that the next day their preference would return to pretzels. They predicted a lasting preference for water.

Busby and Suddendorf (2005) simply asked 3- to 5-year-old children what they had done yesterday and what they were likely to do tomorrow. Discounting general, scripted answers (e.g., “I played”) there was a marked improvement in children’s ability to report past and likely future episodes. Suddendorf and Nielsen (2009) tested 3- and 4-year-olds on a simple problem (e.g., use a triangular key to open a box). When the key was not available children went to another room and played for 15 min. Before returning to the first room they were given the choice among three different keys (one triangular) and asked which one they would like to bring with them to the room. Three-year-olds chose at random, while 4-year-olds chose the triangular key above chance (with plenty of room for improvement).

Atance and Jackson (2009) assessed different aspects of future thinking in 3- to 5-year-old preschoolers, including mental time travel (Atance & Meltzoff, 2005; Busby & Suddendorf, 2005), delay of gratification, planning (simple version of Tower of Hanoi, Carlson, Moses, & Claxton, 2004), and prospective memory (remembering to do something after finishing something else: Kvavilashvili, Messer, & Ebdon, 2001). Children improved on all these tasks and performance between all tasks was correlated but correlations were dependent on age and receptive vocabulary. Ford, Shum, and Driscoll (2009) found that prospective memory performance in 4- to 6-year-olds was related to inhibitory ability and understanding false beliefs.

Finally, Russell, Alexis, and Clayton (2010) let 3–5 year old children play blow-football for which one needed to bring a straw for blowing and as player on the blue (as opposed to the red) side a box to stand on as that side of the table was otherwise too high for children at this age. Children were asked questions framed in the past, present and future about what they themselves or another child needed to play on the blue side. The variable was whether they thought of both required items or not. In case children gave different answers for themselves than for their peer, the authors reasoned, this is a likely sign that children bring to bear their own perspective when imagining how they had played, or were going to play, i.e., project themselves into the role of player. Systematic comparison of self and other was only carried out for present (Experiment 4) and future (Experiments 2 and 3). In particular, 4-year-olds gave similar answers for self and other in the present and for other in the future, but ignored their own needs for the future. This was interpreted as evidence that between 3 and 5 years children become able to project themselves into future situations. Supportive evidence for this age trend comes from Prencipe and Zelazo (2005), who found that 3 year olds make wise choices of delaying an immediate small reward for a larger one later when choosing for another person but not for themselves. This discrepancy ceases around 4 years.

Autism

The spectrum of autism consists of a developmental disorder that affects the very triad of theory of mind, retrospection, and prospection. Children with autism spectrum disorder tend to have problems with basic theory of mind tasks (Baron-Cohen, Leslie, & Frith, 1985; Perner, Frith, Leslie, & Leekam, 1989) or the least impaired cases (Asperger’s syndrome) have deficits appreciating subtle mental interactions like sarcasm, jokes, etc. (Happé, 1994). They also have executive deficits (Ozonoff, Pennington, & Rogers, 1991; for a review, see, Hill, 2004), which tend to be more pronounced for planning (prospection) than for inhibition. When tested for memory their free recall seems to be specifically impaired over cued recall and recognition (Boucher & Lewis, 1989; Boucher & Warrington, 1976; Bowler, Gardiner, & Grice, 2000; Tager-Flusberg, 1991) and even the highest functioning individuals with Asperger syndrome give fewer “remember” judgments in recognition tasks (Bowler, Gardiner, & Gaigg, 2007) than unimpaired people.

Sharpening the issues

The neurophysiological and developmental evidence leaves little doubt that there is some common ground for theory of mind, episodic remembering and prospective abilities. Whether the co-emergence of these different abilities and their sharing of common cerebral structures is good evidence for mental time travel, i.e., for the ability to re- and pre-experience past or future events as the common basis remains an open question. The evidence for that claim is weak in two respects: few of the reviewed studies attempt to isolate specific cases of episodic remembering (retrospection: re-experiencing a past situation by projecting oneself into the past) as opposed to merely retrieving knowledge about the past, and of projecting oneself into a future experience (prospection) as opposed to knowing what is likely to happen in the future or knowing what is needed in an imagined future situation. Consequently, the developmental synchrony could easily be explained by quite general features, e.g., thinking about the past and future.

The contrast between free and cued recall does provide some measure for teasing out the retrospective aspects of memory. Its usefulness is based on Tulving’s (1985) argument that free recall is more dependent on the availability of episodic traces than is cued recall. Hence, if free recall correlates specifically with progress in theory of mind when controlling for cued recall, then that correlation ought to be due to changes in episodic remembering (Perner, 1990) and not some more general ability to think about or retrieve information about the past. The link between free recall and episodic remembering, one has to admit, is rather weak. Free recall does not depend completely on episodic remembering (even in free recall a few items can come to mind automatically). Moreover, free recall is not exclusively helped by episodic remembering, which can also enhance cued recall. The contrast between memory of experienced events and indirectly conveyed events (Perner et al., 2007) is somewhat sharper in this respect. It takes advantage of the fact that only experienced events can possibly be re-experienced. Indirectly conveyed events should, therefore, not profit at all from the developing ability to remember episodically.

