Language and thought
An age-old philosophical tradition emphasizes the close links between language and thought. In the Theaetetus (189e–190a) Plato said that thought is ‘the conversation [logos] the soul has with herself’. (In Ancient Greek, in fact, logos means both ‘sentence, word, conversation, discourse, language’ and ‘thought, reason’.) In a clear echo of Plato, the German philosopher Johann Gottfried von Herder (1744–1803) claimed that language is the ‘common understanding of the soul with itself’ (Herder 2002 [1771]: 90). For Herder, language and reason (we would say ‘cognition’) are intimately connected, perhaps the same thing.
This understanding of the connection between language and thought had its most influential recent advocate in the linguist Benjamin Lee Whorf (1897–1941). Whorf believed that the grammatical and lexical categories of one’s language determine the categories in which we think:
We dissect nature along lines laid down by our native languages. The categories and types that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, the world is presented in a kaleidoscopic flux of impressions which has to be organized by our minds – and this means largely by the linguistic system in our minds. We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way – an agreement that holds throughout our speech community and is codified in the patterns of our language. (Whorf 1956: 213)
For Whorf, in other words, language itself shapes the categories we use to reason about the world. Our conceptual categories are derived from the semantic categories of our native language. This idea is known as linguistic determinism or the linguistic relativity hypothesis, and is often para phrased as the proposal that language determines thought, which thus varies from one language to another. Obviously, this proposal could mean many different things. ‘Thought’ is an extremely vague expression: it covers conscious and subconscious mental processes, reasoning, the holding of beliefs and desires, and so on. No investigator would be willing to claim that every thing we call ‘thought’ is determined by language. In particular, we need to distinguish thinking in general from thinking for speaking. This latter term refers to the particular types of cognitive process involved in preparing and uttering language. Slobin (1996, 2001) and Levelt (1989) emphasize the extent to which the types of semantic distinctions encoded in language may direct the speaker to explicitly engage in certain thoughts. For example, a language which obligatorily encodes a perfective/imperfective contrast on the verb will require the speaker to subconsciously determine the relevant aspectual construal of the event being referred to in the lead-up to the utterance. Similarly, a language with a definite/indefinite contrast on NPs requires the correct definiteness value to be chosen for every NP, which means that speakers have no choice but to subconsciously attend to this contrast. This process of thinking for speaking means that the grammatical categories of a language must determine thinking for speaking.
It is during first language acquisition that the effects of thinking for speaking are most noticeable. In learning their native language, the child gradually learns what kind of conceptual distinctions are relevant in framing messages:
In learning the language, the speaker (the child) must surely have realized that the language requires him to attend to certain perceptual or conceptual features when he encodes a message. And . . . the child makes characteristic errors that reveal his successive hypotheses about the conceptual properties required for the assignment of his language’s morphology. (Levelt 1989: 104–105)
Each native language, in other words, ‘has trained its speakers to pay different kinds of attention to events and experiences when talking about them’ (Slobin 1996: 89). Languages without an explicit perfective/imperfective contrast, for example, do not require speakers to attend to this dimension of an event, whereas languages with an explicitly coded definite/indefinite contrast will require speakers to determine the definite ness values of the NPs they mention. Building on Levelt’s and Slobin’s proposals, Gentner and Boroditsky (2001) qualify the extent to which language exerts an influence on conceptualization in child learning. Based on evidence from language acquisition, they suggest that the conceptualizations referred to by verbs, prepositions, and other relational predicates are linguistically influenced, ‘whereas concrete nouns are in many cases simply names for preexisting cognitively natural referents’ (2001: 241). According to Gentner and Boroditsky, the denotations of concrete nouns tend to follow natural partitions – naturally preindividuated perceptual groupings. Nouns like rock, apple and box denote entities which are highly demarcated from their environment. Acquisition of grammatical distinctions relevant to these nouns, like number, does not require learners to attend to aspects of the objects which are not already salient perceptually: it is obvious from just looking whether there is one apple, or more than one. Relational terms, however, like concrete verbs and prepositions, are more linguistically influenced. In order to learn them, speakers have to enter the system of semantic distinctions that their language uses. Learners are forced to go beyond the most perceptually salient aspects of the referents, and have to actively attend to those specific aspects of the event relevant to their subsequent linguistic coding.
