Tactile object perception

From Scholarpedia
Guy Nelinger et al. (2015), Scholarpedia, 10(3):32614. doi:10.4249/scholarpedia.32614 revision #154541 [link to/cite this article]
Jump to: navigation, search
Post-publication activity

Curator: Guy Nelinger

It is commonly assumed that object perception is the combination of sensory features into unified perceptual entities. Tactile object perception may therefore be defined as the perception of objects whose feature information is acquired via touch. Consequently, research relevant to the topic of tactile objects has focused on exploring the primitives of the tactile system, their interrelation, and how they may be bound together. The current discussion does not explicitly rule out, nor does it address, kinesthetic sensation. As such, tactile perception is used here interchangeably with haptic perception (Lederman & Klatzky, 2009).


The Concept of an Object in Different Modalities

The question of objects has received a great deal of attention in philosophical treatises spanning many centuries. Perhaps most accurate in representing modern science’s stance was Kant (1781/1999). Kant argued that the existence and nature of objects as a self-contained entity, unperceived, is unknowable to us. Therefore, we should not attempt to know the ‘object in itself’ (i.e., noumenon), but reformulate our topic of interest altogether. The question at issue should be what determines the objects of the senses (i.e., phenomenon). More specifically, how do sensory and cognitive faculties constrain the way that sensory information is obtained, processed and combined into compounded entities? Accepting the Kantian perspective, it may be appropriate to preface the discussion of tactile objects with a more general theoretical enquiry. Explicitly, is the concept of an object different for separate modalities, and if so - how?

To answer this question, it is useful to consider the cases of olfaction and audition, in which the idea of a modality-specific object is still a matter of debate. For audition, conjunction of features into an object has been suggested to occur in a hierarchical fashion. Processing in the cochlea is based on putative two-dimensional time-frequency primitives. Features are then extracted, compared to templates, and integrated with other modalities, as one ascends the primary and secondary auditory cortices (Griffiths & Warren, 2004). For olfactory objects, it has been suggested that the molecular structure of the entire olfactory ‘landscape’ is essentially projected to a single dimension, and this dimension corresponds to the inherent pleasantness of the odor. The percept of an odor object is then given by integration of this pleasantness value with the perceiver's internal state (Yeshurun & Sobel, 2010).

The cases of audition and olfaction clearly demonstrate two contrasting views of object perception. Audition suggests that feature conjunction into objects is achieved by hierarchical convergent processing of isolated feature information, a framework widely accepted in vision. Studies in olfaction appear to suggest a common representation across all levels of processing, questioning the involvement of a hierarchical processing framework. This chapter examines the perception of tactile objects in light of this theoretical divide, through comparisons to visual object perception, concluding by reviewing the suggestion that objects in touch are best understood not through comparison with vision, but within a unique framework of active sensing which takes into account both sensory and motor variables.

Co-processing of Tactile Features

In order to characterize the ways in which features are combined by a perceptual system, it is imperative to first probe the fundamental units, or “primitive units”, of feature processing by that system. It is quite possible that these primitive units do not correspond to the units of description used by humans when referring to the aforementioned features. In the following we thus refer to description-derived units, such as texture, length or softness, as ‘features’ and to the, yet unknown, fundamental units of processing as ‘processing-primitive units’. A specific individual feature can be considered as corresponding to a specific individual primitive unit if the processing of that feature is not affected by simultaneous processing (“co-processing”) of other features. Behavioral studies that have examined the degree to which co-processing of different tactile features occurs may provide the first clues for the identities of such processing-primitive units.

Both examples of co-processing and independent processing have been presented. For example, Corsini and Pick (1969) demonstrated co-processing of texture and length. They showed that sheets shorter in length than a reference were better discriminated when they had a coarse texture, and those longer than the reference – when they had a fine texture (for similar examples of tactile illusions, readers are encouraged to turn to Hayward, 2008; Lederman & Jones, 2011). Sinclair, Kuo & Burton (2000), on the other hand, demonstrated that texture and vibration frequency are perceptually separable from duration, even when processed simultaneously.

Klatzky, Lederman & Reed examined similar questions in a series of studies. They presented subjects with objects which could differ from one another by one or more tactile features, and taught them to classify these objects into groups. The classification rule was based on the values of a single feature (e.g., groups A, B, and C contain small, medium and large objects, respectively). However, for some groups of subjects an additional feature covaried with the group-determining feature (e.g., small, medium and large objects had fine, semi-rough and rough texture). They found that this feature covariation significantly improved response times in the classification task (Klatzky, Lederman, & Reed, 1989; Reed, Lederman, & Klatzky, 1990). These redundancy gains in performance appeared to result from covariation in practically any two features (Lederman, Klatzky, & Reed, 1993). However, covariation of an additional third feature (i.e., covariation in texture, shape and hardness) did not yield significant response time improvement compared to covariation of any two features (Klatzky et al., 1989).

Task improvement achieved by co-processing of covarying features is, in itself, not a trivial finding. However, the opposite question may be more informative. That is, can subjects ignore information from one tactile feature while processing another, when such information adversely affects performance? To examine this question, Klatzky, Lederman & Reed used a paradigm they termed ‘withdrawal’. Here, as before, subjects learned to classify objects by a sorting rule which relied only on one tactile feature as a determinant. Again, the values of a second tactile feature initially covaried with those of the first. This time, however, at some point during testing, the covariation of the second feature was stopped and it was fixed to a given value (i.e., it was ‘withdrawn’). If response times were lengthened by this manipulation, one can assume that the two features were co-processed despite explicit motivation to avoid such integration.

