Language And Thought Research Paper

‘Twas brillig, and the slithy toves

Did gyre and gimble in the wabe.

All mimsy were the borogoves,

And the mome raths outgrabe.

And in French:

Il brilgue: les toves lubricilleux

Se gyrent en vrilland dans le guave.

Enmımes sont les gougesbosqueux

Et le momerade horsgrave.

Finally in German:

Es brillig war. Die schlichten Toven

Wirrten und wimmelten in Waben.

Und aller-mumsige Burggoven

Die mohmen Rath’ ausgraben.

The poem means nothing, other than the images it evokes. In a Pulitzer Prize winning series of essays on thought, Douglas Hofstader (1979, p. 372) maintained that each translation will evoke different thoughts in the user, because terms like the (according to Hofstader) earthy ‘slithy’ (slime, slither, etc.) evoke different associates than the associates evoked by the intellectual-sounding French lubricilleux. To the English ear, Die schlichten To en sounds ominous. Is language controlling our thought? More particularly, do the forms of the approximately one thousand natural languages spoken today exert any influence upon the thoughts of their speakers?

Leonard Bloomfield (1933), a prominent linguist during the 1920s and 1930s, said that thinking was ‘talking to oneself.’ One way to interpret this statement is to say that thinking is an internal communication that takes place in the same language that external communication does. If so, the denotative properties of external language should influence internal thought: if you can’t say it, you can’t think about it. The anthropologist Edward Sapir (1921) took a stronger position. He maintained that people could only perceive those categories of the external world that are coded in their language, i.e., that if you cannot say it you cannot see it! Sapir’s student, Benjamin Lee Whorf, adopted a rather more flexible view. For brevity, I shall refer to the position that language exerts a controlling influence on thought as the Whorfian hypothesis. (Some authors maintain that the term Sapir–Whorf hypothesis would be more accurate, but I believe that Whorf’s amplification was great enough to justify the single reference.)

The Whorfian hypothesis is closely related to Bloomfield’s notion of thinking as an internal conversation, so it is appropriate that we first look at this idea. Bloomfield’s picture of the relation between language and thought was consistent with behaviorist theories that dominated psychology in the 1920–50 period. When the behaviorists spoke about thinking at all they spoke about the manipulation of response produced internal stimuli, i.e., subvocal words. The strong position that all thought is internal language cannot be maintained, for there is ample evidence that people can think by manipulating internal visual images, and perhaps other sensory images as well. It has also been denied that the forms of the external language have any pervasive influence on thought at all.

One of the major theses of the ‘cognitive revolution’ that swept over psychology and linguistics in the 1950s was the contention that language and thought are independent. This argument derives from the seminal work of Noam Chomsky (1957, 1963) on the role of generative grammar in language. The clearest exposition of the language–thought distinction may be in the writings of Jerry Fodor (1975). Following Chomsky, Fodor argued that internal thought takes place in a special language, ‘mentalese,’ that is roughly equivalent to the predicate calculus. Expressions in mentalese are then translated into expressions in external languages according to the rules of the particular external language involved. An important assumption of the Fodor–Chomsky approach is that any expression in mentalese can be expressed by at least one (and usually many) expressions in any natural language. Therefore, any expression in language is translatable into an expression in any other language. Because the thoughts behind these translations occur in mentalese, the form of the external expression plays little role in thinking. This will be called the intertranslatability hypothesis. It will play a crucial role in the argument to be presented here.

The intertranslatability hypothesis is a statement about the competence of two languages to express each other’s ideas. It says nothing about the difficulty of language-specific information-processing actions required to go from the mentalese expression to the external statement in the two languages. The distinction is important.

To begin the current discussion, let us consider how language is used. One way is as a communicative act: a speaker (or writer) tries to create a thought in another person. Many studies have shown that it is possible to overwhelm the information-processing capacities of a receiver by presenting linguistic messages that are too complex to understand. Therefore, language contains devices to call attention to different aspects of the message to be communicated. For instance, attention can be focused on a topic by the device of first mention. The sentences

(a) John loved Mary.

