#symbodiment should be #symbodimeaning: Do we need concepts?
Tuesday, June 14, 2016
How are the meanings of words, events, and objects represented and organized in the brain? This question, perhaps more than any other in the field, probes some of the deepest and most foundational puzzles regarding the structure of the mind and brain.
…so begins Mahon and Hickok’s introduction to this collection of papers “on issues of fundamental significance to a theory of meaning representation.” The collection is a tour de force that I expect will become required reading in the field for many years.
Rather than adding my own voice to the debate (which most closely aligns with Yee & Thompson-Schill’s contribution; see Lupyan & Bergen, 2015; Lupyan, 2015; Casasanto & Lupyan 2015), I wanted to bring attention to a certain word.
Although the deep puzzle mentioned by Mahon and Hickok is about the nature of meaning, much of the debate (including Mahon and Hickok’s) is about the nature ofconcepts. The word “concept” and its variants occurs 1732 times in the text and references. “Meaning” and its variants occur just 232 times. (In “normal” language, the ratio is 3:1 in favor of “meaning”).
Although colloquial usage of “concept” tends to mean something like an abstract idea or notion, its usage in cognitive science is often more specific: “Concepts are constituents of thoughts. […] This much is relatively uncontroversial” proclaims the Stanford Encyclopedia of Philosophy.
Might it be that on the road to understanding the representation of meaning (or else the representation of knowledge (“knowledge” fares a bit better at 409), our field has taken a wrong turn at the “concept/meaning” intersection? Might it be that although much of the work reviewed in this issue has brought us closer to understanding how meaning is represented, such progress has been in spite of rather than because of our focus on concepts as static constituents? Could we do better if we retired the concept of concepts and focused on studying meaning and knowledge instead? Here are three reasons why we can, and we should
1. A focus on concepts leads to searching for static representations of knowledge.
Because so much of what we know is acquired in the course of our interactions with the world, any theory of meaning/knowledge must be concerned with the role of experience. Although I can recognize a canonical picture of a violin as well as Itzhak Perlman can, my knowledge of violins is clearly different from Itzhak Perlman’s. I know this because Perlman knows more about violins than I do. Also, he can play one and I can’t. I can kick a ball, but my knowledge of kicking is different from a professional soccer player’s.
Yet, in many of the papers in the #symbodiment volume, is an assumption that somewhere deep down, a concept is the same from one person to another, and from one time to the next. For example, in responding to the finding that the phrase “Kick the ball” activates the motor system more than the phrase “Do not kick the ball”, Mahon and Hickok appear to assume the word “kick” is accessing the same concept of kicking: “the meaning of “kick” is the same in both the affirmative and negative sentence contexts.” But how do they know? The answer cannot be “because that’s the meaning of negation” without being circular. It also cannot be “because we use the same word in both contexts” without committing to a rather strong version of linguistic determinism.
Examples of behavioral and neuroimaging work showing the ways in which context affects neural representations abound in Yee and Thompson-Schill’s contribution (see also Casasanto & Lupyan, 2015). Researchers prescribing to the more traditional view of concepts readily accept such effects of context, and indeed, offer their own. For example, Leshinskaya and Caramazza in the #symbodiment issue mention a study in which participants viewed names of social groups (e.g., atheists, evangelicals, economists) while making pairwise comparisons in either political orientation (liberal vs. conservative) or spiritualism (spiritualist vs. materialist). The authors found patterns of activation in the right precuneus sensitive to the political orientation in one context, and to spiritualism in the other. They conclude that “part of right precuneus explicitly represents these belief trait categories.” (Perhaps it does, although such analyses may be less likely to reveal underlying representational and computational processes than often assumed). “The precuneus represents concepts…”, the authors continue. But does the concept of liberalism only exist in the precuneus when it is task-relevant? That’s what the data seem to suggest! Yet, contrary to the described data, the interpretation is that the concept is simply “there”. Again, how do we know?
Focusing on concepts as static entities (ones that you acquire, access, and lose in the case of dementia) makes context dependence an uncomfortable caveat rather than a critical feature of our mind. Apropos of Goldinger et al’s engaging #symbodiment discussion of the surprising difficulty in processing unfamiliar faces, consider the frequent difficulty people have with recognizing a familiar face in an unexpected context (or else the surprising ease in an expected context!).
One conclusion is that the concept of the person is the same, but what differs is access. Another interpretation is that the representation of one’s friend is indeed different (not completely, but still different) with context. As put by Yee and Thompson-Schill: “the concepts themselves are inextricably linked to the contexts in which they appear, so much so that the dividing line between a concept and a context may be impossible to clearly make out.” If that conclusion seems counterintuitive as applied to seeing an acquaintance in varied contexts, consider a longer timescale: Is a mother’s representation of her son as an infant, a toddler, a teenager really the same?If not, what makes it the same concept?
2. A focus on concepts leads to a focus on idealized rather than actual behavior.
One reason why psychologists care about concepts is that they care about the behaviors such as categorization, inference/induction, and reasoning that concepts ostensibly make possible. For example, Murphy writes in #symbodiment, “The purpose of having concepts is to make inductions”.
