The Curious Case of Color Words:
Considering Adjectival Placement in Discrimination Learning
We present a formal analysis of word learning in terms of cue competition, and consider two possible ways in which words might be learned: by learning to predict a label from the features of objects in the world, and by learning to predict features from a label. This analysis predicts differences in learning depending on the sequencing of objects and labels.
Discrimination learning is facilitated when objects predict labels, but not when labels predict objects.
We have previously documented a computational simulation and adult experiments that confirm these differences, revealing the existence of a significant ordering effect in learning.
Here we present an expanded theoretical account of this effect and examine how it relates to children’s notoriously labored acquisition of color words. Although 4-month olds can
perceptually distinguish basic color categories (Bornstein, Kessen, & Weiskopf, 1976), young children struggle to learn to map the appropriate label to a given hue. Indeed, younger sighted children’s use of color words is much like that of blind children (Landau & Gleitman, 1985).
Three-year olds who correctly identify a blue object in one situation may confuse “blue” with
“red” in another (Sandhofer & Smith, 1999), and even at age four, some children still struggle to discriminate color words appropriately despite hundreds of explicit training trials (Rice, 1980).
Consider, however, that in most contexts in which children hear color words, they also see a wide array of colors present in the surrounding environment. Without some way of reducing the available perceptual cues, they will encounter very few situations that serve as optimal contexts for learning to discriminate between the various hues that might be associated with different color words. Color learning may be facilitated, however, by using language to narrow the child’s focus from the environment as a whole to a specific object, thereby reducing the number of conflicting perceptual cues. This is so, first, because objects are much more likely than colors to systematically co-vary with their labels, and second, because once a child has learned the name of a given object, she can more readily learn to pick out the individual
properties of that object. For these reasons, postnominal placement of color words appears to be optimal for discrimination learning.
Why is learning English color words so difficult then? Perhaps it is because in English, color words occur postnominally only about 30% of the time in child-directed speech (Thorpe &
Fernald, 2006). When color words occur prenominally, the child’s attention has not yet been narrowed to focus on the object when the color word is heard. To test this hypothesis, we taught children color words in two conditions: postnominal placement and prenominal placement. Our results strongly suggest that learning to match a color word to a specific color is affected by the sequencing of words and the predictive relationships that sequencing builds between words and objects or words and other words. In postnominal (object to color) conditions, the narrowing of the child’s focus to the named object facilitated cue competition and learning, while in
prenominal (color to object) conditions, the lack of focus and ubiquity of perceptual cues made learning far more difficult.
500 Words
Figures and Tables
Figure 1. When features predict their labels, the non-discriminating features will be dissociated from the labels through cue competition.
Figure 2. The absence of cue competition when labels predict features will result in the conditional probability of a feature given a label being learned. In this situation, the outcome of learning will simply be a representation of the probability of the features given the labels.
Figure 3. A depiction of color-word learning in a natural environment. In word-learning, prediction error enables the learner to detect systematic covariance between features and labels. However, if color words are heard in contexts where most colors are available as cues, there will be little systematic covariance to discover. In this example, “red”
is heard at time (i) and “blue” is heard at time (ii). In this context, a child will learn to associate both red and blue (and all of the colors present) with the labels “red” and “blue” indiscriminately. This may be why children more easily learn labels for objects than textures and colors (see Bloom, 2000).
Training Type Question Type Pre-Test Mean
Post-Test Mean
Consistency Mean
Consistency vs. chance (33%)
FL-trained Post-Nominal 59% 71% 55% t(51)=3.065
p<0.005
FL-trained Pre-Nominal 57% 55% 37% t(51)=0.574
p>0.5
LF-trained Post-Nominal 59% 53% 39% t(51)=0.852
p>0.4
LF-trained Pre-Nominal 55% 49% 37% t(51)=0.574
p>0.5
Table 1. Children’s performance on the three alternate forced choice color matching task, broken down by testing and training type.
Prenominal Postnominal
# of tokens M (SD) M (SD)
Big 821 .73 (.06) .27 (.06)
Blue 138 .71 (.10) .29 (.10)
Red 207 .67 (.16) .33 (.16)
Pretty 96 .53 (.22) .47 (.22)
Good 740 .52 (.10) .48 (.10)
Nice 423 .51 (.14) .49 (.14)
Dirty 10 .42 (.12) .58 (.12)
Hot 147 .18 (.12) .82 (.12)
Cold 109 .14 (.12) .86 (.12)
Wet 96 .14 (.24) .86 (.24)
Broken 180 .11 (.11) .89 (.11)
Table 2. The distribution of the pre- and post-nominal forms of 11 common adjectives in an analysis of two corpora from the CHILDES database. While children generally master adjectives with post-nominal bias easily, they struggle with pre-nominally biased adjectives like big, red and blue (Clark, 2003; Sandhofer & Smith, 1999).
References
Bloom, P. (2000). How children learn the meanings of words. Cambridge, MA: MIT Press.
Bornstein, M. H., Kessen, W., & Weiskopf, S. (1976). Color Vision and Hue Categorization in Young Human Infants, Journal of Experimental Psychology, 2, 115-19.
Clark, E. V. (1993). The lexicon in acquisition. Cambridge: Cambridge University Pres Darwin, C. (1877). Biographiche skizze eines kleinen kindes. Kosmos, 367 –376. Translation
from Bornstein, M.H., (1985). On the development of color naming in young children: Data and theory. Brain and Language, 26, 72-93.
Landau, B., & Gleitman, L. R. (1985). Language and experience: Evidence from the blind child.
Cambridge, MA: Harvard University Press.
Rice, N. (1980). Cognition to language. Baltimore, MD: University Park Press.
Sandhofer, C. M., & Smith, L. B. (1999). Learning color words involves learning a system of mappings. Developmental Psychology, 35, 668-679.
Thorpe, K., & Fernald, A. (2006). Knowing what a novel word is not: Two-year-olds "listen through" ambiguous adjectives in fluent speech. Cognition, 100, 389-433.