As a result, acquisition of relational terms is hypothesized to only come later, after the child has had more exposure to the language. In the same study, Gentner and Boroditsky show evidence that the hypothesized acquisition sequence is indeed the one that occurs – concrete nouns are acquired earlier than verbs. This is true even if the nouns are morpho logically more complex, a feature which could be expected to disfavour their early adoption. Gentner and Boroditsky describe the process as follows:
Consider the child’s initial task in its simplest terms, as one of attaching words in the stream of speech to their referents in the stream of experience. . .Concrete objects and entities have already been individuated prelinguistically. . .Given a salient potential referent, part of the child’s task of finding word-referent connections is already solved; it remains only to find the correct linguistic label. In contrast, for verbs and other relational terms, isolating the word is only part of the job. The child must also discover which conflation of the available conceptual elements serves as the verb’s referent in her language. (2001: 219)
If language plausibly influences conceptual development through the demands of thinking for speaking, what happens later? By adulthood, Levelt suggests that thinking for speaking no longer actively happens:
although conceptualizing and grammatical encoding are interacting for the language-acquiring child, the mature speaker has learnt what to encode when preparing a message for expression. He knows by experience whether his language requires a category of medial proximity, number, tense, object shape, or whatever is needed, and he will select the appropriate information in building his preverbal messages. (Levelt 1989: 105)
This would explain some of the difficulties of second language acquisition. The process of learning our first language ‘sets’ the mind in a particular way that sensitizes it to certain distinctions while accustoming it to ignoring others. Mastery of a second language therefore requires sensitization to unfamiliar categories, and is correspondingly difficult.
What about cognition that is not purely geared towards speaking? Can we discern any wider influence of language on non-linguistic thinking (‘thinking for action’, such as planning sequences of events, or navigating on a map)? The idea that our thinking in general is influenced by the language we speak seems both plausible and implausible. Its plausibility comes from the following fact: the thought processes of which we are most consciously aware are precisely the explicitly linguistic ones. The experiences we call ‘thinking’ usually take place subconsciously. If you read a sentence that you don’t understand, for example, and stop to think about it, you usually aren’t aware of any explicit thoughts unfolding in your mind as you try to work out what it means. Instead, what happens is that you reread it, or go over it mentally until, suddenly, you understand. This sudden understanding usually just happens: the pieces just fall together all at once. We’re not aware of any of the mental processing that must be going on in the background: it is all below the level of conscious ness. Sometimes, though, our thought surfaces in an explicitly conscious form. Examples of conscious thought include visualizing scenes in the imagination, performing thought experiments like rotating geometrical figures, or doing mental arithmetic. One of the most obvious forms of this self-aware thought is inner speech. Thinking quite often takes the form of silent talk to oneself. For me, this is often the case for planning sequences of events. When thinking about what order to do things in, I will often mentally construct sentences: if I go to the library first, it will be too late to buy bread, but if there’s no bread there’ll be nothing for lunch tomorrow, so I should go and get bread first. Experiences like these are apparently common. Because of them, the idea that thinking actually is linguistic in form seems to make a lot of sense.
From another point of view, however, the idea that thought is in language, and hence determined by whatever language we speak, is most implausible. We’re frequently aware of how inadequate language is to the ideas we want to express. Often, for example, words are ambiguous, and only one of the possible interpretations corresponds to the meaning we want to convey. Cases like this seem clear evidence of the non-identity of language and thought: we have a particular intention or meaning that we want to express, and it just so happens that our language expresses this meaning ambiguously. Surely this shows that the categories of language and thought are separate. More evidence of language–thought separability comes from coinages and borrowings. We resort to coinages and borrowings because it seems that the resources of our native language aren’t adequate on their own to the ideas we want to express.
The variations in meaning discussed in the previous section provide a good test case for the linguistic relativity hypothesis. If general cognition really is determined by the semantics of natural languages, there ought to be experimental evidence of a correlation between a speaker’s native language and their performance in non-linguistic tasks in the relevant domain. If this correlation was found to exist, it would not be direct evidence that language influences cognition, but would establish a correlation between the two. Further studies would be required to show whether the correlation was the result of cognition influencing language, or language influencing cognition. But if no correlation between language and thought is found, then that will be a clear refutation of linguistic determinism.
As part of the same study into spatial reference discussed in the previous section, a team of investigators from the Max Planck Institute for Psycholinguistics set out to test the linguistic relativity hypothesis. This research is discussed in a string of books and papers (see the further reading section at the end of the chapter); here, we will rely mainly on the account in Pederson et al. (1998). After establishing the features of spatial reference in the languages concerned, Pederson et al. conducted an experiment designed to reveal how speakers of different languages behave in spatial reasoning tasks. Subjects were presented with three different toy animals arranged in a row. They were asked to memorize the order of the animals ‘just as they are’, then turned around 180 degrees and after a thirty second delay asked to rearrange the animals in the ‘same’ order. This task thus required subjects to store the order of the animals in memory, and to draw on these stored memory representations in order to reproduce the original array. The hypothesis was that the type of spatial frame of reference characteristic of the language would influence the way subjects behaved in non-linguistic spatial reasoning tasks.
In principle, two types of result in these non-linguistic tasks are possible, corresponding to the relative and non-relative (intrinsic/absolute) frames of reference described in 11.4.5. A subject observing a relative frame of reference will reconstruct the row of animals in the same left right order as the original. If the original array was in the order pig on the left, horse in the middle and cow on the right, a subject observing a relative frame of reference will reconstruct the array as pig–horse–cow from left to right, mirroring the original scheme. This preserves the animals’ orientation with respect to the subject’s own body. On the other hand, a subject observing an absolute frame of reference will keep the order of the animals fixed with respect to external anchoring points like the cardinal points or landmarks. Because of the 180 degree rotation, the order will be inverted on the transverse (left-right) axis, but fixed with respect to the external bearings. For such a subject, the animals will be rearranged in the order cow left, horse middle and pig right.
