Introduction
Reversible or ambiguous figures like the Rubin’s face/vase picture or the Necker cube have been used to study how people spontaneously alternate between two mutually exclusive interpretations of objectively stable pictures. The ability to reverse ambiguous figures depends on a combination of top–down and bottom–up processes (Intaitė, Noreika, Šoliūnas, & Falter,
2013; Long & Toppino,
2004), including recurring neural fatigue (review in Long & Toppino,
1981,
2004), gaze orientation (Ruggieri & Fernandez,
1994), mental imagery (Doherty & Wimmer,
2005) and context effects (Intaitė et al.,
2013). A critical factor that determines whether participants are able to reverse an ambiguous figure is the amount of information given about the two potential interpretations (Mitroff, Sobel, & Gopnik,
2006). (1) When no information is given, one needs to be aware of the ambiguity (i.e., uninformed reversal). Adults rarely ever reverse spontaneously (Rock & Mitchener,
1992). (2) When aware of the ambiguity, one needs to explore or generate potential alternative interpretations (i.e., ambiguity-informed reversal). Under these conditions, about 50% of adults are able to recognize the second interpretation (Girgus, Rock, & Egatz,
1977). (3) When both ambiguity and content are known, the perceiver must be able to conceive of a figure having more than one interpretation to flexibly alternate between those two (i.e., content informed reversal), a requirement that was only initially evident in children (Doherty & Wimmer,
2005; Gopnik & Rosati,
2001).
Children up to 5 years almost never show uninformed reversals (Girgus et al.,
1977; Rock, Gopnik, & Hall,
1994; Rock & Mitchener,
1992)—and even struggle with content informed reversals (Gopnik & Rosati,
2001; Rock, Hall, & Davis,
1994; Wimmer & Doherty,
2011). In fact, even the majority of 5- to 9-year olds fail to reverse ambiguous figures spontaneously (Mitroff et al.,
2006). Although reversal rates increase throughout childhood (Holt & Matson,
1976), they still do not reach adult level by age 10 (Ehlers, Strüber, & Basar-Eroglu,
2016).
This developmental phenomenon is compelling and requires further investigation for two reasons. A late onset may either reflect an ontogenetically emerging basic, low-level perceptual process or a cognitively demanding, late-manifesting process. Previous research has produced mixed findings. Gopnik and Rosati (
2001) showed that reversals were closely linked with tasks that elicited non-perceptual, multiple representations (e.g., false belief task), suggesting that the ability to reverse depends upon late-developing cognitive abilities. Yet, more recent evidence (Doherty & Wimmer,
2005; Wimmer & Doherty,
2011) does not find a direct relation between false belief performance and switching interpretations. The current set of studies was designed to further investigate this question.
Another reason to investigate this phenomenon developmentally is that children, even when informed about the ambiguity, are unlikely to reverse. Ambiguity-informed reversals explicitly require children to explore or generate potential alternative interpretations. Children, however, fail to switch between interpretations—even when they have acknowledged their existence beforehand (Doherty & Wimmer,
2005). While children may simply be reluctant to switch interpretations in view of physically unchanged objects, this seems unlikely to explain their difficulties, given that children were encouraged to report any changes they saw (Gopnik & Rosati,
2001). Alternatively, once children understand that two interpretations are possible, they may struggle to (voluntarily) inhibit one interpretation over the other. An increased ability to reverse would then be an indicator of bottom–up development through successful inhibition (Bialystok & Shapero,
2005; Wimmer & Doherty,
2011).
Also, earlier studies reported an association between content informed reversal and theory of mind (Gopnik & Rosati,
2001; Mitroff et al.,
2006) which was not confirmed in later studies (Doherty & Wimmer,
2005; Wimmer & Doherty,
2011). Studies that found a positive association argued that considering someone else’s belief and understanding ambiguity both require children to have an abstract understanding of perspective (for a discussion see Perner, Stummer, Sprung, & Doherty,
2002).
In the current study, we used a continuous version of the ambiguous figures task. In this task, an animal (e.g., a duck) morphs over 15 iterations into another animal (e.g., a rabbit) with picture #8 depicting a well-known ambiguous figure (e.g., duck–rabbit, Wittgenstein,
1953). This picture morphing task (Stöttinger et al.,
2014; see also Burnett & Jellema,
2013) measures how many morphs participants need before they switch to the new interpretation (i.e., rabbit). Unlike in previous studies, children were not informed about the content of the second interpretation, only about its potential for another interpretation. Participants have to generate alternative perspectives (i.e., it could transform into a cat, a dog, a fish, etc.), and compare or rank perspectives (i.e., at some point it cannot be a fish anymore, given the ears) to report the target animal. Thus, possible alternatives cannot be chosen randomly but need to be consistent with the actual stimulus. Otherwise, the amount of potential alternative explanations would be overwhelming, ultimately hampering efficient updating.