But as much as I believe in the importance (indeed, primacy) of induction, does it require concepts of the traditional constituent kind? Induction requires categorization: the process of representing non-identical instances as being the same in some way. This is what we do when we categorize a pyramid-shaped equilateral triangle and an obliquely oriented scalene triangle as both being triangles. How do we do this? One answer is that the (highly variable) perceptual input is “mapped onto” or “accesses” a common stable concept.
But how do we know that there is such a thing as “the concept” TRIANGLE? Well, as I have shown recently, when we ask people to define it, their definitions are quite similar (Lupyan, 2013). Almost everyone tells us that triangles are three-sided shapes/polygons/figures.
And yet, people consistently fail to use this definition in a context-free way. People who apparently “have” the right concept reject scalene triangles as “not real triangles” and say things like “equilateral triangles are the trianglest.”
People with the correct definition of even numbers often fail to classify numbers like “798” as even (with occasionally dire consequences). One response might be that such results are just about performance. People really have the same concept, they are just not using it correctly. But if the only measure of correctness is the verbal definition and all behavior points to gradedness and moment-to-moment context-dependence (Lupyan, 2016), where is the evidence for the stable and unchanging concept?
People’s classification, production, inference, and reasoning all demonstrate contextual and perceptual influences (as when people make more mistakes categorizing scalene triangles as triangles). Focusing on concepts leads to overlooking these systematic behavioral patterns in favor of idealized competences.
3. A focus on concepts obscures individual differences
Our ability to communicate and collaborate is sometimes taken as evidence that wemust have the same concepts. How could someone pass me the salt if they have a different concepts of “salt”?
But such logic leads to odd conclusions, as when Leshinskaya and Caramazza mention the “the blind [..] wield[ing] rich concepts, even those about vision—such as shiny or bright.” Leshinskaya and Caramazza seem to be convinced that blind individuals have the same concepts despite (clearly) having very different knowledge of brightness and shininess. It is indeed remarkable that language provides enough structure to allow blind people to understand something about the meanings of the words “shiny” and “bright.” But in what sense is the meaning of shininess and brightness the same in blind and sighted people?
There is certainly value in searching for commonality of knowledge in light of differences in experience. But such a commitment can lead to researchers overlooking individual differences that help us understand how meaning is actually represented.
About a month ago, Blake Ross, the co-creator of the Firefox web browser posted a fascinating personal essay about finding out that other people can form mental images.
I have never visualized anything in my entire life. I can’t “see” my father’s face or a bouncing blue ball, my childhood bedroom or the run I went on ten minutes ago. I thought “counting sheep” was a metaphor. I’m 30 years old and I never knew a human could do any of this. And it is blowing my goddamned mind.
It has been known since Galton that some people lack the ability to use explicit mental imagery and have no subjective experience of a “mind’s eye” (and in many cases a mind’s ear, nose etc.”). Ross’s ability to use language, recognize objects, and to create a program used by 100s of millions of people might be taken to mean that differences in mental imagery have nothing to tell us about language comprehension, memory, recognition, etc. But not so fast. Ross writes about reading novels:
Some books are so fleshy they’re opaque: Lord of the Rings numbs. But Lord of the Flies gnaws, because I could meditate on the idea of society-gone-wild forever. Animal Farm is awesome. 1984. The splendor of Hogwarts is lost, but the idea of a dementor is brain fuel.
And about not being able to remember what he did two hours ago without rehearsing the memory:
I’ve always chalked this up to having “bad experiential memory,” a notion I pulled out of thin air because “bad memory” doesn’t fit: I can recite the full to-do list of software I’m building. On a childhood IQ test, my best performances were on Coding and Digit Span, both memory driven. Given an increasingly long string of random numbers, I hit the test ceiling by repeating and then reversing 20 digits from memory on the fly. My three worst performances were on Picture Completion, Picture Arrangement, and Object Assembly. I couldn’t put the damn images in order to save my life.
To say that Blake Ross has all the “same concepts” is to miss the very things that concepts are supposed to do: enable recognition, memory, and so on.
Consider the research questions we can ask about how differences in mental imagery may influence the content and format of mental representations. Rid the theory of meaning of concepts, and individual differences become interesting again!
Taking a detour to visit William James
It is a testament to our discipline that although the idea of concepts as “things in the mind” might be entirely wrong, the work reviewed in the papers in the#symbodiment issue is, I believe, bringing us closer to understanding human behavior and how the brain represents meaning.
Rather than criticize or convince, the goal of this commentary is simply to suggest the possibility that we might make faster progress in developing theories of meaning and knowledge if we focus on meaning and knowledge and the ways that they interact with behavior of interest (categorization, induction, reasoning, etc.).
Instead of searching for the static concepts that somehow enable our hugely flexible behavior, we should shift the focus to understanding the computational principles that dynamically transform current and past experiences into representations useful for current goals. As much as I would like to claim credit for this pragmatic turn, the credit really goes to William James (1890):
…the word concept is often used as if it stood for the object of discourse itself; and this looseness feeds such evasiveness in discussion that I shall avoid the use of the expression concept altogether, and speak of conceiving state of mind, or something similar, instead. The word conception is unambiguous. It properly denotes neither the mental state nor what the mental state signifies, but the relation between the two, namely, the function of the mental state in signifying just that particular thing.