When the covarying features used in this method were size and shape, withdrawing either significantly worsened performance times (symmetrical deficits), indicating these features tend to be co-processed. However, withdrawing size damaged performance more than withdrawing shape (Reed et al., 1990). When hardness and texture were the covarying features, withdrawing either yielded comparable, significant response time deficits (Klatzky et al., 1989; Lederman et al., 1993). As a putative rule of thumb, it appears that when features are related (i.e., both properties of either structure or material), co-processing is more likely to occur. Cases in which features were non-related (e.g., size and hardness, shape and texture, etc.) produced only asymmetrical deficits, if at all, after withdrawal. This is not in accord, of course, with the hypothesis that these features were being co-processed by default. These findings were virtually unaffected by whether subjects were explicitly told that classification depends only on one feature (Klatzky et al., 1989; Reed et al., 1990) or not (Klatzky et al., 1989).

However, one final research in this line of studies had quite different results (Lederman et al., 1993). This study found that covariation of shape (a structural property) and texture (a material property), followed by withdrawal of either, significantly worsened reaction times. That is, deficits were symmetrical, implying co-processing. These results were also replicated using a method called ‘orthogonal insertion’, which is sort of a mirror-image of withdrawal. In ‘orthogonal insertion’, the non-focal feature was first fixed to a given value and then varied orthogonally compared to the feature of interest (Lederman et al., 1993). These results do not fit with those of a previously mentioned study, in which only the withdrawal of shape significantly damaged performance after a period in which shape and texture covaried (unidirectional deficits, Klatzky et al., 1989). This discrepancy can be accounted for by the fact that evidence of co-processing was found using 3D ellipsoids as shapes. With these objects it was possible to extract both texture and shape information (in terms of curvature) from any local region; exploratory procedures for shape and texture extraction (contour following and lateral motion, respectively) could be performed together in a more compatible manner. Section 4 expands on this point. Other explanations are, of course, possible. For example, it was recently shown that different cognitive styles (namely, the tendency to employ different types of imagery) may affect the co-processing of texture and shape, suggesting a meaningful role of individual differences in these tasks (Lacey, Lin & Sathian, 2011).

Taken together, the studies cited in this section support the idea that individual features corresponding to everyday experience may not necessarily correspond to individual processing primitive units. Yet, it appears that simple behavioral paradigms for testing co-processing of features should be further developed to achieve more meaningful results. This can be performed in two ways: either by analyzing more carefully the effect that task demands and specific parameter values have on the convergent results of these studies, or by approaching their results from the perspective of a guiding theoretical framework. As explained in section 1, the following sections attempt to do the latter: section 3 discusses results in line with existing models in vision, and section 4 discusses results in line with a unique theory of object perception in touch as an active-sensing process.

Tactile Objects and Visual Objects

'Feature Integration Theory’ and Tactile Objects

Arguably the most widely accepted model of feature binding developed in vision is Treisman’s ‘Feature Integration Theory’ (FIT). According to FIT, features that are processed by primitive units are pre-attentively and automatically registered in separate feature maps (Treisman & Souther, 1985; Treisman, 1998a; Treisman, 1998b). An additional master-map of locations is formed, containing information of the boundaries of all these features (where features are), but lacking the information of feature identity (which features are where; Treisman & Souther, 1985; Treisman, 1998a; Treisman, 1998b). A scalable window (Treisman, 1998a), or ‘spotlight’ (Treisman, 1998b), of attention may then define a specific location, exclude features outside of it, and thereby bind all features falling within its range to a given location (Treisman, 1998a; Treisman, 1998b). A priori, FIT does not intuitively apply to touch, mainly because touch is a proximal sense and at any given moment information is available only from a small set of locations in the surrounding world. Nonetheless, this does not preclude FIT, or a variant of it, from adequately describing feature binding in touch.

Much of the appeal of FIT stems from its clearly defined experimental paradigms and extremely specific predictions. A particularly relevant example is FIT’s predictions regarding search time when subjects are required to determine whether a target is present or absent within an array of distractors. If the target is defined by the presence of a unique single feature which is absent from distractors (e.g., a red ‘Q’ in an array of red O’s; ‘disjunctive search’), it should ‘pop-out’ immediately, independent of the number of distractors (up to ~3-4ms per distractor; Treisman & Souther, 1985; Treisman, 1998b). This is because information about the target is immediately present within the activity of the dedicated feature map (or the lack of such activity; Treisman, 1998b). Therefore, if the target is present, it will ‘pop-out’, and if not its absence will also be effortlessly detected. In neuronal terms, this simply requires perceivers to examine the pooled response of feature detectors and determine if feature detection has taken place. However, if the target differs from distractors by a conjunction of features (e.g., a red ‘Q’ among red ‘O’s and blue ‘Q’s’; ‘conjunctive search’), a serial self-terminating approach is required, in which the features of each element are bound and only then is the element accepted or rejected as the target (Treisman, 1998b).