The picture morphing task, therefore, provides several advantages over the standard ambiguous figure task: (1) the continuous measure makes it possible to test children younger than 5 years, who would have otherwise rarely reversed (Gopnik & Rosati,
2001). (2) It introduces variance into a typically dichotomous measure. (3) Most importantly, it allows to capture gradual developmental improvements and to quantify any specific deficits in this form of reversal.
In Experiment 1, we assessed children’s developmental progression on the picture morphing task and compared it to performance of adult participants exposed to the same stimulus material. We were particularly interested in whether younger children would need more morphing stages until they named the emerging animal compared to older children and adults. This task has never been administered in children before. It is therefore difficult to make any definitive predictions about how they will perform. In a similar vein, we can only speculate which other factors may influence recognition of the second object. We tested whether children who master the standard false belief task (Wimmer & Perner,
1983) will identify the second object in the picture morphing task sooner. In addition, we tested whether executive functioning and selective attention would influence children’s performance in the picture morphing task—both of which were previously hypothesized to affect performance on the ambiguous figures task. Increased inhibitory control (i.e., the ability to withhold the previous representation of the stimulus and exchange it for the conflicting option) could help children to ascribe a different meaning to the picture (Bialystok & Shapero,
2005). Wimmer and Doherty (
2011) found that performance on the day/night Stroop task (Gerstadt, Hong, & Diamond,
1994), a measure of inhibitory control, predicted ambiguous figure switching. We, therefore, used the same task in our study.
For exploratory purposes, we also recorded children’s eye movements to test whether picture morphing effects relate to individual differences in visual inspection of the pictures. In adult participants, impoverished control of selective attention led to lower sampling of informative parts of the stimulus (Tsal & Kolbet,
1985) and could explain variance in the picture morphing task (Ruggieri & Fernandez,
1994). Also, fixations recorded prior to object recognition predicted which interpretation of an ambiguous image was reported (Kietzmann, Geuter, & König, 2011). Wimmer and Doherty (
2007), however, failed to find any difference in eye movement patterns between children who reversed an ambiguous figure and children who failed to do so.
Results of Experiment 1 revealed that children in all age groups required more morphs to switch interpretations than adult controls, with no significant improvements within the child group. In Experiment 2, we tested older children (up to 9 years) to examine at which age the ability to report the second object in the picture morphing task improves.
Discussion
Children overall reported the second object much later (i.e., at picture #12) than adults (i.e., at picture #8) and only when the picture displayed predominantly the second object (Fig.
1a). Also, there was no clear developmental improvement despite older children outperforming younger children on theory of mind and inhibitory control (Table
1). While these factors have affected the probability of content informed reversal in previous studies (Bialystok & Shapero,
2005; Doherty & Wimmer,
2005; Gopnik & Rosati,
2001; Mitroff et al.,
2006; but see Ropar Mitchell, & Ackroyd,
2003), none of these measures affected performance on the picture morphing task (i.e., when children were naïve as to the second interpretation, but were continuously shifted towards that interpretation).
It is plausible to assume that children switched so late due to the rather impoverished silhouette type displays of objects. While we cannot fully rule out that children struggled more than adults with the stimuli, all children were able to name animals based on line drawings in the word naming check prior to the picture morphing task. Although stimuli in the middle were highly ambiguous and suggestive of more than one interpretation, “other” reports for children (and adults) did not peak at the most ambiguous figure. Instead, they were equally distributed across all picture positions (Fig.
1b). Critically, adults gave similar “other” reports as did children (e.g., “dinosaur” in the cat–swan picture set) and a comparable proportion of sets had to be excluded in both participant groups. Also, children and adults found it harder to reverse the horse–seal picture set compared to the other sets.
One may argue that the word naming check prior to the picture morphing task primed children to search for those same animals. This, however, should have made the task easier for children and cannot explain their delayed recognition of the second object. Children who used more extensive search strategies—as suggested by longer and a higher number of fixations as well as longer saccades—identified the second object earlier. For one of the picture sets, qualitative analysis of fixation patterns revealed an interesting trend. Children who switched later focused mainly on the eye of the figure, while children who recognized the second object earlier were also able to direct their attention to other regions. They still, however, did not reach adult level performance. Children in all age groups required more pictures before they reported the second object compared to adult participants, with no significant age improvements.
