Introduction

Human infants naturally develop an understanding of the physical world through observation of regularities which extend and/or confirm innate predispositions [1–3]. Playful exploration, including manipulating objects and bringing them in relation to each other, is an important basis for this development and allows detection of the objects’ affordances [4,5]. During such exploration, infants spontaneously search for causal explanations for their observations and engage in “hypothesis testing” activities to confirm their expectations [5–9]. In addition, exploration involves practicing motor sequences that can serve as precursors to later problem-solving behaviour [4].

Also for animals it has been suggested that playful object exploration may lead to affordance learning and honing of mechanical skills [10–12]. Some support for this suggestion is provided by a comparative analysis that linked object play to extractive foraging and tool use in primates [13]. In addition a causality bias during associative learning has recently been described also for chimpanzees (Pan troglodytes): they learn about causal cues more quickly than about arbitrary cues [14]. Opportunities to manipulate objects, to learn about their affordances and to detect causal relationships may therefore have substantial influences also on animals’ cognitive skills in the physical domain.

A wealth of data has accumulated in recent years that documents substantial differences between species in their physical problem-solving abilities, which is typically explained by differing selection pressures exerted by the ecology and environment of the species together with morphological and phylogenetic constraints [15–20]. In addition, many studies have described sometimes vast individual differences in performance in cognitive tasks also within species (e.g. [21–24]). Exploring the basis of these individual differences is of prime relevance for our understanding of animal intelligence [23].

As learning plays an important role for problem-solving performance, it appears likely that differences in prior experience are the basis of at least some of the observed individual differences in problem-solving performance. Consequently, it is increasingly becoming standard practice to report prior participation in similar experiments for subjects tested in problem-solving tasks (e.g. [20,25–27]). In addition, where the opportunity arose, researchers investigating cognitive abilities of animals in the physical domain have started testing whether subjects with more experience in similar tasks, typically in terms of participation in earlier experiments, outperform more naïve subjects (e.g. [28–31]). These analyses showed for example that chimpanzees with prior experience with a two-trap box outperformed their less experienced conspecifics in two other trap tasks [30], or that great apes with pre-experience with an obstacle task performed better in a trap-platform task and a barrier-platform task than subjects without the same experience [31]. In addition, Cunningham and colleagues found that prior exposure to the rake led to improved performance in a subsequent raking task in female (but not in male) gibbons (Hylobatidae) [32]. Similarly, an early study by Thompson and Heron [33] found that Scottish terriers raised under impoverished conditions performed poorly in barrier and maze tests compared to terriers raised as normal pets. In contrast to these studies, Brent found no effect of prior experience with reaching tools on the performance of chimpanzees in a tool-reaching task [34], and Osthaus and colleagues found no effect of prior training with two short parallel strings on the subsequent performance of dogs with other string-pulling arrays [29].

As outlined, the extent to which physical problem-solving abilities are influenced by (or dependent on) prior experiences currently still remains unclear. This is at least partly due to the fact that many of the existing analyses stem from studies that were not specifically designed to address the question of the influence of prior experiences on individual problem-solving performance. Instead they describe post-hoc analyses of experiments, in which limitations in availability necessitated the use of subjects with different experimental histories. In addition, most of the existing studies investigating effects of prior experience are short-term (i.e., testing for transfer between tasks within studies or between consecutive studies; e.g. [28–32,35]). However, subjects accumulate experiences not only in the short-term but over their lifetime. In addition, opportunities to learn about objects’ affordances already in early life may be particularly relevant for the development of an understanding of the physical world. Indeed, differences in rearing conditions, in particular in terms of access to diverse manipulanda, have been suggested as a likely explanation for striking population differences in physical problem-solving skills, for example between different captive populations of chimpanzees (e.g. [36,37]), or between captive and wild kea (Nestor notabilis) [17]. To evaluate the extent to which experiences with object manipulation influences future physical problem-solving skills, it is necessary to compare groups of subjects that differ in their experiences accumulated over the lifetime, with type and amount of experience manipulated by the experimenters and random assignment of subjects to the different groups. This approach was therefore implemented in the present study.

