Explanations of animal problem-solving often represent our choices as limited to two: first, we can explain the observed behavior as a product of trained responses to sensory stimuli, or second, we can explain it as due to the animal’s possession of general rules utilizing general concepts. My objective in this essay is to bring to life a third alternative, namely, an explanation in terms of imagistic cognition. The theory of imagistic cognition posits representations that locate objects in a multidimensional similarity space. It proposes that an animal’s expectations can be explained on the basis of the similarity of novel objects to objects previously encountered. The animal can predict the behavior of the novel object by producing a mental movie of the novel object by morphing it into an object, the behavior of which has previously been observed.
After criticizing the theory of concept abstraction, I will identify some of the key elements of imagistic cognition. Then I will attempt to illustrate the utility of this conception of cognition by using it to explain the combination of successes and failures observed in monkeys in tool-mediated retrieval tasks (Fujita et al. 2003; Fujita et al. 2011) and in great apes in trap tube and trap table experiments (Martin-Ordas et al. 2008; Martin-Ordas et al. 2012).
Many students of animal cognition regard their task as that of exploring the extent to which nonhuman animals can form abstract concepts. The philosophical and psychological literature contains a variety of attempts to explain what abstract concepts might be and how they might be formed (e.g., Barsalou 1999; Gärdenfors 2000; Mandler 2004). If one wants to reject these, one has to explain carefully why they fail. I cannot do that here, but the basic problems should be familiar enough that a reminder will suffice to motivate a hunt for alternatives. (For a thorough critical review of a wide range of theories of concepts, see Gauker 2011.)
Historically, the common thread in attempts at defining abstraction, going back to Locke (1975 [1689]), is the idea that abstract ideas are formed by a process of abstraction from perceptual representations. There are at least three questions about this process that, as far as I can see, are never squarely addressed. The first is: How does the mind select a class of perceptions from which to make an abstraction? Why, for instance, might the mind abstract from perceptions of four poodles, rather than from three poodles and a cat? The answer cannot be that the mind recognizes that the poodles but not the cat have something in common, since the capacity to recognize that commonality is supposed to be the product of the process, not its impetus.
The second question is: Of all of the abstract ideas that might be abstracted from the perceptions of, for example, four poodles, how does the mind choose the ones to abstract? Some of the endless possibilities are: one-of-these-four-things, poodles, furry things, pets, barky pets, mammals, self-mover. The third question is: If the idea to be abstracted is not already present in the perceptual representation, how can the mind abstract it? And if it is already present in the perceptual representations, how did the mind acquire the capacity to form such concept-containing perceptual representations in the first place? Of course, these questions are even more difficult when we cannot suppose that spoken language mediates the process.
The red herring of abstraction leads to false dichotomies. Many authors in the field of animal cognition cast the choice of explanations as exclusively a choice between the supposition that the animal relies on associations between sensory experiences, and the supposition that the animal grasps general rules by means of abstract concepts. For instance, after describing some tool-using behavior in animals, Seed and Byrne write:
Behavior like this raises the intriguing possibility that animals represent the physical properties and forces involved in the tool-using event in an abstract, conceptual way: in terms of properties such as rigidity, continuity and connectedness. The simpler alternative is that the animals’ thinking is grounded in perceptual features of the objects (their shape, feel and spatial orientation).
(Seed and Byrne 2010: R1034)
(Compare, for example, Hauser 1997: 289; Call 2010: 83; Seed et al. 2011: 90; Mayer et al. 2014: 1; Albiach-Serrano et al. 2015: 176.) The dichotomy is false, because, as we will see, it ignores the possibility of distinctively cognitive activity at the level of imagistic representation.
There are no off-the-shelf theories of imagistic cognition. Early work on mental rotation, especially that by Shepard and Metzler (1971), reawakened the field of psychology to the possibility of imagistic problem solving. The ensuing debate, represented, for example, in Pylyshyn’s critique (1973) and Kosslyn’s defense (1975), was focused on the question of whether mental imagery is real. Regrettably, this debate never blossomed into a research program aimed at identifying the kinds of problems that can be solved by means of imagistic cognition.
