Different Conceptions of Space

The role of the dorsal stream in visuomotor control is now well established, and even those scientists who subscribe to the spatial vision theory of dorsal stream function acknowledge the contributions of the dorsal stream to sen-sorimotor control (e.g., Boussaoud et al. 1990; Haxby et al. 1993). Nevertheless, proponents of the spatial vision story view visuomotor behavior as a component of dorsal stream function that is subsumed under the broader function of spatial vision. From their point of view, the primary function of the dorsal pathway is the perception of the spatial relations between, and locations of, objects in the environment (for review, see Milner and Goodale 1995). Thus, optic ataxia (disordered reaching to visual targets that the subject can "see") is regarded not as a visuo-motor deficit but as a visual disorientation—a problem in the perception of space.

In Goodale and Milner's (1992; Milner and Goodale 1993, 1995) original interpretation of ventral stream function, the term "perception" is used in a restricted sense to refer to phenomeno-logical visual experience. This use of "perception'' is different from the more general meaning of perception as sensory processing. As we emphasized earlier, it is the ventral stream that mediates our conscious perception of the world. In this context, the processing of the spatial relations of objects (or the parts of objects) is as central to perception as the processing of the intrinsic features of the objects, such as their shape and surface properties. Indeed, without percep tion of the relative position of objects in a scene, we could make little sense of our visual world.

In contrast to the conscious perception of spatial layout provided by the ventral stream, the computation of spatial location carried out by the dorsal stream is entirely related to the guidance of specific visuomotor actions, such as grasping an object, walking around obstacles, or gazing at different objects in a scene. As a consequence, the dorsal stream mechanisms do not compute the "allocentric" location of a target object (i.e., its location relative to the locations of other objects in the scene), but rather the "egocentric" coordinates of the location of the object with respect to the location of the observer. Indeed, the egocentric coordinates are in the particular frame of reference for the action to be performed. For example, to control a grasping movement, the dorsal stream must eventually transform visual information about the object into arm- and hand-centered coordinates. Moreover, we have no conscious access to the visual information that is transformed into the coordinate frames for action. To reiterate, allocentric spatial information about the layout of objects in the visual scene is computed by the ventral stream mechanisms (which mediate conscious perception), while precise egocentric spatial information about the location of an object in a body-centered frame of reference is computed by the dorsal stream mechanisms (which mediate the more automatic visual control of action). This viewpoint represents a marked departure from traditional spatial vision accounts (e.g., Ungerleider and Mishkin 1982), which appear to assume that both kinds of spatial analysis (allocentric and egocentric) are the domain of the dorsal stream.

Consider the following argument. Perceptual systems are biased toward stability: Objects in the world retain their identity and their spatial relations in spite of dramatic changes in the retinal stimuli that are produced when, for example, the observer moves from one viewpoint to another. Thus, the pen lying on our desk, independent of viewpoint, appears to remain in the same place on the desk in spite of dramatic rotations about the axes of our favorite swiveling chair. Visuomotor systems, on the other hand, must compute the precise location of objects in the environment in a metrically accurate fashion with respect to the effector being used. Information about the pen's relative position with respect to the corner of the desk is not adequate for guiding an accurate and efficient grasping movement toward it. The visuomotor system must have the position of the pen in precise body-centered coordinates. In short, both dorsal and ventral visual systems compute information about spatial location, but in very different ways.

As we reviewed earlier, much of the evidence for the perception-action account of the ventraldorsal distinction came from work with neurological patients such as DF. This patient, who has ventral stream damage, is able to produce well-calibrated grasping movements toward objects that she cannot visually discriminate. It was this kind of result that led Goodale and Milner (1992) to suggest that object attributes (i.e., dimensions or shape) are processed by two relatively independent visual mechanisms, one specialized for the perception of objects and the other specialized for the control of actions directed at objects. But although these findings provide support for differential processing of object features for perception and action, they say nothing about how spatial information might be handled by perceptual and action systems.

Of course, as we saw earlier, there have been some studies, admittedly few in number, that have examined the ability of patients with damage to the dorsal stream to perform allocentric and egocentric spatial judgments. Thus, Perenin and Vighetto (1988) noted that even though their optic ataxic patients were unable to reach accurately to objects presented in different positions in their contralesional visual field, they could nonetheless make normal judgments about the relative locations of these objects within the same visual field. In other words, Perenin and Vighetto's patients could use spatial information

Figure 12.2

The two tasks used to test DF's spatial encoding abilities. Panel A shows the pointing task, in which the subject is required to point to the different colored tokens according to a predetermined order provided by the experimenter. Panel B shows the copying task, in which the subject is required to construct a copy of the arrangement, using an identical set of colored tokens. Subjects are allowed as much time as they need to construct the copy.

