Summary of Findings

We ran three sets of experiments to investigate cotton-top tamarins’ spatial representations, hoping to explore themes of nonlinguistic concepts, interspecies differences, and domain-specific learning. In Chapter 3, we found that in contrast to toddlers and rats, tamarins use featural cues while reorienting, and may be able to conjoin geometric and nongeometric information in the absence of language. While some animals prefer geometric cues while reorienting, tamarins attended to and reoriented by a featural cue. It was clear that preferential attention to some landmark features and not others may be due to interspecies differences, and that not all animals necessarily rely only upon configurational properties to reorient themselves. The second experiment (Chapter 4) showed that tamarins could learn to search above or below a landmark, attending to both the featural properties of that object and its configurational relationship to the goal. We found that the representation was relatively specific in the sense that they distinguished between the familiar landmark and instances in which one feature had been altered, and also flexible in the sense that they could apply it to novel situations. "Landmarkness" appears to be like a category with graded membership; shape and identity are crucial features while changing an object’s orientation or color does not appear to diminish its reliability as a landmark. These results also suggest that tamarins are capable of abstract conceptualization, so we set out in Chapter 5 to test this possibility, specifically asking whether they understood the concept of "middle." Preliminary results from this experiment suggest that some tamarins may be able to represent "middle" and learn to search midway between two landmarks. Various probe conditions revealed that under some conditions, some monkeys successfully generalize the rule to novel scales, positions, and orientations, but within limits. Interestingly, the monkeys can do this despite lacking the elaborative effect language can have on symbolic encoding.

Landmarks: Spatial Objects

One issue that unites all three experiments presented concerns how tamarins use landmarks: whether they attend to their nongeometric or geometric information; which featural properties are most important; if they use landmarks as single objects or within a configuration. We have learned more about the constraints on the tamarins’ spatial domain of cognition and what features the tamarins pick out of their environment. The second experiment investigating the concepts "above" and "below" shows that some nongeometric features are more important to "landmarkness" than others, and that to be a "good" landmark, an object must maintain these properties over time. In particular, overall identity and shape appear to be the most salient characteristics, while tamarins may perceive orientation and color as incidental features. Not only are there other features that can be tested in tamarins, but it will also be interesting to test other animal species with the properties that we have studied here. Interspecies differences commonly appear throughout the literature on spatial cognition, so it is reasonable to assume that our evidence may only speak to the tamarins’ domain of spatial cognition, and not animal spatial knowledge universally. Unfortunately, one property that we did not test was positional constancy; this was varied across trials and we took the monkey’s ability to succeed in spite of the variability as a sign of an "abstract" concept. However, given the preference for the geometric features of landmarks in many animal species, including rats, human infants, and other monkey species, this is probably an interesting area of study. Further experimentation should take this characteristic into account, and would hypothetically find that in addition to shape and identity, positional constancy may be a crucial feature. Interestingly, shape as a feature is related to geometric properties, in the sense that it is the geometric distribution of parts within the object.

Positional constancy over time may not merely be one other landmark feature, but a necessary condition for an object to be considered a "landmark." In the second and third experiments, we used moveable landmarks, which may more appropriately be called "goal-signs," in that they unambiguously indicate the presence of a target but bear no fixed relation to the global environment. It is possible that because of this, the monkeys were essentially forced to attend to the landmark’s features. If a stable landmark had been used instead, the tamarins may not have attended to the object at all and simply encoded the location with respect to the entire apparatus. Altering landmark features in this case may not have produced the same pattern of preferences that we found in our study. However, many experiments have shown that some animals attend to both the location and distinguishing characteristics of immoveable landmarks while foraging (e.g., Greene and Cook, 1997). It is therefore unlikely that monkeys would not notice any of the landmark’s features, though the interesting possibility that the most salient features for goal-signs may not correspond to the most important landmark characteristics should be studied in greater detail.

