Posted by In his comment to my post on conceptual metaphor theory (CMT), reposted here, Dr. Gibbs writes:
The topic of why conceptual metaphor theory arouses such vehemence is one that greatly interests me and is again the subject of my in progress book. My own opinion is that conceptual metaphor theory, and other related ideas from cognitive linguistics are threatening to traditional scholars because it is contrary to prevailing, enduring beliefs that the mind is literal and disembodied. Put simply, many cognitive scientists could not continue to work in the ways they do if they embraced some of these alternative views. Might this be the reason why some scholars seem resistant to looking at the real evidence?
There’s probably something to this point. Mainstream cognitive science has traditionally been dominated by functionalist approaches to mind, and the corresponding computer metaphor of mind. These approaches generally involve discrete representations — usually described as existing in the form of propositions, or a mental language (mentalese) — and a syntax, or set of rules, for manipulating those representations. This view allows mental representations, and the rules used to process them, to be amodal or disembodied. Theoretically, you can use the same representations and rules with any hardware, whether it’s another brain or an actual computer.
This traditional view of mind has been challenged from several directions within cognitive science. The most well known of these challenges is connectionism, or neural networks. Connectionism, instead of using propositional representations, uses units, connections of different strengths between those units.
Since neural nets are designed to simulate the human nervous system, it readily lends itself to the idea that the “mind” is embodied, and thus in addition to providing a different modeling paradigm, connectionism can also provide a different theoretical paradigm for studying cognition. Over the years, connectionism and computationalism (or symbolism), as the traditional view is often called, have become largely integrated within mainstream cognitive science, though there are some connectionists who still treat the two paradigms as irreconcilable. Other challanges to computationalism have come from dynamic systems theory, which has been very successful in modeling some motor processes, though entirely unsuccessful at modeling higher order cognitive and perceptual processes; from perceptual symbol systems theory (and related “empiricist” theories of representation); and as Dr. Gibbs notes, from cognitive linguistics.
The foundation of cognitive linguistics is CMT, which argues that our concepts are structured (i.e., they get their form and some of their content) through metaphorical mappings to the representations of more concrete concepts. If you follow the train of metaphorical mappings all the way down, you eventually get to direct embodied experience. So at base, all concepts are grounded in embodied experience, either directly (through direct mappings to that experience, for highly concrete concepts) or indirectly, through mappings to more concrete concepts. A common example of a conceptual metaphor is “ARGUMENT is WAR.” CMT argues that ARGUMENT is structured through a mapping to WAR, which allows arguments to have sides, winners, etc. Other examples include GOOD is UP, ANGER is a HEATED FLUID IN A CONTAINER, and NATION is a FAMILY. In each case, the first concept (often called the target, in the metaphor literature) is the abstract concept, and the second concept (often called the base or vehicle) is the more concrete, structuring concept.
Obviously, the focus on the embodied grounding of concepts is inconsistent with traditional functionalism and computationalism, and thus, inconsistent with much of the theoretical work in cognitive science over the last 50 years. I suppose this is a good candidate, then, for explaining why researchers have largely ignored CMT (and when they’ve paid attention, either dismissed it or argued vehemently for rejecting it) since Lakoff and Johnson first described CMT in Metaphors We Live By.
But I don’t think it’s the right explanation. For one, cognitive science has grown up a lot in the last 27 years. Connectionism forced computationalists to closely examine many of their assumptions, including their functionalism. I don’t know any mainstream cognitive psychologists today who are the least bit wedded to the idea of a disembodied mind. To see this, one only has to look at the debate centering around the perceptual symbol systems theory I mentioned above. This theory, championed by Larry Barsalou (see the paper linked above), argued that concepts are represented perceptually (i.e., in the perceptual systems of the brain), rather than propositionally. Insiders will recognize that this is not a new position. One of the most heated debates in the history of cognitive science — the great imagery debate — took place in the 1970s between image-representation theorists and propositional-representation theorists (analogous to “amodal” representation theorists). The debate ended largely in a stalemate, with Zenon Pylyshyn, a member of the proposition camp, arguing convincingly that the two types of representations couldn’t be distinguished empirically (that is, any data would be consistent with both). But Barsalou’s work, and that he’s inspired, has infused the debate with new energy, and though no data has yet been produced that is inconsistent with “amodal” representations, the focus on the perceptual aspects of representations has produced some interesting experimental results. So despite the fact that perceptual symbol systems theory, which requires an embodied mind, is inconsistent with traditional functionalism/computationalism, its productiveness has led researchers to seriously engage it. And in doing so, they’ve readily admitted that at least some of the content of representations has to be perceptual, and therefore embodied.
