Tuesday, May 31, 2011

The World Color Survey

2009. By Paul Kay, Brent Berlin, Luisa Maffi, William Merrifield, and Richard Cook. CSLI Publications, Stanford, California.

Contains a brief history of the development of the B-K hypothesis since 1969, including substantial modifications of the theory in response to criticisms and further data. There follows a summary of the massive dataset of the WCS and a general conclusion, along with discussion of so-called "Emergent Hypothesis" languages that do not encode hue. (Ancient Greek is claimed to be such a language by Lyons.)

Some highlight quotations:

[2] "BK operated on a tacit assumption, which was retained in the UE models throughout the 70s, 80s and 90s, that every language has a basic color term system, in the sense that every language has a small set of words, or word senses, of pure color meaning whose significata partition the subjective color space. Let us call this the assumption [3] of Partition. (Recent UE abandonment of Partition as exceptionless is discussed briefly below and in greater detail in Section 5. Exceptions to Partition are real but discussion of them can profitably be deferred for the moment.) According to Partition, a language with only two terms, shown as 'black' and 'white' in diagram (1), 'extends' these two terms so that jointly their significata cover all visible colors. BK denoted such extended colors in capital letters, BLACK and WHITE, stating that BLACK comprises 'black plus most dark hues' and WHITE comprises 'white plus most light hues' (BK: 17). The first major amendment to the UE model was taking note, based mainly on the work of E. Rosch (Heider 1972a, 1972b, Heider and Olivier 1972), that two-term systems in fact contain one term for black, green, and blue and other so-called cool colors and one term for white, red, yellow and other 'warm' colors (Berlin and Berlin 1975, Kay 1975)."

[7] "As noted, contrary to the BK model shown in (1), derived (intersective) categories can occur before all the composites have been resolved into their constituent primaries. MacLaury (1986) suggested that the composite-to-primary evolutionary path and the less regular evolutionary path of the derived categories be considered independent, [8] albeit largely non-overlapping, sequences rather than successive portions of a single sequence. This suggestion is adopted in Kay, Berlin, Maffi and Merrifield (1997, hereafter KBMM), who consequently drop stages VI and vn and restrict the typology of basic color term systems to what they term 'basic stages', namely those five evolutionary stages which represent successive partitions of the original two maximal composites, W /R/Y and Bk/G/Bu12, ending with separate basic color terms for each of the six primaries. There are thus five basic stages, representing two-, three-, four-, five- and six-term systems. Two stages, m (four-term systems) and IV (five-term systems) admit of more than one type of basic color lexicon. KBMM noted that languages were quite often better classified as intermediate between two types than as belonging to a single type. Also, color term systems can sometimes best be described as recently entering a certain stage or as about to enter transition into another stage. The actual typology for basic stages and the resulting evolutionary model of KBMM need not be discussed in further detail here, because the more recent typology of Kay and Maffi (1999), described below, is nearly identical and is the one employed as a descriptive framework in this work."

[8] "Kay and Maffi (1999) present a slightly revised version of the UE model of KBMM
which, however, emphasizes the degree to which universal aspects of the evolution
of basic color term systems can be motivated by presumably language-independent
facts regarding color appearance.13 The UE hypotheses of universality and evolution
are expressed there as follows:
I There exists a small set of perceptual landmarks (that we can now identify with the Hering primary colors: black, white, red, yellow, green, blue) which individually or in combination form the basis of the denotation of most of the major
color terms of most of the languages of world.
II Languages are frequently observed to gain basic color terms in a partially fixed order. Languages are infrequently or never observed to loose [sic] basic color terms."

