argo industri

tol binjai medan

NSM and the meaning of color words

PAUL KAY

Uwe Durst (D) is to be commended for a clear exposition of the Natural Semantic Metalanguage (NSM) theory of Anna Wierzbicka (W) and her associates. In addition to laying out clearly the assumptions of the NSM theory, D states that

The .. .NSM model... has turned out to be a most useful theoretical and methodological framework for semantic analysis in various linguistic... domains.

In defending this claim, D discusses a substantial array of studies of particular languages and cross-language studies of a variety of semantic domains. This commentary would not be the place, nor I the appropriate commentator, to evaluate the degree to which the full range of studies of particular phenomena performed by NSM researchers substantiate the claim that NSM is a useful theoretical and methodological framework for semantic analysis. Consequently, as far as factual matters are concerned, I will largely confine my attention to NSM work on a lexical domain with which I have prior experience, color terms.   As I understand D, the key notions ofthe NSM approach are, or at least include, the following.

1. Natural language meanings must be represented in natural language.    The familiar trappings of semantic analysis: features, mathematical    concepts, logical languages, models of worlds or situations, etc. are    out of place in proper semantic analysis. 2. There exists a universal, finite and relatively small set of semantic    primitives. 3. Each of these primitives is expressed in every natural language by    a single morpheme, a single word or a single fixed phrase, modulo    polysemy.

Theoretical Linguistics 29 (2003), 237-245

0301^(428/03/0029-0237    © Walter de Gruyter

238 Paul Kay

4. The NSM notion of a semantic analysis or definition can be character-    ized in terms of (1) the [semantic] primitives, (2) [semantic] molecules    and (3) [natural language] paraphrases. If a word can be accurately    paraphrased by an expression containing only primitives, that word    is a molecule. A word is a [semantic] molecule iff it can be accurately    paraphrased by a natural language expression containing only pri-    mitives or molecules. (Recursive application of this definition can    produce increasingly complex molecules.) An accurate paraphrase of    a target word is a semantic analysis or definition of that word iff it    contains only primitives and molecules. 5. The primary goal of NSM semantics is to provide analyses, as just    defined, for as many words in as many semantic domains in as many    languages as possible. (There are additional uses to which NSM    semantics has been put and which are discussed by D; these will not be    considered here.)

From the point of view of the common garden-variety semanticist, the most notable thing about this semantic approach is that it appears to take no interest in analyzing the meanings of sentences. Something like natural language sentences are employed as analyses of words, but sentences are never taken as the obfects of analysis. Thus, the goals of this theory are in complementary distribution with those of the compositional semantics tradition, going back through Montague and Frege at least to Aristotle. In this respect, the NSM tradition is more closely related to the ethno- graphic semantics movement in anthropology than to most linguistic semantics and in fact anthropologists and anthropological linguists are prominent among the practitioners of NSM semantics. This observation is by no language sentences for granted. What most semanti- cists consider the central problem of their discipline - how the meanings of sentences are put together from the meanings of the words that comprise them - is assumed in the NSM approach to be a solved problem, or perhaps simply not a problem. means a criticism. Every scholar has the right to study his or her chosen part of the elephant. Epluribus unum.    A related characteristic of the NSM approach does invite criticism. In using natural language sentences to provide analyses (express the meanings) of individual words, the NSM method takes the combinatorial semantics of natural Take for example W's (1990: 127) analysis of the English word yellow.

NSM and the meaning of color words 239

X is yellow

when one sees things like X one can think of the sun at some times people can see everything when one sees things like X one can think of times of this kind

Let us assume for purposes of argument that W, D and every English speaker are perfectly agreed on the meanings of all the individual words: when, one, sees, things, like, and so on. According to NSM theory, the meanings of these words are "known to everybody, including children" (D: 000)'. Let us grant that. W and D believe that the three sentences following Xis yellow express in clear and simple terms the meaning of "X is yellow" that yellow and the three-sentence analysis are "ways of saying the same thing in other words" (W 1996: 107, quoted by D 000). There is at least one potential English-speaking informant who disagrees with this empiri- cal claim (the writer of this commentary), and I suspect others abound. But my (and probably others') disagreement with W and D about the accuracy of the offered paraphrase can never be reconciled if we have no common theory of how the meanings ofthe three sentences are constructed from the meanings of the individual words. Assuming we agree on the meanings of the individual words, if we disagree on the meanings ofthe sentences, then at least one of us has failed to correctly compose the meanings of the sentences. But unless we have some rational account of how the meanings of sentences are composed from the meanings of their constituent words, it's just one person's intuition against the other's. This is not a desirable scientific outcome.