If researchers are willing to admit that concepts have perceptual (i.e., embodied) content in discussions of perceptual symbol systems theory, then it’s difficult to explain the rejection of CMT as being a result of the rejection of theories that highlight the embodied aspects of cognition. Furthermore, in conversations about cognitive linguistics, I’ve heard over and over from metaphor and analogy researchers that they find the theory interesting, and believe that its major forms (CMT, blending, etc.) could benefit from what we’ve learned about analogical and metaphorical mappings in mainstream cognitive science, particularly what we’ve learned about the process of structural alignment. As far as I can tell, little if any work has been done by cognitive linguists to integrate their theories of concepts with research on analogical/metaphorical mapping, however.
So it seems that basic philosophical and theoretical differences is insufficient to explain the rejection of CMT in mainstream cognitive science. A better candidate might be methodological differences. CMT was inspired by linguistic evidence that people use words from one domain (e.g., the domain of WAR) to talk about other domains (e.g., the domain of ARGUMENT), and for most of its history, cognitive linguists have relied on this evidence to support CMT. The argument from linguistic evidence to a theory of concepts does require a theoretical commitment, namely that the words we use closely reflect the way we conceptualize things, and that’s not a trivial commitment. However, I think most cognitive scientists would be willing to admit that there’s some truth to it. There is even evidence from experimental research that the names we use to refer to concepts are privileged over other features of those concepts2. So the theoretical commitment is not a big problem. However, the methodology is. Arguments from linguistic data are notoriously circular, as Greg Murphy pointed out in his 1996 paper arguing against conceptual metaphor theory3. He wrote:
In [Metaphors We Live By], a cultural metaphor is identified on the basis of various idioms and collocations, such as I destroyed her argument; he lambasted me in class; she undermined my position. Then a metaphoric representation is proposed on the basis of these data, such as ARGUMENT IS WAR. What predictions or consequences are derived from this metaphoric representation? In [Metaphors We Live By], it is further idioms and collocations: He can’t defend against that argument, etc. There is an absence of other psychological data given in support of this view. Lakoff (1993, pp. 205, 246) identifies five types of evidence for the metaphoric representation view: Four of them are linguistic) and one of them is psycholinguistic experiments. Notably, none of them provides a nonlinguistic measure of conceptual structure. Standard theories of concepts have had implications for findings in induction, problem-solving, object recognition, conceptual development and memory, among other areas. It would be useful to see evidence for metaphorical concepts from these domains, in order to escape the linguistic circularity. (p. 183-184)
And there’s the problem. As I’ve previously noted, the only experimental work on issues like induction, problem-solving, etc. within CMT has been with temporal metaphors [UPDATE: Note that I'm qualifying my previous claim here. There is data from psycholinguistic experiments that is consistent with CMT, but none in the areas that Murphy mentions in the above quote.]. There, Lera Boroditsky and her colleagues have shown that when primed with a particular temporal schema, that schema can influence people’s interpretations of temporal statements, which is a form of induction4. But that work is being challenged from several different directions, and even if in the end those results stand up, the use of a mapping between temporal concepts and spatial ones is hardly evidence for a general theory of all concepts. It could very well be that time is a unique conceptual domain.
There are other methodological issues as well. Boaz Keysar has shown that people have a hard time guessing the original meaning of conventional idioms like those used as evidence for CMT5. If people can’t access the original meanings of those idioms, and when asked to interpret them, produce what amounts to guesses that differ depending on the context, then it’s unlikely that either the linguistic data can provide a reliable source of evidence for CMT (because the idioms are opaque for CMT researchers too), and it’s unlikely that in interpreting those idioms, people are using mappings between the domains in the idioms. Furthermore, research on conventional metaphors has shown that conventional metaphors (like the war metaphors used to talk about arguments) are in fact “dead” metaphors. That is, they no longer involve mappings between the concepts involved, but are instead individual lexical entries with their own meanings that are specific to the new domain (perhaps in the form of new categories formed during the mapping process)6. In a way, conventional metaphors involve polysemous senses of the original meaning of the words. Thus, finding that conventional metaphors are ubiquitous in everyday speech cannot provide evidence for CMT, because those metaphors may no longer involve the mappings that CMT requires. At least, linguistic can’t distinguish between dead and “living” metaphors.