The self-other contrast used by Russell et al. (2010) helps isolate the prospective aspects of anticipating future needs. It also allows for projection into the past, unfortunately, not for remembering any past event but only for figuring out what was needed for a past action as can be seen from their example test question: “… point to the two things you think the little girl had to have to play blow-football on the blue side?” (Section 2.1.3). Even for prospection this method depends on the risky background assumption that the difference between predicting one’s own future needs and predicting someone else’s future needs reflects an inability of projection, which is primarily applicable to oneself and not to others. For believers in simulation theory projection is also applied in the case of others. More generally, the difference can be due to factors that tend to interfere more strongly in the case of self than other.

In sum, the claim for a specific developmental relationship between the ability to retrospect (episodic remembering) and prospection rests on evidence that can be given a more general interpretation (e.g., thinking about different times) or fails to provide the needed correlations: Perner and Ruffman (1995), Naito (2003) and Perner et al. (2007) only provide evidence for retrospection and ToM but not prospection. Russell et al. only provide evidence for prospection. Our prime objective is to provide evidence for a specific developmental relationship between retrospection (episodic remembering) and prospection.

To capture retrospection we use the free-cued recall contrast in our first experiment and the experienced-indirectly conveyed contrast in our second experiment. For measuring children’s prospective abilities (the study by Russell et al. was published well after our research was conceived and conducted) we relied on children’s ability to engage in mental rotation. We used the child appropriate simplification by Estes (1998) of the original tasks by Shepard and Metzler (1971). Children were shown pairs of teddy bears (monkeys in Estes’ original study) with one arm raised. Children had to judge whether the monkeys had raised the same side arm (e.g., both their left arm) or an arm on different sides (one left the other right). This is easy if the creatures stand side by side but increasingly difficult if one of them is rotated sideways (see Fig. 1). An easy solution is to take the tilted picture and rotate it back to upright. However, children are not allowed to do this physically. In that case one can rotate it mentally by forming an image of what one sees and rotate the image and then compare the rotated image of the tilted figure with the perception of the other figure. This procedure is an instance of prospection: one pre-experiences an actual rotation of the tilted figure. And the reaction times tell us whether a child used this method because mental rotation results in a linear increase of reaction time with the degree of rotation (Shepard & Metzler, 1971). Estes (1998) reported that children start using this method between 4 and 6 years.

Fig. 1

Examples of stimuli used for the mental rotation task in both experiments.

Experiment 1

Method

Participants

Fifty-nine children (25 girls and 34 boys) participated in the study. Children came from a nursery school in Upper Austria and two after-school care clubs, one in the city of Salzburg and one in Upper Austria. Most participants came from a middle-class background. Children’s ages ranged from 4, 11 (years, months) to 8, 7 (M = 6, 8, SD = 13.67 months).

To analyze and display age trends, we divided the children into three approximately same sized age-groups: Twenty children from 4, 11 to 6, 0 (M = 5, 5, SD = 3.99 months), 17 children ranging in age from 6, 1 to 7, 4 (M = 6, 8, SD = 5.51 months), and 22 children ranging in age from 7, 5 to 8, 7 (M = 7, 11 SD = 4.4 months).

Design

Each child was tested individually in a quiet room of the nursery school or after-school care club. Children were given two memory tasks (one with free and one with cued recall). In addition, children received a computerized mental rotation task based on Estes (1998) and an age-appropriate measure of verbal intelligence (verbal subtest of KISTE, Häuser, Kasielke, & Scheidereiter, 1994; or HAWIK-III, Tewes, Rossmann, & Schallberger, 1999).

All tasks were administered in two sessions a few days apart. Each session started with presentation of items for the memory task and ended with recall of memory items. Between presentation and recall of memory items, half of the children were given the mental rotation task in the first session and the verbal intelligence measure in the second session; the other half received these tasks in the opposite order. Half of the children received the free recall task in the first session and the cued recall task in the second session; the other half started with the cued recall task.

Procedure and materials

Mental rotation task (Estes, 1998)

In a warm-up phase, children were presented with an odd one out game. They were shown three bears in upright position (two of them both raising the same arm and one of them raising a different arm). Children were asked to select the bear that was different from the other two.

Then, children were asked to play a computerized mental rotation task presented on a notebook computer using the software package Presentation (Neurobehavioral Systems Inc., http://www.neuro-bs.com). Children were told to press the button with the smiling face if two bears both raising the same arm appeared on the screen and to press the button with the sad face if two bears each raising a different arm appeared.

