Color term knowledge does not affect categorical perception of color in toddlers

https://doi.org/10.1016/j.jecp.2004.10.001Get rights and content

Abstract

Categorical perception of color is shown when colors from the same category are discriminated less easily than equivalently spaced colors that cross a category boundary. The current experiments tested various models of categorical perception. Experiment 1 tested for categorical responding in 2- to 4-year-olds, the age range for the onset establishment of color term knowledge. Experiment 2 tested for categorical responding in Himba toddlers, whose language segments the color space differently from the way in which the English language does so. Experiment 3 manipulated the conditions of the task to explore whether the categorical responding in Experiments 1 and 2 was equivalent to categorical perception. Categorical perception was shown irrespective of naming and was not stronger in those children with more developed color term knowledge. Cross-cultural differences in the extent of categorical perception were not found. These findings support universalistic models of color categorization and suggest that color term knowledge does not modify categorical perception, at least during the early stages of childhood.

Introduction

Color is perceived categorically. The color spectrum is continuous; however, this continuum is perceived as a number of discrete categories, with the members of categories resembling each other more than they resemble members of other categories. This has been termed “categorical perception” (Harnad, 1987). Stimuli from within a category (within-category stimuli) are perceived as more similar than stimuli that straddle a category boundary (between-category stimuli), even when stimulus separation sizes for within- and between-category stimuli are equal. For example, in Fig. 1, the three stimuli A1, A2, and B are equidistant in color space, two stimuli (A1 and A2) belong to the same linguistic color category (e.g., blue), and the third stimulus (B) belongs to an adjacent linguistic category (e.g., green).

Categorical perception is shown when the stimulus pair A2–B1 is discriminated faster, more easily, or more accurately than the stimulus pair A1–A2. Categorical perception of color has been evidenced using same–different judgment, recognition memory, and two-alternative forced-choice tasks (2-AFCs) (e.g., Bornstein and Korda, 1984, Pilling et al., 2003, Roberson and Davidoff, 2000, Uchikawa and Shinoda, 1996).

The origin of categorical perception of color is under debate. On the one hand, universalists argue that categorical perception is “hardwired” into the visual system and that categorical perception of color is an innate, universal, and perceptual effect. On the other hand, linguistic relativists argue that categorical perception of color is constructed through language—an idea that has its roots in the Sapir–Whorf hypothesis that language determines thought (Sapir, 1921, Whorf, 1956). Language “warps” perceptual space, creating compression of within-category perceptual space and expansion of between-category perceptual space.

The universalistic theory leads to the prediction that categorical perception should be shown during infancy—before color terms are learned. Studies that show categorical perception of color in 4-month-olds (e.g., Bornstein et al., 1976, Franklin and Davies, 2004) support this prediction. Using a habituation technique and monochromatic lights, Bornstein and colleagues (1976) showed that infants respond categorically across blue–green, yellow–green, and yellow–red boundaries. Using a novelty preference technique and the Munsell color metric,1 Franklin and Davies (2004) showed that infants respond categorically across blue–green and blue–purple hue boundaries and across a pink–red boundary defined by differences in lightness and saturation.

The linguistic relativity theory leads to the prediction that categorical perception will vary as language varies. Cross-cultural studies of categorical perception in adults (e.g., Kay and Kempton, 1984, Pilling and Davies, in press, Roberson et al., 2000) support this prediction. For example, Roberson and colleagues (2000) tested for categorical perception (using a 2-AFC task) in two populations whose languages segment the color space differently from each other. Whereas English distinguishes between blue and green, the Berinmo language (Papua New Guinea) does not. In contrast, Berinmo makes a distinction (nol/wor) that English does not. Category effects were shown by English and Berinmo speakers, but only at the bluegreen boundary for the English speakers and only at the nolwor boundary for the Berinmo speakers.

In summary, there is evidence to support both sides of the debate. On the one hand, prelinguistic infants show categorical perception across a range of color boundaries. On the other hand, cross-cultural differences in categorical perception are found. These two streams of evidence seem to contradict each other. How is it possible that prelinguistic infants show categorical perception, whereas adults do not show categorical perception when their language does not mark the category boundary? The apparent contradiction is based on two crucial, but untested, assumptions: (a) that infant categorical perception is hardwired and universal and (b) that infant categorical perception is equivalent to adult categorical perception. Infant color categorical perception has been found in British and American infants, but no such tests have been made on infants from other language groups. The behavioral markers of infant categorical perception are based on either dishabituation (Bornstein et al., 1976) or novelty preference (Franklin & Davies, 2004), whereas adult categorical perception is evidenced typically using 2-AFC discrimination tasks (e.g., Roberson et al., 2000). Thus, infant and adult behaviors may be based on different processes. However, if these two crucial assumptions are accepted, it seems to imply that categorical perception disappears unless it is supported by linguistic distinctions; some kind of “perceptual reorganization” must occur.

