Tool use and the two visual streams hypothesis 14

An attractive aspect of the two visual streams proposal is in how the princi defined for vision-for-perception and vision-for-action so sensibly differ. The frame o reference for action must be with respect to the body and must take into account real world metrics; an actor must compute the real size, shape, distance, and orientation of objects with respect to themselves, in ‘egocentric’ coordinates. In fact, the brain is able to transform visual information within and between multiple coordinate frames tied to particular body parts (Andersen & Buneo, 2002; Buneo & Andersen, 2006; Graziano & Gross, 1998). The frame of reference for perception does not require consideration of the absolute metrics of objects or relations between them, but instead, to be most adaptiv perception must operate in relative metrics. Recognition of objects independent of viewing conditions, individuation of objects in a cluttered scen

can be achieved much more efficiently and reliably if the operations are performed in relative terms. The basic concept is well appreciated by considering how easily we make sense of images on TV; objects on the screen are perceived and understood in relation to one another, not with respect to real world dimensions (Milner & Goodale, 2006, pp. 239-240). A second operating principle

it makes little sense to store the computational outcomes of previous actions to guide and control new ones, given that, in general, actors and objects in the world are in constant flux; for example, the parameters needed to accurately transport and shape the hand to grasp a given object will change as the actor and/or object moves. Thus, it makes more sense for the action system to compute parameters de novo; from the bottom-up each time a new action is carried out. In contrast, the perceptual system needs to construc

operate within long standing representations of the world in order to recognize ob including people and places, despite various changes in appearances and viewing conditions. To recognize old objects as they change over time, and to categorize new ones, the perceptual system must ultimately compare what is seen with what is stored in memory. The systems underlying perception and action appear to operate on

fundamentally different timescales and frames of reference (Goodale, 2001; Goodale & Haffenden, 1998

t and jects,

).

sion-for-perception and vision- , -use otor l l tool ry o With learned tool use, the separation between vi

for-action would seem to reach its limits. Familiar tools are bound to action plans that stretch beyond what is available on the ‘surface’, defined instead by previous experience unlocked only with successful recognition. How the hand is shaped when grasping-to tools depends not only on their physical aspects (e.g. size, shape, and orientation), but also on stored knowledge of function and use. Once the actions of tools are known, identity, and thus visual object recognition, represents an efficient route to learned m plans. The implications with respect to the two streams hypothesis are clear: familiar too use is likely to involve explicit cooperation between ventral and dorsal streams. Further, the ventral stream is expected to act as a first step in retrieving and activating stored motor representations. After all, stored action plans represent the hallmark of skilfu use. That is to say, for tool use, as with any other learned motor skill, procedural memo storage, retrieval, and implementation is essential to lasting improvements in motor performance. To summarize, the actions of learned tool use extend beyond physical object properties and the timeframe by which routine dorsal stream function is thought t operate, and instead appear to rely on stored object representations and successful recognition as mediated by the ventral stream.

In their original account of the two streams dichotomy, Milner and Goodale predicted that familiar tool use would require the cooperation of both streams, and that the ventral stream would lead the way (Milner & Goodale, 1995, pp. 202-204). In particular, in conjunction with recognizing tools, they suggested that the ventral stream would also take part in selection of where and how to grasp tools in accordance with intended use and known function. Notably, they were also clear to indicate that this process of action specification, solving the ‘how to’ part of the puzzle, would critically depend on ventral stream cooperation with specialized areas of inferior parietal co Finally, the end products of this processing would reach the dorsal stream, where actu sensorimotor transformations critical to motor implementation and control are computed and carried forth. The basic scheme fits well with their broader account of how

streams are likely to interact in general. With everyday actions, individual objects are parsed from others i

rtex. al the two n busy crowed environments by the ventral stream, selected for potential action, while the dorsal stream then specifies particular motor outputs and

d

n

well, tes. ly awkward grasping postures if this will enable more comfortable postures upon completion of end goals. The tendency to grasp governs online control (Goodale & Humphrey, 1998; Milner & Goodale, 2006, pp. 231- 233).

What evidence is there for ventral stream involvement in familiar tool use? The strongest support has come from testing functional grasping in patient DF (Carey, Harvey, & Milner, 1996). When asked to grasp and use familiar tools, DF shaped an oriented her hand perfectly well with respect to the metrical properties of tools; however, she often failed to posture her hand in a way that reflected knowledge of function and use. The most obvious example of this functional orienting of the hand when grasping tools can be seen with handled tools, when the handle is faced away from the actor. I this situation, people will typically rotate their hand to end up in a final posture well suited for use (see Chapter 2, Figure 2.1A). While control participants were seen to follow this tendency, DF did not. The tendency is part of a broader ‘rule’ in motor control, identified by Rosenbaum and colleagues (Rosenbaum, van Heugten, & Cald 1996; Rosenbaum, Vaughan, Barnes, & Jorgensen, 1992), whereby the kinematics of early aspects of multistep actions are selected in anticipation of comfortable end-sta Most clearly, people will reliably adopt relative

handled tools in a way that is appropriate for use was specifically addressed in a stud Creem and Proffitt (2001), and their findings provided additional evidence for ventral stream involvement in tool use. They found that when participants were asked to gr tools while simultaneously performing a task involving semantic processing, grasping actions appropriate for tool use were made less frequently than when participants eith grasped without performing the concurrent semantic task or when they instead

simultaneously performed a spatial imagery task. The results were taken to indicate that planning functional grasps to tools necessarily involves semantic systems, presumably including ventral stream areas. When the semantic task was being performed at the same time as grasping, grasping did not tend to incorporate functional knowledge, presumably because resources from semantic systems were not fully available. Insofar as semantic processing involves the ventral stream, the findings nicely converge with those of functional grasping in patient DF. Apparently ventral stream functioning is needed to grasp tools appropriately for use, precisely in line with the scheme initially proposed Goodale and Milner reviewed above.