Here is a partial list of cognitive problems that might be solvable by means of mental imagery: 1) Figuring out how objects come apart and go together. If I need to replace a faulty washer in a faucet, I can take the washer apart, record a mental movie of the parts coming apart, and then play that mental movie in reverse in order to put the faucet back together again. 2) Object tracking. Within limits, we can keep track of objects as they move around in space, even while they undergo certain changes (Scholl 2001). 3) An elementary grasp of causal relations. Our imagistic grasp of certain patterns of motion (such as those studied by Michotte 1963) can qualify as an elementary grasp of causal relations. 4) An imagistic representation of similarities. On the basis of an imagistic representation of an unfamiliar thing x and its behavior, we may represent x as more like a familiar thing y than like a familiar thing z, and on that basis form an imagistic expectation of what it will do.
Any deeper understanding of imagistic cognition will rely on an account of imagistic representation. Imagistic representation, I suggest, has two main aspects. The first aspect consists in the representation of spatial configuration. Spatial configurations consist of discrete entities, their shapes, their parts, their surfaces, and the configuration of the parts of each object and the arrangement of the objects relative to one another. We may suppose that spatial configuration is represented by virtue of an isomorphism between the elements of the representation and their relations to one another, and the elements of the scene represented and their spatial relations to one another.
The second aspect consists in the representation of an object’s location in a many-dimensional space of graded qualities. The representation itself can be said to have a location in a perceptual similarity space. A perceptual similarity space is an aspect of, or model of, cognitive function, although it does not correspond directly to neurological properties. Each dimension is a measure of some more-or-less continuously variable, perceptible quality that an observable object or arrangement of objects might have. For example, there will be a number of dimensions that measure the various aspects of color. There will be dimensions that measure various aspects of shape. Beyond these, there will be dimensions that measure qualities that less readily come to mind, such as jerkiness of motion. The motion of a squirrel is jerkier than the motion of a cat. (Mandler [2004] emphasizes the role that jerkiness of motion plays in an infant’s representation of animacy.)
My assumption is that perception can be modeled, in part, as the recording of a mark in perceptual similarity space. Points in perceptual similarity space correspond to points in objective quality space, the dimensions of which measure qualities that the perceived object actually has. Accordingly, a perception, considered as a mark in perceptual similarity space, can be said to represent the location of the perceived object in objective quality space. If mark x is closer to mark y than to mark z in perceptual similarity space, then the mind represents x as more similar, all things considered, to y than to z. If an act of perception results in a mark’s being recorded in a biologically abnormal way, then the mark may be said to misrepresent. Further, the geometry of perceptual similarity space may not exactly match the geometry of objective quality space, and that disparity can be the source of persistent illusions, such as the Müller-Lyer illusion. (For a fuller exposition of the ideas in this paragraph, see Gauker 2011, Chapter 6.)
Not only perceptions, due to sensory contact with external objects, can be modeled as marks in perceptual similarity space. Also endogenously generated mental images of objects and scenarios can be so modeled. One means by which the mind might generate mental images is to start with a perception and “translate” it some distance across one or more dimensions of similarity space. For instance, a perception of a blue cube may be translated along the color dimensions to produce a mental image of a red cube. A perception of a slinkily moving cat may be translated across the jerkiness-of-motion dimension to produce a mental image of a jerkily moving cat. Call this process of generating mental images by translating perceptions across dimensions of perceptual similarity space imagistic morphing.