In addition or alternatively to prior experiences, also other individual-specific factors may influence a subject’s problem-solving performance. In particular the subject’s level of inhibitory control is assumed to have consistent effects on problem-solving performance: subjects with better inhibitory control should perform better in problem-solving tasks than those with poor inhibitory control, who will be easily distracted by irrelevant but salient cues and/or will fail to take into account the complete setup of the task as their attention is captured by a single salient stimulus (such as a food reward) [38]. In humans, the level of inhibitory control has been linked to a variety of cognitive performance measures in children and adolescents [39–43], and a recent comparative analysis suggested a robust evolutionary relationship between the level of inhibitory control and brain size across a wide range of animal taxa [44]. Therefore, we expected that an individual’s performance in physical problem-solving tasks would also be influenced by its level of inhibitory control at the time of testing. This association could appear in at least two ways: poor inhibitory control may interfere with an individual’s ability to apply its knowledge to the problem at hand or with its ability to learn to solve the task.

Domestic dogs are a suitable species to investigate factors affecting individual problem-solving performance. They are available in sufficient numbers and they are highly amenable to experimental studies, which made them one of the most popular species for studies of animal cognition in recent years [45,46]. While the main focus of dog cognition research has been on their skills in the social domain, which may exceed even those of great apes in some tasks [47–49], an increasing number of studies have also investigated the performance of dogs in physical problem-solving tasks. The first studies suggested that the abilities of dogs in the domain of physical cognition are poor compared to other mammals due to their history of domestication, which may have led to alleviated selection pressures in this domain [50,51]. Recent evidence does not support this hypothesis, however, as wolves did not perform better than dogs in string-pulling tasks [52] or object permanence tasks [53] (see also [54] for similar evidence in domestic pig and wild boar). Instead, another explanation may account for the comparably poor performance of modern domestic dogs in many physical problem-solving tasks: limited opportunities to gain experiences in this domain. Compared to primates (including humans), young pet dogs typically have fewer opportunities to learn about means-end connections and other contingencies that may facilitate problem-solving performance in the physical domain (indeed, many dog owners and dog trainers consider it undesirable if a dog manipulates objects such as doors, drawers etc.). If experience plays a relevant role for the problem-solving performance in dogs, we therefore predict that offering them ample opportunities to learn about physical contingencies in their early lives will lead to an improved performance in physical problem-solving tasks later-on. As the baseline level of physical cognition skills in dogs is rather low compared to other species, the potential for improvement is high. In addition, as dogs reach adulthood in less than two years, studies manipulating their experience over prolonged periods with subsequent testing as (young) adults are feasible.

Here, we report the results of a 3-year study, in which dogs were assigned to one of three different experience groups as puppies. For all these dogs, the level of inhibitory control was assessed with three impulse control tasks and they were tested in four physical problem-solving tasks as young adults. Two groups of puppies were supplied with a series of manipulative toys between the age of 3 and 15 months. One (the enriched group) was supplied with a set of toys that offered the dogs opportunities to learn about means-end connections, effects of gravity and other contingencies involved in the physical problem-solving tasks presented to them as adults. In order to differentiate between effects of specific relevant experiences and effects of purely manipulative experiences (performing certain actions/movements), the second group of puppies (the manipulative group) was supplied with a set of similar but perceptually opaque toys that did not offer the same learning opportunities. The third group of puppies (the control group) was not supplied with extra toys but grew up with only balls, ropes, rubber toys and similar objects for stimulation, as is customary for pet dogs in Austria. We predicted that the extensive experiences of the subjects in the enriched group would result in an improved performance in the physical problem-solving tasks compared to the control group. If the dogs in the enriched group learned about relevant contingencies, we predicted that they would outperform also the subjects in the manipulative group. If, in contrast, the extensive manipulative experience led to an improved performance in the physical problem-solving tasks, we predicted no difference between the two experimental groups and that the subjects in both experimental groups would outperform the subjects in the control group. In addition, we predicted that dogs with better inhibitory control scores would perform better in the physical problem-solving tasks, as better inhibitory control would give them a better chance to perceive relevant perceptual information before making a decision.