After experimenter and child had jointly completed six training trials, children were asked to play on their own. The test phase comprised 56 trials. Each trial consisted of a pair of bears. The bear on the left was always upright, whereas the bear on the right appeared in seven different orientations. He was either upright (i.e., 0° rotation) or rotated clockwise in 30° increments up to 180°. In addition, each bear raised one of his arms. For half of the trials, both bears raised the same arm (i.e. both right or both left). For the other half of trials, the bears raised different arms (i.e. the left bear raised its right arm and the right bear raised its left arm, or the reverse). In total, there were 28 different stimulus pairs: two “same” and two “different” pairs for each of the seven different orientations. Each child was given these 28 different stimulus pairs twice in a fixed random order. After the first 28 trials, there was a short 2 min break.

Children had to press as fast as possible the button marked with a smiling face if the two bears had raised their same side arm and the button marked with a sad face if the bears had raised different side arms. Children had to keep their index fingers positioned on the response buttons throughout the trial block. Children were asked if they were ready before the experimenter initiated the next trial. Correct responses were followed by a brief tune, errors by silence. No other feedback was given. After Trials 5, 28, and 56, children were asked how they could tell if the two bears had raised their same arms or not. Sample stimuli are shown in Fig. 1.

Memory tasks

For the memory tasks two sets of 20 coloured pictures (21 × 29.7 cm) of familiar objects or animals were created (similar to Perner & Ruffman, 1995, Exp. 1). Each set comprised five categories, with four items in each category (see Appendix A for list of items). Pictures were presented in a fixed random order. Each child received both sets. Half of the children were given Set 1 for the free recall task and Set 2 for the cued recall task. For the other half the two sets were exchanged.

Children were told to look carefully at each picture they are going to see, because they will be asked about them later. Each picture was shown for 5 s and the child had to name it. Slight misidentifications (e.g., calling the farm a barn) were accepted. Completely wrong answers were corrected.

After the intervening task (mental rotation or verbal intelligence measure), children were reminded: “I’ve shown you some pictures earlier on. Can you remember?” In the free recall condition, children were then simply asked “What was in these pictures?” In the cued recall condition, they were asked for each category, e.g.: “There were some animals, what were they?”

The number of correctly recalled items and of false alarms (items that had not been on the learning list) were recorded. The difference between the number of correct recalls and the number of false alarms was used as a measure of recall accuracy.

Results and discussion

First, children’s performance on the memory tasks and on the mental rotation task was analyzed separately, and then we looked at the relationship between these tasks.

Memory tasks

Recall accuracy (hits minus false alarms) on the two memory tasks was analyzed by an analysis of variance with age-group (younger, middle, older) as between participants factor and recall (free, cued) as a within participants factor. There was a significant main effect of recall, F(1, 56) = 49.61, p < .001, partial η2 = .47. No other effect was significant p > .50. Correlation between free and cued recall accuracy was r = .17, p > .19.

Mental rotation

We first looked at children’s number of correct answers. There was a clear bimodal distribution. One group of 24 children had values from 24 to 38 normally distributed around the mean of 30.21 correct, which is just two items more than the guessing level of 28 (of 56 items): t(23) = 3.35, p < .003. Presumably these are the children who did not resort to rotation, without which they could only judge about two items consistently correctly. There were no children with values 39–41 and the remaining 35 children attained values from 42 to 55 with a mean of 49.89 correct.

To see how children managed to give correct answers they were classified as rotators or non-rotators according to the two criteria used by Estes (1998). For each child a regression analysis was computed for the median reaction times for each angle as dependent and angle of rotation as independent variable. When the regression coefficient (slope) was significantly different from zero the child was classified as a “rotator according to RT”. Children were also asked at the end how they had approached the task. If they indicated that they had rotated the stimuli in their mind (e.g., “I rotated the bear in my head”) they were classified as “rotators according to explanation”. The other children gave no insightful answer (e.g., “I just know it”). Correlation of these measures with the number correct solutions showed r = .64 for explanation, r = .66 for reaction time, and r = .67 for rotation by reaction time and explanation. We use this binary classification as “rotators” (rotators by RT and by explanation) vs. “non-rotators” for all further computations (we checked that the interpretations stay the same if anyone of the other two is used). This way of classifying children into rotators and non-rotators had a marginally significant relation with age (r = .25, p = .052) but a clear relationship with verbal intelligence (r = .40, p = .002).

Interrelations between retrospection and prospection

The argument about the relationship between type of recall and episodic remembering is that both kinds are (can be) helped by the availability of episodic traces. This can be equally strong, so that no interactive effect of the availability of episodic traces on free vs. cued recall can be expected. However, free recall is more dependent on episodic traces than cued recall (which profits to a large degree from the availability of the cues). Availability of episodic traces should, therefore, correlate more strongly with free than with cued recall, since more alternative factors influence cued than free recall. The theoretical claim is that the ability for episodic remembering requires the ability to project oneself into the past and that the same projective ability is required in the mental rotation task by projecting oneself as an observer of a future or hypothetical action (rotating one of the items to be compared). On those grounds we expect that mental rotation should show a significant correlation with free recall even when any correlation with cued recall has been partialled out.