When an object is observed as it undergoes changes over time, these observations leave a trail of marks in perceptual similarity space, which we can call a mental movie. Just as a mental image can be produced by translating a perception across dimensions of similarity space, so too a whole course of events can be imagined by translating a mental movie across some dimensions of perceptual similarity space. For instance, having seen a ballet dancer execute a piroutte, we can imagine a panda bear executing a pirouette by morphing our image of the panda bear into our image of the ballet dancer. Furthermore, if we can form two such mental movies, one of which ends in a given mental image x and the other of which begins with mental image x, then we can link the two to form a mental movie of the one course of event followed by the other. Having imagined a panda bear executing a pirouette turn and a panda executing a fouette turn, we can imagine a panda bear executing first the pirouette and then moving directly on to the fouette.
There will be a distinction between mental morphings that we regard as realistic and those that we regard as fantastic. If we imagine a wine glass falling, shattering, and splattering wine all over the place, we will regard that as something that could happen, even if we have never seen it, and we will take care to make sure it does not. But if we imagine a wine glass falling and, on the way down, turning into a bird and flying away, we do not open the windows to let the bird out. I will assume that, in general, there is a difference between courses of imagination that we regard as realistic and those that we regard as fantastic.
In this section, I attempt to explain the results of a series of experiments carried out by Kazuo Fujita and Hika Kuroshima and colleagues (Fujita et al. 2003; Fujita et al. 2011), in which capuchin monkeys learned to use various hook-shaped tools in order to drag food to themselves. (Fujita et al. 2003 builds on paradigms reported in Hauser 1997 and Hauser et al. 1999.) The interesting observation is that, having learned to solve one sort of problem, the monkeys were quickly able to solve similar problems. I will suggest that their quick transfer may be attributed to imagistic morphing.
In all of the tasks to be reported here, four capuchin monkeys (the same four in all experiments) were confronted with a tray containing two “lanes” in which hook-shaped tools had been laid. In each trial, in one lane a piece of food was positioned so that the monkey could obtain it by pulling on the tool, and in the other lane a piece of food was positioned so that the monkey could not obtain the food by pulling on the tool.
In experiment 1 in Fujita et al. 2003, the monkeys had to choose between two black cane-shaped tools (see Figure 2.1, Exp 1). There were 12 different configurations, training sessions consisted of 12 trials each, and the monkeys reached criterion (10 correct choices out of 12) within 15 to 19 sessions. In experiment 2, the black tools were replaced with similarly shaped red or blue tools. For each color, all four monkeys immediately transferred the skill they had acquired in experiment 1 to the new condition involving tools of a different color.
In experiment 3, the cane-shaped tools were replaced with parabola-shaped tools (Figure 2.1, Exp 3). The monkeys reached criterion in this new task within two sessions. In some of the trials in this experiment, the food was oriented with respect to the tool so that it was inside the parabolic shape of the tool but pulling the tool would not bring the food (see the second example in Figure 2.1, Exp 3). The monkeys reliably chose the correct tool even in trials of this kind. In experiments 4 and 5, tools of two further shapes were used (Figure 2.1, Exp 4 and Exp 5), and the monkeys readily chose the correct tools in these tasks.
In experiments 6 and 7, hindrances were added to the lanes. In experiment 6, the hindrance was a small block secured to the tray, which, depending on orientation, might or might not prevent the monkey from using the tool to obtain the food (see Figure 2.1, Exp 6). In experiment 7, the hindrances were holes in the tray that the food could fall into (see Figure 2.1, Exp 7). The monkeys did not reliably solve these tasks.
The task of pulling food past a hindrance was further explored in Fujita et al. 2011. In experiment 1, the four monkeys learned to obtain the food in trials in which either an obstacle (as in Fujita et al. 2003, experiment 6) or a trap (as in Fujita et al. 2003, experiment 7) could prevent the use of one of the two tools. The monkeys were not immediately successful in these tasks but learned to reliably choose correctly within about 10 sessions. In experiment 2, the three monkeys that were first trained on the obstacle task were tested on a different set of obstacle tasks (different configurations of obstacle, food and tool), and the one monkey that was first trained on the trap task was tested on a different set of trap tasks. Two learned to choose correctly in the new set within one session, and two learned to choose correctly in the new set within two sessions. In experiment 3, the three monkeys that were first trained on the obstacle task were tested in two sessions for each of the two sets of trap tasks, and the one monkey that was first trained on the trap task was tested in two sessions for each of the two sets of obstacle tasks. The combined scores for all four sessions were significantly above chance for all four monkeys. Experiment 4 tested the transfer of the monkey’s skills in avoiding hindrances to a similar task using a tool of a different shape, and again achieved positive results.