D appears to be aware of this problem. But his reaction to that realiza- tion comes across as more a statement of faith than a solution. In his discussion of Combinability (section 2.4), D writes, "What we need is not only a metalexicon but a whole metalanguage. The languages of the world differ enormously in their syntax. .. But what is at issue is not a set of abstract syntactic rules, but universal patterns of combinability of the postulated primes, their semantic valency, so to speak. While the vari- ous versions of NSM are not as different in their surface structures as the languages they are derived from, they are nonetheless substantially

Actually, by NSM theory, some of these words might be semantic molecules rather than primitives, but the point is unchanged.

240 Paul Kay

different from one another. The combinatorial properties of NSM can be seen as a subset ofthe syntax of a particular language, in the same way as the NSM primitives is [sic] a subset of its lexicon" (D 000 italics added). This passage appears to express the faith that some pared-down version of the syntax of each natural language will correspond to a universal metalanguage of semantic composition. Each expression in one of these reduced codes is immediately understood by its speakers and directly translatable, item by item, into a corresponding expression in each other's reduced code. I am aware of no reason to believe this and, unless I have missed something, none is offered.    NSM analysis "rejects the use of any formal language" (D 000). "The meaning of a word is roughly what people 'mean' or 'have in mind' when they use it" (W 1990: 107). Consider the NSM analysis of green (W 1990: 117):

X is green

in some places things grow out of the ground when one sees things like X one can think of things of this kind

It does not seem to me that when I say that a particular car, say, is green what I have in mind is that when I see things similar to this car I can think about things that grow out of the ground. It seems to me that thoughts of things growing out ofthe ground are not usually in my mind on such occa- sions and the thought that I could be having thoughts about things grow- ing out ofthe ground is, if possible, even further from my mind. Of course the NSM theorist can contend that when I say that something is green the thought that I could be thinking of growing things is exactly what I think unconsciously, and that my inability to bring this unconscious thought to consciousness is not a problem for the theory (W 1990: 107). Admittedly, no one can prove that my unconscious thoughts are not what an NSM analysis says they are. On the other hand, the NSM theory has given us nothing beyond assertion in support of the claim that the content of each of its analyses are the contents of the thoughts of speakers when they employ the word analyzed.    The NSM analysis of green also illustrates a persistent obscurity of NSM analyses, which arises from the NSM avoidance of standard

NSM and the meaning of color words 241

notations. Although referential indices are among the alleged abuses of garden-variety semantics that NSM takes pride in avoiding (D: 000), in this case, as in some others, NSM analyses employ an artificial substitute for a standard convention, which the reader has to figure out is intended to convey what the standard convention conveys. In the analysis of green, coreference is (I assume) expressed by the two tokens of X. Another idio- syncratic convention is the translation of embedded parentheses into a system of graded indentation, which is harder to read for the specialist reader and, I suspect, for the non-specialist as well (Cf. W's 1997; 141 analysis of the Russian svoboda , discussed by D p. 000).^ A third piece of gratuitous notation is the slash symbol appearing, for example, in "good/ bad." Presumably this symbol signals some form of disjunction, but its logical nature (inclusive? exclusive? something else?) and the intended scope ofeach token are left for the reader to intuit. It is hard to believe that whatever this slash is intended to mean is "known to everybody, including children" who speak any natural language. In the opinion of this conmien- tator, the idiosyncratic conventions of punctuation and special symbols adopted by the NSM tradition add nothing in the way of understandability for the non-specialist and introduce significant confusion for the specialist.    Turning to W's discussion of color terms, we have already considered two analyses, those for yellow and green. One more example is

