Elsevier

Brain and Cognition

Volume 67, Issue 2, July 2008, Pages 197-211
Brain and Cognition

The direction of hemispheric asymmetries for object categorization at different levels of abstraction depends on the task

https://doi.org/10.1016/j.bandc.2008.01.003Get rights and content

Abstract

In this study hemispheric asymmetries for categorizing objects at the basic versus subordinate level of abstraction were investigated. As predictions derived from different theoretical approaches are contradictory and experimental evidence is inconclusive in this regard, we conducted two categorization experiments, where we contrasted two experimental paradigms. In the first experiment, subjects had to verify whether a word and a laterally presented picture matched or not. In the second experiment, subjects had to identify laterally presented pictures of animals either at the basic or subordinate level by pressing a corresponding response key. Whereas the first experiment revealed an advantage of the left hemisphere (LH) for categorizing objects at the basic level and of the right hemisphere (RH) for categorizing at the subordinate level, just the opposite brain asymmetry was found in the second experiment. As the stimuli were identical in both experiments, hemispheric asymmetries seem to be strongly task dependent.

Introduction

Each object we confront in our everyday life can usually be categorized at more than only one level of abstraction (Jolicoeur et al., 1984, Rosch et al., 1976). In most situations, objects are classified at the so-called basic level (e.g. dog). Basic categories are perceptually the most distinctive compared to categories at other levels and contain sufficient information for most of our everyday interactions (Rosch et al., 1976). In some situations, however, a more specific classification at the subordinate level can be necessary, for instance, if one refers to a specific object among other exemplars of the same basic category (e.g. a Dalmatian among other kinds of dogs). It is assumed that categorizations at the basic and subordinate level mainly result from perceptual processing (Jolicoeur et al., 1984, Kosslyn et al., 1995, Tanaka et al., 1999).

The main challenge for our cognitive system in this regard is to accomplish different classifications with identical visual input. For basic-level classification visual inputs of different instances have to be mapped to the same output category. In this case, a certain amount of information has to be ignored. For subordinate-level categorization, on the other hand, very similar visual inputs have to be sorted into different subordinate categories, i.e. even small differences in perceptual features have to be considered. How does our cognitive system fulfill these somewhat opposed demands?

Two major kinds of ideas have been proposed as a solution to this problem. One approach, linked to research on object categorization and object representation, focuses on the kinds of representations that are used for categorizations. The other approach, inspired by research on object perception, focuses on the perceptual features that might be differentially diagnostic for different kinds of categorizations. Both accounts also predict hemispheric asymmetries for categorizing visually presented objects at the basic or subordinate level, which are the focus of the present paper. Before we report our own experiments, both approaches and corresponding empirical research will now be considered in turn.

As one solution to the problem of categorizing objects at the basic or the subordinate level it has been proposed that there might be different kinds of representations for different categorization tasks located in different subsystems in the brain. One kind of representation could subserve classification at the basic, and another kind of representation classification at the subordinate level. Within this representation account it is further assumed that the cerebral hemispheres differ with respect to these representations (e.g. Laeng et al., 2003, Marsolek, 1999). According to Marsolek (1999), for instance, each hemisphere has a bias toward storing the shape of objects in a specific representational format. The left hemisphere (LH) stores visual objects in an abstract format in the sense of a prototype. That is, those features are stored that are invariant over the exemplars of a category. These features of an object or object parts are represented independently of each other (Marsolek & Burgund, 1997). Such a representational format should be helpful for categorizing objects at the basic level as a subset of features is always present in all exemplars of a basic category but often in different configurations. For instance, all kinds of pianos possess the feature ‘keyboard’ but this feature can be present in different configurations in an upright piano compared to a grand piano (c.f. Marsolek, 1999). The right hemisphere (RH), on the other hand, is assumed to store specific exemplars. That is, object shapes are represented in a format that preserves the specific configuration of the features of a specific instance of a category. For instance, such a representation would include the information that the keyboard of a piano is arranged in line with the corpus of a grand piano but horizontally to that of an upright piano (c.f. Marsolek, 1999). Therefore, the specific representation of the RH can be used to differentiate between different subordinate categories.