The cross-cultural variation in categorical perception in adults is not necessarily due to language. The cross-cultural studies show an association between language and categorical perception, but they do not show that language causes the variation. The cross-cultural differences in categorical perception could predate language acquisition. For example, perhaps Berinmo infants would not show categorical perception across the bluegreen boundary but would show categorical perception across the nolwor boundary. Moreover, it could be these prelinguistic differences that lead to different color lexicons. However, although this argument is valid logically, its plausibility is low because it questions a major tenet of vision science—the universality of early visual processes—without supporting evidence.

The first stages of chromatic processing are based on three different classes of cones with different spectral sensitivities. Processing within the retina recombines the cone signals into three opponent channels that are carried to the visual cortex on anatomically separate pathways (e.g., Gegenfurtner & Kiper, 2003). The opponent signals are based on the receptive field structure of ganglion and thalamic cells. This basic structure is assumed to be common to all individuals with normal color vision. However, this structure is not fully developed at birth. Rather, retinal and receptive field organization develops, probably over several years (e.g., Atkinson, 1984). Neuronal growth that leads to receptive field structure could be completely hardwired, there could be a random component to it, or there could be a degree of environmental tuning to it (MacLeod, 2003, Yendrikhovskij, 2001). The latter possibility implies that different chromatic environments could lead to different chromatic processing structures. In extremis, restricted chromatic environments during the first year lead to defective color vision and abnormal color constancy in macaque monkeys (Sugita, 2004). However, even if chromatic tuning occurs, it is not clear what chromatic regularities neuronal growth would respond to, and these regularities could be universal (Shepard, 1992).

There are variations in the optical properties of the eye and retinal sensitivity that could produce differences in color vision. Preretinal filtering varies due to different optical densities of the lens and macula pigmentation (Bornstein, 1973), and there are variations in the spectral sensitivities of cone photopigments (e.g., Jameson, Highnote, & Wasserman, 2001). However, there is no evidence that these variations covary with language, culture, or region. Webster and colleagues (2002) did report systematic differences in the identification of focal and unique hues between Indian and American populations, and physiological, environmental, and cultural reasons for the differences were considered. However, the origin of this difference is not clear.

The second part of the assumption is that infant categorical perception is hardwired. The brain structures responsible for color categories are not known, but they almost certainly involve cortical structures beyond the visual cortex. Perhaps these structures are tuned by the chromatic environment such that by 4 months of age, adultlike categorical perception is shown. As with our speculations about tuning of earlier chromatic stages, there is no direct evidence that this occurs, nor is it clear what chromatic regularities would influence tuning. However, one possibility is that infant environments in the industrialized world are dominated by artifacts (e.g., toys, clothes, pictures) in saturated primary colors that are close to category prototypes (Rosch, 1972). Caregivers may also emphasize these prototypes by, for instance, bringing them to the attention of infants. Perhaps such processes tune category formation, leading to categorical perception by 4 months of age. In contrast, infants in tribal societies, such as the Berinmo, have chromatic environments based mostly on natural objects that tend to be in more muted colors. This could lead either to less salient categories or even to different categories.

Both infants and adults behave as though there is something pertinent about categorical differences of colors. However, even on tasks that seem to be formally equivalent, such as adult 2-AFC and infant novelty preference, there are differences that call their equivalence into question. The formal equivalence is that both infant and adult tasks involve 2-AFCs. The infants are familiarized to a standard color and then confronted with the standard paired with a novel color, either from the same category as the standard or from an adjacent category. Categorical perception is evidenced by infants looking more at the categorically different test color than at the same category test color. In adult 2-AFC, a target color is presented, followed after an interval by two test colors: One identical to the target and one different (the foil). As with infants, on some trials the different color is from the same category as the target, and on some trials the different color is from an adjacent category. For adults, accuracy is the performance measure; for infants, the degree of novelty preference, as indicated by direction of gaze, is the index of performance. Even assuming that the two measures are equivalent, adult and infant patterns of behavior indicating categorical perception differ. Adults are more accurate for between-category pairs than for within-category pairs, but their within-category accuracy is well above chance (e.g., Pilling et al., 2003). In contrast, infants show no novelty preference for within-category pairs; they seem to require a categorical change to engage their interest. Although this is sufficient to suggest that infants have something like color categories, it also suggests that they might not yet have developed their adult form. Beyond their formal equivalence, the tasks differ between infants and adults. First, adults are explicitly told that they have to choose the color they think is identical to the target, whereas infants are (necessarily) not instructed. Second, adults are likely to bring to bear on the task a repertoire of strategies that are not available to infants. Perhaps most crucially, adults may use verbal labels for the stimuli to help them remember the target color across the interval (we return to this in the next section).