To summarize these hypotheses in quantitative terms, FIT predicts that slopes of search time in a disjunctive task will be virtually flat, indicating parallel search, while slopes in a conjunctive task will be steeply linear with the number of distractors, indicating serial self-terminating search. Moreover, in the latter task, the ratio of search time slopes between target-absent and target-present trials is expected to be 2:1 (Treisman & Souther, 1985; Treisman, 1998b). This is because the former condition requires scanning of the entire array for a response to be made, while in the latter condition the target will be found after scanning half of the array, on average.

An additional finding of FIT is that when the target is defined by the absence of a unique feature (e.g., a red ‘O’ among red ‘Q’s) search will also be serial (Treisman & Souther, 1985; Treisman, 1998b). This is presumably because the overall activity in any single feature map is highly similar for arrays in which the target is absent and those where it is present, hampering pop-out detection (Treisman & Souther, 1985). This phenomenon, in which a target defined by the presence of a feature pops out from among distractors lacking the same feature, while the opposite arrangement produces serial search, is one example of what is called a ‘search asymmetry’. A search asymmetry of this type strongly suggests that the feature in question is represented by a dedicated map, or equivalently – that it is processed by a unique primitive unit. Many of the studies described below have used search asymmetry to characterize tactile primitives.

Two studies using a tactile search paradigm demonstrated results very close to those obtained by Treisman in vision (Sathian & Burton, 1991; Whang, Burton, & Shulman, 1991). In the tactile paradigm, the individual fingertips of subjects simultaneously received vibrotactile stimulation, in two consecutive intervals (Whang et al., 1991). Each fingertip received its own stimulation, making their combination an array. On one interval all fingertips were presented with distractors, while on another, one of the fingertips was presented with a target. Subjects were asked to report the interval in which a target appeared. In one condition, the target had a segment of altered frequency and distractors had constant frequency (‘presence’ search condition), while in another these roles were reversed (that is, all stimuli but one had a segment of altered frequency; ‘absence’ search condition). Results showed that pre-cuing the targeted fingertip benefited task accuracy in the ‘absence’ condition, but not in the ‘presence’ condition. The fact that frequency change detection does not benefit from cuing might suggest, in the terms of FIT, that it is processed by a specific primitive unit. The same principal finding was obtained, using the same paradigm, for textured stimuli (Sathian & Burton, 1991).

Plaisier, Bergmann Tiest & Kappers (2008) found that texture occurrence (as opposed to texture change, reported above) also produces a search asymmetry. Their subjects performed a free hand sweep over a surface in which elements of texture were embedded. One element served as a target, and had either a rougher or a smoother texture than the other elements. Subjects were required only to report whether a target was present in the array. Rough targets popped-out easily from among smooth distractors; search time in this condition had little dependence on the number of distractors, yielding a close-to-flat search-time slope. Additionally, this condition typically required only a single sweep over the array. Conversely, smooth targets among rough distractors required more complex hand-motions, and the search-time slope of this condition depended on the number of distractors more steeply. This asymmetry was given a mechanical explanation, according to which rough-textured stimuli elicit more friction in the course of hand motion, and therefore pop-out during a single sweep.

An additional asymmetry, one of edge presence, was examined by Plaisier, Bergmann Tiest & Kappers (2009). They had subjects cup (in their hands) an array of objects which either did or did not contain a single target whose shape differed from distractors. Subjects were to report if a target was present, and could release the array and re-examine it in almost any way they wished. If information was not available in the initial grasp, search times lengthened, trivially. Results showed that cubes tended to somewhat pop-out from among spheres (that is, they could be sensed with minimal release of the stimuli) while the opposite did not hold (supporting a search asymmetry). A follow-up experiment tested for search of ellipsoid, cylinder or tetrahedron targets among cube or sphere distractors; it demonstrated that search times for a target were heavily affected by the identity of distractors. To quantify the effect of features on search times, correlation was computed for the slope of search-time from the target-present trials with a variety of geometrical features. The geometrical feature most correlated with search-time was the difference in edge acuteness values for each shape and its distractor in absolute value (with edge acuteness defined as the smallest angle between two planes of a shape). This result suggests that, contrary to the initial results obtained with cubes and spheres, edge acuteness affects pop-out in a symmetrical manner (whether edges are a feature of the target or the distractors). As such, it appears that the differences for search time of a cube among spheres and vice versa, found in the initial experiment, reflect search asymmetry in the domain of objects (cubes vs. spheres) but not necessarily in that of tactile primitive units.

In a similar behavioral paradigm and setup (grasp-and-release), objects colder than body temperature were found to pop-out from among objects warmer than body temperature (Plaisier & Kappers, 2010). The slope of search-time for cold objects was similar to the one found for a tetrahedron target among spheres in the experiment reported in the previous paragraph. However, here, the reverse correspondence of a warm target among cold distractors was not examined, making talk of an asymmetry impossible.