Fujita et al. conclude from the second set of experiments, in their 2011 paper, that the monkeys “abstracted” a general rule that allowed them to choose the tool that allowed them to obtain the food despite the hindrance. However, the rule, as they formulate it, contains two parts, one part pertaining to the obstacle tasks and one part pertaining to the trap task (2011: 16). Since the rule has two parts, it is unclear how it captures the understanding that apparently transfers from the one task to the other. In any case, they describe themselves as having provided evidence for an explanation in terms of general rules, as opposed to “stimulus generalization” (Fujita et al. 2011: 16). Fujita et al. (2011) do not consider the possibility of an explanation in terms of imagistic cognition.
A hypothesis that takes us some distance toward explaining the results obtained is that the monkeys were able to imagistically morph new tasks into tasks that they had already learned to solve. Having learned to use the cane-shaped tool in experiment 1 of the 2003 paper, they were quickly able to learn to use tools of other colors and shapes by imaginatively morphing the color or shape of the tool in the second task to fit the colors or shapes of the tools in the earlier tasks. Granted, the hypothesis is in one way incomplete. In experiment 3, the monkeys were able to choose the correct tool (but not quite as reliably) even when, in both options, the food lay within the curve formed by the tool. It is not easy (but also not impossible) to see how a solution to this problem could be obtained by morphing the new configurations into configurations with the cane-shaped tools.
On this account, it is not surprising that the monkeys did not readily transfer their skills to the tasks that included hindrances (experiments 6 and 7 of Fujita et al. 2003). The effect of the hindrance could not be predicted on the basis of imagistically morphing the arrangements of tool and food in the tasks with hindrances into arrangements without hindrances. However, the imagistic morphing hypothesis can take us some distance toward explaining why, in the experiments in Fujita et al. 2011, the monkeys were able to transfer their skills on one set of hindrance tasks to a new set of hindrance tasks of the same kind (either obstacle or trap tasks) (experiment 2), and why they were able to transfer their skills to a differently shaped tool (experiment 4). Namely, the configurations in the new sets could generally be recognized as morphed versions of configurations in the old sets.
The imagistic morphing hypothesis might likewise explain why the monkeys (three of them) were able to transfer their skills on the obstacle task to the trap task and (in the fourth case) on the trap task to the obstacle task (Fujita et al. 2011, experiment 3). Namely, an obstacle in a lane can be imagistically morphed into a trap in the lane, and conversely. Granted, this explanation must be tempered by the realization that an obstacle and a trap do not behave in every way alike, because an obstacle behind the food will prevent a tool from being pulled to the food, but a trap behind the food will not prevent the tool from being pulled to the food (as Fujita et al. 2011 emphasize, p. 15). It is possible that morphing in these cases provides the start on a solution, which then must be completed by independent learning.
An important line of research into the tool-using abilities of apes and monkeys was initiated by Elisabetta Visalberghi and Luca Limongelli, working first with capuchin monkeys (Visalberghi and Limongelli 1994) and subsequently with chimpanzees (Limongelli et al. 1995). Limongelli et al. (1995) studied whether chimpanzees were able to use a rod to push a reward out of a transparent tube, and in so doing avoid pushing the food into a trap at the bottom of the tube. Only two of five chimpanzees were able to learn to do this, but both of them quickly transferred this skill to a second version of the task in which the trap was not at the center of the tube but displaced from the center.