These differences between adult and infant measures of categorical perception suggest that it would be useful to test for categorical perception prelinguistically using tasks as similar to the adult tests as possible. Studies that test for categorical perception as soon as children are able to be instructed, using tasks such as 2-AFC discrimination, would bridge the gap between the current infant and adult studies of categorical perception.

So far, we have followed conventional use by using the term “categorical perception.” However, although it is clear that there are categorical influences on responding, the perceptual basis for these effects is inferred only indirectly. Most tasks used to measure categorical perception require the use of memory, and for adults performance might also be influenced by labeling. The evidence is consistent with categorical perception arising from perceptual processes, but the evidence is also consistent with it being due to labeling or arising from memory processes.

Labeling could produce apparent categorical perception as follows. If the target color in a 2-AFC task is labeled and the label is retained across the interval to be compared with the labels given to the two test stimuli (target plus foil), then this would support accurate choices when the foil and target are in different categories (e.g., blue–green) but not when they are in the same category (e.g., blue–blue). A simple model based on naming reliability predicts between-category performance well but underestimates within-category performance for delayed 2-AFC and same–different tasks (Pilling et al., 2003). Consistent with this labeling account, Roberson and Davidoff (2000) found that categorical perception is eliminated by verbal interference (presentation of a list of words that had to be read), but not visual interference (presentation of a stationary curved line that had to be tracked visually), during the interstimulus interval (ISI) of a successive 2-AFC task. Categorical perception in infants cannot be due to verbal labeling, but verbal labeling could explain the cross-cultural differences in categorical perception seen in adults. For example, perhaps the Berinmo did not show categorical perception for blue–green because they both have the same label (Munnich & Landau, 2003). With the use of nonlinguistic tasks, cross-cultural differences might not be seen and the underlying perceptual categorization of color might be revealed to be universal.

Categorical perception-like effects could also arise from “distortions” in memory traces. For example, Huttenlocher, Hedges, and Vevea (2000) argued that as uncertainty about a target stimulus increases with factors such as noisy stimuli, reduced attention, and increasing retention periods, it pays to bias judgments toward the category prototype. Because within-category stimuli have the same prototype and between-category stimuli have different prototypes, a shift toward the prototype would make the representations of target and foil more different for between-category pairs but more similar for within-category pairs, mimicking categorical perception. There is evidence that memory errors can be biased systematically (Davies et al., 2004, Petzold and Sharpe, 1998, Prinzmetal et al., 1998). However, Pilling and colleagues (2003) found that color categorical perception and memory bias occurred but were dissociated. Moreover, color categorical perception is found using simultaneous same–different tasks (Özgen & Davies, 2003) and visual search tasks (e.g., Daoutis et al., 2004, Daoutis et al., 2004, Franklin et al., 2004) where the memory load is minimal and labeling is of no use.

It is quite possible that “categorical responding” could arise from perception, and from memory and labeling, and that these different sources are more or less active depending on the particular task and the strategies deployed. If the concern is whether genuine categorical perception is occurring, then studies of categorical perception need to include tasks that can either isolate perceptual processes or exclude memory and labeling mechanisms.

A third theory that combines the ideas of universalism and linguistic relativism could resolve the conflict. The resulting theory, the theory of “perceptual reorganization,” postulates that there is an innate predisposition for category boundaries at certain points in the color space but that language learning modifies the location and extent of categorical perception, reorganizing the representation of perceptual color space. Here the process of learning color terms would draw attention to the similarities of colors given the same term and would draw attention to the differences of colors given different terms. If color perception is “plastic,” then this process could cause compression of perceptual space for the areas of color space that are given the same term and could cause expansion of perceptual space for the areas of color space where there is a linguistic boundary. Such perceptual learning could create categorical perception across newly learned boundaries and could attenuate or eliminate it from within-category regions, altering the structure of perceptual categorization. Thus, categorical perception shown during infancy may later be lost if a language does not mark the boundary, and categorical perception may be accentuated and sharpened if a language does mark the boundary. Finally, categorical perception not present during infancy may later be acquired if a language does mark the distinction between two color regions.