Table 1 shows the correlations between relevant variables in the upper right corner above the main diagonal. Partial correlations with age and verbal intelligence taken into account are shown for the remaining variables in italics below the main diagonal. The correlation between rotation and free recall remains at least marginally significant but the correlation between rotation and cued recall reaches only about half its size. More importantly, when cued recall is partialled out the partial correlation of rotation with free recall remains clearly significant (pr = .31, p = .016). This is as predicted when retrospection (as measured by free recall beyond cued recall) is developmentally linked with prospection (as measured by mental rotation). In contrast, when free recall is partialled out the partial correlation of rotation with cued recall is weaker but does stay marginally significant (pr = .25, p = .06), which attests to the fact, pointed out earlier, that both free and cued recall profit from the ability of episodic remembering to project oneself back into the past. This highlights the need for a more discerning measure of episodic remembering in the next experiment.

Table 1

Correlations and partial correlations with age and verbal intelligence controlled.

	1	2	3	4	5
1. Age	–	.17	.25+	.10	.18
2. Verbal intelligence		–	.40⁎⁎	.29⁎	.39⁎⁎
3. Mental rotation			–	.35⁎⁎	.29⁎
4. Free recall: accuracy			.26+	–	.17
5. Cued recall: accuracy			.14	.06	–

Note: Numbers in italics – partial correlations after partialling out age and verbal intelligence.

p < .06.

p < .05.

p < .01.

Experiment 2

Experiment 1 provided some evidence that the development of prospection, assessed by the ability to engage in mental rotation, is specifically linked to retrospection (free recall) when controlling for general ability to retrieve information about the past (cued recall). The differential performance on free over cued recall is a very weak measure of episodic remembering (retrospection) since both types of recall can benefit to equal amounts from the availability of episodic traces. The only difference is that the overall variance of free recall should be more strongly dominated by episodic abilities than the variance of cued recall. This weakness is also reflected in the controversy over Tulving’s (1985) claim that relatively more items recalled in free recall should be judged as “remembered” than in cued recall (Jones & Roediger, 1995; Roediger & McDermott, 1995; vs. Hamilton & Rajaram, 2003).

In this second experiment we try to get stronger evidence for a specific developmental link by employing a contrast that depends more directly on the availability of episodic traces. A suitable contrast is between (directly) experienced events and events about which one was indirectly informed (e.g., through verbal or pictorial media). Only experienced events can be re-experienced. Though imagined experiences can be mistaken for re-experiences (false memories) this should be relatively rare if children are not instructed to vividly imagine the event about which they are indirectly informed. Under these premises we can predict that development of retrospective abilities (episodic remembering) should uniquely improve recall of experienced events but not at all (except for the occasional false memory) recall of indirectly conveyed events. Following the method used by Perner et al. (2007) we had children put cards with different pictures into a box. In the direct-experience condition the children saw the picture on each card as they put it into the box. In the indirect-information condition they put the cards into the box when being blindfold and afterwards they were shown on a monitor the pictures on those cards. We expect that children’s ability to engage in mental rotation should coincide with improved recall of directly experienced items but not of indirectly conveyed items.

Thirty-one children (16 girls and 15 boys) from two nursery schools in towns near the city of Salzburg participated in the study. Most participants came from a middle-class background. Children’s ages ranged from 5, 0 (years, months) to 6, 4 (M = 5, 10; SD = 4.72 months). For later analysis children were divided into a younger (n = 13, M = 5, 5; ranging from 5, 0 to 5, 10) and an older group (n = 18, M = 6, 1; ranging from 5, 11 to 6, 4).

Each child was tested individually in a quiet room of the nursery school. Children were given two memory tasks (one with direct experience and one with indirect information). In addition, children received the mental rotation task based on Estes (1998) described in Experiment 1. All tasks were administered in three sessions a few days apart. In the first session, children received the mental rotation task and a modality-specificity task in counterbalanced order.2 In the second and third session, one of the two memory tasks was given. In each session, presentation and recall of memory items were separated by a 10-min delay. Half of the children started with the direct-experience condition and were then given the indirect-information condition. For the other half, the order was reversed.

The mental rotation task (Estes, 1998) was administered as described in Experiment 1.

For the memory tasks, four sets of 12 coloured pictures (21 × 29.7 cm) of familiar objects, animals, or human beings were used. Set 1 and Set 2 were administered in the first session. Set 3 and Set 4 were given in the second session. In each session, half of the children were given a particular set as test items and the other set as distractor items. For the other half, the two sets were exchanged, so that each set was equally often used as test or distractor set in the direct experience and in the indirect-information memory task.