(b) Mary was loved by John.

say the same thing, in one sense, but in (a) John is the focus of attention, while (b) is about Mary. Similarly, the provision of undisputed facts can shape meaning in subtle ways. Consider:

(c) President Eisenhower, the former five-star general, spoke to his cabinet.

This suggests that President Eisenhower did not conduct cabinet meetings in the folksy way that Abraham Lincoln is said to have done.

Whorf himself cited examples like (c) to show that the categories available in a language influence thoughts about the world. For instance, early in his professional life, Whorf worked as a safety engineer. He observed people smoking around gasoline drums that were labeled empty, disregarding the fact that a drum that has been emptied of liquid gasoline may contain highly volatile fumes. Whorf argued that the semantic association of the English term ‘empty’ led workers to the incorrect interpretation that the drums were inert. This is very close to Hofstader’s observation that the meaningless terms ‘slithy’ and ‘lubricilleux’ will evoke different associates, and thus a different interpretation of Lewis Carroll’s nonsense poem.

All languages use devices like this, and there is no doubt that thinking can be framed by subtleties of language. What is more questionable, and to the psychologist more interesting, is how the language available to a perceiver can be used to influence the way that perceivers can describe a situation to them-selves.

At this time we return to Sapir’s extreme position that a person literally cannot perceive a physical difference unless the person has a word category that depends on it. This proposition was tested, and apparently falsified by Eleanor Heider (Rosch) (Heider and Oliver 1972). Rosch studied color perception in the Dani, an aboriginal people on the island of New Guinea, in what is now the Indonesian province West Irian. The Dani language has only two terms that are roughly equivalent to ‘dark’ and ‘light.’ However, Rosch found that color discrimination in the Dani was the same as the color discrimination of English speakers. Rosch’s work was widely cited in textbooks as conclusive evidence against Sapir’s position that language can influence perception. Then, almost 30 years later, Roberson et al. (2000) conducted a follow-up study that raises questions about the earlier work. Roberson et al. used somewhat improved experimental techniques in a study of color perception in the Berinmo, a Papua New Guinea group whose language contains only five basic color terms. The Berinmo were most sensitive to color changes at the borders of the color categories in their language, whereas English speakers showed maximal color discrimination at the borders in their language. For instance, English speakers were maximally sensitive to colors that did or did not cross the blue–green boundary line, while Berinmo speakers were maximally sensitive to crossing the nol–wor boundary, which lies within the range of the English term green.

Another aspect of color naming has often been overlooked. Naming can affect memory even though it does not affect perception. Suppose that an English speaker is induced to call a particular object whose color is in the blue–green range either ‘blue’ or ‘green.’ Subsequently, the speaker’s memory for the object’s color will be distorted toward the prototypical hue of the name used to describe the object. This phenomenon can be thought of as a specific case of a much more general phenomenon. Examples of a category are remembered as more prototypical in appearance than they actually are. The general phenomenon has been demonstrated for categories as disparate as the position of a point in a clearly marked region of a page and the personality characteristics of people assigned to social roles, such as ‘professor’ or ‘neo-Nazi.’ Because our thought is heavily dependent on memory, the categories that language provides are important.

The decisions required to construct a linguistic statement can focus attention on selective aspects of a situation. This argument has been developed by Daniel Slobin (1996), who has coined the term ‘thinking in speaking.’ In order to describe a situation to ourselves or to another we have to decide what linguistic constructions (morphemes, syntactical forms, etc.) to use. The decisions required to choose between linguistic constructions focus attention on certain aspects of a situation. If it can be shown that different natural languages force a person’s attention toward different aspects of a situation then we would have evidence for a Whorfian effect.

Whorf himself thought that he had observed such an effect in his studies of the way that the Hopi, a Native American people (and one of the larger tribal groups still extant) thought about time. According to Whorf, Hopi verbs stress cyclic stages (e.g., growth, decay). This sort of linguistic frame is used to discuss time. By contrast, English verb forms specify discrete states; the past, present, and future. Whorf believed that as a result, the Hopi had a different view of time than did speakers of ‘standard European languages.’