X is white

in some places, people can see everything when one sees things like X one can think of places of this kind (W 1990: 116)

All of W's analyses of color words use the formula "when one sees things like X one can think of . ..". W (1990) argues that Kay and McDaniel's (1978) characterization of the meanings of color words as patterns of neural response demonstrates a failure on their part to understand the nature of linguistic meaning.' "Language," according to W, "reflects

How the indentation convention is translated into languages using a radically different writing system (e.g. Chinese), or no writing system at all, is not discussed. The fact that Kay and McDaniel's specific physiological characterization of basic color categories was almost certainly wrong (Derrington et al. 1984, De Valois and De Valois 1993, Abramov and Gordon 1994, Abramov 1997, Kay and Maffi 1999) is irrelevant. W was objecting to any physiological characterization ofthe meanings of color words.

242 Paul Kay

conceptualizations, not 'the neural representation of color...'..." In a move apparently taken to avoid giving color concepts a physiological analysis, W defines color words in terms of non-color experiences, such as day, night, the sun, fire, foliage, the sky, etc. And in order to avoid claim- ing bluntly that, e.g., yellow means 'the color ofthe sun', W intentionally introduces vagueness with the formula "X is [e.g., yellow\, when one see things like X one can think of.. .[e.g., the sun]" (W 1990: 121). But this seemingly extreme move, including the deliberate introduction of vague- ness into one's definitions, was not actually necessary to treat color words within the NSM system. One could simply assume that the phenomenally basic colors are NSM primitives, that, e.g, white means 'white'. (Then, any further analysis of the primitive concept 'white' would have to be con- ducted at some lower-than-conceptual level, say the physiological level.) All the NSM analyst needs to do is take the phenomenally basic colors: black, white, red, yellow, green and blue as primitives and define other color words, such as words for pink or light blue (Russian golubof, Polish neibiski), or light red (but darker than pink = Hungarian/>/ro.s) in terms of these. It is notable in this regard, that while W systematically employs the names of reference obfects in her definitions, in her explanations of those definitions she often uses visual terms like light and dark. ".. .niebieski refers only to light and medium dark blues, not to very dark blues (which in English would still be called blue). To account for this, I would assign to blue, but not to niebieski, an additional reference point: naturally occur- ring 'water places', such as sea, lakes or rivers" (W 1990: 120). At other times W encodes lightness or darkness (whiteness or blackness) in her definitions in a more nearly direct fashion. For example, included in W's analysis of Russian golubof 'light blue' is the clause "at some times people can see everything; when one sees things like X, one can think of times of this kind." This is W's definition of white with times substituted

for places.

It is true that occasionally empirical research on cross-language color naming does turn up a basic color term that fairly clearly corresponds to the color of an object or substance and is seemingly represented conceptu- ally in that way. The only totally convincing example of this know to me is the Warlpiri word walyawalya.

Walyawalya is an interesting term of heterogenous color denotation. It is classified as basic because of its use by [all] twenty-three speakers, the fact that it appears at

NSM and the meaning of color words

243

the 61% agreement level, its accounting for nineteen chips on the 100% naming aggregate and its term map consensus of 83% (in brown). The unreduplicated sub- stantive form walya refers to earth, soil, ground. The colors covered by walyawalya range over deep browns, reddish browns, lighter - yellowish - browns and oranges, yellowish sahnons, pinkish purples and other light purples. This turns out to be just about the range of colors displayed by the soil/dirt of central Australia. Interest- ingly, these colors form an unbroken band in the Munsell (or other) color space, perhaps accounting in part for why a substance displaying such a wide range of colors furnishes a probable basic color term (Kay, Berlin, Maffi and Merrifield forthcoming).

