Evolution of Colour Terms: 6 Categorisation Constraints

Continuing my series on the Evolution of Colour terms, this post reviews evidence for categorisation constraints on colour perception. For the full dissertation and for references, go here.

This section reviews the conflicting evidence for ability of linguistic categories to affect perception, which is crucial for the Cultural implication.  Studies of Embodied Cognition which found evidence that comprehension of size and shape activated perceptual representations.  However, colour is not necessarily processed in the same way as spatial properties and is encoded with less stability (Aginsky & Tarr, 2000).  Several studies have found fragile links between colour concepts and colour perception.  For example, Kay and Kempton (1984) showed that categorisation could affect perception in the domain of colour by measuring the subjective distance of green and blue hues of two cultures – one with a separate term for green and blue and one with one term that covered both.  Participants were presented with triads of coloured chips and asked to identify the one that was most different (‘furthest away’) from the other two.  The subjective distances of the speakers of the language with two colour terms were warped at the category boundary.  That is, their categorisation influenced the way they contrasted colours.   However, this effect disappeared when the task was constrained so as to avoid the “name strategy”.  This strategy involved labelling the chips with colours terms, then basing the decision on those labels rather than the actual spectral properties.  Kay and Kempton suggest only a weak form of the Whorfian hypothesis is supported by their study, namely that “Structural differences between language systems will, in general, be paralleled  by nonlinguistic cognitive differences, of  an unspecified sort, in the  native speakers of the two languages” (See Kay & Kempton, 1984, p. 74).

Ritcher and Zwaan (2009) found that perceptual symbols are activated during the comprehension of colour terms using an interference paradigm.  Participants were shown two coloured squares in sequence, interrupted by a colour term.  The squares were either identically coloured, or very slightly different. The colour term either matched or mismatched the colour of the squares.  Participants performed a lexical decision task on the word, and then judged whether the colours were the same or different.  Responses for the colour discrimination task were slower when participants saw an intervening colour term that mismatched the colour of the squares.  This result held even when the participants were not required to perform the lexical decision task.  This implies that, for the current study’s experiment, perceptual activation is still expected without a comprehension task.  Ritcher and Zwaan interpreted the results as a conflict between the perception of the colour of the first square held in short-term memory and a different colour activated by the mismatched colour word.  This supports the theory that perceptual representations are accessed during comprehension.

In perceptual symbol processing terms, their comprehension of the structure of the object activated perceptual representations of that object’s typical colour.  Olkonnen et al. (2008) suggest that this is due to a top-down effect of expectations.  In other words, they do not rule out comprehension involving abstract symbol manipulation (contra-Embodied Cognition).  However, they do predict that the object must share visual features with its referent in order to effect perception.  Indeed, they find less of an effect with objects stripped of shading or texture.  This suggests that the perception (of a structure) activates a perceptual ‘memory’ (of a colour) whereas Embodied Cognition would argue that there is no difference between perceptions and concepts in terms of the neural substrates involved.

However, there is evidence that, although perceptual symbols may be activated by comprehension, they can occur in separate substrates to direct perceptual activation (Connell 2005, 2007).  This would be a modification of a strict perceptual symbols theory.  Connell (2007) showed participants an image following a sentence that primed the reader to expect a certain colour configuration.  For example, a “steak on a plate” primed a cooked (so brown) steak while a “steak in a butcher’s window” primed a raw (so red) steak.  Participants had to confirm that the image contained the object mentioned in the sentence. There was a significant difference in response times between matching and mismatching conditions, suggesting that the comprehension of colour involves perceptual symbols.  Counter- intuitively, however, the mismatched condition evoked faster responses (participants primed with a steak on a plate responded faster to a red steak than a brown steak).  Connell explains this by suggesting that colour is encoded in a different way to some other object features.  A distinction is made between stable embodied representations, such as size and shape, which are multimodal and salient in visual field configuration and unstable embodied representations such as colour and smell, which are unimodal and not salient in visual field configuration.  If shape is more important for recognition than colour, colour can be ignored without incurring a processing cost.  Connell argues matched colours are harder to ignore, since the neural substrates that represented it are already active (see nurological evidence in Spalek & Thompson-Schill, 2008).

The conflict between Ritcher and Zwann (2009) and Connell (2007) could be due to separate levels of representation.  For example, the perceptual representation for the abstract term ‘red’ and the one for the more concrete red of a raw steak could belong to different orders of an abstraction hierarchy.  Connell and Lynott (2009) suggest that separate representations can be

Returning to the motivational paper for this study, Hansen et al. (2006) found that perceptions are influenced by the knowledge of an object’s ‘typical’ colour (bananas always look yellow).  maintained when considering the interactions between typicality and context.  Response times were tested for objects presented either in their typical or atypical colour (e.g., red tomato versus green tomato), crossed with matching and mismatching primes (e.g., ‘Dan ate the tomato’ versus ‘Dan tasted the tomato before it was ready to eat’).  Responses were fastest both for typically coloured objects and matched primes but there was no additional effect for these conditions combined.  This suggests that typical colours (e.g., ‘red’ for tomatoes) are activated, regardless of context, but a representation of the contextual colour – the ‘actual’ colour - can be held in parallel.  An interpretation of Hansen et al.’s findings, in the light of possible dual representations, is that the expected colour of an object (yellow for a banana) can interfere or integrate with the actual colour of an object (a grey banana with a blue tint) when participants are forced to give a single response.  Conell and Lynott suggest that holding parallel representations may be advantageous for error-correction, in the same way maintaining many possible parses of a sentence is advantageous (see Mitchell, 1994).  However, when required to sequentialise concepts, as for expression in language, a fusion of expected and actual representations is required, causing the ‘memory colour’ effect.