The experimenter presented the 24 pictures (test and distractor items), one after the other, alternating between items from each set. As each picture was produced, the child was asked to name it. If necessary (which was rarely the case), the experimenter provided the correct label.

In the direct-experience memory task, children were then asked to place the 12 test items into a (26 × 40 × 4 cm) box. They were allowed to look at each picture for 2 s and were instructed to keep these pictures in mind. In the indirect-information memory task, children were also asked to place the 12 test items into a box but they were blindfolded so that they could not see the pictures. After having placed the 12 items into the box, they were shown “what these 12 pictures had depicted” by means of a computerized presentation. Each picture was shown for two seconds, and children were instructed to keep these pictures in mind. After a 10-min delay, children were asked in both memory tasks, “Do you remember the pictures you put into the box?” If children answered with an item that was not put inside the box it was scored as a false alarm.

On both memory tasks, the number of correct recalls (hits) and the number of false alarms (if children mentioned items that were not put inside the box) were recorded. Mostly, false alarms comprised items from the distractor set used in the familiarization phase. The distractor set was introduced to get a more precise measure of remembering the pictures being put into the box as opposed to mere familiarity with the pictures. The contrast between recall of items placed inside the box with false alarms (mostly items that children were familiarized with but did not put inside) sharpens the detection of episodic memories in contrast to mere familiarity answers. For instance, a child who recalls two correct items and no distractors is comparable with one who recalls 12 targets and 10 distractors. Both children vastly differ in recall of familiar items (pictures) but are similar in terms of memory for items placed inside the box. This is critical, because our manipulation of direct experience and indirect knowledge pertains to the placing of cards into the box, not to familiarity with the pictures (they are directly experienced in both conditions). For this reason it is essential to rely not only on the quantity of pictures recalled (number of placed pictures recalled in relation to all pictures placed into the box) but also check for accuracy (number of placed pictures recalled in relation to all pictures recalled; as this terminology is used by Koriat and Goldsmith (1996, p. 177)). A measure that captures both these aspects common in signal detection theory is d′ which is based on the difference between hits and false alarms. This is typical for forced choice recognition tasks and rarely used with recall (Koriat & Goldsmith, 1996, p. 183). Following this approach, we use the difference between number correct items recalled minus number false alarms as our critical indicator of episodic recall and refer to it, for want of a shorter label and in line with signal detection theory, as: recall accuracy.

Results

First, children’s performance on the memory tasks and on the mental rotation task was analyzed separately, and then we looked at the relationship between these tasks.

Recall accuracy (hits minus false alarms) was subjected to an analysis of variance with age-group as between participants factor and experience (direct, indirect) as a within participants factor. There was a significant interaction between age-group and experience, F(1, 29) = 4.31, p < .047, partial η2 = .13. Whereas the recall of directly experienced items increased with age from 2.69 to 3.56 items, recall of indirect items declined with age from 3.46 to 2.50 items. This unexpected decline will be discussed below when looking at the relationship with mental rotation (prospection). No other effect was significant p > .70. Correlation between direct and indirect recall accuracy was r = .42, p = .019.

The results mirror those of Experiment 1. The numbers of correct answers range from 28 to 54. There was again a strong correlation (r = .57, p = .001) between number correct solutions and rotation by explanation (r = .38, p = .033), rotation by reaction time (r = .46, p = .01), and rotation by reaction and explanation (r = .58, p = .001). As in Experiment 1 we use rotation by reaction time and explanation for our further analyses. This classification had a marginal correlation with age in Experiment 1 but this time it had no correlation with age at all (r = .00), which is probably due to the much narrower age range in this experiment.

The relation between retrospection and prospection

The contrast between directly experienced versus indirectly conveyed events works differently than the contrast between free and cued recall in Experiment 1. While free and cued recall can profit from episodic remembering, only directly experienced events can do so, because only experienced events can be re-experienced. This means that the emerging ability for episodic remembering should enhance recall of directly experienced events but not recall of indirectly conveyed events. Moreover, re-experiencing past events (episodic remembering) and pre-experiencing imagined events for mental rotation require the same ability of projecting oneself as an observer. Hence, we expect that children who can use mental rotation will have better recall of directly experienced events than of indirectly conveyed events. Children, who fail to use rotation, will show no, or at least a much reduced, difference in recall.

An analysis of variance of recall accuracy (number of hits minus false alarms) with rotation (rotation by RT and explanation: rotators vs. non-rotators) as a between subjects factor and experience (experienced vs. indirectly conveyed events) showed significant main effects of rotation (F(1, 29) = 5.16, p = .031, partial η2 = .15) and experience (F(1, 29) = 4.51, p = .042, partial η2 = .13), and a highly significant rotation × experience interaction (F(1, 29) = 7.83, p = .009, partial η2 = .21). This interaction was predicted. However, the form of the interaction as displayed in the left panel of Fig. 2 harbours a surprise. The ability to mentally rotate does not lead to enhanced memory for experienced events but has a detrimental effect on trying to remember indirectly conveyed events.