Perhaps the most detailed test of FIT’s hypothesis of conjunction using tactile features was performed by Lederman, Browse & Klatzky (1988). Subjects placed their hand in a passive manner, with each finger receiving stimulation individually. In a ‘disjunctive search’ condition, subjects were instructed to search for a target, either rough or vertical, among smooth horizontal distractors. In a ‘conjunctive search’ condition, the target was both rough and vertical, and it appeared among rough-and-horizontal as well as smooth-and-vertical distractors. In both conditions, 50% of trials contained no target at all. Subjects were only asked to indicate if a target was present in each trial. The ratio of slopes between the target-absent and target-present conditions was approximately 2:1 in the conjunctive search task, as predicted by findings in vision, indicating a serial self-terminating search. For the disjunctive search task, slopes were not completely flat as predicted. Thus, while disjunctive tactile search may have a meaningful parallel-search component, it does not precisely fit the quantitative predictions derived from search tasks in vision (different interpretations are therefore possible for these findings). In the absence of a clear ‘pop-out’ effect in this study, neither orientation nor roughness can be confidently considered to be processed by tactile primitive units (this is further complicated by the fact that the slopes show some profound qualitative differences from equivalent slopes in vision).

In general, thus, FIT-inspired methodology rarely produced unequivocal results in touch. Still, several pieces of data seem to be valuable for tracing the fundamental components of tactile object perception. The tactile perceptual system seems to be quite sensitive to presence over absence, whether the present element is an edge, or roughness of texture (Plaisier et al., 2008). It is also sensitive to change over constancy, whether the change is in texture (Sathian & Burton, 1991), or vibration frequency (Whang et al., 1991). Cold items were shown to pop-out from among warm ones (Plaisier & Kappers, 2010). Also, the tactile system was demonstrated to more quickly find targets defined by either orientation or texture than their conjunction (Lederman et al., 1988). While these studies cannot indubitably indicate tactile primitives, their collective summary suggests that edges, texture, vibration and temperature are all strong candidates. Perhaps the most important finding supporting FIT in touch was that search for a conjunction of texture and orientation was slower and, presumably, more effortful than search for either component (Lederman et al., 1988)

Neuronal Conjunction: Convergence, Population Firing & Synchrony

Classical electrophysiological findings have demonstrated that, at least in the passive mode, some neurons in the visual cortex respond selectively to specific features of a given sensory stimulation (Hubel & Wiesel, 1962). Furthermore, as one ascends the cortical hierarchy, specific cells tend to respond to complex combinations of such features (Barlow, 1972; Hubel & Wiesel, 1962; Kobatake & Tanaka, 1994). Convergence of feature information to a single neuron was thus suggested to occur in a hierarchical manner. Simultaneous appearance of features should lead to simultaneous activation of several ‘feature detector’ neurons. Input to a single higher-level neuron from several such feature detectors would allow it to act as a coincidence detector and represent the combination of features. In the extreme case, such a cell could correspond perfectly with a specific compounded percept of an object (Figure 1A; Barlow, 1972). Indeed, the existence of such extremely selective cells, colloquially termed ‘grandmother cells’, has been documented in vision (Gelbard-Sagiv, Mukamel, Harel, Malach, & Fried, 2008; Quiroga, Mukamel, Isham, Malach, & Fried, 2008; Quiroga, Reddy, Kreiman, Koch, & Fried, 2005).

Similarly, single neurons displaying highly selective responses to complex tactile features have also been reported. Jörntell and collaborators have found that when the glabrous pads of decerebrate cats were stimulated in different fashions and velocities, cuneate neurons showed very different selectivity in their response patterns (Jörntell et al., 2014). Iwamura and collaborators have also documented specific neurons with strikingly selective responses, located in or around the post-central gyrus of monkeys. For example, some neurons, which had rather complex receptive fields spanning large parts of the hand, did not respond to passive stimulation with a probe object nor did they respond to positioning of the joints in a form matching the object’s shape. They did respond, however, to a combination of the two, when the object was actually grasped (Iwamura & Tanaka, 1978). Other neurons in this area appeared to be highly tuned to specific features such as softness, mobility, and even familiarity, reflecting varying degrees of sensory integration and abstraction (Iwamura, Tanaka, Hikosaka, & Sakamoto, 1995). Saal & Bensmaia (2014) argued further that when the relevant literature is inspected more closely, with an emphasis on appropriate and naturalistic stimulation paradigms, it appears likely that a typical cortical tactile neuron receives inputs from various mechanoreceptor types and can potentially encode information from a diversity of features.

With time, however, the idea that feature conjunction is represented via the activity of a single cell has mostly fallen from grace. This is mainly because such a representation would clearly be noise-sensitive and vulnerable, and the system as a whole would require a superfluous number of such explicit elements (Barlow, 1972; Quiroga et al., 2005; Treisman, 1996; Von Der Malsburg, 1995). This does not mean that convergent anatomical hierarchy cannot represent or compute feature conjunction. It only suggests that such a representation would be implemented at the level of a neuronal population response (Figure 1B).
Figure 1: Three suggested schemes by which neurons can represent the existence of external objects via feature binding. Each row shows one scheme. The left and right panels show how two different objects (a cube and an apple, respectively) can be differently encoded within each scheme. A) “Grandmother cells”: neurons send their outputs in a converging manner, such that each layer has more selective response properties than its predecessor. The activity of specific cells in the topmost layer represents the existence of a specific external object (active neurons appear in yellow, inactive ones in blue). B) Population coding: each specific pattern of activation across the entire population of neurons represents the existence of a specific external object. C) Binding by synchrony: the firing rate of each single neuron may hold some information about the object. The complete bound representation, however, is achieved by compounding those neurons whose firings are synchronized temporally (marked here in yellow).

