In referring to tactile object recognition as “one major function of the somatosensory system” (sect. 3.1, para. 1), Dijkerman & de Haan (D&dH) echo a theme that has guided our research for over 20 years. However, we would substitute the term haptic for tactile: Whereas the latter refers primarily to skin sensation (Loomis & Lederman Reference Loomis, Lederman, Boff, Kaufman and Thomas1986), object recognition by touch is fully active, affording valuable kinesthetic, as well as cutaneous information. That very activity is difficult for the present authors to handle, because it points to an essential link between perception and action that they seek to decouple.
D&dH's arguments are weakened, we claim, for two reasons. First, they confuse object primitives at proximal and distal levels, as evidenced, for example, by the target article's combining vibration, pressure, and roughness under the rubric of “simple stimulus features” (sect. 3.1, para. 2). Second, they underestimate the role that material properties play in haptic object recognition. The first error leads to over-emphasizing the role of spatiotemporal integration (via the posterior parietal cortex [PPC]). The second error leads to discounting the richness of material properties.
We have developed a model for haptic object recognition that makes certain fundamental points. As a preliminary, it is important to distinguish among distal and proximal descriptions of objects accessed by the sense of touch. We differentiate between proximal object primitives that arise at the receptor level (e.g., heat flow, vibration) and distal properties of external objects (e.g., apparent warmth, roughness) (Klatzky & Lederman Reference Klatzky, Lederman, Gopher and Koriat1999). Transitioning from proximal to distal representations requires computation by neural systems. Further, we have made a critical distinction between two broad categories of distal object properties available to the sense of touch. There are geometric properties, including size and shape features (e.g., curvature) and material properties (e.g., roughness, compliance, apparent warmth).
Given this partitioning, we can move to the first point our model raises, which concerns how properties contribute to object recognition. Generally, recognition refers to the assignment of an instance to a category with a common name. It has been demonstrated that there is a basic level of categorization, corresponding to high-frequency names, which hierarchically dominates subordinate categories (Rosch et al. Reference Rosch, Mervis, Gray, Johnson and Boyes-Braem1976). Members of a basic-level category tend to look alike; however, object categories at both basic and subordinate levels share non-shape features as well, and these are effectively perceived by the sense of touch. Lederman and Klatzky (Reference Lederman and Klatzky1990) asked subjects to rate the diagnostic features of objects that were to be categorized haptically at basic and subordinate levels. We found that, at the basic level, shape was deemed important for touch, like vision, but texture was also highly diagnostic; moreover, at subordinate levels, categories were often defined by material properties. An important implication of this work is that haptic object recognition relies on material properties as well as, and in some cases instead of, geometric ones.
Our second point is that when an object is recognized by touch, correlations among object properties can be exploited. A china plate is not only round and thin, but also cool, smooth, and hard. The conjunctions of these properties diminish the diagnostic value of shape and can lead to a rapid identification response without overt exploration. Although sequential manual exploratory movements can be linked to perceived object properties, they are not necessary for object recognition by touch.
We have observed (Lederman & Klatzky Reference Lederman and Klatzky1990) a common two-stage process for haptic object recognition. In Stage 1, people begin by grasping and, where feasible, lifting the object. This encompasses the exploratory procedures we call enclosure and unsupported holding, which are sufficient to provide coarse information about many geometric and material object properties (Lederman & Klatzky Reference Lederman and Klatzky1987). For objects being named at the basic level, the grasp–lift movement is often sufficient. Indeed, brief fingertip contact, without grasping, can suffice for recognizing many objects (Klatzky & Lederman Reference Klatzky and Lederman1995). Stage 2, which comprises extensive exploration, occurs when initial contact, grasp, or a grasp-lift combination is insufficient. Particularly when object categories are designated below the basic level (a dull pencil), one observes exploratory procedures being used that are optimal (most precise) for obtaining the desired diagnostic information.
Our model casts a new light on the present proposal that the anterior parietal cortex (APC)–PPC route guides action, including exploratory procedures, whereas the APC–insula (+ right PPC) route provides perceptual descriptions leading to recognition. The model suggests that we need to further consider two aspects of an object-recognition task. One is whether the individual intends to perceive or to act; the other is whether he or she performs purposive exploration, or merely Stage-1 contact.
With respect to whether the performer's intention is to perceive or to act, we note that similar actions can arise from different motivations. When we close our fingers around an apple, we may want to feel it to see whether it is ripe, or grasp it in order to eat. Although these intentions are distinct, both have perceptual consequences. Contact with an object has the consequence of feeling, just as breathing has the consequence of smelling. Importantly, the perceptual consequences may not be the same, given equivalent patterns of action. Just as the brain has been found to gate the processing of odorants according to whether the intent was to breathe or to smell (Zelano et al. Reference Zelano, Bensafi, Porter, Mainland, Johnson, Bremner, Telles, Khan and Sobel2005), it might gate the processing of touch sensations arising from a grasp, depending on whether the actor's goal was to sense or to act.
Considering the second task feature, namely, whether the performer uses Stage-1 contact or, alternatively, extended purposive exploration as the mode of touch, our model suggests that sequential exploratory procedures will be observed only in the case where there is an intention to perceive that is not satisfied by grasping and lifting. Because the exploratory action is invoked by the perceptual system, the ensuing processing will then be gated toward object identification.
Our analysis suggests, ultimately, that two aspects of the performer's action – intention and extent of exploration – define three distinct cases that are likely to modulate the balance between the two pathways identified by D&dH. When there is an intention to act, not perceive, extended purposive exploration will generally not be observed, and the APC–PPC pathway will predominate. When purposive exploration is directed toward perceptual consequences, the APC–insula pathway would predominate, although the PPC would be invoked for purposes of guiding action. Finally, if perceptual consequences were intended but achieved through minimal exploration, a more equal balance between these two pathways would be found.