Fig. 2

Results for rotators and non-rotators on three different memory measures in Experiment 2.

This decline in accuracy for indirectly conveyed events persists for rotators-by-RT but not for rotators-by-explanation (p = .016), although the pattern of results is still very similar. This sharp decline (by 3.2 items) in accuracy (hits minus false alarms) is, as the centre panel in Fig. 2 shows, less pronounced (1.6 items) for hits (items correctly recalled) but still significant (interaction: (F(1, 29) = 4.81, p = .037, partial η2 = .14)). At the same time the number of false alarms increases more sharply for indirectly conveyed items than experienced items (F(1, 29) = 4.45, p = .045, partial η2 = .13) as the right panel in Fig. 2 shows.

Discussion

The ability to engage in mental rotation had a strong differential effect on children’s ability to recall directly experienced events and to recall indirectly conveyed events. The finding, however, contained a great surprise: the differential effect did not consist in rotators’ better recall of experienced events but was due to their worse recall of indirectly conveyed events. This unexpected decline in recall of indirectly conveyed items does, nevertheless, confirm a trend observed in two experiments by Perner et al. (2007), where recall of indirectly conveyed items declined with increased theory of mind competence. What seemed initially a curious and haphazard finding now looks increasingly robust: As children improve their theory of mind and develop mental rotation skills their recall of indirectly conveyed information declines.

It is difficult to find an obvious explanation for this decline. One could suspect that the reason lies in the unusual procedure of the indirect condition. But why should the more sophisticated rotators become confused by this procedure, when adults find it unusual but clear and perform as well as in the more normal experience condition (Stöttinger, 2006). The only plausible but highly speculative explanation has been briefly alluded to by Perner et al. (2007, p. 480). The younger children use their non-episodic recall (knowledge retrieval) for both conditions more or less successfully. Then when their theory of mind competence increases and become able to project themselves as past perceivers of experienced events they switch to using episodic recall. This switch may (Perner et al., 2007) or may not (current Experiment 2) help them immediately to recall more items than with non-episodic recall. In any case, this switch does not work for indirectly conveyed items and children’s performance declines. Presumably some time after discovering episodic recall children will realize that it does not work in every case and will readjust their strategy for indirect information.

Eventually they may also discover the possibility of indirect episodic recall for indirect information. That is, when told about an event they should not try to re-experience (episodically remember) the event (e.g., putting the crocodile into the box) but to remember the information event (i.e., remember the crocodile appearing on the monitor and the instructions that pictures on the monitor were on the cards put inside the box). It is left to future studies to explore this development.

General discussion

We have presented some data showing that prospection (as measured by children’s ability to engage in mental rotation) is related to retrospection (episodic remembering: as measured by variance shared with free recall when controlling for cued recall and by the difference between recall of experienced events (which can be re-experienced) versus recalling indirectly conveyed events (which cannot be re-experienced). Our expectation was that recall of indirect information would be unaffected while recall of experienced events, which can be re-experienced, would increase with children’s ability to mentally rotate.

Contrary to expectations recall of experienced events was unaffected while children’s recall of indirect information dropped drastically with their ability to engage in mental rotation. Although, retrospectively (sic!), this trend was already apparent in investigations of children’s theory of mind competence (Perner et al., 2007), the sharpness of this trend with mental rotation came as a surprise. The only plausible explanation we found was that acquisition of mental rotation marks a change in recall strategy. Instead of retrieving knowledge as before, children expect to find answers by trying to re-experience the relevant events. This is a successful replacement for the old strategy in the case of experienced events but fails miserably for indirect information about events. Despite these unexpected results we were able to demonstrate a developmental link between retro- and prospection that is Difficult to reduce to a more general connection, e.g., understanding time, which could account for existing data as reviewed in the Introduction.

The specific connection we identified is children’s ability to project themselves mentally as experiencing events. This characterisation is commensurable with the neurocognitive suggestions of self-projection (Buckner & Carroll, 2007) and scene-construction (Hassabis & Maguire, 2007). In fact, on our view these two abilities are necessary complements: To project oneself as an observer of events at different times one has to be able to project one self into different times AND be able to reconstruct the scenes that one is to observe.

Our results are less compatible with the ages observed in earlier developmental studies. Results from many of the studies reviewed earlier that looked at prospection suggest that the main development takes place between 3 and 5 years, though not from all studies. For instance, Atance and Meltzoff’s (2006) report that 5-year olds still could not imagine/anticipate that their current desire for water induced by overfeeding on salty pretzels will have reverted next day to their usual desire for pretzels. The fact that prospection as measured with mental rotation points to later development might mean that the studies that see the development completed by 5 years did not assess children’s ability to project themselves as observers (experiencers) of events at different times but assessed their knowledge about events in the past or future.