Methods

Subjects

All subjects lived as pet dogs with their owners, who volunteered to participate in this study and provided written consent. The study was restricted to a single breed to exclude possible variations due to breed differences. Sixty-three young Border Collies were recruited. Of these, 22 dogs dropped out at some point during this long-term study, mostly due to time limitations of their owners. The remaining 41 dogs (16 males and 25 females) completed the study (though not all subjects completed all tests, cf. Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Number of subjects in the respective tasks.

https://doi.org/10.1371/journal.pone.0147753.t001

Results

Effects of experience

Performance in the physical problem-solving tasks did not differ between the three treatments (enriched, manipulative, control; Fig 1). This was true for performance in the occluded condition of the size constancy task (GLM: χ²₍₂₎ = 1.89, p = 0.39) as well as for the on-off task (initial performance: χ²₍₂₎ = 0.09, p = 0.96; probability to reach the learning criterion: χ²₍₂₎ = 0.20, p = 0.90). In addition, we also found no interaction between half (first vs. second half of trials) and treatment group on the performance in the size constancy task (GLMM; χ²₍₂₎ = 2.43, p = 0.30). These results remained unchanged if subjects were excluded from the enriched group if they had not mastered the toys corresponding to the respective physical problem-solving task. In particular, in the case of the size constancy task, the result did not change if the four subjects that had not mastered toy 11 and/or toy 3 (cf. S1 Table) were excluded from the enriched group. As mentioned above, performance in the blocked tube task and in the trap tube task were ultimately not included in the analyses due to minimal variation in performance between individuals (floor effects).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Physical problem-solving performance of the three treatment groups.

Proportion of correct choices in a the occluded condition of the size constancy task and b the first two sessions of the on-off task. Boxplots indicate median, inter-quartile range, minimum and maximum.

https://doi.org/10.1371/journal.pone.0147753.g001

Even though similar physical contingencies are underlying the two tasks, performance in the 4-string task was not correlated with performance in the on-off task: Initial performance in the on-off task was not correlated with initial performance in the string-pulling task (GLM: χ²₍₁₎ = 0.01, p = 0.93) and was not significantly different for the subjects that had reached the learning criterion in the string-pulling task than for the subjects that had not reached the criterion (GLM: χ²₍₁₎ = 0.08, p = 0.78). Furthermore, subjects that had reached the learning criterion in the string-pulling task were not more likely to reach the learning criterion in the on-off task than subjects that had failed in the string-pulling task (GLM: χ²₍₁₎ = 1.42, p = 0.23).

Effects of inhibitory control

Inhibition scores ranged between 1 and 6 and did not differ systematically between the three treatment groups (Kruskal-Wallis test: χ²₍₂₎ = 3.53, p = 0.17). The subjects’ inhibition score was a significant predictor of their performance in the size constancy task and the on-off task. The inhibition score was negatively correlated with performance in the occluded condition of the size constancy task (GLM: χ²₍₁₎ = 4.00, p = 0.046; Fig 2A) and positively correlated with initial performance in the on-off task (GLM: χ²₍₁₎ = 4.65, p = 0.03; Fig 2B). That is, subjects with poor inhibitory control performed poorly in the first sessions of the on-off task, but well in the size constancy task. Whether a subject reached the learning criterion for the on-off task, however, was not predicted by its inhibition score (GLM: χ²₍₁₎ = 0.75, p = 0.39). The negative association between inhibition score and performance in the size constancy task cannot be explained by higher ball motivation of dogs with poor inhibition as the proportion of no-choice trials (the dog did not go searching for the ball when released, see also S1 File) did not predict performance in the occluded trials (binomial GLM: χ²₍₁₎ = 0.08, p = 0.93) and was not correlated with the subjects’ inhibition score (binomial GLM with correction for overdispersion: F_{1, 34} = 0.06, p = 0.81).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Physical problem-solving performance and inhibition scores.