Figure 12.2

The two tasks used to test DF's spatial encoding abilities. Panel A shows the pointing task, in which the subject is required to point to the different colored tokens according to a predetermined order provided by the experimenter. Panel B shows the copying task, in which the subject is required to construct a copy of the arrangement, using an identical set of colored tokens. Subjects are allowed as much time as they need to construct the copy.

for the purposes of making perceptual judgments about the locations of objects in their environment, but they could not use spatial information about object location to make accurate visually guided movements.

A natural question to ask then is this: If patients with dorsal stream damage show preserved allocentric spatial processing in the face of a profound disturbance in egocentric spatial processing, will patients like DF with ventral stream damage show the opposite pattern of behavior? If one subscribes to the "spatial vision" theory of dorsal stream function (Ungerleider and Mishkin 1982), then DF should do well at tasks requiring her to analyze the spatial relations among objects in the environment—her visuomotor abilities indicate that her dorsal stream is intact. In contrast, a strict interpretation of Goodale and Milner's (1992) perception and action story would lead one to expect that DF would show deficits in her ability to perceive the relative location of objects in a visual scene even though she can use information about their location to direct her actions toward them.

Allocentric and Egocentric Spatial Processing in DF

Recently we carried out a set of experiments in which we examined DF's ability to use spatial information about a small group of objects in order to make decisions about where those objects were located relative to her (egocentric) and relative to one another (allocentric) in space (Murphy et al. 1998). For these tasks we presented small groups of differently colored circular tokens that could be easily discriminated on the basis of color but could not be distinguished on the basis of shape or edge-based cues. Thus, relatively pure spatial responses to the group of tokens could be examined. In our egocentric task DF was required to point to all available tokens in the target array in a specified sequence (figure 12.2a). In our allocentric task DF was required to copy the spatial arrangement of tokens in the target array as precisely as possible, using another set of colored tokens (figure 12.2b). Given the fact that DF appears to show normal visually guided movements, which require egocentric spatial processing, we expected that she would show normal sensitivity to the spatial locations of all of the tokens in a given array when she had to point to them. We expected, however, that she would do poorly on the copying task, since to do this, she would have to use an allocentric frame of reference; in other words, she would have to process information about the positions of targets with respect to one another.

As expected, DF showed no difficulty whatsoever in pointing to each of the targets in turn. She pointed acccurately even when she was told

Figure 12.3

Examples of the reproductions of the token arrays made by DF and one of the control subjects. Although DF shows some sensitivity to the relative position of the tokens, her performance is clearly far from normal.

Figure 12.3

Examples of the reproductions of the token arrays made by DF and one of the control subjects. Although DF shows some sensitivity to the relative position of the tokens, her performance is clearly far from normal.

to point to the tokens in a specified order that she was told about just before she began a sequence of pointing movements (e.g., "Point first to the red token, then to the blue, the green, the yellow, and the black, in that order."). Except for the fact that she was slightly slower, her performance was indistinguishable from the control subjects. This means that DF can use egocentric spatial coding of target location to control visuomotor acts such as manual aiming movements. In fact, her normal visuomotor performance on this task is consistent with her past performance on a broad range of visually guided behaviors (for a review, see Milner and Goodale 1995).

DF's normal performance on the pointing task, where egocentric coding was demanded, can be contrasted with her poor performance on the copying task, where allocentric coding had to be used. On the copying task, she showed large displacements of the position of objects with respect to one another when she attempted to reproduce the target arrays. As figure 12.3 illustrates, her copies looked quite different from the actual target patterns she was presented with. The accuracy of all her attempts at copying the token arrays is summarized in figure 12.4, alongside the performance of two control subjects. DF's poor performance in spatial percep tion cannot be ascribed to an overarching deficit in "spatial vision,'' since she had no problem directing her hand to any of the targets. Her problem was with allocentric, not egocentric, coding of spatial position.

It is important to note that DF's poor performance on the copying task cannot be explained by suggesting that this task was more "cogni-tively challenging'' than the other tasks we used. The copying task, for example, did not require much working memory, since the model was always available. In contrast, the pointing task, on which DF did as well as normal subjects, did put a load on working memory, since subjects were required to point to the tokens in a predetermined order. Although DF's copies tended to be displaced slightly toward the top half of the square white background, this shift in placement was unaccompanied by any evidence for a neglect of or inattention to the bottom part of the target display. She always copied the entire array of five tokens. More important, she reproduced the vertical arrangement of the tokens as well (or as poorly) as she did the horizontal.

A recent study by David Milner and his colleagues (Dijkerman et al. 1998) has shown that DF is also unable to use an allocentric frame of reference to guide her grasping movements. When faced with a disk in which two finger holes

Figure 12.4

Graph summarizing the mean displacements of the positions of the tokens in the reproductions from the positions of corresponding tokens in the sample array for DF and two control subjects. DF is far less accurate than the other two subjects, even in a second session, after she has had more practice.

