Effects of Language on Colour Constancy

EFFECTS OF LANGUAGE ON COLOUR CONSTANCY

The Effects of Colour Term Acquisition on Categorical Colour Constancy

Candidate Number: 79015

University of Sussex

Word count: 5794

Acknowledgements

I would like to thank Dr Anna Franklin and Dr Christoph Witzel for their much valued support through the project. As my project supervisor Dr Franklin provided support and opportunity wherever needed and provided the initial inspiration that drove my enthusiasm for colour. Dr Franklin, Dr Witzel and I collaborated on the design of the experiment and Dr Franklin and Dr Witzel provided feedback following the pilot. Dr Witzel also assisted with some of the data collection,suggestedsomeinitial analyses and produced the figures not possible with the usual programs. I would particularly like to thank Dr Witzel for the weekend he spent learning carpentry to produce the frame used in each of the nurseries to support the window filters.For the large part of data collection Ms Chen accompanied me, as second experimenter, freely giving her time, for which I am also truly grateful. Stimuli, remaining analyses and figures were independently produced.

Enormous thanks to the nurseries, their patient staff, kind parents of the child participants, and of course to all the truly wonderful children who agreed to participate.

Abstract

Colour constancyoccurs whena colour appears the same despite changes in illumination which cause the light reaching the eye to be different. Few have studied the development of colour constancy, in particular the impact that language has on categorical colour perception. This study tested fourteen 36-51 month olds using a sorting task to sort 163 coloured stimuli under four illuminant conditions. The consistency with which these were categorised was analysed, showing that categorisation of colours was related to adult-like category maturity and not to age. When illumination changes were considered, consistency again reflected adult consistency, showing robust categorical colour constancy exists in children of this age group. Results also showed that some colour categories were more stable than others, and prototypical colours were named more consistently than those at the boundary of a colour category. The findings from this study suggest that language may indeed influence categorical colour constancy. It also suggests adult categories emerge from conceptual linguistic categories which may in turn have their origins in the most perceptually stable colours.

The Effects ofColour Term Acquisition onCategorical Colour Constancy

Introduction

Colour constancy describes the way perception of surface colour remains the same, despite alterations in illumination which alter the light reaching the eye. Without this, an object that is blue outdoors in sunlight appears brown under typical indoor lighting, making object identification rather difficult. On a day to day basis, we move from indoor environments to outdoor environments, effortlessly maintaining a constant perception of the colours we see, despite dramatic changes in the colour of illumination (Romero, Hernandez-Andre, Nieves, & Garcia, 2002) which in reality alter the spectrum of light reflected from the surface, which finally reach the eye. The ability to bridge this gap between the message our eye receives and our visual perception of colour is colour constancy.

Remarkably high levels of colour constancy can be found in adults. Most studiessuggest the visual system needs to discount the effects of any illuminants to calculate the reflectance properties and therefore the colour of the surface in question (Foster, 2003; Palmer, 1999). Various models have been proposed to account for colour constancy, employing algorithms to estimate surface reflectance (Land, 1983, 1986), spatial averaging to estimate the illuminant (Hurlbert, 1986), or the use of relative frequencies of colours appearing together as a predictor (Long & Purves, 2003). However, no single theory seems able to account fully for colour constancy (Foster, 2011; Kraft & Brainard, 1999).

Understanding the mechanisms which support colour constancy is addressed as a perceptual question on the whole. In reality, colour for humans does not exist in the physical uniform wavelength spectrum, but as fairly well-defined categories; each given a colour name. The origin of colour names has been the source of much research, with evidence that colour categories are universal to all humans (Berlin & Kay, 1969). There is also evidence of categorical perception of colour in infants as young as 4 months old, broadly reflectingcategories of adults (Franklin & Davies, 2004). This suggests languages are shaped in response to our perceptual representations. The Universal approach sees colour categories as being arranged around focal colours (Berlin & Kay, 1969). Philipona and O’Regan (1996) have suggested that prototypical examples of categories are so because they are the most stable, and that this fact has led to their similar categorisation across languages. Conversely, evidence that language shapes the way we see colour from cross cultural studies led to a far more conceptual explanation(Roberson, Davies & Davidoff, 2000). This view sees shared representationsas resulting in colour categories, with colours built from boundaries negotiated through language. Considerable research has made the arguments less dichotomous, with an answer that lies somewhere between the two.

Categorical colour constancy was assessed using a sorting paradigm with adults across various differently illuminated conditions (Olkkonen, Hansen & Gegenfurtner, 2009; Olkkonen, Witzel, Hansen & Gegenfurtner, 2010). Theyconcluded by linking colour constancy with the ability to consistently name colours. They based their conclusions on findings that only minor boundary changes occurred with illumination changes, and that colour categories changed little between observers, or illuminations. This correlation between naming consistency between individuals and illuminations in particular suggests colour constancy and categories for colour naming are linked.