This does not fit with a strict perceptual symbol account, which predicts only one representation for both perceptions and concepts.  Neither does it fit with a Symbolist account, which predicts two representations, but one of them being non- perceptual.

Furthermore, it has been suggested that these parallel representations can be driven by different domains of constraint, possibly confounding the Universalist/Relativist debate.  Claidière, Jraissati and Chevallier (2008) asked participants who had 11 colour categories in their native language to sort Munsell chips into 4 categories.  Participants’ choices reflected their native linguistic categories more closely than four-colour-category languages from the WCS, suggesting that participants were using their lexical categories, rather than perceptual ones.  This was interpreted as supporting a relativist theory.  However, the same results were obtained when participants sorted the chips while doing verbal shadowing (mindlessly repeating sentences played to them through headphones), which has been shown to interfere with lexical access and categorical perception (Winawer et al., 2007).  Claidière et al. concluded that colour categories are both language-specific (a cultural constraint supporting Relativism) and perceptually motivated (a perceptual constraint supporting Universlaism).  Furthermore, they conclude that the Relativist/Universalist dichotomy is too narrow and restricts research and experimental paradigms.  Claidière et al. suggest that the influence of categorisation on cognition is task-dependent and that not all perceptually determined categories may be perceptually grounded (e.g., ‘red’ vs. ‘orange’ may be salient, but ‘orange’ on its own is not).  This runs against the idea of Embodied Cognition, which hypothesises a single process of comprehension.

However, an alternative interpretation of Claidière et al.’s finding is that linguistic and perceptual categories are synchronised.  That is, participants in the verbal shadowing task were sorting the colours perceptually without influence from their linguistic categories, but their perceptual space had been warped by their linguistic categories and so the results were the same.  This synchronisation of linguistic and perceptual categories is exactly what is predicted by the dynamic outlined at the beginning of this section.  That is, a feedback loop between a single conceptual/perceptual space and linguistic labels.   Furthermore, the current study will show that people see colour term as achromatic only when they are adjusted to directly oppose the best exemplar of that colour term.  That is, the concept and the percept align directly.  This suggests single, not dual representations.  However, the current study does find differences in the strengths of this effect for different colours, potentially agreeing with Claidière et al.’s suggestion of graded perceptual grounding.

Next, a look at perceptual warping ->

Aginsky, Michael, V., & Tarr, M. J. (2000). How Are Different Properties of a Scene Encoded in Visual Memory? Visual Cognition, 7 (1-3), 147-162 DOI: 10.1080/135062800394739

Kay, P., & Kempton, W. (1984). What Is the Sapir-Whorf Hypothesis? American Anthropologist, 86 (1), 65-79 DOI: 10.1525/aa.1984.86.1.02a00050

Zwaan, R., Stanfield, R., & Yaxley, R. (2002). Language Comprehenders Mentally Represent the Shapes of Objects Psychological Science, 13 (2), 168-171 DOI: 10.1111/1467-9280.00430

Olkkonen, M., Hansen, T., & Gegenfurtner, K. (2010). The structure of color space is largely invariant under illuminant changes Journal of Vision, 8 (6), 574-574 DOI: 10.1167/8.6.574

Connell, L., & Lynott, D. (2009). Is a bear white in the woods? Parallel representation of implied object color during language comprehension Psychonomic Bulletin & Review, 16 (3), 573-577 DOI: 10.3758/PBR.16.3.573

SPALEK, K., & THOMPSONSCHILL, S. (2008). Task-dependent semantic interference in language production: An fMRI study☆ Brain and Language, 107 (3), 220-228 DOI: 10.1016/j.bandl.2008.05.005

Hansen, T., Olkkonen, M., Walter, S., & Gegenfurtner, K. (2006). Memory modulates color appearance Nature Neuroscience, 9 (11), 1367-1368 DOI: 10.1038/nn1794

Claidière, N., Jraissati, Y., & Chevallier, C. (2008). A Colour Sorting Task Reveals the Limits of the Universalist/Relativist Dichotomy: Colour Categories Can Be Both Language Specific and Perceptual Journal of Cognition and Culture, 8 (3), 211-233 DOI: 10.1163/156853708X358155

Winawer, J., Witthoft, N., Frank, M., Wu, L., Wade, A., & Boroditsky, L. (2007). Russian blues reveal effects of language on color discrimination Proceedings of the National Academy of Sciences, 104 (19), 7780-7785 DOI: 10.1073/pnas.0701644104