In their study of monkeys, Visalberghi and Limongelli (1994) had asked what would happen if the trap tube were rotated 180 degrees so that the trap was still present but upside down and no longer functional. Only one monkey had learned to solve the original task with the functional tube, and that monkey continued to avoid pushing the food past the trap even when it was no longer functional, rendering it uncertain whether the monkey understood the function of the trap. Limongelli et al. did not try this version of the task on their chimpanzees. Reaux and Povinelli (2000), after successfully training one chimpanzee to solve the original trap-tube test (three others failed to learn it), administered the upside-down version of the trap-tube test to that one chimpanzee and found that she likewise continued to avoid the trap, even when it was no longer functional.
The fact that an animal avoids even the nonfunctional trap does not show that the animal does not understand the function of the trap when it is functional. Silva et al. (2005) showed that even adult humans are strongly biased to insert the rod in the end of the tube furthest away from the reward, even when there is no functional trap that has to be avoided. We would not want to infer from that that the humans do not understand the function of the traps. Moreover, subsequent researchers were able to obtain greater success in teaching great apes to avoid traps when certain predispositions were accommodated and complications were minimized. Mulcahy and Call (2006) obtained better results in the trap-tube test when they allowed their chimpanzees to rake the reward toward them through the tube rather than requiring them to push it away from themselves. Seed et al. (2009) showed that chimps could learn to avoid traps more readily when they were allowed to push the food along the length of the tube using their fingers (inserted through finger holes placed along the length of the tube) rather than using a tool.
Povinelli and Reaux (2000) also pioneered the use of a different kind of trap test, involving tables with two lanes along which an animal can rake food toward itself. One of the lanes has to be avoided because a trough runs across it into which the food will fall. Povinelli and Reaux had only limited success in training chimpanzees to perform this task, but a subsequent study by Girndt et al. (2008) showed that chimpanzees’ performance on this sort of task could be improved by giving the chimpanzee only one tool and letting it decide for itself which lane to apply it to.
The paradigm of trap tables was extended by Martin-Ordas et al. (2008), who invented the trap platform, which is a U-shaped platform divided by a trap (see Figure 2.2). A reward can be placed at the “bottom” of the U and, provided no hindrance is present, raked in along either branch. The trap platform contains a break between the two branches positioned off-center from the bottom of the U. The subject must rake the food along the branch that does not take it over the trap. A population of 20 apes (chimpanzees, bonobos, orangutáns and gorillas) received three 12-trial sessions with the trap platform and three sessions with a version of the trap-tube test. Half experienced the trap-platform task first and half experienced the trap-tube test. Twelve out of their population of 20 great apes learned to solve the trap-platform task. The apes were slower and less successful in solving the trap-tube test. No significant correlation between successful learning in the trap-platform test and successful learning in the trap-tube test was found. On this basis, Martin-Ordas et al. report failure to find transfer of skills from one task to the other (2008, p. 245).
A major step forward in demonstrating transferable skills was taken by Martin-Ordas et al. (2012) when they compared ape performance on three tasks: a hindrance-free U-shaped platform, the trap platform described above, and a barrier platform in which a barrier was placed across the width of the platform, over which the reward could not be dragged (see Figure 2.2). Martin-Ordas et al. discovered that apes who had been trained on the trap platform quickly learned to succeed on the barrier platform and conversely, but apes who had been trained on a platform with no hindrance did not as quickly learn either the trap-platform or the barrier-platform task. In other words, skill in avoiding one kind of hindrance transferred to the other kind of hindrance.
Why did the skill in avoiding the trap on the platform not transfer to the trap-tube task (in the Martin-Ordas et al. 2008 study), while skill in avoiding the trap on the platform transferred to the barrier platform, and vice versa (in the Martin-Ordas et al. 2012 study)? The answer, I would like to suggest, lies in imaginative morphability.