There is some evidence that supports the model. The flexibility of categorical perception in general is supported by category training studies (Goldstone, 1994, Goldstone et al., 2001, Guenther et al., 1999, Livingston et al., 1998, Özgen and Davies, 2002). For example, Özgen and Davies (2002) provided clear evidence of the plasticity of color discrimination and the modifiability of category structure. First, they showed that color discrimination could be improved by training and that the learning was restricted to the training stimuli. Second, they showed that learning to subdivide a preexisting basic color category induced categorical perception across the new boundary. The boundary was at the center of either the blue or green category, and before training discrimination in this region was the minimum for the category. Training reversed this pattern, with the old category center now having peak discriminability.

The idea that universal hardwiring is at some point molded by linguistic or environmental input can be found in other domains. For example, there is evidence that some speech contrasts discriminated by infants are later lost if a language does not encode them (Werker and Lalonde, 1988, Werker and Tees, 1983, Werker and Tees, 1984). Hespos and Spelke (2004) showed that 5-month-olds reared in an English-speaking environment make a conceptual distinction that is marked by the Korean language, whereas English-speaking adults do not make this distinction. The mechanisms that are at play in these studies are not necessarily the same (Bloom, 2004) and are not necessarily the mechanisms that would be needed for perceptual reorganization of color. However, these studies, as well as Goldstone’s (1994) and Özgen and Davies’s (2002) studies, may suggest that there is a degree of plasticity to discrimination and categorization (see also Fahle & Poggio, 2002).

To summarize, there is no current consensus about the origin or nature of categorical perception. Prelinguistic infants show categorical perception, but it is not clear whether the category effect is equivalent to categorical perception in adult studies or whether the category effect is perceptual. It is also not clear whether prelinguistic categorical perception is universal. Cross-cultural differences in categorical perception exist, but it is not clear whether cross-cultural differences are a result of language learning or whether these differences exist before language. It is also not clear whether these differences are truly perceptual or whether these differences are due to verbal labeling. Therefore, there are various issues that need to be resolved. First, can categorical perception be shown prelinguistically using tasks equivalent to the tasks in the adult studies, and what is the impact of language learning on the extent of categorical perception? Second, if categorical perception is shown prelinguistically, is it universal? Third, is categorical perception truly perceptual? Before these issues are resolved, a clear understanding of the origin and nature of categorical perception cannot be reached.

The set of experiments presented here combine developmental and cross-cultural approaches to investigate each of these issues further. In Experiment 1, category effects are tested in 2- and 3-year-olds at different stages of color term knowledge. The three boundaries of Franklin and Davies’s (2004) investigation of categorical perception of color in infants are tested. A 2-AFC task (equivalent to the task used in adult studies of categorical perception) is used. Experiment 2 uses the same technique as Experiment 1 and tests for category effects in a group of toddlers from rural Namibia who have not yet learned color terms and whose parental language has 5 basic color terms (as opposed to the 11 terms in English).

Each of the three models of categorical perception would predict different results for Experiments 1 and 2. The universalistic model would predict that all toddlers, irrespective of color term knowledge or parental language, would respond categorically. The linguistic relativity model would predict that only the English toddlers in Experiment 1, who have learned to linguistically mark category boundaries, would respond categorically. The perceptual reorganization model would predict that all toddlers, irrespective of color term knowledge or parental language, would respond categorically but that the English toddlers in Experiment 1 (who have learned their color terms) would show a stronger category effect.

Experiment 3 looks at the nature of the category effect by varying the memory load of the task to ascertain whether the category effect found is based on a memory process. If the locus of the category effect is memory, then the size of the category effect should decrease as the memory component of the task is reduced. If the locus of the category effect is perception, then a category effect should be found even when there is no memory component to the task. Until it is clear whether the locus of any category effect found is perceptual, the term “categorical responding” will be used instead of “categorical perception.”