Some studies have specifically examined the existence of anatomical areas co-processing tactile features by using imaging techniques. Roland, O’Sullivan & Kawashima (1998) used PET imaging on 3 groups of subjects who were asked to discriminate two objects differing either by length, shape, or roughness. Results showed that the lateral parietal operculum (LPO) was more significantly activated during roughness discrimination compared to both shape and length discrimination (using a Boolean intersect of contrasts). This suggests that the LPO is more sensitive to properties of structure than those of material. Indeed, recent studies have found further support for the involvement of the parietal opercular cortex in haptic texture processing (Stilla & Sathian, 2008; Sathian et al., 2011). The intraparietal sulcus showed significant activation in the exact opposite circumstances. That is, it was significantly less activated by roughness discrimination than by either length or shape discrimination. Hence, the intraparietal sulcus may actually encode relatively abstract features, such as the combination of length and shape (i.e., representation of structure, and not of material). This, of course, would support the idea that conjunction of features is coded by population activity of higher processing areas.

However, a different study using a slightly more relevant paradigm found no support of the existence of such convergent areas (Burton et al., 1999). In this second study, subjects were asked to discriminate which of two stimuli had either greater roughness or was longer. Additionally, subjects were either cued to notice a specific tactile feature (selective attention condition) or were not cued (divided attention condition). A PET scan was performed during the task, allowing researchers to examine if any areas were activated when subjects attended both features (which was the case during the non-cued, divided attention condition). An activation contrast showed that one area was indeed more significantly activated during divided attention than during selective attention: the orbitofrontal region. However, in light of existing literature, the authors suggest that this activation should be attributed to memory demands. Indeed, when subjects in the divided attention condition awaited the appearance of the second stimulus in each trial, the task required them to keep both features of the first stimulus in memory. Subjects in the selective attention condition only needed to memorize one feature.

Taken together, the findings outlined above do not strongly support the idea that convergence along anatomical areas is a correlate of perceptual compounding of tactile features in the somatosensory cortex. One intriguing suggestion has been that some of these features may be processed by non-somatosensory cortices, such as the visual cortex (Lacey & Sathian, 2015). Another suggestion raised by another group of studies is that binding of features into compounded representations may be achieved through synchronized or temporally correlated responses of population firing (Figure 1C; Damasio, 1989; Treisman, 1996; Von Der Malsburg, 1994; Von Der Malsburg, 1995). This may be achieved by phase-coherent activity of different areas processing fragments of the entire object (Damasio, 1989). It has been further postulated that it is specifically neuronal discharges at the band of gamma oscillations that yield the perceptual binding of co-processed features (Fries, Nikolić, & Singer, 2007; Tallon-baudry & Bertrand, 1999).

In the framework of synchronization (or correlation), it is quite possible that no single brain area would selectively and consistently respond to a complex combination of features, which would explain why no such area was found by Burton et al. (1999) in the study cited above. This would be expected when the processing of individual features has spatial signatures (i.e., via the firing of dedicated neuronal circuits), and object binding has only temporal signatures (i.e., via synchronization).

At least one study is in line with the involvement of neuronal synchrony in tactile perception (Steinmetz, Roy, Fitzgerald, Hsiao, & Johnson, 2000). In this study, neurons were recorded in the secondary somatosensory cortex of monkeys that performed both a tactile and a visual task and were cued to alternate attention between them. Synchronization analyses were performed on pairs of neurons. Of the 427 neuron pairs which had significant cross-correlogram peaks during either task, 17% (79/427) demonstrated changes in firing synchrony between the two conditions. 80% of these pairs (59/74) displayed significantly more synchronous firing when attention was focused on the tactile task compared to the visual one. Tactile stimulation was continued whether monkeys were attending and performing the tactile task or not (i.e., were performing the visual task). Hence, the rise in synchrony does not result purely from stimulation. Rather, it corresponds to attention and to perception of the tactile stimuli. This would of course be in line with the relation between attention and feature binding proposed by FIT, as reviewed in the previous subsection.

As an ending remark to this subsection, we note that the assumption of distributed representations, either over space or time, does not satisfy a solution to binding per-se; the questions of how correlations in neuronal firing are computed by the brain, or how semantics emerge from such a correlation, remain open (Roskies, 1999).

Feature Processing from an Active Sensing Perspective

In everyday experience, the importance of action in touch is surely quite prominent. While ‘foveating’ objects in touch (i.e., exploring them with the fingertips) interaction often affects both the object and the perceiver. This is to be expected, as the tactile fovea and the body’s main instrument of fine-tuned external manipulation are essentially one and the same: the hand. Given this consideration, it stands to reason that specifically in the tactile modality, the way in which sensors are moved and the perception derived from sensory input should be tightly coupled. Indeed, the way in which one chooses to explore a tactile stimulus has been shown to have a meaningful effect on resulting percepts (Lederman & Klatzky, 1987; Locher, 1986). While free tactile exploration of everyday stimuli leads to virtually perfect recognition (Klatzky, Lederman, & Metzger, 1985), constrained exploration under short exposure time hampers recognition severely (Klatzky & Lederman, 1995).