In the Martin-Ordas et al. 2008 study, the trap platform and the trap tube look quite different. The former is flat and broad. The latter is a transparent tube into which a tool has to be inserted. In the platform task, the trap was just a gap dividing the surface of the platform. In the Martin-Ordas et al. 2008 version of the trap tube, the trap was a large black box beneath the tube. It would take a kind of morphing genius to imagine the trap-laden tube morphing into the gappy platform. It is not surprising that the chimpanzees could not do this.1
By contrast, the trap platform and the barrier platform were visually quite similar. From a topological point of view, the trap platform and the barrier platform are fundamentally different, since the former and not the latter contains a gap. But for imaginative morphing, they are really quite similar. Both the trap and the barrier represent an edge at which motion of the food (dragged across the surface) comes to an end. One might even say that the barrier and the gap can be perceived as raised or sunken surfaces that may be imaginatively morphed into one another by imagining the upper edge of the barrier to descend below the surface of the platform, and conversely. Thus, on the hypothesis that chimpanzees can engage in imaginative morphing, it is perhaps explicable that an acquired skill in performing the trap-platform task would transfer to the barrier-platform task, and conversely.
At this point, the hypothesis that monkeys and apes solve problems by imaginative morphing must be deemed highly speculative. However, it does suggest directions for future research. A possible test of the theory would be to test whether the morphing tendencies of monkeys and apes can be exploited in order to fool them into making wrong choices. A further question would be whether monkeys and apes can learn to solve difficult problems if they first learn to solve simpler problems, such that the solution to the difficult problem might be obtained by imaginatively morphing the solutions to the simpler problems. Another question would be whether monkeys and apes can be educated to utilize their imaginative morphing skills more fully.
1 Taylor et al. (2009) report that New Caledonian crows transfer their ability to solve a trap-tube task to a trap platform. But the construction of their trap-platform task is such that, from the point of view that the crows must adopt in solving these two tasks, the tasks look quite similar.
Albiach-Serrano, A., C. Sebastián-Enesco, A. Seed, F. Colmenares, and J. Call (2015) “Comparing Humans and Nonhuman Great Apes in the Broken Cloth Problem: Is Their Knowledge Causal or Perceptual?,” Journal of Experimental Child Psychology 139: 174–189.
Barsalou, L. W. (1999) “Perceptual Symbol Systems”, Behavioral and Brain Sciences 22: 577–609.
Call, J. (2010) “Trapping the Minds of Apes: Causal Knowledge and Inferential Reasoning About Object-Object Interactions,” in E. Lonsdorf, S. Ross, and T. Matsuza (eds.), The Mind of the Chimpanzee: Ecological and Experimental Perspectives, Chicago: University of Chicago Press, pp. 75–84.
Fujita, K., H. Kuroshima, and S. Asai (2003) “How Do Tufted Capuchin Monkeys (Cebus apella) Understand Causality Involved in Tool Use?,” Journal of Experimental Psychology: Animal Behavior Processes 29: 233–242.
Fujita, K., Y. Sato, and H. Kuroshima (2011) “Learning and Generalization of Tool Use by Tufted Capuchin Monkeys (Cebus apella) in Tasks Involving Three Factors: Reward Tool, and Hindrance,” Journal of Experimental Psychology: Animal Behavior Processes 37: 10–19.
Gärdenfors, P. (2000) Conceptual Spaces, Cambridge, MA: MIT Press.
Gauker, C. (2011) Words and Images: An Essay on the Origin of Ideas, Oxford: Oxford University Press.
Girndt, A., T. Meier, and J. Call (2008) “Task Constraints Mask Great Apes’ Ability to Solve the Trap-Table Task,” Journal of Experimental Psychology; Animal Behavior Processes 34: 54–62.
Hauser, M. (1997) “Artifactual Kinds and Functional Design Features: What a Primate Understands Without Language,” Cognition 64: 285–308.
Hauser, M., J. Kralik, and C. Botto-Mahan (1999) “Problem Solving and Functional Design Features: Experiments on Cotton-top Tamarins, Saguinus oedipus oedipus,” Animal Behavior 57: 565–582.