Marsolek (e.g. 1999) provides some evidence for this representation account in perceptual priming experiments. In a priming phase line drawings of objects were presented centrally. In the following test phase line drawings of either the same objects or of other instances from the same basic categories were presented either in the left visual field (LVF; i.e. projected to the RH) or the right visual field (RVF; i.e. projected to the LH) for a very short time. The task for the subjects was to name these objects. Marsolek found that for stimuli presented in the RVF/LH performance was independent of whether the same exemplar had been used as prime or another exemplar of the same basic category. That is, priming was independent of the specific shape of an object. For LVF/RH-stimuli, in contrast, there was an advantage in performance when the prime and the test object were the same exemplar and a disadvantage in performance when the prime and test object were different exemplars of the same basic category. In this case, priming was sensitive to the specific shape of an object. This pattern of results could be produced with different stimulus material such as word forms (Marsolek et al., 1992, Marsolek et al., 1996, Marsolek et al., 1994), pseudo-word forms (Burgund & Marsolek, 1997), letter-like forms (Marsolek, 1995) and with line drawings of objects (Koutstaal et al., 2001, Marsolek, 1999, Saneyoshi et al., 2003). This indicates that the assumed representational systems are used for storing any kind of visual shapes, e.g. word forms as well as object shapes, and are therefore independent of specific semantic content (c.f. Marsolek, 1995).

The question is whether or not these findings could be transferred to object categorization at the basic and subordinate level. Therefore, Laeng et al. (2003) tried to replicate the findings from these priming experiments in a categorization experiment, using a picture–word verification task. They presented black-and-white drawings of objects (animals, artifacts as well as photos of faces of famous persons) either in the RVF/LH or in the LVF/RH. Directly after picture presentation a word which described the objects either at the basic level (e.g. ‘dog’) or at the subordinate level (e.g. ‘collie’) was presented acoustically through a loudspeaker. The task of the subjects was to indicate whether picture and word matched or not by pressing a corresponding response key. When subjects had to match basic category labels (e.g. ‘dog’) to pictures, responses were faster for RVF/LH-stimuli than for LVF/RH-stimuli. In contrast, when subjects had to match subordinate category labels (e.g. ‘collie’) to pictures, responses were faster for LVF/RH-stimuli than for RVF/LH-stimuli. Laeng et al. concluded from these observed visual field (VF)-effects that hemispheric asymmetries in categorizing objects at different levels of abstraction can be attributed to the fact that each hemisphere stores objects in a specific kind of representation which is in accordance with Marsolek’s representation account (Marsolek, 1999).

Another solution to the problem of categorization at the basic and the subordinate level, which we call the feature account, assumes that specific features might be diagnostic for certain kinds of categorizations (e.g. Schyns, Bonnar, & Gosselin, 2002). Hemispheric asymmetries should then be observable when the processing of these features is lateralized. In this regard, Schyns and colleagues (e.g. Schyns et al., 2002) have recently shown that stimuli need not necessarily be completely encoded and represented in a categorization task, but that task demands influence which features of an object are extracted and used to perform the task at hand (c.f. Schyns, 1998). It has been argued that the cognitive system might, therefore, rely on different aspects of the information that is present in the stimulus when having to categorize an object at the basic or the subordinate level (Archambault et al., 1999, Morrison and Schyns, 2001, Schyns and Oliva, 1997). At least for the domain of living things, there is some evidence that the global and local features of an object play a crucial role. For instance, early work of Eleanor Rosch (Rosch et al., 1976) has already shown that the overall shape of an object is diagnostic for categorizing at the basic level as categories at this level are most distinctive in this respect. For example, the global shape of a dog is quite distinct from the shape of a bird. Recent experimental work has confirmed this idea (Archambault et al., 2000, Collin and McMullen, 2005, Hayward, 1998, Large and McMullen, 2006). In contrast, when an object has to be categorized at the subordinate level, then categorization must rely on local features, because objects of the same basic category are very similar in global shape (Archambault et al., 2000, Collin, 2006, Collin and McMullen, 2005).

The involvement of global and local processing in object categorization also suggests that hemispheric asymmetries should be observable. Research on global/local processing has shown that the RH has a bias for processing the global features of an object, whereas the LH is biased toward the processing of local stimulus features. Corresponding asymmetries were found in lesion studies (e.g. Delis et al., 1986, Robertson and Lamb, 1991), electrophysiological studies (e.g. Heinze and Münte, 1993, Malinowski et al., 2002, Proverbio et al., 1998, Volberg and Hübner, 2004), imaging studies (e.g. Fink et al., 1996, Heinze et al., 1998, Martinez et al., 1997) as well as response time studies (e.g. Hübner, 1997, Hübner, 1998, Martin, 1979). For a recent review see Hübner and Volberg (2005).