A major theoretical contribution to the question of action in touch is Lederman & Klatzky’s definition of exploratory procedures (EPs; Lederman & Klatzky, 1987). Noting that haptic exploration typically involves a stereotypical set of hand gestures, these researchers hypothesized that each such procedure specifically relates to the extraction of a given tactile property (Figure 2). Of the identified procedures, five were most prominently explored in later works: lateral motion (for texture), pressure (for compliance), static contact (for temperature), enclosure (for global shape or volume) and contour following (for exact shape or volume). Preliminary validation showed that the profile of EP frequencies in a trial could generally be predicted by the feature required for discrimination, and vice versa (classification of the trial type by EP profile; Lederman & Klatzky, 1987).
Figure 2: Illustration of exploratory procedures, as described by Lederman & Klatzky (1987). Out of 8 procedures described in the original article, 6 prominently explored ones are presented here. For each such procedure, there exists a haptic feature which it extracts in an optimal manner, appearing in parentheses.

EPs form a meaningful explanatory variable. For example, such is the case with the withdrawal studies mentioned above (subsection 2). When two features covaried redundantly, and one was then withdrawn, EPs were found to mostly correspond with performance time. Specifically, Klatzky et al. (1989) found that for most pairs of features (e.g. shape and hardness), performance after withdrawal suffered only slightly and EPs relevant to the withdrawn features were shown to be minimized. However, when the pair of covarying features was texture and hardness, not only was performance hampered by withdrawal, but persistence of EPs relevant for the withdrawn feature was evident. Recall also the fact that covariation in one property of material and one of structure (e.g., texture and shape, respectively) inconsistently led to symmetrical deficits after withdrawal, discussed in section 2. Symmetrical deficits were found in Lederman, Klatzky & Reed (1993), but not in Klatzky, Lederman & Reed (1989). Under the framework of EPs, this difference is readily understood. In the former experiment all shapes were 3D ellipsoids. As such, information of shape, just like information of material, could be assessed at any regional section of the object. The same cannot be said of stimuli in the latter experiment. Thus, in the former experiment, EPs for extraction of shape and texture (i.e., contour following and lateral motion, respectively) were more compatible with each other. That is, in this experiment both procedures could be optimally performed on interior homogenous sections of the object. This is presumed to have led to stronger co-processing of relevant features. Taken together, these results suggest that feature co-processing (and as a direct result, conjunction) is highly affected by EPs.

EPs are, despite their major contribution, quite a crude description of tactile exploration. Over the last decades, behavioral studies have shifted to using more elaborate analyses, which are designed to help uncover the complex dynamic nature of many phenomena. In an influential line of works, Turvey has suggested that during object hefting and wielding, the tactile perceptual system can extract information by exploiting a variety of rotational factors. Particularly, the inertia tensor about a point (i.e., the object’s resistance to rotation in different axes) is an invariant property of the object, which corresponds with subjects’ perception of objects magnitudes (e.g., length, width and weight) as well as orientation (summarized in Turvey, 1996). Klatzky & Lederman (1992) have found that exploration often starts with a generalized sequence (grasp-and-lift) followed by more detailed exploration. Taking these observations together, it is expected that exploration in a given trial will develop over time, affected by slight variations in movement, object properties, and their interaction. This is because even when correspondence between physical and perceptual properties is strong, information is gathered in a piecemeal manner.

To address the actual complex motions of exploration, methods that explicitly model development over time are required. Indeed, methods for fine-grained modeling of exploratory procedures are gradually being adapted, as discussed in Klatzky & Reed (2009). One such method is called recurrence quantification analysis (RQA), which enables the detection of recurring patterns in a time series. Riley and colleagues (Riley, Wagman, Santana, Carello, & Turvey, 2002) found that dynamics of wielding under perceptual intent were more complex and more stable (in terms of response to perturbations) than those of free wielding. Another task examined dynamics of wielding for identifying either height or width. The objects used did not have a very large ratio of height to width, but the magnitude of the principal moment of inertia corresponding to height was at least five times greater than that corresponding to width. Consequently, height was predicted to be a much more salient property for perceivers than width. Results showed that discriminations of width led to wielding dynamics more deterministic, complex, and stable than those of height, suggesting fine-tuning of exploratory manipulations. These results affirm the idea outlined above. That is, they extend the concept suggested by Lederman & Klatzky (1987) to show that exploration is anticipatory not only at higher levels of description, but also over very short timescales, presumably adapting itself in an iterative manner.

This last property is in line with the behavior expected of closed-loop systems, systems in which signals affect their sources via feedback. Indeed, studies in humans and rats seeking out features have shown that gradually convergent changes of motor and sensory variables occur during perceptual tasks (Horev et al., 2011; Mitchinson, Martin, Grant, & Prescott, 2007; Saig, Gordon, Assa, Arieli, & Ahissar, 2012). It is quite plausible, then, that EP compatibility has an even more profound role in tactile object perception then originally proposed. Importantly, EP dynamics likely vary greatly over short time intervals, given the complex interactions between intent, exploratory performance, and sensory information received. We thus predict that works extensively examining co-processing of features from the perspective of dynamical systems are imminently the next step within the field of tactile object perception.