Numerous studies have shown that the geometric properties of landmarks – their positional relationships to other objects and goals – are most salient for many animals. In spite of this phenomenon, only a few animal species use small configurations of objects to locate goals. For instance, pigeons and gerbils use single landmark bearings as opposed to geometric relationships to locate a target (see Chapter 5). However, despite the failure to use other local cues, the animals may instead be using the object’s relationship with more reliable global cues. The failures of pigeons and gerbils show that they do not merely search at an absolute distance from a landmark, but they do search at a vector from that landmark determined from training. In this sense, they are just using a different geometric relationship relating one local cue to global features rather than to other local cues.

Additional experiments could further explore the role of proximity in determining the relative importance of objects as landmarks. Cheng (1986) had found that rats are unable to use distal cues to locate food in the reorientation test, and thus can only use cues proximal to a store of food. It will be interesting to see which contexts and tasks motivate animals to use proximal versus distal cues, and local versus global features. We think that stability may be an important factor here: global cues tend to be more stable and so prevail in such tasks as reorientation. In contrast, navigating to a goal while oriented may demand that the animal draw on its representation of local features of the environment.

In summary, while we did investigate identity, shape, color, and orientation, we did not explicitly test positional constancy and proximity as important landmark features. Also, testing a variety of species in a variety of tasks will show how the situation influences whether a landmark is considered as part of a configuration or solely in its relation to global cues. Future studies could elaborate on how the context determines tamarins’ preference for some features, and how these preference relate to the behavior of animals of other species. Our studies suggest that besides species differences, individual differences introduce another factor determining feature preferences. There seems to be an interaction between the context and the animal that determines the rules and constraints of the spatial domain, and is a fruitful area for investigation.

Interspecies Comparisons: The ABC’s of Abstract Concepts

By comparing the responses of nonhuman animals to the behavior of human adults on the same tasks, we can identify abilities that may not require language. The Whorfian Hypothesis that linguistic patterns determine how individuals perceive and think about the world begs the question of what kinds of thoughts are possible in animal minds. The strong claim that cognitive abilities are dependent on the acquisition of linguistic labels is challenged by evidence that primates use abstract concepts without access to language. For instance, in the second experiment, cotton-top tamarins appeared to understand the concepts of "above" and below," and in the third experiment, some monkeys may have understood "middle." These abstract notions exist in the absence of language, suggesting that human conceptualization may not entirely rely upon linguistic development. Observing the behavioral and cognitive flexibility of human adults may lead one to suspect that language is what separates us from other animals. However, it may be more productive to search for common mechanisms underlying abilities shared by two species to understand how those processes have been elaborated in humans.

Xu (1999) suggests that labels help humans "glue" featural properties of objects to their spatiotemporal addresses, allowing for the integration of the "what" and "where" pathways. She accedes to the ability of nonhuman primates to integrate these systems, but claims that even if nonhuman primates’ and humans’ developmental pathways for the individuation of objects are similar, "human infants simply possess a different mechanism for integrating these two types of information, namely language" (p. 133). In support of this idea, she points to the fact that animals are "alinguistic" while human infants are "prelinguistic." However, considering the relationship of common ancestry to homologous traits, it does not seem parsimonious to assume that closely related species solve the identical problem with completely different mechanisms. It may be true that language elaborates on mechanisms already in place, but given animal abilities, it seems questionable that in humans language drives the individuation of objects and the integration of object properties over time.

Earlier, we also discussed Hermer-Vazquez et al.’s (1999) "Language Hypothesis," whereby grammar and lexicon give humans access to linguistic structures that unite modular information into an integrated representation. Our first experiment showed that tamarins may be able to search "to the right of a striped wall," a representation combining nongeometric and geometric information. This shows that animals may be able to conjoin information from different modules without language. We suggest that behavioral flexibility in spatial tasks may derive from abstract concepts that do not depend on linguistic structures.

Interspecies Comparisons: An Evolutionary Perspective

In addition to informing strong claims about language and cognition, interspecies comparisons also shed light on the role of ecological pressures in the development of cognitive abilities and behavioral adaptations. In the first experiment, we saw that cotton-top tamarins use a featural cue to reorient, in contrast to rats and toddlers who use global geometry to redetermine heading. It is clear that one cannot make generalizations about the "domain of spatial cognition" in all animals, or even about the process of reorientation across all species. Natural species preferences may arise because of differing ecological niches, and prevent conclusions about animals in general.