The functional hemispheric asymmetries for global/local processing suggest that corresponding hemispheric differences should also be observable for categorizations at different levels of abstraction. The specific hemispheric asymmetry should occur depending on the level of the features that have to be processed. This feature account of hemispheric asymmetries for categorization should at least hold for the domain of living things. As the classification of these objects at the basic level predominantly relies on global stimulus features, the RH should be better in performing basic-level classification than the LH. In contrast, classification of objects at the subordinate level is mainly based on local features. Thus, subordinate-level classification should be better performed by the LH than by the RH.

To our knowledge, up to date there has been no attempt to test this feature account directly. However, there is some indirect evidence from studies of object recognition which usually corresponds to a classification at the basic level. If laterally presented objects have to be classified at the basic level, then, according to the feature account there should be a LVF/RH advantage for this task. Although older studies are inconclusive in this respect (e.g. Bryden and Rainey, 1963, Levine and Banich, 1982, Schmuller and Goodman, 1980, Sergent and Lorber, 1983, Wyke and Ettlinger, 1961), and also exhibit a number of methodological problems (c.f. Biederman & Cooper, 1991), more recent studies seem to support the feature account (e.g. Brooks and Cooper, 2006, Koivisto and Revonsuo, 2003, Laeng et al., 1999, McAuliffe and Knowlton, 2000). An advantage of the LVF/RH was shown for the identification of laterally presented pictures of animals (Brooks and Cooper, 2006, Laeng et al., 1999) and when subjects had to detect familiar objects with nonsense shapes as distracters (Koivisto and Revonsuo, 2003, McAuliffe and Knowlton, 2000). In contrast, when the object recognition task required an analysis of local details, an advantage for the RVF/LH has been observed (e.g. Koivisto and Revonsuo, 2003, Laeng et al., 1999, Lag et al., 2006, Laws and Neve, 1999, Perani et al., 1995). Taken together, these results support the feature account.

As this short overview shows, there are two kinds of predictions about the pattern of hemispheric asymmetries that can be made for categorizing objects at the basic and at the subordinate level. Interestingly, at least for the domain of living things, where the global shape is an indicator for basic-level categorization and local features for subordinate-level categorization, the pattern of hemispheric asymmetries predicted by one approach is contrary to that predicted by the other. Whereas the feature account predicts that the RH and the LH have an advantage for categorizing objects at the basic and subordinate level, respectively, the representation account states exactly the opposite. The aim of the present study was to investigate the origin of the contradicting observations. Why is there evidence for both the feature and the representation account?

One conceivable reason could be that different experimental methods are applied in the two research areas. Research on categorization is traditionally based on experiments using word–picture verification tasks. That is, a word and a picture are presented and subjects have to indicate whether both match or not. This task seems to require to compare two object representations with respect to whether or not they are from the same (basic or subordinate) category. Therefore, functional hemispheric differences according to the representation account might be expected. On the other hand, research on global/local processing and object recognition prefers identification tasks. That is, a picture is presented and subjects have to identify it, e.g. by pressing a corresponding response key. In such a task, subjects are prepared in advance at which level the following object will have to be categorized. Therefore, they might prepare to look for certain diagnostic features that allow to categorize the shown object. There are, however, also differences in the employed stimulus sets. In global/local tasks, a limited number of stimuli is usually presented repeatedly whereas in categorization experiments a large stimulus set is employed where each stimulus is only shown once or at least only a few times. It is conceivable that subjects are able to extract diagnostic features only when they see the same stimuli repeatedly but not when each stimulus is shown only once.

To see whether these methodological differences account for the opposite results, we conducted two experiments in which the stimulus set was largely the same but the task for the observers varied. In the first experiment, we applied an experimental task similar to the one used by Laeng and colleagues (2003), i.e. the subjects had to match pictures and words. The aim was to examine whether Laeng et al.’s (2003) result can also be obtained with a stimulus set whose relatively small size corresponds to that usually used in global/local experiments. The results of this task were then compared to those of a second experiment, in which the same stimuli were used but where a more global/local like task was applied (e.g. Hübner & Malinowski, 2002).

Our stimuli were shaded three-dimensional drawings of animals. In order to be able to clearly distinguish the predictions of the two approaches, we selected animals as stimuli which could be distinguished by their global shape at the basic level and by a local surface pattern at the subordinate level (for examples, see Fig. 1). That is, the membership to a given basic category was defined by the global object shape, whereas the membership to a given subordinate category was defined by a combination of global shape and local texture. If the feature account is valid, then there should be a LVF/RH advantage for categorizing our stimuli at the basic level and of the RVF/LH for categorizing them at the subordinate level. If, on the other hand, the representation account is valid, then the pattern of hemispheric asymmetries should be reversed.