Whorf’s fieldwork with the Hopi has been criticized on the grounds that he did not ‘really’ understand Hopi. This claim is almost impossible to evaluate. Whorf did his fieldwork in the 1920s. The modern Hopi are much more integrated into the standard US culture than they were in 1920, and few speak the language as a first language. Therefore, we cannot interview the current Hopi and, with confidence, either object to or confirm Whorf’s fieldwork. However, a similar claim has been made for the contrast between two of the most widely spoken languages in the modern world; Spanish and English. In fact, because of a unique feature, the claim is for Spanish speakers against at least all romance and Germanic languages.

Spanish has two forms of the verb ‘to be,’ ser and estar. The ser form is generally used to indicate permanent states and properties, while estar is used for continuing actions or temporary states. Two examples are:

(d) Soy un hombre =I am a man.

(e) Estoy en mi oficina =I am in my office.

Sera and her colleagues (Sera et al. 1997) have found that the use of ser or estar assists Spanish-speaking children in distinguishing between permanent and temporary properties of objects, in a manner that is unavailable to English-speaking children.

The use of classifier systems provides a second example of how ‘thinking in speaking’ can produce a Whorfian effect. Some of the languages distributed around the Pacific Rim require the use of classifiers before certain types of nouns. For example, Japanese requires the use of the term bo before things that are long and narrow, as in:

(f) I have three bo pencils.

In this example, shape classification, the property upon which classification is based, is perceivable. This is not always the case. Chinese uses a classifier for ‘things that can be grasped by the hand’ (Zhang and Schmitt 1998). Dyribll, an Australian aboriginal language, has a classifier for things derived from certain myths. It is applied, among other things, to ‘women, fire, and dangerous things’ (Lakoff 1987).

Languages that use classifier systems force the speaker to attend to certain features of objects. As a result, speakers of languages that use classifiers give higher similarity ratings to pairs of objects that share a classifier than do people whose language does not require classifiers. For example, Chinese speakers in Hong Kong, whose language classifies door keys, rulers, and pliers as ‘things that can be grasped by the hand,’ rate these objects as being more similar to each other than do English speakers, who do not use classifiers (Zhang and Schmitt 1998).

‘Thinking in speaking’ can provide effects if the development of appropriate linguistic forms calls attention to a nonlinguistic relation that is important in the world. Miller and his colleagues (Miller et al. 1995) report a more subtle effect involving number names. In every language, digit names up to the base term are arbitrary. That is, there is no reason other than history that 1= one, 2= two and so on up until 10. By contrast, the formation of English number names from twenty (2 x 10) onwards incorporate a mathematical relation; twenty-one =20 +1, and even three hundred and forty-five =300 +40+5. The number names for 11–19 are irregular in two ways. ‘Eleven’ and ‘twelve’ are arbitrary names, and a special regularity, thirteen =3+10, etc. is used instead of stating the most significant digits first, as is done for integers greater than twenty. Spanish is irregular also, although in a slightly different way. The integers from 11 to 15 have special names (once, doce, trece, catorce, quince), followed by the regular form with the most significant digits first (dieciseis, diecisiete 10 +6, 10 +7). By contrast, in Mandarin Chinese, the spoken number names for all digits follow a regular convention. The Chinese equivalents of the digits eleven and twelve are 10+1, 10 + 2, as if English speakers said ‘teenone, teentwo’ or Spaniards said ‘dieciuno, deicidos.’

Miller et al. observed that when children in Englishand Spanish-speaking countries learn elementary arithmetic they have an unusual amount of trouble with operations involving numbers whose names have special terms, 11–12 in English and 11–15 in Spanish. Chinese-speaking first and second graders have no such trouble. Miller attributed this to the fact that irregular number names failed to reinforce numerical relations in the way that regular number names do.