Pederson et al. found a statistically very highly reliable correlation between the type of response in the non-linguistic animals-in-a-row task and the prevailing frame of reference used in the subject’s native language. Speakers of Arrernte, Tzeltal and Longgu, which all have absolute frames of reference, were likely to reconstruct the animals in an inverted order which preserved their orientation with respect to the external bearings. In contrast, speakers of Dutch and Japanese, languages with relative frames of reference, were likely to preserve the left–right order of the animals, inverting their order with respect to fixed external bearings. These differences of behaviour were independent of other variables such as literacy, schooling, sex or age (Levinson et al. 2002: 161). These results do not mean that speakers are locked into any one form of reasoning. Anyone is able to reason in any of the three ways at different times, and context may play a large role in determining which style of reasoning will be adopted at any one time. The original Max Planck Institute findings are still controversial, but have stood up to challenge (by, for example, Li and Gleitman 2002: see Levinson et al. 2002).
This experiment therefore provides evidence of a correlation between language type and non-linguistic cognition. The frame of reference used in a language is correlated with the way people conceptualize spatial relations in non-linguistic reasoning. This is enough to keep the linguistic relativity hypothesis in the game, but it is not yet enough to confirm it. The experiment tells us nothing about the direction of any influence between language and thought. Do speakers behave as they do in the memory task because their language has moulded the concepts they use to reason spatially? Or does the frame of spatial reference characteristic of a particular language derive from patterns in the way its speakers think? Many researchers think the former conclusion is the more likely. Levinson et al. (2002: 161–162) construct the argument that language moulds thought like this. Neighbouring, closely related cultures can use an entirely different mix of reference frames: Mopan, for example, uses intrinsic only, while the neighbouring Tzeltal, another Mayan language, has absolute and intrinsic frames. In a case like this, there simply is no other source for the observed differences in spatial reasoning techniques than the individual’s native language. As Levinson (2003: 214) puts it, ‘linguistic determinism seems the most likely explanation for the correlation . . . it would seem to take a communicative system to induce cognitive uniformity throughout a community in such an abstract psychological domain’. In the same vein, for Pederson et al. the language structure manifested in language use provides individuals with a system of spatial representation:
use of the linguistic system, we suggest, actually forces the speaker to make computations he or she might otherwise not make. Any particular experience might need to be later described, and many are. Accordingly many experiences must be remembered in such a way as to facilitate this. Since it seems, based on our findings, that the different frames of reference cannot be readily translated, we must represent our spatial memories in a manner specific to the socially normal means of expression. That is, the linguistic system is far more than just an AVAILABLE pat tern for creating internal representations: to learn to speak a language successfully REQUIRES speakers to develop an appropriate mental representation which is then available for nonlinguistic purposes. (Pederson et al. 1998: 586; emphasis original)
Pederson et al. express confidence that similar effects will be observed in other areas of language:
We do, however, feel optimistic that these correlations between language and thought will generalize to some other domains as well – when these are investigated in the manner described here. The domain of these spatial relations seems especially basic to human experience and is quite directly linked to universally shared perceptual mechanisms. Since linguistic relativity effects are found here, it seems reasonable that minimally they could be found in other, less basic domains as well. Finally, there must be a mechanism at work that creates mental representations consistent with social language use. It seems improbable that such a mechanism would be specific only to this one domain. Rather, such a mechanism would potentially operate across many areas of human cognition. (1998: 586)
Even if Pederson et al. are right that similar effects of language on cognition exist elsewhere, there is still a substantial body of evidence from other domains suggesting that Whorfian effects are not pervasive. Papafragou (2002), for example, investigated path and manner distinctions of the type studied by Talmy (11.4.4). She showed that the differences between English and Greek in the lexicalization of motion don’t correlate with any differences in the behaviour of Greek and English speakers in memory and classification tasks based on these variables. Subjects don’t differ in their memory or classification for path and manner distinctions, in spite of the differences between their languages. Malt et al. (1999) studied perceptions of container similarity for bottle and jar-like objects among speakers of languages which draw the boundaries between these categories in very different ways. Speakers of Argentinian Spanish, Chinese and American English were asked to undertake sorting tasks in which they had to sort photos of objects into piles that were physically similar, functionally similar, and similar over all. Here again, no significant linguistic relativity effect was found. Malt et al. conclude as follows:
our correlations suggest that linguistic categories are not even the primary determinant of perceived similarity. Our data, if anything, suggest that perception of the similarity among objects remains relatively constant despite wide variation in linguistic category boundaries. (1999: 258)
Perceptions of similarity among objects are thus not influenced by striking cross-linguistic differences in the way objects are categorized. This is consistent with the results of Gentner and Boroditsky mentioned earlier. The challenge for linguistic research is to clarify exactly which domains seem to involve Whorfian effects like those uncovered in the spatial domain.