Section snippets

Experiment 1

In our first experiment, we used a word–picture verification task similar to the one used by Laeng et al. (2003). The question was: Can the pattern of hemispheric asymmetries predicted by the representation account be obtained with a stimulus set that is smaller and that is clearly defined by global and local features with such a task? The subjects had to decide whether a visually presented word describing an object at the basic or at the subordinate level matched a subsequently presented

Experiment 2

The identification task used in this experiment was similar to that used in global/local studies. We expected that identification would lead to hemispheric asymmetries that differ from those in the previous experiment. According to the feature account, individual feature levels (global or local) are diagnostic for specific categorizations and, therefore, determine the direction of the hemispheric asymmetries. In particular, we hypothesized that the RH should have an advantage for categorizing

General discussion

In the present study, two experiments were conducted in which we investigated hemispheric asymmetries for the categorization of natural objects (animals) at the basic level and at the subordinate level. In this context, we contrasted two theoretical approaches—one that is more related to research on object categorization and another one that is more related to research on object perception. Interestingly, both accounts predict hemispheric asymmetries for categorization, but in opposite

Acknowledgments

We are grateful to Stephen Christman and four anonymous reviewers for helpful comments on an earlier version of this paper. We also thank Patrick Kleiner, Maddalena Brunetti, and Jan Schlösser for collecting the data.

References (66)

  • K.R. Laws et al.

    A ’normal’ category-specific advantage for naming living things

    Neuropsychologia

    (1999)
  • C.J. Marsolek et al.

    Computational analyses and hemispheric asymmetries in visual-form recognition

  • M. Martin

    Hemispheric specialization for global and local processing

    Neuropsychologia

    (1979)
  • A.M. Proverbio et al.

    Electrophysiological evidence of a perceptual precedence of global vs. local visual information

    Cognitive Brain Research

    (1998)
  • L.C. Robertson et al.

    Neuropsychological contributions to theories of part/whole organization

    Cognitive Psychology

    (1991)
  • J. Schmuller et al.

    Bilateral tachistoscopic perception, handedness and laterality: II. Nonverbal stimuli

    Brain and Language

    (1980)
  • P.G. Schyns

    Diagnostic recognition: Task constraints, object information, and their interactions

    Cognition

    (1998)
  • J. Sergent et al.

    Perceptual categorization in the cerebral hemispheres

    Brain and Cognition

    (1983)
  • E.E. Smith et al.

    Alternative strategies of categorization

    Cognition

    (1998)
  • K.F. Sullivan et al.

    The roles of stimulus repetition and hemispheric activation in visual half-field asymmetries

    Brain and Cognition

    (1985)
  • G. Volberg et al.

    On the role of response conflicts and stimulus position for hemispheric differences in global/local processing: An ERP study

    Neuropsychologia

    (2004)
  • S.W. Allen et al.

    Specializing the operation of an explicit rule

    Journal of Experimental Psychology: General

    (1991)
  • A. Archambault et al.

    A natural bias for the basic level?

  • A. Archambault et al.

    Blind to object changes: When learning the same object at different levels of categorization modifies its perception

    Psychological Science

    (1999)
  • F.G. Ashby et al.

    Human category learning

    Annual Review of Psychology

    (2005)
  • F.G. Ashby et al.

    Category learning and multiple memory systems

    Trends in Cognitive Sciences

    (2005)
  • B.E. Brooks et al.

    What types of visual recognition tasks are mediated by the neural subsystem that subserves face recognition?

    Journal of Experimental Psychology: Learning Memory and Cognition

    (2006)
  • M.P. Bryden et al.

    Left–right differences in tachistoscopic recognition

    Journal of Experimental Psychology

    (1963)
  • C.A. Collin

    Spatial-frequency thresholds for object categorisation at basic and subordinate levels

    Perception

    (2006)
  • C.A. Collin et al.

    Subordinate-level categorization relies on high spatial frequencies to a greater degree than basic-level categorization

    Perception & Psychophysics

    (2005)
  • E.L. Cowin et al.

    Categorical versus coordinate spatial processing: Effects of blurring and hemispheric asymmetry

    Journal of Cognitive Neuroscience

    (1994)
  • G.R. Fink et al.

    Where in the brain does visual attention select the forest and the trees?

    Nature

    (1996)
  • M.S. Gazzaniga et al.

    Cognitive neuroscience. The biology of the mind

    (2002)
  • Cited by (0)

    View full text