[9] "• First, what proportion of the world's languages are non-partition languages, that is, fail to have lexical sets of simple, salient words whose significata do partition the perceptual color space?
• Secondly, in the case of partition languages, to what extent and in what manner do they conform to generalizations I and II above?
• Thirdly, in the case of non-partition languages, to what extent and in what manner do they correspond to generalizations I and II?
Regarding the first question, it appears that in the ethnographic present non-partition languages are rare. The data from most languages studied in the WCS give no indication of non-partition status. (The exceptions are discussed in Section 5.) Also, most reports on color term systems in the literature a11d in personal communications received by the authors give no suggestion that the language being reported fails to provide a simple lexical partition of the color space. The EH enthusiast might reply that such reports merely betray an unreflecting assumption, based on the report~r' s own language, that every language partitions the color space with a simple lexical set. Such a conjecture is neither provable nor disprovable. In m1y case, the apparent paucity of non-partition languages in the ethnographic present may not be representative of human history. Specifically, just as there are no Stage I (two-term) languages in the WCS sample and very few reported in the literature17 , the relative lack of non-partition languages in the ethnographic present may reflect to an unknown degree the (putative) facts that (1) some extant partition languages were non-partition languages in the past and (2) some extinct non-partition lm1guages may have left no non-partitioning descendants, or no descendants at all. Again, it is not obvious how empirical evidence may be brought to bear on such conjectures. Philological reconstructions of data on extinct languages (e.g., Lyons 1995, 1997 on Ancient Greek) and some exegetical reanalyses of reports that were originally aimed at different goals (e.g., Lyons 1997 on Hm1un6o, Lucy 1996, 1997 on Hantm6o a11d Zuni, Wierzbicka, 1996: 306-308 on Hm1un6o) were suggestive of the EH, if not probative. The field study of Levinson (2000) was aimed directly at EH issues, and that study, together with some of the WCS data to be discussed in Section 5, lend substantial support to that hypothesis. (For further discussion, see Kay 1999b [10] The second question - "How do color-space-partitioning languages satisfy I and IT?" -is addressed by the individual analyses of the 106 partitioning languages (out of 110 total WCS languages) that constitute the bulk of Section 5 of this book."

[10] Evolution of color systems acc. to the revised BK hypothesis:

Methodology:

[13] "In the WCS procedure, no preliminary interview was administered to establish a set of basic color terms, and in the naming task the 330 individual color stimuli were shown to each cooperating speaker, one by one, according to a fixed random order, and a name [14] elicited for each. Field workers were instructed to urge subjects to respond with short names (although, depending on particular field circumstances and particularities of the language itself or the local culture, there was considerable variation in the degree to which the field investigators were able to satisfy these desiderata). Identification of the basic color terms for the purpose of eliciting the best examples was therefore decided by the fieldworker after - rather than before - the naming task, using criteria specified in the inshuctions (See Appendix 1). The nature of these criteria required a certain amount of judgment from the fieldworker. In later analysis, we substituted our own judgments of basicness for those of the fieldworkers, taking account both of the fieldworkers' original judgments and of certain statistical summary information regarding the responses of the speakers, which was not available to the workers in the field and is discussed below. The WCS mapping task only required informants to map focal points -best example choices -not boundaries. The boundaries of categories emerging from the naming task were not available to the fieldworkers because the naming data took the form of responses to a random order of individual chip presentations, with no key between this order and the stimulus palette. The ranges of the various named categories in each language were only established at a later, post-field, stage, when responses to the chips named in the random presentation order could be assigned to locations on the stimulus palette.
The best example (focus) responses were elicited in the same way in both studies. Once a set of basic color terms was isolated, the collaborating speaker was presented with the full stimulus palette (in WCS, an improved version, devised by Collier et al. (1976), of the original Munsell chip board) and asked, for each term, to indicate the chip or chips that represented the best example(s) of that term. Unlike the BK data, WCS focal responses were highly variable and in the case of some languages unreliable on their face. For example, best example choices that fall outside the range of naming responses of the term they are supposed to typify are impossible to interpret. Not infrequently terms were provided with focal choices that did not appear in the naming data, and some terms that had high consensus naming mysteriously did not appear in the focus data. Possibly these cases represent refusals of the collaborating speakers to make the requested judgment, but they might also represent oversights on the part of the fieldworker, or other causes. Focal choices were unusable for a significant subset of native speakers. We cannot say with assurance why this occurred, but fatigue on the part of the speaker, the interviewer or both, following the naming of 330 individual chips should not be ruled out as a possible factor. Although poor quality of the focus data does not apply to every language, it does apply to enough languages and to a sufficient degree that focus data could not be used systematically in analysis of the individual languages. Aggregated over the WCS as a whole the focal choices show clear patterning (MacLaury 1997, Regier and Kay 2004, Regier, Kay and Cook 2005). For several languages, however, they are not usable for the purpose of analyzing the language. As a consequence, none of the individual language analyses of the next section rely on focal choice data."