When exploring the interaction of language and colour, research has turned to developmental studies to address this type of question (e.g. Franklin, Clifford, Williamson & Davies, 2005; Franklin, Wright & Davies, 2009). Few studies to date (Dannemiller, 1989; Dannemiller & Hanko, 1987) have examined the development of colour constancy in infants. These suggest colour constancy is present in early life, with evidence of colour constancy in infants at around 4-5 months of age. The development of categorical colour constancy poses a number of important questions which, if answered, could add considerable understanding to current theory. If categorical colour constancy patterns are unchanged through the lifespan, this suggests perceptual origins, supporting universal theories. If however, these patterns change through the acquisition of language, this would suggest categories are conceptual, and formed by shared representations in language.This raises the question of what impact learning the names of colours has on colour constancy.

Early studies suggested children cannot name colours consistently and reliably until about 4-7 years old (Bornstein, 1985). In fact, more recent work shows children can consistently comprehend and name the first nine main colours at around the age of three (yellow, blue, green, purple, red, orange, black, pink and white), with brown and grey following over the course of the next nine months (Pitchford & Mullen, 2002 & 2005). The acquisition of colour terms is thought to be more difficult than acquiring other words such as object words (Pitchford, 2006) due to its abstract nature (Kowalski and Zimiles, 2006) or an attentional bias (Soja, 1994) toward shape. Interestingly though, this late mastery of colour names stands in stark contrast to the early evidence of colour constancy in infants. This timing difference between early perceptual colour representations,with those emerging later from language, provides an opportunity to explore how they interact.

Bonnardel and Pitchford (2006) compared children aged 26-57 months with adults in a colour sorting study. Stimuli were 100 Munsell[1]chip samples which participants were asked to sort into eight colour categories (red, pink, orange, brown, yellow, green, blue, and purple). Children placed the chips into boxes decorated with a picture of a teddy bear and a Munsell chip as an example of each colour category. Children were grouped according to colour naming ability.

This study enabled comparison of the colour categories and boundaries from adults and children with ‘beginning’, ‘developing’ or ‘accurate’ colour naming abilities. Bonnardel and Pitchford concluded that there was little evidence for language influencing categorization of colours, except with brown, which appeared to require some conceptual understanding before consistent boundaries were found.

The current study aims to determine if adult categories are the result of concepts learned as children, or have their basis in perception. The relationship between categorical colour constancy and acquisition of colour termswill be examined. Perceptual origins will be supported by little change during language acquisition; conceptual origins will be supported by the emergence of adult like patterns. Children with partial colour naming abilities will be asked to sort colours under different illuminations. Five key questions will be used to guide the research.

How Consistently Can Children Categorise Colour?

Children’s categories would be expected to be less consistent than adults and with more variability between illuminant changes, as categories are forming. As children are still learning some of the key terms some colours may be incorrectly named and categorised.

Does Children’sConsistency Change During Language Acquisition?

Theoretically, if categorical colour constancy is conceptually driven and shaped by linguistic colour categories, then this similarity to adult categories would be expected to emerge over time as acquisition of colour terms becomes more stable. If categorical colour constancy has a perceptual basis then it would be logical to expect learning colour words to have little impact on these categories. The distribution of consistencies and the mapping of the categories should be similar, even though categorization is likely to be ‘noisier’ than with adults, resulting in lower consistency.

How Consistent Are Children When Illuminations Change?

The available research suggests children do have colour constancy, but no literature to date suggests how well developed this is. It may not be as robust as that in adults which will be demonstrated by the level of consistency between illumination conditions. High consistency despite changing illuminations will suggest good colour constancy. If consistency between illuminations and between individuals is high this would suggest that colour constancy mechanisms are related to category formation.

Are Some Categories More Stable Than Others?

Individual colour categories may be more or less consistent than others; perhaps some will be more stable under changing light conditions for children. Certainly there is evidence this is the case in adults (Olkkonen et al., 2009; Olkkonen et al., 2010) after controlling for category size. Larger categories would be expected to be more consistent as they have a smaller proportion of hues as borders than as central (or focal) hues.

Are Boundaries Less Consistent Than Prototypes?

The Universalist view would predict that categories are built around some key focal points (Berlin & Kay, 1969; Collier et al., 1976) most likely due to higher saturation. The opposing view would suggest that boundaries are the critical decider, linguistically determined, with prototype locations being influenced by where boundaries are set. Either way, if prototypical chips represent the best example of each category, then it seems likely that they will be located fairly centrally within categories. Categorisation consistency may differ between children and between illuminants.

Method

Participants

Fourteen children (seven boys) aged36-51(M = 41) months, with English as their first language and with no known visual deficiencies were tested. Seven completed all four conditions, two completed three and five completed two or less.