Our experiments showed that tamarins attend to the nongeometric characteristics of landmarks in spite of ambiguous global geometry, and that they also made use of representations that extended into three-dimensional space. These observations are related to the ecology of cotton-top tamarins’ natural environment. Their success in the "above" and "below" experiment indicates a readiness to use objects to locate goals, and may be related to the fact that in their natural environment, cotton-top tamarins must exploit multiple food sources and pick out appropriate plant species using the features of those trees. Garber (1989) claims that tamarins must have a detailed map of their feeding ranges that allows them to forage efficiently by moving directly between multiple trees of the same species that bear fruit in synchrony. The place representations on this kind of map would encode not only a tree’s geometric relation to other environmental features, but also its relevant nongeometric characteristics (e.g., salient visual and olfactory cues; perhaps even classification as a plant species). It is therefore not surprising that the tamarins were able to use a small colored object to locate food efficiently, as they may be predisposed to attend to the visual features of the environment. Also, concepts such as "above" may be ecologically relevant to tamarins; to an arboreal species, three-dimensional relationships may be key to niche exploitation. Similarly, the third experiment investigating tamarins’ notion of "middle" showed that the monkeys could generalize this concept to rotations between the horizontal and vertical axis, confirming the idea that ecological pressures may demand specialized tools, such that general conclusions about the "spatial domain" are infeasible.

Another factor complicating general inferences about animals is that even when species behave similarly, the underlying mechanisms may be different. For example, in Chapter 5 we discussed experiments that showed that Clark’s nutcrackers, like people, can use a concept of "middle" at a variety of scales. However, even bees are capable of this to some degree, and can locate a goal with respect to an array of three landmarks (Cartwright & Collett, 1983). The bees could generalize their strategy to two interesting conditions: (a) the positions of the landmarks were held constant, but all three landmarks were enlarged; and, (b) the configuration of the array was expanded by a factor of two with the landmark size held constant (see Figure 6-2). Bees tested with only one enlarged landmark search at a distance proportionally larger than the training distance, as predicted by a "snapshot model." In contrast, the fact that they search at the appropriate location in the three enlarged landmarks condition suggests that they used the entire configuration and not perceptual mechanisms. However, the authors showed that the bees still attempted to match the stored representation with the current retinal image of the three landmarks by matching the familiar angular distance between the objects. Thus, they were still using a snapshot representation, though the overt behavior insinuated a more abstract concept. Although bees, nutcrackers, and human beings were all able to search at a location specified by a geometric configuration and generalize within certain limits, this is not evidence that the three different species use the same mechanism. Bees are still attempting to correct mismatches between stored and current retinal images – in this case relying on angular over size information – while humans and nutcrackers may be using representations of geometry abstracted over a series of perceptual events.

Not All Cotton-Top Tamarins Are Created Equal

One major obstacle in interpreting much of our data derives from the limited sample size, particularly in the first and third experiments, in which we were able to run only five and four monkeys, respectively. In the first experiment, we were therefore unable to show conclusively that monkeys can use the geometry of a room to reorient, though a variety of species tested on this paradigm can. In the third study, we could not make general inferences about tamarins as a group, but could only show that some tamarins may be able to represent the concept of middle. Clearly, we would have to run more monkeys to be able to draw more general conclusions.

In spite of these problems, by showing that a few monkeys have a particular ability, we demonstrate that this is within the range of species possibilities. The logic is similar for researchers who study only one or a few chimpanzees or parrots. Nonetheless, it is dangerous to make sweeping generalizations based on a few animal geniuses. The strongest arguments based on these data are those that use comparisons to enhance or disprove a position. For example, if one were to make the claim that animals without language cannot have concepts of "middle," then we can safely counter by showing that some of our subjects could, as language was unavailable to them yet they solved the task. We cannot conclude, however, that all tamarins understand that correspondence between the actual environment and scaled versions of reality.