Research on children’s retrospection (episodic remembering) also finds development in this age bracket but clearly also points to improvements beyond the age of 5 years. The original studies by Perner and Ruffman (1995) and also Naito (2003) included children up to 7 years and their more difficult theory of mind tasks tended to show the stronger correlations with measures of episodic remembering. These findings are, therefore, more in line with the developments documented in the present studies.

We want to end with more general considerations about the methodological difficulties in assessing young children’s and non-verbal creatures’ explanations and subjective phenomenology. The autonoetic awareness of re-experiencing a formerly experienced event is something very subjective, in the sense that evidence for it comes exclusively from introspective verbal reports. With adult participants in memory experiments the relevant experience has been assessed by the “R–K” test. People have to judge their subjective experience of recall or recognition as one of merely knowing (K) what had happened (presentation of an item) or remembering (R) this event. This test is less than straight forward with adults, its value highly controversial (Donaldson, 1996; Dunn, 2004; Gardiner & Richardson-Klavehn, 2000; Stöttinger, Aigner, Hanstein, & Perner, 2009; Stöttinger, Kaiser, & Perner, 2009) and its use for young children beyond contemplation. The only solution seems to be an objectively measurable indicator of these subjective phenomena.

This enterprise requires a fine instrument in order to distinguish between knowing what happened and re-experiencing (remembering) the event retrospectively, and between knowing what will happen and pre-living the likely event in prospection. The instrument has to be precise because both kinds of mental processes can, in principle, produce the required information about the world. They are “computationally equivalent”: both can give the correct answer to the questions posed, e.g.: What happened? Which item was presented? The difference must primarily lie in how the system arrives at the answer. Cognitive Psychology can, supposedly, provide the relevant instruments. So, no wonder that one of its most impressive findings, mental rotation, has such appeal in this context. The critical point is not that mental rotation can provide the correct same–different judgments—there may be infinitely many computationally equivalent ways. It is the pattern of reaction times that tells us that the answer is arrived at by mental rotation as we subjectively experience it. However, even the reaction time pattern is not a fail safe indicator of prospection.

Several species of animals can give correct responses, e.g., pigeons, but do not show the same linear pattern of reaction times with angle of rotation (Hollard & Delius, 1982). Their visual system directly provides rotation invariance and they do not have to mentally simulate perceiving a hypothetical rotation. However, the reaction times of baboons (Vauclair, Fagot, & Hopkins, 1993), a sea lion (Stich, Dehnhardt, & Mauck, 2003) and a lion-tailed macacque (Burmann, Dehnhardt, & Mauck, 2005) also show some signs of the linear relationship with angle of rotation, though their times deviate from the human pattern in ways adaptive for their respective perceptual environment. For argument’s sake let us assume their reaction times showed exactly the same pattern. We could still not conclude that these animals project themselves as perceivers of hypothetical rotation events (mental time travel). Their visual system might provide rotation invariance but processing time happens to depend on angle of rotation. Thus, the ultimate justification for interpreting the linear reaction time pattern in humans as a sign of mental time travel hinges on the fact that this linearity goes hand in hand with the subjective experience of mental rotation in adults. This relationship between observable answer and subjective experience in adults is not so tight for other measures. For instance, when asked what to pack for a trip to a place shown in a hibernal scene (Atance & Meltzoff, 2005) my answer “winter coat” could be given by projecting myself travelling to that place and noticing the need for a winter coat. However, it could equally well be based on just knowing that in places like this you need a winter coat. Mental rotation (provided the linear reaction time pattern does indicate it) is a much less vulnerable indicator of prospection.

For investigating episodic remembering (retrospection) in animals and young children several proposals have been made. The technically most sophisticated has been developed by Yonelinas (1999) for distinguishing between recall based on familiarity versus recollection, which can be profitably used even with rats (Sauvage, Fortin, Owens, Yonelinas, & Eichenbaum, 2008). Unfortunately the distinction between familiarity of items and recollection of associative information is not quite the same as between knowing and remembering. Associative information can be known or remembered. We need methods that drive a wedge between knowing the past and remembering the past. Perner and Ruffman (1995) followed Tulving’s criterion that free recall produces relatively more experiences of remembering than cued recall. Unfortunately the link to the one direct method of assessment of the remember–know paradigm remains tenuous (Hamilton & Rajaram, 2003; Jones & Roediger, 1995; Roediger & McDermott, 1995). Perner et al. (2007); our Experiment 2) relied on an objective prerequisite for episodic remembering, i.e., the direct experience constraint (Stöttinger, Aigner, et al. 2009; Stöttinger, Kaiser, et al., 2009): only an experienced event can be re-experienced.