Miller’s research can be looked upon as a case in which different languages vary in the facility with which they express facts about the external world, but the facts can still be expressed. Research on terms for spatial relations provides an example of a situation in which a language may not be able to express certain facts. Most languages have words for expressing relationships in both an egocentric frame of references (right, left in English) or in an abstract geographic frame of reference (north, south, etc.). However, there are some languages that only have terms for one of these frames. Speakers of languages that are restricted to geographic frames interpret ‘looks the same’ to mean that two views of objects maintain their geographic orientation, while speakers of languages that have egocentric terms interpret ‘looks the same’ to mean that egocentric orientations are retained in two views (Levinson 1996). On the other hand, if a language only has terms for egocentrically defined relations it can be difficult to describe spatial layouts to people who have not observed them (Hunt and Agnoli 1991).

Hunt and Agnoli (1991) offered another argument for the Whorfian hypothesis. They pointed out that working (short-term) memory capacity is a major bottleneck in speech comprehension. Furthermore, the bottleneck seems to depend on the number of word forms that are to be held in short-term memory rather than the complexity of the concepts to which the words refer. Therefore, specialized vocabularies evolve to provide a convenient way to let people talk about concepts of interest. Borrowing an argument form Whorf, Hunt and Agnoli suggested that having such specialized vocabularies facilitates thinking about theme. To a computer specialist the statement:

(g) The hacker put a cookie on the client’s machine. is certainly more compact than the equivalent in more standard English:

(h) The person attempting to break the security of a computer system for nefarious purposes put a program that surreptitiously records and reports the activity of a computing machine on the machine of a person who is using a desktop computer to gain entry to the Internet via another machine, called a server.

There has been a running argument between supporters and opponents of the Whorfian hypothesis over whether or not the existence of specialized vocabularies such as these constitutes a facilitation of thought by language. Whorf himself believed it did, arguing that Eskimos (Inuit) were better able to think about snow than English speakers. (Whorf overestimated the richness of the Inuit vocabulary for snow, but they do make distinctions that English does not, so his point is still valid.) Hunt and Agnoli do the same, arguing that words are used to package ideas, as illustrated by examples (g) and (h). As these examples show, it is possible to translate computer jargon into normal English, but the result is a pragmatically incomprehensible sentence.

Granted that examples like this can be made up, do they occur? The answer is that they can, and in situations in which the topic may be quite serious. In the late 1999–early 2000 period, a serious political dispute broke out between the government of Taiwan, which calls itself the Republic of China (ROC) and the mainland government, which refers to itself as the People’s Republic of China (PRC). The two had been separate and hostile for over 50 years, but in the 1990s fairly extensive commercial relations developed between Taiwan and the mainland. Taiwanese (ROC) politicians said that, therefore, it was time to begin formal relations on a state-to-state basis. PRC officials reacted in fury, and literally threatened war if Taiwan attempted to establish two Chinese nations.

According to New York Times reports of the incident, part of the problem was that the Mandarin Chinese language uses the same word for ‘state formal machinery of government’ and ‘nation unitary cultural and social entity.’ On a de facto basis, PRC officials had been dealing with the ROC as a separate state for years. Unacknowledged contacts already existed between the formal machineries of the two governments. On the other hand, PRC leaders have an official and deeply held claim to be the only legitimate Chinese nation. To them the ROC is the government of a temporarily renegade province. Therefore, any claim to separate nationhood was, in the PRC ideology, a cause for war. ROC officials then claimed that they had only proposed government-to-government, rather than nation-to-nation, discussions. In order to expand on this claim, some officials resorted to English, even when speaking to other Chinese speakers.

The anti-Whorfian psychologist has two rejoinders to examples like these. One is to claim that these examples refer to the communication of ideas rather than the formation of them. This is consistent with the notion that ‘basic thinking’ takes place in an alinguistic, unobservable mentalese. Bouchard (1997) has remarked, in quite another context, that you can always explain an observation by appealing to unmeasured variables. The remark applies here. The notion of mentalese has a great deal of appeal because it fits with certain notions about generative models of language. However, it is not clear whether or not mentalese could be replaced with a reduced form of natural language terms, permitting thought directly in a speaker’s language. And it is even less clear how this question could be investigated objectively.