In sum, the research of tactile object perception is a multi-faceted field of study, with various sub-fields and paradigms contributing unique insights into its functions and mechanisms. In some cases these contributions are synergistic. For example, attention’s role in tactile perception is emphasized both by behavioral studies and by electrophysiological recordings of firing synchrony in neurons (e.g., Steinmetz et al., 2000). The dependency of object perception on motor patterns is demonstrated in primate and rodent by behavioral studies (e.g., Horev et al., 2011; Lederman & Klatzky, 1987; Mitchinson et al., 2007; Riley et al., 2002; Turvey, 1996) and supported by electrophysiology studies (e.g., Iwamura & Tanaka, 1978). Action, in general, appears to be indispensable in interpreting results of behavioral paradigms, including those manipulating attention (Riley et al., 2002), as predicted decades ago by Gibson (Gibson, 1962). The accumulated evidence suggest that action, sensation and neuronal activity form a loop of variables affecting each other, and by doing so determining tactile object perception.

We hence conclude by suggesting that study of tactile objects from the perspective of a dynamical system may help complete the unifying framework that the field is currently lacking. When research questions are reformulated under this approach theoretical difficulties may simply dissolve; Questions related to attention may find their resolution in kinematic descriptions, and the ambiguity of primitives might be resolved by considering motor-sensory contingencies (Bagdasarian et al., 2013; O’Regan & Noë, 2002). Importantly, experiments addressing these questions must allow natural-like employment of motor patterns.