Compare this with W's characterization of the water reference in the meaning of English blue that is absent from Polish niebieski.

in some places one can see water not because people did something in these places when one sees things like X one can think of these places

The factual basis of the association of Warlpiri walyawalya and earth is manifest. The association of English blue with the water of sea, lakes and rivers is tenuous at best. The two cases are sufficiently different that confusing them should be avoided. Basic color categories based on the colors of objects, substances or other forms of non-color experience are extremely rare. For example, unless we take it as a matter of a priori prin- ciple that red, green, yellow and blue must be based on associations with some other visual experiences, one does not find any actual evidence in the writings of W for the claim that their meanings are based on associations to fire and/or blood, foliage, the sun, and the sky and/or water, respec- tively. W bases her argument on the observation - or at least the empirical claim, not supported with experimental evidence - that across languages not all terms we might gloss as, say, 'light blue' have exactly the same extension. To the extent it is factually accurate, this observation supports a legitimate argument against the particular universal model posited by Kay and McDaniel (1978), but it does not support the claim that color term meanings are invariably based on associations with non-color prototypes. Assuming that the meanings of color words must be based on some non- color prototypes W makes a plausible case for her particular choices, but the argument rests entirely on the supposition that red, green, yellow and

244

Paul Kay

blue can not simply mean 'red', 'green', 'yellow' and 'blue' at the perceptual (or 'conceptual') level.    Interestingly, W's ideas regarding changes in the non-color prototypes of color concepts over time undercut her assumption that white can't mean 'white', and so on. She writes:

It seems to be a universal feature of language that color perceptions are described, at some stage, in terms of locally salient referents, such as certain characteristically-looking minerals, animals or plants...

But words of this kind, like any words, are subject to semantic change. For example, the fact that the Russian word goluboj 'light blue' is etymologically related to the word for pigeon, or that the PoUsh word czerwony 'red' is etymolo- gically related to the name of a particular red worm, does not mean that the associations in questions [sic] are synchronically alive. They are definitely not: in present-day Russian, goluboj is clearly associated with the color of the sky, not the color of pigeons (W 1990:139).

W fails to explain bow Russian speakers can be capable of entertaining a concept 'light blue' that they at first associate with pigeons but later disso- ciate from pigeons (and re-associate with the sky) and yet are incapable of entertaining a concept 'light blue' simpliciter. Otherwise put, W gives no reason why the conceptual object, call it LIGHT BLUE, which can be switched over from a pigeon association to a sky association can't be enter- tained on its own, that is, no reason why the meaning of golubof couldn't be simply 'light blue', the meaning of white simply 'white', and so on. In addition to lacking empirical motivation, the claim that the meanings of color words are invariably based on non-color prototypes introduces an unnecessary theoretical complication.    Despite the disagreements I have expressed with several general prac- tices and certain specific analyses of the NSM approach, I must conclude by acknowledging that useful and interesting empirical semantic work has been done within this framework. To cite just a single example, I find intriguing W's speculation that the frequent grouping in languages with few color terms of the long wavelength ("warm") colors, red, orange, yellow, pink, etc., with white may be related to the fact that the sun and fire provide low technology societies almost the unique sources of both light and heat.

University of California at Berkeley

NSM and the meaning of color words

References

245

Abramov, L (1997) Physiological mechanisms of color vision. In Color Categories in Thought    and Language, C.L. Hardin and L. Maffi (eds.) Cambridge: Cambridge University Press. Abramov, I. and J. Gordon (1994) Color appearance: on seeing red - or yellow, or green, or    blue. Annual Review of Psychology 45,451-485. Derrington, A.M., J. Krauskopf and P. Lennie (1984) Chromatic mechanisms in lateral    geniculate nucleus of macaque. Journat of Physiology 357,241-265. De Valois, R. L. and K. K. De Valois (1993) A multi-stage color model. Vision Research 33,    1053-1065. Kay, P. and L. Maffi (1999) Color appearance and the emergence and evolution of basic color    lexicons. American Anthropologist 101, 743-760. Kay, P. and C.K. McDaniel (1978) The linguistic significance ofthe meanings of basic color    terms. Language 54:61-646. Kay, P., B. Berlin, L. MafTi, and W. Merrifield. (forthcoming) The World Color Survey,    Stanford, California: CSLI Publications. Wierzbicka, A. (1990) The meaning of color terms: semantics, culture and cognition.    Cognitive Linguistics 1,99-150. Wierzbicka, A. (1996) Semantics: Primes and Universals, Oxford/New York. Oxford    University Press.