Proportion of correct choices in a the occluded condition of the size constancy task and b the first two sessions of the on-off task depending on the subject’s inhibition score. Shown are raw data and fitted curves with 95% confidence limits.

https://doi.org/10.1371/journal.pone.0147753.g002

As we found evidence for learning across the twelve trials of the occluded condition in the size constancy task (cf. S1 File), we further explored whether inhibition score predicted overall performance or learning in this task. We found evidence for the latter rather than for the former, reflected by a significant interaction between half (first vs. second half of trials) and inhibition score (GLMM; χ²₍₁₎ = 3.86, p = 0.049). To characterize this interaction, we split the dataset into the first and the second half of the occluded trials and found that the inhibitory control score was a significant predictor for performance in the second half (GLM: χ²₍₁₎ = 7.88, p = 0.005) but not for performance in the first half of the trials (GLM: χ²₍₁₎ = 0.002, p = 0.97).

The scores for the three inhibitory control tasks correlated positively with each other. However, these correlations were rather weak (in one case even not significant, cf. Table 2). We therefore explored which of the three tasks predicted performance in the two physical problem-solving tasks best. Performance in the size constancy task was near-significantly related with the score for the middle cup task but not with the scores for the other two inhibition tasks (Table 3). In contrast, initial performance in the on-off task was significantly related with the score for the leash task, but not with the scores for two other two inhibition tasks (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Spearman correlation coefficients for scores of the three inhibitory control tasks.

https://doi.org/10.1371/journal.pone.0147753.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Relation between inhibitory control scores and performance in problem solving tasks (binomial GLMs).

https://doi.org/10.1371/journal.pone.0147753.t003

Discussion

The main aim of this study was to test whether experiences gathered over a prolonged period during manipulation of objects can lead to improved cognitive skills in the physical domain in animals in similar ways as they do in human infants. Our results do not support this hypothesis in the case of the domestic dog. While individual performance of our subjects varied considerably in two of the four physical problem-solving tasks, we found no differences in performance between the three treatment groups, even though the subjects had very different levels of experience with manipulative toys over a prolonged time period. Even experiences with conceptually very similar toys in the enriched group did not subsequently lead to improved performance in the physical problem-solving tasks. In addition, the subjects that had successfully solved a string-pulling task earlier [24] were not more likely to solve the conceptually similar on-off task. Taken together, these results suggest that dogs do not transfer knowledge about physical rules from one physical problem-solving task to another, but rather approach each task as a novel problem that they may (or may not) learn to solve over time. Our results also suggest that the reported poor performance of dogs in physical cognition tasks compared to primates [49–51] is only to a minor extent, or not at all, due to lacking experiences with manipulative toys (and hence fewer opportunities to learn about contingencies in the physical world).

While we found no evidence to support the prediction that experience with object manipulation leads to improved skills in the domain of physical cognition, we cannot rule out that experiences gathered closer in time to the tests (e.g. only few days or hours earlier) would influence the performance of dogs in physical problem-solving tasks. Also, while the toys offered to the enriched group gave opportunities to learn about physical contingencies (e.g. the concept of support or the relevance of size differences) that could have helped the subjects to solve the later physical problem-solving tasks, the setups of the toys and the later physical problem-solving tasks were quite different (for example in their spatial dimensions). It therefore remains possible that transfer between tasks would occur if the setups of the tasks resemble each other more closely and thus their similarities are recognizable to the subjects more easily.