Kosslyn, S. (1975) “Information Representation in Visual Images,” Cognitive Psychology 7: 341–370.
Limongelli, L., S. Boysen, and E. Visalberghi (1995) “Comprehension of Cause-Effect Relations in a Tool-Using Task by Chimpanzees (Pan troglodytes),” Journal of Comparative Psychology 109: 18–26.
Locke, J. (1975 [1689]) An Essay on Human Understanding, ed. P. Nidditch, Oxford: Oxford University Press.
Mandler, J. (2004) The Foundations of Mind: Origins of Conceptual Thought, Oxford: Oxford University Press.
Martin-Ordas, G., J. Call, and F. Colmenares (2008) “Tubes, Tables and Traps: Great Apes Solve Two Functionally Equivalent Trap Tasks But Show No Evidence of Transfer Across Tasks,” Animal Cognition 11: 423–430.
Martin-Ordas, G., F. Jaek, and J. Call (2012) “Barriers and Traps: Great Apes’ Performance in Two Functionally Equivalent Tasks,” Animal Cognition 15: 1007–1013.
Mayer, C., J. Call, A. Albiach-Serrano, E. Visalberghi, G. Sabbatini, and A. Seed (2014) “Abstract Knowledge in the Broken-String Problem: Evidence From Nonhuman Primates and Preschoolers,” Plos One 9: 1–7.
Michotte, A. (1963) The Perception of Causality, trans. T. R. Miles and Elaine Miles, Methuen, MA: Routledge.
Mulcahy, N., and J. Call (2006) “How Great Apes Perform on a Modified Trap-Tube Task,” Animal Cognition 9: 193–199.
Povinelli, D., and J. Reaux (2000) “The Trap-Table Problem,” in Daniel J. Povinelli (ed.), Folk Physics for Apes: The Chimpanzee’s Theory of How the World Works, Oxford: Oxford University Press, pp. 132–148.
Pylyshyn, Z. (1973) “What the Mind’s Eye Tells the Mind’s Brain: A Critique of Mental Imagery,” Psychological Bulletin 80: 1–25.
Reaux, J., and D. Povinelli (2000) “The Trap-Tube Problem,” in Daniel J. Povinelli (ed.), Folk Physics for Apes: The Chimpanzee’s Theory of How the World Works, Oxford: Oxford University Press, pp. 108–131.
Scholl, B. (2001) “Objects and Attention: The State of the Art,” Cognition 80: 1–46.
Seed, A., and R. Byrne (2010) “Animal Tool-Use,” Current Biology 20: R1032–R1039.
Seed, A., J. Call, N. Emery, and N. Clayton (2009) “Chimpanzees Solve the Trap Problem When the Confound of Tool-use Is Removed,” Journal of Experimental Psychology: Animal Behavioral Processes 35: 23–34.
Seed, A., D. Hanus, and J. Call (2011) “Causal Knowledge in Corvids, Primates, and Children: More Than Meets the Eye?,” in T. McCormack, C. Hoerl, and S. Butterfill (eds.), Tool Use and Causal Cognition, Oxford: Oxford University Press, pp. 89–110.
Shepard, R., and J. Metzler (1971) “Mental Rotation of Three-dimensional Objects,” Science 171: 701–703.
Silva, F., D. Page, and K. Silva (2005) “Methodological-Conceptual Problems in the Study of Chimpanzees’ Folk Physics: How Studies With Adult Humans Can Help,” Learning and Behavior 33: 47–58.
Taylor, A. H., G. R. Hunt, F. S. Medina, and R. D. Gray (2009) “Do New Caledonian Crows Solve Physical Problems Through Causal Reasoning?” Proceedings of the Royal Society 276: 247–254.
Visalberghi, E., and L. Limongelli (1994) “Lack of Comprehension of Cause-Effect Relations in Tool-Using Capuchin Monkeys (Cebus apella),” Journal of Comparative Psychology 108: 15–22.