Deciding whether or not a color term is basic and assigning a basic stage:

[21]"The 'basic stage' of a language is assigned according to the typology of Figure 1 of Section 1, after a decision is made regarding which terms of the language are basic color terms. Derived categories and heterogeneous or theoretically anomalous categories also appearing to be basic are identified as such. The judgment of whether a terms is basic or not takes into account, in addition to the field worker's decision, an intuitive weighting of four criteria (1) the percentage of speakers using the term, relative to those of other terms, (2) the range of chips named by the term on the modal array, relative to the ranges of other terms, (3) the level of agreement at which the term first appears on the naming aggregates: the higher the level the more established the term is inferred to be, again relative to the same criterion assessed with respect to the other terms of the language, and (4) the relative clarity of definition of the term map. Because different languages display a wide range of variability in the degree of consensus they show overall, the 'strength' of a given term on any of these criteria can only be assessed relative to other terms in the same language. The complexity of this situation and the difficulty of specifying to objective criteria for many of the componenet judgments have prevented the elaboration of an algorithmic procedure to decide which terms are basic and which are not." [NB: these criteria are quite different from those of the original BK thesis, but note that "abstractness" is not one of them.]

[24] "The amount of information carried by the colors of objects may affect the salience of the color domain. In a technologically simple society, color is a more predictable property of things than in a technologically complex one. Except perhaps for a few pairs of closely related species of birds or of fish, it is rare that naturally occurring objects or the artifacts of technologically simple societies are distinguishable only by color. In technologically complex societies, on the other hand, artifacts are frequently to be told apart only by color. The limiting case is perhaps color-coding, as used in signal lights, electric wires and other color-based semiotic media. But almost every type of material thing a member of a technologically complex society encounters in daily life: clothing, books, cars, houses, ... presents the possibility that two tokens of this type will be distinguishable only, or most easily, by their colors. As the colors of artifacts become increasingly subject to deliberate manipulation, color becomes an increasingly important dimension for distinguishing things and hence for distinguishing them in discourse. As technology develops, the increased importance of color as a distinguishing property of objects appears to be an important factor in causing languages to add basic color terms, i.e., to refine the lexical partition of the color domain (Casson 1997)."

[27] "A distinction between 'warm' and 'cool' colors has long been recognized by color specialists from both the arts (e.g., art critics and historians and teachers of painting) and the sciences. Red, yellow and intermediate orange are 'warm'; green and blue are 'cool.' Hardin (1988: 129££) provides an excellent discussion of both experimental and philosophical considerations of the warm/ cool distinctions, beginning with Hume and concluding, in part, 'These explanations [of the warm I cool hue associations and cross-modal associations] are of varying degrees of persuasiveness, but they should at least caution us not to put too much weight on any single analogical formulation. However, they should not blind us to the striking fact that there is a remarkable clustering of oppositions which correlate with this hue division' (Hardin 1988: 129). Early experiments (e.g., Newhall 1941) established red as a warm hue. More recent experiments (Katra and Wooten 1995), controlled for brightness and saturation, have shown that English-speaking subjects' judgments of warm color peak in the orange region and cover reds and yellows, while judgments of cool color peak in the blue region and cover non-yellowish greens and blues. Judgments of warmth/ coolness also correlate with saturation (saturated colors are judged warm), but not significantly with lightness. These groupings of basic hue sensations into warm and cool agree with those common in the art world. A recent study of color term acquisition in two-yearolds, besides finding surprising control of color terms in very young children, found no significant differences among colors in the age at which they were acquired but did find that 'there was some evidence that our subjects maintained the warm-cool boundary; in general they make more within- than across-boundary errors' (Shatz et al. 1996: 197). Both artistic tradition and recent experimental evidence thus point to an affinity between red and yellow on the one hand and between green and blue on the other. A recent color model based on observed cone frequencies (De Valois and De Valois 1993, 1996) posits an intermediate stage of chromatic information processing that consists of two channels: one red/yellow and one green/blue (See Kay and Berlin 1997 for discussion of the possible relevance of this model to cross-language color naming). Psychological color space, so-called, is notoriously lacking in a reliable long-distance metric31
• We take the facts mentioned in this paragraph to indicate, albeit indirectly,
that red and yellow are experienced as in some respect similar and that green and blue are experienced as similar in that same respect."

[31] "A single trajectory, which we call the main line of color term evolution, accounts for the vast majority of WCS languages. Ninety-one of the 110 WCS languages (83%) belong either to one of the five stages of Trajectory A or to a transition between two of these stages, as shown in Figure 2, where a boxed numeral represents the number of WCS languages found at the corresponding type, and a circled numeral represents the number of WCS languages found in transition between the types indicated."