Figure 12.4

Graph summarizing the mean displacements of the positions of the tokens in the reproductions from the positions of corresponding tokens in the sample array for DF and two control subjects. DF is far less accurate than the other two subjects, even in a second session, after she has had more practice.

and a thumb hole had been cut, she was unable to guide her hand successfully so that her two fingers and thumb went into the holes when she reached out to grasp the disk. Thus, when the distance between the forefinger and thumb holes, and the orientation of the line formed by them, were independently varied from trial to trial, she could not adjust either her grip aperture or the orientation of her hand to reflect those variations. She could adjust the orientation of her hand when there were only two holes (one for the thumb and one for the finger), but even here, she could not correctly adjust the aperture between her finger and thumb. This result suggests that DF is unable to use allocentric coding, not just to guide her behavior in "perceptual" tasks such as the copying task described earlier, but also to guide her motor output in an "action" task.

Of course, under other circumstances, DF apparently can locate two or more positions when she is engaged in a motor task. Thus, she can point to two colored tokens simultaneously with her left hand and right hand (Dijkerman et al. 1998). But in this case, each hand is performing a separate act of egocentric localization, and no allocentric coding is required. Similarly, DF can reach out and pick up objects that vary in outline shape, placing her forefinger and thumb on appropriate "opposition" points on the surface of nonsymmetrical, smoothly contoured objects (Goodale et al. 1994). Clearly, some sort of object-based coding is needed here. It is when the grasping task requires an analysis of the location of separate objects (e.g., holes in a disk) for controlling the posture of one hand that DF's use of allocentric cues for the control of action breaks down. The dorsal stream, by itself, appears unable to perform this analysis.

It is generally believed that patients with a visual agnosia (like DF) have suffered damage to the ventral visual cortical pathway (Milner and Goodale 1995). Typically, such patients have been characterized as having a deficit in the perception of the shape of objects. Rarely has the ability of such patients to perceive the spatial location of objects been investigated. This may be the case for two reasons. First, "spatial vision" is thought by some researchers to be mediated exclusively by the dorsal visual cortical pathway. Second, the severity of the object perception/ recognition deficits overshadows any possible observations of deficits in space perception, particularly when there is an absence of visuomotor disturbance.

Neurological patients who present with spatial perceptual impairments as a primary feature (i.e., patients with hemispatial neglect, dressing dys-praxia, and constructional apraxia) typically have damage in the parietal lobe, in particular the inferior parietal region. We have not observed any of the classically described visuospatial impairments, often reported after parietal lobe damage, in DF's behavior, and this is consistent with the fact that her MRI scan does not show evidence of parietal lobe injury. Nonetheless, we felt it was important to address the issue of constructional apraxia (i.e., the inability to perform familiar movement sequences when engaged in the act of making or preparing something) and to explain why we thought that DF's difficulty in reproducing the allocentric location of an object cannot be accounted for by appealing to some sort of constructional apraxic deficit. First, DF's drawings of objects from memory are remarkably intact and do not show any evidence of constructional apraxia (Servos et al. 1993). Second, DF can perform image construction tasks and is able to use visual imagery to manipulate objects in order to fashion new objects in her mind's eye just as well as normal subjects (Servos and Goodale 1995). In short, DF does not have constructional apraxia; but she does have a fundamental deficit in allocentric spatial coding.

DF's fundamental deficit in allocentric coding cannot be explained in terms of the old idea that all spatial processing resides in the posterior parietal cortex. In fact, regardless of the exact interpretation of the nature of the spatial deficits, DF's perception of spatial relations should be intact if all forms of visuospatial processing, and not just visuomotor control, depended on mechanisms in the dorsal stream. The results of the allocentric spatial tasks demonstrate that this is clearly not the case. Instead, as argued elsewhere (Goodale and Milner 1992; Milner and Goodale

1995), the characterization of dorsal stream function as visuomotor rather than visuospatial predicts the obtained data.

As was discussed earlier, other accounts of single cases support the dissociation of spatial encoding for perceptual representation from spatial encoding for visuomotor control. Recently, Stark et al. (1996 ) described a patient, with normal visual object recognition abilities, who was impaired at making spatial judgments about the locations of visual (or auditory) targets but could accurately localize these targets with a pointing movement. Balint's (1909) patient had poor visuomotor control when he used his right hand but not his left, a finding hard to reconcile with the idea of an overarching visuospatial deficit (Harvey and Milner 1995). And as we have already seen, Perenin and Vighetto (1988) have argued that poor manual localization of targets in patients with dorsal stream damage is largely unrelated to the degree of visuospatial dysfunction measured in perceptual tests. In summary, it seems that although the dorsal stream plays a major role in spatial processing, that processing is confined to transforming visual information about a goal into the egocentric frames of reference required for a particular action directed at that goal. Allocentric spatial encoding is not a dorsal stream function, and appears to depend more on perceptual mechanisms in the ventral stream.

Was this article helpful?

0 0

Post a comment