The other anti-Whorfian rejoinder is that all these examples show is that language adjusts to cultural demands. Words and their related concepts are easily added to the lexicon. Modern English defines ‘hacker’ in a different way than it was originally used in computer science. More dramatically, a gay bachelor of the 1920s would not be gay in the sense the word is used in 2000. The anti-Whorfian argument is that cultural demands force language to adjust to cultures, rather than letting language restrict cultural change. Therefore, any restriction imposed on thought by lack of lexical entries for concepts like hacker or state versus nation is only temporary and probably not very important (Pinker 1994).

This objection is a value judgment rather than a hypothesis about facts, so it cannot be supported or refuted by scientific argument. It also brings us to another important point. We have been discussing specific points at which the form of a natural language may influence the thinking of its speakers. There is a ‘romantic’ version of the Whorfian hypothesis that claims that language may reinforce and/or restrict certain very broad cultural ways of thinking. This assertion is based on the assumption that there are definable cross-cultural ways of thinking, itself an assumption that requires analysis.

No one doubts that different cultures vary in their beliefs and value systems. This is part of the definition of culture. At the same time, modern psychologists and neuroscientists are confident that the content-free aspects of information processing, such as the distinction between short-term and long-term memory, work in the same manner for all normal human beings. If the differences between languages are regarded as a matter of setting parameters to govern the operation of a species-general universal grammar, then it is unlikely that the parameter setting would determine behavior outside of language. This is the approach taken by linguists and psycholinguists working in the generative grammar tradition. On the other hand, if both linguistic abilities and cultural knowledge are due to general learning experiences, mutual interactions are more likely.

Some popular beliefs appear to acknowledge a belief in the romantic Whorfian hypothesis. French is said to be the language of lovers, and German the language of soldiers and engineers. In the extreme, this is silly, for differences between natural languages certainly do not preclude broad categories of activity or belief. Frenchmen built the Suez Canal, and the German romantic song ‘Lili Marlene’ was hummed by homesick soldiers on two sides in two wars.

What we need is more precise evidence of a linguistic effect. Let us take a fairly frequent accusation against US society. European commentators have remarked that contemporary US society mixes superficial personal familiarity with an almost cold treatment of close acquaintances. It is also true that English, unlike many other languages, uses the same form of address, ‘you,’ for formal meetings and for conversations between close friends. We have no objective measure showing that French people or Spaniards, whose language does include different forms of personal address, are in fact more formal in business and more intimate with their friends than are English speakers.

The closest that we can get to scientific evaluation of romantic Whorfianism is to look at correlations between definable properties of a language and global impressions of a culture. For instance, A. Wierzbicka, a Polish expatriate to Australia, has written extensively on the contrast between the Polish and English languages and what she sees as the relative emphasis on group solidarity and intimacy in Polish society and the personal speaking autonomy in the English-speaking societies. One of her favorite examples is the English convention of using indirect requests, as in

(i) Could you close the window?

This would normally be taken as a request for action rather than a request for information. Wierzbicka (1991) claims that Polish speakers prefer the more direct

(j) Close the window.

The problem is that we do not have any objective definition of ‘emphasis on personal autonomy.’ The problem is compounded by the fact that different subgroups in a large society may use their language in different ways. They may also vary in general cultural attitudes. Several of my Polish-speaking acquaintances dispute Wierzbicka’s assertion about Polish, saying they would normally use a form like (i) rather than (j). Similarly, I, an English speaker, regard both (i) and (j) as appropriate uses of English, depending on the circumstances. I also suspect that the relative frequencies of direct and indirect requests vary a great deal across social groups and situations within English-speaking cultures. Besides, even if you could establish a correlation between language uses and situations, the causal relations would remain unknown. Does the culture adjust to the language or the language to the culture?