In Experiment 2, we, therefore, extended the age (up to 9 years) to examine at which age the ability to report the second object in the picture morphing task improves. In addition, we measured theory of mind and inhibitory skills, using a different type of measurement. Advanced theory of mind skills were measured using Happé’s strange stories (O’Hare, Bremner, Nash, Happé, & Pettigrew,
2009). This task was previously used to assess mentalizing skills in middle childhood (Devine & Hughes,
2013) and was consequently regarded as an appropriate task for our sample. Further, because the day/night Stroop task by Gerstadt et al., (
1994) was not related to the picture morphing task, we substituted the Stroop task with the Simon task (picture test) by Davidson, Amso, Anderson, and Diamond (
2006). This task is also appropriate for the age range tested but assesses slightly different skills. While the Stroop task in Experiment 1 requires the inhibition of conceptually incompatible stimuli, which is expected to be developed in middle childhood, participants in the Simon task need to inhibit responses that are spatially incompatible with the stimuli. In other words, it examines competition at the stimulus response level (Liu, Banich, Jacobson, & Tanabe,
2004), which we deem relevant in our task where local changes increasingly fail to match the global representation of the animal.
Although qualitative analysis of fixation patterns revealed an interesting trend, we did not include eye-tracking metrics in the follow-up experiment. We were able to confidently measure number and duration of fixations as well as saccade length in Experiment 1. Yet, the restricted number of valid sets for which we had eye tracking data only allowed us to link fixation location with the actual interpretation of the picture in one picture set in Experiment 1. While we would have had enough data for a similar analysis in two other picture sets (i.e., snail–whale: n = 27; horse–seal: n = 29), these sets did not produce enough variance in responses to separate groups into early and late switchers. That is, more than 80% of participants reported the second object at—or later than—picture #11. For meaningful insights from an ROI analysis we would have to select new picture sets that produce more variance. Because we further wanted to establish at which age children would reach adult level performance, we opted against changing the picture sets in Experiment 2 that would have allowed for such analysis.
General discussion
To date, there are only a handful studies investigating how children reverse ambiguous figures. These studies typically find that young children never spontaneously reverse (Girgus et al.,
1977; Gopnik & Rosati,
2001; Mitroff et al.,
2006; Rock & Mitchener,
1992; Wimmer & Doherty,
2011) and even 10-year olds still show lower reversal rates than adults (Ehlers et al.,
2016). Children until the age of five are only able to alternate between two interpretations of an ambiguous figure when they are informed about the second interpretation beforehand (i.e., content informed reversal). The present set of studies used a continuous version of the ambiguous figures task to measure how children switch to a second interpretation under conditions where they were naïve as to the second object and where each change in the picture increasingly supported the alternative interpretation.
In both experiments, we found that children switched much later than adult participants. Performance in the picture morphing task improved with age—as demonstrated in Experiment 2—with 8.5-year olds (
M = 9.49) starting to approach the level of adult participants (
M = 8.38). Our results are, therefore, in line with earlier findings showing an improvement with age on reversals of ambiguous figures (Ehlers et al.,
2016; Girgus et al.,
1977; Gopnik & Rosati,
2001; Mitroff et al.,
2006; Rock & Mitchener,
1992; Wimmer & Doherty,
2011).
We further investigated whether children who possessed advanced theory of mind understanding and strong inhibitory skills would recognize the second object sooner in the picture morphing task. None of these measures predicted performance in our continuous ambiguous figures task. Previous research has found inconsistent evidence for a link between reversals and false belief, with two studies showing a link (Gopnik & Rosati,
2001; Mitroff et al.,
2006); while more recent evidence (Doherty & Wimmer,
2005; Wimmer & Doherty,
2011) failed to find a direct link between false belief and reversals. Studies that found a relationship interpreted the link as reflecting a common reliance on actively contrasting perspectives. The late switch to the second object in the picture morphing series, however, suggests that switching requires a strong external drive in young children (i.e., when the picture represents predominantly the second object) which may not rely on the ability to abstractly contrast perspectives.