Another concern is the possibility that our laboratory animals do not resemble the other members of their species in their natural environment. We have shown that tamarins attend to featural properties of landmarks, and in fact prefer that information while reorienting. However, one argument is that in the artificial environment of the laboratory, after having been run in numerous experiments in which colors and shapes mean something, our tamarins may be predisposed by experience to look for featural information. Tamarins in the rain forest, in contrast, may perform differently, reorienting by the macroscopic shape of the environment as opposed to local cues. As an example, Garber and Dolins’s (1996) tamarins were unable to use red flags as moveable landmarks to locate food in a subset of 13 platforms, whereas our tamarins were able to use colored poles to find food in one of eight possible locations.

Again it seems as though the safest conclusion is that our results are within the realm of possibilities for the tamarins: they may not immediately use moveable landmarks to locate goals, but they could if given sufficient experience. If we were to show in subsequent experiments that they can use geometry to reorient themselves, as other animals do, in addition to featural information, we would demonstrate that the tamarins behave flexibly, using cues appropriately, conjoining information when necessary. Future studies could overcome these difficulties. For instance, these experiments or similar tests can be conducted on tamarins in the wild. Alternatively, studies on captive tamarins could be run on subjects of different ages, tracking developmental patterns and thus more fully exploring the contribution of experiential factors.

A Final Return to Some General Questions

Earlier, we distinguished between domain specific and general learning mechanisms and hypothesized that animal probably has a variety of both processes available to them. In the domain of spatial learning, it seems clear that animals are guided by rules and constraints intrinsic to the domain, and that constraints probably enhance rather than hinder learning, by motivating the navigator to attend to the most relevant information. We have shown that our tamarins seem to naturally prefer certain features of landmarks that they use to locate a goal. Ignoring irrelevant information tends to make learning more efficient, such that not being influenced by unimportant changes in the environment may aid an animal. Over time, superficial characteristics of the world change, and it may be difficult to survive using an inflexible representation that must match current perception. In this sense, maintaining a category with fuzzy boundaries may provide for behavioral flexibility.

This possibility relates to a distinction we made previously between a "cognitive map" and simpler representations such as a "snapshot" system or a route map. In the latter cases, the navigator is forced to match the current perceptual information with a stored representation. Abstract geometric concepts are thus impossible: changes in scale, orientation, and position throw a wrench into the system. Under experimental conditions, these limits become absurd errors; but in natural conditions, reliable landmarks do not tend to change so drastically, and animals using these systems can account for these changes by simply minimizing the discrepancy with their encoded spatial concept. Even so, the cognitive map provides greater behavioral complexity. The experiments in this thesis have shown that tamarins respond flexibly to spatial tasks, generalizing learned rules to some novel scales and contexts. Their successes and failures help clarify the limits of their abstract conceptualization of environmental relationships, and insinuate the possibility of a tamarin cognitive map.

Gallistel (1990) explains that the cognitive map is created by the integration of "two interrelated sets of processes," one that "constructs a metric representation in egocentric coordinates of the relative positions of currently perceptible points, lines, surfaces – a representation of the geometry of what is perceived," and dead reckoning that "provides a representation in geocentric coordinates of the vantage points and the angles of view (headings)" (p. 106). Key to this idea is that the animal encodes the landmarks as relative positions and in interlocking geometries. Geometric concepts are thus fundamental to the formation of the cognitive map. Perhaps only animals that demonstrate cognitive map-like behavior can access abstract representations of geometry that are flexible under variations of scale and orientation.

Another aspect of this theory of cognitive maps is that the final representation is an integration of processes that may occur in two or more separate modules. After forming both egocentric and geocentric representations, animals must compute transformations mapping the egocentric coordinates onto the geocentric framework. Information collected during one process, dead reckoning, must be integrated with information derived from various perceptual experiences with distinct places. A place representation within the map contains both an address and the featural information that distinguishes that landmark from other places, recalling Xu’s (1999) discussion of how the integration of the "what" and "where" pathways allows for the individuation of objects. Representations of landmarks within a map derive from conjoining landmarks’ geometric and nongeometric properties.

There is a wide variety of solutions animals have discovered to solve spatial problems, and even the tiniest brains are capable of computations that elude calculus students. While human representations seem so flexible and abstract, it is clear that even without language animals are capable of astounding feats. Exploring animal abilities further promises to shed light on how we solve the problems we do, providing insight into the fundamental tools needed for abstract spatial concepts.