My point is not to dispute Wierzbicka’s particular assertion. I am more concerned with the general point. Scientific hypotheses about the relationship between language and thought have to point to specific features of both language and cognition. Without adequate operational definitions we cannot investigate general assertions about relationships between language and thought.

Where, then, are we with respect to the Whorfian hypotheses? Linguistic effects on thought have been shown in specific domains, such as spatial reasoning, color perception, and memory, and the control of attention. The latter effect is particularly important, because the deployment of attention determines what goes into memory, and what goes into memory very largely determines how we think about things. The effects are generally rather small, but they occur every day a speaker uses language. Whether or not language exerts an important control on thought depends upon how one defines ‘important.’

Grand romantic statements about language controlling, or at least heavily influencing, thought are harder to prove or disprove. This is at least as much because of vagueness in specifying what objective acts of thought are to be controlled as it is because of vagueness in specifying what aspects of language do the controlling.

I do not think that this state of affairs would have surprised Whorf. In his own work he was very much a student of the particular. The broader notions of what I have called romantic Whorfism come largely from his interpreters. The Whorfian hypothesis is not a proposition to be verified or disproven. It is a scorecard, telling us the extent to which language and thought are entwined.

Bibliography:

1. Bloomfield L 1933 Language. Holt, Rinehart, & Winston, New York
2. Bouchard T 1997 IQ Similarity in twins reared apart. In: Sternberg R J, Grigorenko E L (eds.) Intelligence, Heredity and Environment. Cambridge University Press, Cambridge, UK, pp. 126–60
3. Chomsky N 1957 Syntactic Structures. Mouton, The Hague, The Netherlands
4. Chomsky N 1963 Formal properties of grammars. In: Luce R D, Bush R R, Galanter E (eds.) Handbook of Mathematical Psychology. Wiley, New York, Vol. 2, pp. 323–418
5. Fodor J 1975 The Language of Thought. Thomas Crowell, Cambridge, MA
6. Heider E R, Oliver D C 1972 The structure of the color space in naming and memory for two languages. Cognitive Psychology 3: 337–54
7. Hofstader D R 1979 Godel, Escher, Bach: An Eternal Golden Braid. Basic Books, New York
8. Hunt E, Agnoli F 1991 The Whorfian hypothesis: A cognitive psychology perspective. Psychological Review 98(3): 377–89
9. Lakoff G 1987 Women, Fire, and Dangerous Things. University of Chicago Press, Chicago
10. Levinson S C 1996 Language and space. Annual Review of Anthropology 25: 353–82
11. Miller K F, Smith C M, Zhu J, Zhang H 1995 Preschool origins of cross-national differences in mathematical competence: The role of number-naming systems. Psychological Science 6(1): 56–60
12. Pinker S 1994 The Language Instinct. William Morrow, New York
13. Roberson D, Davies I, Davidoff J 2000 Color categories are not universal: Replications and new evidence from a stone-age culture. Journal of Experimental Psychology: General 128(3): 369–98
14. Sapir E 1921 Language, an Introduction to the Study of Speech. Harcourt, Brace, New York
15. Sera M D, Bales D W, del Castillo-Pintado J 1997 Ser helps Spanish speakers identify ‘real’ properties. Child Development 68(5): 820–31
16. Slobin D I 1996 From ‘thought and language’ to ‘thinking for speaking.’ In: Gumpertz J J, Levinson S C (eds.) Rethinking Linguistic Relativity. Cambridge University Press, Cambridge, UK, pp. 70–96
17. Whorf B L 1956 Language, Thought, and Reality. MIT Press, Cambridge, MA
18. Wierzbicka A 1991 Cross-cultural Pragmatics: The Semantics of Human Interaction. Mouton de Gruyter, Berlin
19. Zhang S, Schmitt B 1998 Language-dependent classification: The mental representation of classifiers in cognition, memory, and ad evaluations. Journal of Experimental Psychology: Applied 4(4): 375–85