Instead, stronger inhibitory skills facilitated reversals in a previous set of studies (Wimmer & Doherty,
2011) which suggested that only if the current interpretation can be inhibited, can the second interpretation be identified. Critically, performance can either benefit from a top–down inhibitory insight (i.e., knowing what to inhibit) or from a more general bottom–up inhibitory strength (i.e., having enough power to inhibit; Wimmer & Doherty,
2011). If performance was predicted by the Simon task (Experiment 2), but not by the Stroop task (Experiment 1), it would favor the latter explanation. While the Stroop task assesses the ability to inhibit cognitive interference, the Simon task requires interference resolution at the stimulus response level. Neither the correlation with the Stroop task (Experiment 1) nor the correlation with the Simon task (Experiment 2) remained significant after controlling for age, which suggests that improvements in the picture morphing task are not explained by improvements in inhibitory strength.
The late shift in our child sample may instead reflect limitations in working memory capacity, particularly as working memory measures are typically correlated with inhibition measures (Davidson et al.,
2006). Indeed, switching to a new interpretation requires mental manipulation. In addition to perceiving the picture, participants have to go through different interpretation alternatives at the same time, a process that particularly young children may find too taxing. Data from a pilot study, however, speak against this explanation. There we presented 5- to 6-year-old children
6 with a manual version of this task where we asked them to sort the pictures into a “rabbit” or a “duck” box. Despite a clear reduction in working memory load (i.e., children had to simply compare the picture with the target pictures on the boxes), children still needed significantly more pictures before they reported the second object compared to adult participants reported in Stöttinger, Guay, Danckert and Anderson (
2018); [
F(1, 115) = 21.50,
p < 0.001,
η2 = 0.16]).
The exceptionally late shift could potentially reflect a motivation issue, especially in our youngest children. Odic, Hock, and Halberda (
2014) showed that a history of difficult, low-confidence perceptual decisions resulted in degraded performance in subsequent easier decisions in 4- to 5-year-old children. This perceptual hysteresis (i.e., “…the persistence of the initially established percept despite the evidence reaching values that favor the alternative percept”; p. 2; Odic et al.,
2014) was explained by a tendency to guess without paying attention to the actual sensory input. Only when the task was easy enough did they re-attend. Although a lack of trying could potentially account for our data, Odic et al. (
2014) only found evidence for perceptual hysteresis when children were provided with explicit (i.e., correct/incorrect) feedback. This effect disappeared after explicit feedback was removed. Given that no explicit feedback was provided in our picture morphing task, we deem it unlikely that low motivation due to constant discouraging feedback could fully account for our data. Also, our eye tracking data confirm that children were paying attention to the stimuli throughout the procedure.
It is also possible that the delayed switch to the second interpretation in the picture morphing task reflects a deeper conceptual issue in children. French, Menendez, Herrmann, Evans, and Rosengren (
2018), for example, showed that children (in particular) and adults (to a lesser extent) had persistent problems reasoning about certain types of changes in biological organism that occur over the life-span (e.g., the metamorphosis of a caterpillar changing into a butterfly). Confronted with an unfamiliar animal, children and adults apply the default cognitive constraint that physical features remain stable across the life-cycle. This constraint helps us to maintain an image of a stable world in which ducks do not change into rabbits. The acknowledgement of such metamorphosis increases with age as children become more familiar with the concept. This could explain why younger children retained a stronger bias to resist such changes than older children and adults.
Alternatively, eye tracking data of Experiment 1 showed that children who employed a more extensive exploration strategy (i.e., longer and higher number of fixations in Experiment 1) identified the second object earlier. Typically, a larger number of fixations are associated with being a novice to a scene (Kelly, Rainford, Darcy, Kavanagh, & Toomey,
2016), thus requiring more fixations to cover the visual space. In this respect, it is plausible that more fixations indicate acknowledging novelty to a stimulus in our task. However, none of our children were experts, and it is, therefore, possible that a larger number of fixations have led to a benefit of finding crucial cues. A focus on specific parts of an ambiguous stimulus (e.g., the rabbit’s ears) will favor one interpretation (rabbit) over the other (Tsal & Kolbet,
1985). Although Wimmer and Doherty (
2007) failed to find evidence for a causal relation between search patterns and reversals of ambiguous figures in 3- to 5-year olds, our qualitative ROI analysis did show an interesting trend: children who recognized the duck earlier showed a broader allocation of attention than children who identified the duck later. It remains unclear why some children adopted a more extensive looking pattern than others. None of the indicators that we had in mind (i.e., false belief and inhibition) correlated with eye-tracking measures.