Neither of these methods is apt to produce positive demonstrations of a child enjoying a re-experience of a past episode. These methods only allow an objective test of developmental hypotheses: if some other ability (theory of mind, mental rotation) can be claimed to index the onset of re-experiencing then the method allows for a testable prediction. The development of the index ability should have a differential effect on recall conditions (free vs. cued; recall of directly experienced vs. indirectly conveyed events).

Although the method falls short of giving a direct existence proof of episodic remembering it does provide an empirical test of hypotheses about episodic remembering that could also be used on animals. As it has been applied here it requires some linguistic proficiency for the indirect-information condition. It would not work without being able to tell children that the items shown on the monitor are the same as on the cards they could not see. Indirect knowledge through verbal communication is difficult to use with animals but there are dogs (Miklósi, Polgárdi, Topál, & Csányi, 2000) and goats (Kaminski, Riedel, Call, & Tomasello, 2005), who understand pointing gestures, and there are chimpanzees who can infer from seeing that one of two containers is empty that the object must be in the other container (Call, 2004). So, if we can find a theory of which individuals (ontogenetic development, enculturation, evolution) are able of episodic remembering, then we can provide testable predictions of how these individuals differ in recalling directly observed locations vs. gestured or inferred locations.

Clayton and Russell (2009) made a new suggestion of how to differentiate knowledge of the past from remembering the past empirically without use of language. Their minimal requirement for remembering is evidence during re-experience of the subjective perspective of the original experience. This is clearly a highly relevant aspect—but an important limitation has to be pointed out. Like the direct experience constraint, this criterion cuts only one way: features of subjective perspective at recall do not licence claims about re-experiencing. For, perspective features could result from how knowledge of the past is encoded. To use an example from Davies, Russell, and Russell (2009), children observed how to fix a toy boat by performing on a left or right lever either a pumping or levering motion. By 3 years most children imitated after some delay a series of two consecutive actions in the correct spatio-temporal order. The authors suggested that this shows earlier episodic remembering than commonly expected, because children reproduced the temporal perspective of their original experience. The problem with this conclusion is that the study confounds the subjective order of experiences with the objective knowledge of the order in which the actions were performed.

What one might be able to do is the following. Let us assume children below 3 years can remember sequences of events to some degree. One would then have to show that at the purported age of 3 years children show a marked change in the order in which they recall witnessed events but do not show any change in the order in which they report the events having happened. This method put together with the direct experience vs. indirect information contrast (Perner et al., 2007) may provide a more powerful combination. Reports of events need not follow the order of the events. So, particularly strong evidence would be forthcoming if at the critical age children began recalling events in the order they had been reported rather than the order in which they happened, e.g., when told “he ate after his nap” the younger ones recall, “he slept”–“he ate,” while the older ones recall, “he ate”–“he slept”. This kind of reversal would provide an interesting analogy to our finding that rotators started to have problems with indirectly conveyed events because—our suggestion—they switched from knowledge retrieval to episodic remembering.

There has been a certain resentment among animal and infancy researchers against insisting on Tulving’s (1985) and Wheeler et al. (1997)) criteria for episodic memory because they seem to be impossible to meet for non- or low verbal creatures. With our research and this final discussion we want to make the case that to capture the phenomenon of episodic remembering we need to insist on these criteria but that there is a way of testing theories of when episodic remembering develops or evolves.

Set 1	Set 2
Tiere (animals)	Pflanzen (plants)
– Schaf (sheep) – Pferd (horse) – Schwein (pig) – Kuh (cow)	– Baum (tree) – Sonnenblume (sun flower) – Rose (rose) – Getreide/Weizen (wheat)
Gemüse (vegetables)	Obst (fruit)
– Karotte (carrot) – Erbse (pea) – Zwiebel (onion) – Paprika (red pepper)	– Birne (pear) – Weintrauben (grapes) – Kirschen (cherries) – Apfel (apple)
Möbel (furniture)	Kleidung (clothes)
– Bett (bed) – Tisch (table) – Stuhl (chair) – Schrank (wardrobe)	– Pullover (sweater) – Schuhe (shoes) – Socken (socks) – Hose/Jeans (trousers/Jeans)
Gebäude (buildings)	Fahrzeuge (vehicles)
– Kirche (church) – Burg (castle) – Haus (house) – Scheune (barn)	– Bus (bus) – Auto (car) – Lastwagen (lorry/ truck) – Motorrad (motorbike)
Werkzeuge (tools)	Musikinstrumente (musical instruments)
– Säge (saw) – Schaufel/Spaten (spade) – Zange (pliers) – Hammer (hammer)	– Trompete (trumpet) – Klavier/Piano (piano) – Geige/Violine (violin) – Trommel (drum)