So there appear to be good methodological reasons for rejecting the evidence for CMT, and as the last few paragraphs suggest, there is actually empirical evidence from controlled experiments that argues against CMT. In fact, there’s pretty strong experimental evidence that people are not using those mappings even when using Lakoff and Johnson’s favorite CMT examples (e.g., ARGUMENT is WAR)7. And there’s a lot of evidence for theories of concepts and categorization that don’t involve on-line mappings from one domain to another. For a discussion of these theories, see the posts here, here, here, and here. Lakoff adopts a prototype view of concepts (see the third link in the previous sentence), but not one based on existing prototype models, or on data about how prototype-based concepts are processed. Suffice it to say, there’s no evidence from concept research that prototype-based representations include, much less require, mappings between concepts. Of course, there are problems with prototype representations, too, and if prototype theory ultimately fails, it’s not clear what would happen to CMT. But even if it’s doesn’t, empirical work designed to show how conceptual mappings affect prototype-based representations of concepts and their use are pretty much nonexistent.
My belief, then, is that CMT has been rejected not due to deep philosophical differences between cognitive linguists and mainstream cognitive scientists, but due to serious methodological issues in cognitive linguistics, and a fair amount of disconfirming evidnece (in the form of direct tests of CMT predictions, work on conventional metaphors, and the huge body of research on concepts and categories, none of which suggests conceptual mapping is involved). But I do think that deep philosophical differences are at play in the “metaphor wars” between cognitive linguists and the rest of cognitive science. Specifically, I believe that because Lakoff and Johnson started from the position that CMT required a second cognitive revolution, an anti-objectivist revolution in which all cognition is grounded in embodied experience, and because as a result, cognitive linguists have continued to feel that their theories are fundamentally incompatible with virtually all of the rest of cognitive science, cognitive linguists have been loathe to engage the rest of cognitive science seriously. As I said earlier in the post, I and others think that CMT and blending would benefit greatly from research on analogy. If these theories had a testable model of how mappings between conceptual domains are formed, interpreted, and used — a model that structure mapping theory can provide — it would be possible to make some concrete, experimentally testable predictions about things like inference, object recognition, etc. Furthermore, if cognitive linguists engaged data supporting other theories of concepts (exemplar theories, e.g.) and produced an experimentally testable model to compete with such theories, they would be able to test CMT in ways that the rest of cognitive science would not be able to ignore. And I can only think of two explanations for why they haven’t done these things yet: either they can’t come up with such models, or they are so committed to the perceived (but, as I’ve argued, not actual) deep philosophical differences underlying CMT and more mainstream theories of analogy and concepts, they believe that engaging those theories would be unproductive. Where people like Dr. Gibbs are concerned, I’m inclined to believe it’s the latter, because he’s clearly willing to engage people outside of cognitive linguistics. I’d just like to see a deeper engagement, involving the theoretical and empirical engagement of the wealth of data on metaphor, mapping, and concepts already existing, because that data and the theories it’s produced don’t require disembodied, dualistic (i.e. literal) theories of mind, and most of the researchers working with them aren’t wedded to such theories. I know I’m not.
1Title taken from the book Dr. Gibbs is writing, which he mentions in his comment.
2See e.g., Yamauchi, T., & Markman, A.B. (2000). Inference using categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 776-795; Gelman, S.A., & Heyman, G.D. (1999). Carrot-eaters and creature believers: The effects of lexicalization on children’s inferences about social categories. Psychological Science, 10(6), 489-493.
3Murphy, G. 1996. On metaphorical representation. Cognition, 60, 173-204.
4For example, Boroditsky, L. (2000). Metaphoric Structuring: Understanding time through spatial metaphors. Cognition, 75(1), 1-28, & Boroditsky, L. (2001). Does language shape thought? English and Mandarin speakers’ conceptions of time. Cognitive Psychology, 43(1), 1-22.
5 Keysar, K, & Bly, B. (1995). Intuitions of the transparency of idioms: Can one keep a secret by spilling the beans? Journal of Memory and Language, 34, 89-109.
6Bowdle, B., & Gentner, D. (2005). The career of metaphor. Psychological Review, 112(1), 193-216.
7See e.g., Keysar, B., Shen, Y., Glucksberg, S., & Horton, W. (2000). Conventional Language: How Metaphorical Is It? Journal of Memory and Language, 43, 576-593 & McGlone, M. S. (1996). Conceptual metaphors and figurative language interpretation: Food for thought? Journal of Memory and Language, 35, 544-565. I discussed the experiments in these two papers here.
June 18, 2007
June 19, 2007
June 19, 2007
June 19, 2007
June 19, 2007
June 19, 2007
June 21, 2007
June 24, 2007