However, even if children benefited from a higher number of fixations, and a more diverse looking pattern, only the oldes children started to approach adult level performance. Assuming that scanning strategies were not the main causal factor for this difference, successful recognition of the emerging animal may additionally require high-level (cognitive) search strategies. Adult participants consistently report the target object at a point when the picture still represents more the first object than the second. Interestingly, this is only the case when the picture is presented in a gradual context compared to when the same picture is presented outside of the morphing context (Egré, Ripley, & Verheyen,
2018; Stöttinger et al.
2018; Stöttinger, Sepahvand, Danckert, & Anderson,
2016). This effect is reversed after damage to the right hemisphere: right-brain-damaged patients need significantly more pictures to identify the emerging object when it is presented in the gradual compared to the individual condition (Stöttinger et al.,
2018). It is speculated that healthy adult participants use high-level “exploratory” search strategies (e.g., it started out as a duck, but what else could it be?)—an exploration strategy that may be impaired after damage to the right hemisphere (Danckert et al.,
2012; Mohammadi Sepahvand, Stöttinger, Danckert, & Anderson,
2014). Such exploration strategies may also be late developing in children.
The very late shift in children and in right-brain-damaged patients might be the result of a fundamentally different exploration strategy compared to healthy adults. When one object gradually morphs into another in the picture morphing task, focusing on locally changing elements does not support identification of the object. Only the ability to impose a holistic, “Gestalt-like” template (Van de Cruys et al.,
2014) of an imagined object onto the current item enables the exploration of potential fits (see
Rock et al. 1994; Wimmer & Doherty,
2011 for the same argument in the context of ambiguous figures). A local bias can explain why children and right-brain-damaged patients struggle to recognize the second object. The suggestion here is that for the patients, damage to the right hemisphere has disrupted global processing; while for the children, these processes have yet to fully develop (Moses et al.,
2002 for a review). For example, when confronted with a shape (e.g., the letter H) composed of smaller elements (e.g., small As), right-brain-damaged patients are able to remember details but not the overall pattern, while results are reversed in patients with a damage to the left side of the brain (Delis, Robertson, & Efron,
1986). Similarly, right-brain-damaged patients typically comment on the local changes in the picture morphing task (e.g., duck, duck with beak open, duck with a wider beak, duck looking up, etc.) but struggle to shift to a new interpretation (i.e., rabbit; Stöttinger et al.,
2014). The same local over global bias is found in children. For example, 6-year olds use a local “piecemeal approach” when asked to copy a Rey–Osterrieth Complex Figure. That is, they tend to copy one small detail after the other instead of starting with the global shape and adding details afterwards (Akshoomoff & Stiles,
1995a; Akshoomoff & Stiles,
1995b; Martens, Hurks, & Jolles,
2014). In a similar vein, when confronted with the Kanizsa illusion (i.e., black “pacman” shapes arranged to induce a contour illusion of a bright white triangle in the middle) 4-year olds focus more on the local—illusion inducing “pacman” shapes, while 7-year old children show an adult-like bias towards the global “illusory middle shape” (Nayar, Franchak, Adolph, & Kiorpes,
2015). This shift from a local to a global processing style is accompanied by a developing hemispheric specialization in the brain. Moses et al. (
2002), for example, showed that children who performed like adults in a hierarchical figure task (i.e., global shapes composed of smaller elements) also showed the same adult-like right hemispheric preference for global processing. While some studies report a transition from local to global processing around 6 years of age (Dukette & Stiles,
1996; Kimchi, Hadad, Behrmann, & Palmer,
2005; Martens et al.,
2014; Poirel, Mellet, Houdé, & Pineau,
2008), others find evidence for development further into the teenage years (see Nayar et al.,
2015 for a review). Qualitative ROI pattern analysis of our eye-tracking data supports the idea that a local bias protracts identification of the second object in the picture morphing task. Children who struggled to identify the duck in the “duck–rabbit” picture set did indeed focus mainly on one detail (i.e., eye; Fig.
3).
In two separate studies, we showed that children gradually improve beyond pre-school age in a continuous version of the ambiguous figures task—a finding that was highly reliable. We speculate that a local over global exploration style can account for our findings. Future studies will have to investigate whether it is indeed a local processing style that hinders children to switch to the second interpretation, or whether children’s ability to explore and generate plausible alternative interpretations or the default heuristic of assuming feature stability is causing the delay in their performance.