Section snippets

Experiment 1: The effect of color term acquisition on categorical responding to color in toddlers

Early studies of color term knowledge in children reported that color terms were not reliably acquired until around 4–7 years of age (Bornstein, 1985). However, more recent studies have shown that color terms can be reliably acquired at as young as 2 years of age (e.g., Andrick and Tager-Flusberg, 1986, Shatz et al., 1996). Reliable knowledge of the first nine basic color terms is established at around 3 years of age, with the acquisition of brown and gray being established roughly 6 to 9

Experiment 2: Categorical responding in English and Himba toddlers

Experiment 1 investigated categorical responding in English toddlers. The toddlers showed a category effect for blue–green, blue–purple, and pink–red boundaries. Experiment 2 investigated categorical perception effects in toddlers in a different population-the Himba. The Himba live in the Kaokoveld region of northwestern Namibia, which is remote, mountainous, sparsely vegetated, and arid. The Himba lead a traditional, pastoral, and nomadic life and make up less than 1% of the Namibian

Experiment 3: The nature of the category effect in toddlers

The findings of Experiments 1 and 2 suggest that language may have a minimal role in the origin and early development of the category effect. However, it is possible that the effect reflects memory factors rather than categorical perception. If this were the case, then one would expect poorer categorical perception performance on tasks with high memory demands and would expect better categorical perception performance on tasks with low memory demands. This hypothesis was tested in Experiment 3.

Summary of the main findings

Categorical perception was shown by toddlers across a range of boundaries on a 2-AFC task. The extent of categorical perception was not affected by linguistic categorization, and the strength of the category effect did not increase with increased general color term knowledge. Himba toddlers who had acquired no color terms showed categorical perception across the blue–purple linguistic boundary—a boundary that is not marked linguistically by their parental language. There was no significant

Conclusions

The experiments presented here have provided evidence that language is not the origin of categorical perception and that language does not modify categorical perception in toddlers. Therefore, these experiments provide further support for universalistic theories of perceptual color categorization. No evidence was provided to support the relativistic claim that language is the origin of the categorical perception effect. No evidence was provided to support the theory that language modifies or

Acknowledgments

This research was supported by an Economic and Social Research Council postgraduate studentship (R42200124191) and an Economic and Social Research Council postdoctoral fellowship (PTA-026-27-0110) to the first author. We owe thanks to Amy Riddett for assistance with some of the data collection in Namibia and to Keemu Jakarama for translation. We are very grateful to the toddlers who participated in this research. We are also grateful to Gavin Bremner, Paul Sowden, Christine Daoutis, Samantha

References (63)

  • Androulaki, A. (2003). Colour term acquisition and the development of working memory in children: A cross-linguistic...
  • J. Atkinson

    Human visual development over the first 6 months of life: A review and a hypothesis

    Human Neurobiology

    (1984)
  • P. Bloom

    Children think before they speak

    Nature

    (2004)
  • M.H. Bornstein

    Color vision and color naming: A psychophysiological hypothesis of cultural difference

    Psychological Bulletin

    (1973)
  • M.H. Bornstein et al.

    Color vision and hue categorization in young infants

    Journal of Experimental Psychology: Human Perception and Performance

    (1976)
  • M.H. Bornstein et al.

    Discrimination and matching within and between hues measured by reaction times: Some implications for categorical perception and levels of information processing

    Psychological Research

    (1984)
  • L. Boroditsky et al.

    Sex, syntax, and semantics

  • M. Bowerman et al.

    Shaping meanings for language: Universal and language-specific in the acquisition of spatial semantic categories

  • Daoutis, C. A., Franklin, A., Riddett, A., Clifford, A., & Davies, I. R. L. (2004). Category effects in children’s...
  • Daoutis, C. A., Pilling, M., & Davies, I. R. L. (2004). Categorical effects in visual search for colour. Unpublished...
  • I. Davies et al.

    Categorical similarity may affect color pop-out in infants after all

    British Journal of Developmental Psychology

    (2002)
  • I.R.L. Davies et al.

    The relationship between categorical perception and memory bias towards the prototype

    Perception

    (2004)
  • D. Dedrick

    Color categorisation and the space between perception and language

    Behavioral and Brain Sciences

    (1997)
  • M. Fahle et al.

    Perceptual learning

    (2002)
  • A. Franklin et al.

    New evidence for infant color categories

    British Journal of Developmental Psychology

    (2004)
  • A. Franklin et al.

    Category effects in visual search for colour: Evidence from eye-movement latencies

    Perception

    (2004)
  • D.J. Gabriel

    The spirit of childhood

    (2001)
  • K.R. Gegenfurtner et al.

    Color vision

    Annual Review of Neuroscience

    (2003)
  • S. Glucksberg et al.

    Words in utterance contexts: Young children do not confuse the meaning of same and different

    Child Development

    (1976)
  • R.L. Goldstone

    Influences of categorisation on perceptual discrimination

    Journal of Experimental Psychology: General

    (1994)
  • Cited by (90)

    View all citing articles on Scopus
    View full text