It seems likely that our finding of a minimal or absent effect of prolonged experience on performance in physical problem-solving tasks in dogs, which conforms with the absence of experience effects in an earlier dog string-pulling study [29], extends to other species. Indeed, for animals that typically rely on perceptual cues to solve physical problem-solving tasks, as has been found for dogs (e.g. [67]) as well as for other mammals and birds (e.g. [22,72,80]), experience with perceptually similar tasks may even lead to worse performance in a subsequent task, if the previously learned associations do not lead to success in the latter. However some other non-human animals may also be able to profit from such experiences. Particularly likely candidates are primates, for which a number of studies have found indications for just that [30–32]. Investigating experience effects on performance in physical problem-solving tasks in other non-human animals will be necessary to determine whether the virtual absence of experience effects found in this study is the rule among non-human animals, and the experience effects found in some primates the exception, or vice versa. In addition, conducting studies similar to ours with primates can be useful to determine to what extent experience effects in non-human primates involve learning about physical concepts via playful object exploration, as is the case in human infants.

The second aim of this study was to test whether individual differences in problem-solving performance are linked to individual differences in inhibitory control. As predicted, the dogs’ performance in the physical problem-solving tasks was related to the individual level of inhibitory control. However, the effect was in the predicted direction only for the on-off task, where dogs with better inhibitory control performed better in the initial trials. In the size constancy task, in contrast, dogs with poorer inhibitory control scores performed better. Therefore, our results suggest that the relationship between individual differences in problem-solving performance and individual differences in level of inhibitory control is not straight-forward. We found not only that the direction of the relation differed between tasks, but also that the relation was apparent either only in the initial trials (on-off task) or only in the later trials (size constancy task), and that performance in the size constancy task and performance in the on-off task were best predicted by the scores of different inhibition tasks.

The rather complex interplay between inhibition scores and physical problem-solving performance in our study is likely due to the fact that inhibitory control is not a uniform but a polymorphic phenomenon. Studies in humans have shown not only domain-general variation in impulsivity [87–91], but also domain-specific variation between individuals [91–93]. For example, a recent study with 4th to 8th graders found that schoolwork-related impulsivity and interpersonal impulsivity are correlated with different personality traits and only moderately correlated with each other [93]. Similarly, a recent study on inhibitory control in dogs described domain-general aspects of inhibitory control [94], while other recent studies found that performance is only weakly (or not at all) correlated between different inhibition tasks in dogs [95,96], as we found in our study as well. These weak correlations between performances in different inhibition tasks suggest that the tasks may tap into different aspects of inhibitory control. Alternatively, for some of inhibitory tasks differences in performance may not primarily reflect the individual level of inhibitory control at all. For example, the wait-for-treat and the leash detour task in our study may reflect trainability (a frequently reported personality trait in domestic dogs [97,98]), as much or more than inhibitory control (that is, good trainability could have helped some subjects to reach high scores in these tasks), and the leash task also has a physical problem-solving component. Similarly, performance in tasks with transparent containers or barriers (widely used to assess inhibitory control, e.g. [40,52,78,79], this study) may be influenced by differences in visual perception (there is probably no task that purely measures inhibitory control and is not confounded with any other trait). Finally, as we tested for an association between the inhibitory control score and performance in problem-solving tasks with three different response variables, and the two significant cases cleared the conventional significance level only by a narrow margin, the possibility needs to be considered that one of them may be a false positive (it appears unlikely that both would be false positives).

To conclude, we found no evidence for an influence of manipulative experiences gathered over a long time period on the performance of dogs in physical problem-solving tasks. Instead, we found evidence that individual performance in such tasks is correlated with the subject’s level of inhibitory control. Further studies will be needed to elucidate the apparently complex relationship between inhibitory control scores and performance in problem-solving tasks, to determine under which conditions problem-solving performance is facilitated or hindered by good inhibitory control and which aspects of inhibitory control are influencing problem-solving performance under these conditions. In addition, complementary studies to the one presented here will be needed to determine whether the conspicuous absence of knowledge transfer between physical problem-solving tasks we found in our subjects is the rule or the exception for mammals.

Supporting Information

Acknowledgments

We thank Angela Gaigg, Alina Gaugg, Amelie Göschl, Elisabeth Pikhart, Billy Scaf, Serena Tommasi and Magdalena Weiler for help with the experiments, the dog owners for participation in this study, and the anonymous reviewers for comments on the manuscript.