Materials

For the initial colour naming and comprehension task rabbit flashcards were produced(Pitchford & Mullen, 2002) on grey backgrounds and laminated. For colour naming, individual rabbits wearing coloured clothes (focal colours from each of the eight selected colours; red, green, blue, purple, yellow, orange, pink and brown) were used (Appendix A) and for colour comprehension a central rabbit was pictured with eight coloured jumpers in a circle around it(Appendix B) were used.

One hundred and sixty three glossy chromatic Munsell sample chips were used as stimuli. These were always the highest chroma available, across 40 hues covering all hue groups every 2.5 stepswith lightness values of 3, 5, 6 and 8. These are shown in Figure 1.Card shapes (jumper, shorts, or hats) with dimensions of approximately 4cm² were painted grey (N5 Munsell colour) and the glossy Munsell (1966) colour chips attached. The reverses of stimuli were coded for identification. A wooden board was painted in the same grey paint and used to display all stimuli, including eight toy animals.

Figure 1. shows the Munsell chip collection. A further 3 were added to provide some prototypical reds at value 4.

For the filtered light conditions, red and green Lee filters ( were used to cover windows, either using a frame, or stapled directly over the window frame and natural light was used for the daylight conditions.

Ishihara and Tritan plates were used to test for colour deficiency.

Illumination levels were measured using a specular reflectance standard using a Minolta colour meter. These were averaged to provide CIE xyY values shown in table 1.

Table 1. Lee filter specifications with the corresponding average CIE xyY values. Means and standard deviations are taken from across all testing sessions for each illumination
Illuminant / Filter / x mean (SD) / y mean (SD) / Y (cd/m²) mean (SD)
Daylight / - / 0.307 (0.021) / 0.327 (0.02) / 125 (89)
Red / 35 Light pink / 0.345 (0.009) / 0.301 (0.005) / 78 (39)
Green / 138 Pale green / 0.257 (0.123) / 0.353 (0.016) / 65 (28)

All data were recorded using templates (Appendices C,D,E & F)

Design

A pilot was conducted with two children, which highlighted the need for three additional colour chips for ‘red’. Subsequently 163 were used.

Naming and comprehension. The comprehension task was completed by showing children the rabbit flashcards surrounded by different colour clothes, asking them to ‘point to the colour red’ (yellow, green, etc.) and recordingtheir responses. The naming task was completed by showing children flashcard single rabbits and asking them to name the colour of each of the eight rabbits.Their responses were again recorded.

Main task. A repeated measures design was used across a maximum of 4 different illuminant conditions; daylight, red, green and a second daylight. The dependent variable was the consistency with which each chip is re-categorised across varying illuminations and participants. Illumination conditions were separated in most cases by one week and no child completed more than one condition in a day. The first condition was always daylight, but subsequent conditions were counterbalanced across participants.

Each room was cleared to remove as many distractions as possible. The board was positioned to take advantage of the best natural light; ensuring stimuli would be well lit for the child. Animals were placed in plastic boxes along the back of the board (Figure 2.). The arrangement of the animals and ‘favourite colour’ allocated for each condition and participants were counterbalanced. Colour name labels were positioned at the rear for the use of experimenters during testing. Stimuli were laid out with the hue circle running from one side to the other, darkest chips closest to the child. The hue circle was either centred with pink or green, and this was alternated across conditions for each child.

Children were asked to sort the colour chips into categories and then identify a prototype from each category. Once children had completed the task and left, stimuli from each category were turned over and photographed as a category, so that recording and coding of data could take place later.

Colour deficiency tests were administered before the first condition by asking children to trace the coloured line on Ishihara plates and to point to the different corner on Tritan plates. Participant information and data was collected using forms (Appendices 3-6).

Procedure

Materials were arranged as described above and light measurements taken and recorded. Children for whom consent had been obtained were approached and asked if they would like to come and play. Assenting children were first presented with the comprehension task, followed by the naming task and colour deficiency tests.

Main colour task. Children were asked if they would like to play a game with the animals. The experimenter explained that each animal would like to collect a particular colour and children were asked to sort the stimuli into colour categories. On completion, the experimenter asked the child to select the best example from each category for the animal. Children were thanked and given a picture to colour in. Light levels were retested and recorded once the child had left. All testing sessions were conducted between the hours of 9.30 and 3.30 to ensure that lighting conditions were as consistent and light as possible.

The procedure for the main task was repeated for the four illuminations where children agreed. On the final session, the naming and comprehension test was repeated.

Ethical considerations.Each nursery was visited to provide details of the study (Appendix G), and show the stimuli, flashcards and toys. Parents were provided with an information (Appendix H) and consent letter (Appendix I) regarding the experiment and requesting consent for their children’s participation. Children whose parents’ had consented to their participation were approached in the nursery setting and asked if they would like to participate in some games. Children’s assent was sought separately for each of the conditions. Children were thanked after each session and informed that they would be asked again next week. All participant data was anonymised. Ethical approval was obtained from the University of Sussex Ethics board and British Psychological Society guidelines were complied with at all times. A satisfactory CRB check was obtained for experimenters.