  • Bagdasarian, K et al. (2013). Pre-neuronal morphological processing of object location by individual whiskers. Nature Neuroscience 16(5): 622-631.
  • Barlow, H B (1972). Single units and sensation: A neuron doctrine for perceptual psychology. Perception 1(4): 371-394.
  • Burton, H et al. (1999). Tactile attention tasks enhance activation in somatosensory regions of parietal cortex: A positron emission tomography study. Cerebral Cortex 9(7): 662-674.
  • Corsini, D A and Pick, H L (1969). The effect of texture on tactually perceived length. Perception & Psychophysics 5(6): 352-356.
  • Damasio, A R (1989). The brain binds entities and events by multiregional activation from convergence zones. Neural Computation 1(1): 123-132.
  • Fries, P; Nikolić, D and Singer, W (2007). The gamma cycle. Trends in Neurosciences 30(7): 309-316.
  • Gelbard-Sagiv, H; Mukamel, R; Harel, M; Malach, R and Fried, I (2008). Internally generated reactivation of single neurons in human hippocampus during free recall. Science 322(5898): 96-101.
  • Gibson, J J (1962). Observations on active touch. Psychological Review 69(6): 477-491.
  • Griffiths, T D and Warren, J D (2004). What is an auditory object? Nature Reviews Neuroscience 5(11): 887-892.
  • Hayward, V (2008). A brief taxonomy of tactile illusions and demonstrations that can be done in a hardware store. Brain Research Bulletin 75(6): 742-752.
  • Horev, G et al. (2011). Motor-sensory convergence in object localization: A comparative study in rats and humans. Philosophical Transactions of the Royal Society of London B: Biological Sciences 366(1581): 3070-3076.
  • Hubel, D H and Wiesel, T N (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of Physiology 160(1): 106-154.
  • Iwamura, Y and Tanaka, M (1978). Postcentral neurons in hand region of area 2: Their possible role in the form discrimination of tactile objects. Brain Research 150(3): 662-666.
  • Iwamura, Y; Tanaka, M; Hikosaka, O and Sakamoto, M (1995). Postcentral neurons of alert monkeys activated by the contact of the hand with objects other than the monkey’s own body. Neuroscience Letters 186(2): 127-130.
  • Jörntell, H et al. (2014). Segregation of tactile input features in neurons of the cuneate nucleus. Neuron 83(6): 1444-1452.
  • Kant, I (n.d.). Critique of pure reason. P Guyer and A W Wood (Eds.). Cambridge, MA: Cambridge University Press.
  • Klatzky, R L and Lederman, S J (1992). Stages of manual exploration in haptic object identification. Perception 52(6): 661-670.
  • Klatzky, R L; Lederman, S J and Metzger, V A (1985). Identifying objects by touch: An “expert system.” Perception & Psychophysics 37(4): 299-302.
  • Klatzky, R L; Lederman, S J and Reed, C L (1989). Haptic integration of object properties: Texture, hardness, and planar contour. Journal of Experimental Psychology: Human Perception and Performance 15(1): 45-57.
  • Klatzky, R L and Lederman, S J (1995). Identifying objects from a haptic glance. Perception & Psychophysics 57(8): 1111-1123.
  • Kobatake, E and Tanaka, K (1994). Neuronal selectivities to complex object features in the ventral visual pathway of the macaque cerebral cortex. Journal of Neurophysiology 71(3): 856-867.
  • Lacey, S; Lin, J B and Sathian, K (2011). Object and spatial imagery dimensions in visuo-haptic representations. Experimental Brain Research 213(2-3): 267-273.
  • Lederman, S J; Browse, R A and Klatzky, R L (1988). Haptic processing of spatially distributed information. Perception & Psychophysics 44(3): 222-232.
  • Lederman, S J and Jones, L A (2011). Tactile and haptic illusions. IEEE Transactions on Haptics 4(4): 273-294.
  • Lederman, S J and Klatzky, R L (1987). Hand movements: A window into haptic object recognition. Cognitive Psychology 19(3): 342-368.
  • Lederman, S J and Klatzky, R L (2009). Haptic perception: A tutorial. Attention, Perception, & Psychophysics 71(7): 1439-1459.
  • Lederman, S J; Klatzky, R L and Reed, C L (1993). Constraints on haptic integration of spatially shared object dimensions. Perception 22(6): 723-743.
  • Locher, P J (1986). Aspects of haptic perceptual information processing. Bulletin of the Psychonomic Society 24(3): 197-200.
  • Mitchinson, B; Martin, C J; Grant, R A and Prescott, T J (2007). Feedback control in active sensing: Rat exploratory whisking is modulated by environmental contact. Proceedings of the Royal Society B: Biological Sciences 274(1613): 1035-1041.
  • O’Regan, J K and Noë, A (2002). A sensorimotor account of vision and visual consciousness. Behavioral and Brain Sciences 24(5): 939-973.
  • Plaisier, M A; Bergmann Tiest, W M and Kappers, A M L (2008). Haptic pop-out in a hand sweep. Acta Psychologica 128(2): 368-377.
  • Plaisier, M A; Bergmann Tiest, W M and Kappers, A M L (2009). Salient features in 3-D haptic shape perception. Attention, Perception, & Psychophysics 71(2): 421-430.
  • Plaisier, M A and Kappers, A M L (2010). Cold objects pop out! Lecture Notes in Computer Science 6192: 219-224.
  • Quiroga, R Q; Mukamel, R; Isham, E A; Malach, R and Fried, I (2008). Human single-neuron responses at the threshold of conscious recognition. Proceedings of the National Academy of Sciences of the United States of America 105(9): 3599-3604.
  • Quiroga, R Q; Reddy, L; Kreiman, G; Koch, C and Fried, I (2005). Invariant visual representation by single neurons in the human brain. Nature 435(7045): 1102-1107.
  • Reed, C L; Lederman, S J and Klatzky, R L (1990). Haptic integration of planar size with hardness, texture, and planar contour. Canadian Journal of Psychology 44(4): 522-545.
  • Riley, M A; Wagman, J B; Santana, M V; Carello, C and Turvey, M T (2002). Perceptual behavior: Recurrence analysis of a haptic exploratory procedure. Perception 31(4): 481-510.
  • Roland, P E; O’Sullivan, B and Kawashima, R (1998). Shape and roughness activate different somatosensory areas in the human brain. Proceedings of the National Academy of Sciences of the United States of America 95(6): 3295-3300.
  • Roskies, A L (1999). The binding problem. Neuron 24(1): 7-9.
  • Saig, A; Gordon, G; Assa, E; Arieli, A and Ahissar, E (2012). Motor-sensory confluence in tactile perception. The Journal of Neuroscience 32(40): 14022-14032.
  • Saal, H P and Bensmaia, S J (2014). Touch is a team effort: Interplay of submodalities in cutaneous sensibility. Trends in Neuroscience 37(12): 689-697.
  • Sathian, K and Burton, H (1991). The role of spatially selective attention in the tactile perception of texture. Perception & Psychophysics 50(3): 237-248.
  • Sathian, K et al. (2011). Dual pathways for haptic and visual perception of spatial and texture information. Neuroimage 57(2): 462-475.
  • Sinclair, R J; Kuo, J J and Burton, H (2000). Effects on discrimination performance of selective attention to tactile features. Somatosensory & Motor Research 17(2): 145-157.
  • Steinmetz, P N; Roy, A; Fitzgerald, P J; Hsiao, S S and Johnson, K O (2000). Attention modulates synchronized neuronal firing in primate somatosensory cortex. Nature 404(6774): 187-190.
  • Stilla, R and Sathian, K (2008). Selective visuo‐haptic processing of shape and texture. Human brain mapping 29(10): 1123-1138.
  • Tallon-baudry, C and Bertrand, O (1999). Oscillatory gamma activity in humans and its role in object representation. Trends in Cognitive Science 3(4): 151-162.
  • Treisman, A (1996). The binding problem. Current Opinion in Neurobiology 6(2): 171-178.
  • Treisman, A (1998a). Feature binding, attention and object perception. Philosophical Transactions of the Royal Society of London B: Biological Sciences 353(1373): 1295-1306.
  • Treisman, A (1998b). Features and objects: The fourteenth Bartlett memorial lecture. The Quarterly Journal of Experimental Psychology Section A 40(2): 201-237.
  • Treisman, A and Souther, J (1985). Search asymmetry: A diagnostic for preattentive processing of separable features. Journal of Experimental Psychology: General 114(3): 285-310.
  • Turvey, M T (1996). Dynamic touch. American Psychologist 51(11): 1134-1152.
  • Von Der Malsburg, C (1994). The Correlation Theory of Brain Function (pp. 95-119). New York: Springer.
  • Von Der Malsburg, C (1995). Binding in models of perception and brain function. Current Opinion in Neurobiology 5(4): 520-526.
  • Whang, K C; Burton, H and Shulman, G L (1991). Selective attention in vibrotactile tasks: Detecting the presence and absence of amplitude change. Perception & Psychophysics 50(2): 157-165.
  • Yeshurun, Y and Sobel, N (2010). An odor is not worth a thousand words: From multidimensional odors to unidimensional odor objects. Annual Review of Psychology 61: 219-241.
Personal tools

Focal areas