Concepts : A Thing or An Act?

Oct 08 2010 Published by under From the Melodye Files

There is a temptation to see a concept as a static ‘thing’ somehow stored in mind. One might imagine, for instance, that our concept of the word ‘hammer’ consists of visual memory for a hammer or series of hammers (or a ‘prototype’ or ‘schema’ of a hammer, whatever that should be).  That we can think how this might be so, is not a good reason to adopt this idea.  Indeed, it would be far better if should we discard this notion altogether.  If we take language to be a skill, like tennis or painting, we see quickly how this idea breaks apart at the seams.  For one does not have a static ‘concept’ of how one paints a landscape, or a fixed ‘concept’ of how one approaches a serve; rather one learns, over time, the toss of the ball, the arc of the back, the gentle shiting of weight, the flex of the wrist; and in all of this, there is the demand of sustained practice and coordination, the reproof and rebuke of time, and as ever, a great number of processes – physiological and mental – that contribute to the execution of the act (which is, almost certainly, imprecise).  If a concept is a skill too, then it is a learned process; an active engagement; one of a suite of ways of representing the world.

On our confusion with words (and our idea that a word 'stands' for a thing), Wittgenstein said :

"This [confusion] is connected with the conception of naming as... an occult process.  Naming appears as a queer connection of a word with an object. --[But] you really [only] get such a queer connexion when the philosopher tries to bring out the relation between name and thing by staring at an object in front of him and repeating a name or even the word "this" innumerable times.  For philosophical problems arise when language goes on holiday.  ...We can [of course] say the word "this" to an object, or as it were, address the object as "this"--[but this] is a queer use of the word, which doubtless only occurs in doing philosophy."

10 responses so far

  • Alternatively, verbs and nouns are represented differently in the mind/brain. Just sayin'.

    There has been a lot of interesting work on concepts (the classical theory, atomism, theory theory, prototype & exemplar models, etc.). But nothing says that the same theory has to account for everything ... esp. without a good theory-external definition of "concept."

  • Avery Andrews says:

    Another issue is variation. One possible explanation for how children learn so many word meanings with such a high degree of accuracy so fast is that they don't: the first guess at the meaning of many common words such as 'dog(gie)' is known to usually be wrong (e.g. 'big living thing with four legs', or something like that); I suspect that people don't later converge on shared concepts to the extent that orthodoxy says they do, and because people are furthermore constantly accomodating to other speakers, the boundaries of individual's concepts might well shift around a lot too.

    • melodye says:

      I agree. In review, we often get people saying "but what about 'fast mapping'?" And I always want to respond "Um, what about it? It's made up..."

      To give an example, there's this interesting phenomenon that children use color labels before they understand "what they mean." So, for example, there are a lot of three year olds (blind ones too) who will say "blue sky," "green grass" and so on, because they know that - distributionally - those words go together. But they don't actually know what blue is, and if you give them a swatch test and have them pick between blue / red / green, they can't tell you which one is "blue."

      There are a lot of developmental psychologists who say that children learn words like color or number from one exposure, or "all at once." It's just lunacy.

  • Sabrina Golonka says:

    I like the analogy here and I think it works on another level as well. I'd suggest that, just as it doesn't make sense to discretely represent tennis swings because of the action's tight coupling to the specific environmental context, it doesn't make sense to discretely represent concepts because what we mean by a linguistic label at any given time is strongly constrained by context.

    It's easy to see that there is no such thing as a prototypical tennis swing since what is required from that action varies so much from time to time. A tennis swing is an action with certain characteristic dynamic properties. But, as you point out, because we assign things discrete linguistic labels, it's more difficult to conceive that we don't have discrete representations of concepts.

    Also, as is often the case, we mix up describing a phenomena with explaining it. We can mathematically describe variability in certain tasks through a prototype or exemplar model, but this is one formal solution to the problem and it can't stand in for a psychological explanation (although this is usually how they're interpreted).

    • melodye says:

      Powerful elaboration, Sabrina. Re:

      "we mix up describing a phenomena with explaining it. We can mathematically describe variability in certain tasks through a prototype or exemplar model, but this is one formal solution to the problem and it can’t stand in for a psychological explanation (although this is usually how they’re interpreted)."

      I think this happens all to often in cognitive modeling (e.g., with many connectionist & Bayesian papers I've read). You can build a great model that gives a fantastic description of observed data, but it hardly matters if what you're modeling isn't psychologically plausible. It's why I like the learning model we use (Rescorla-Wagner); it's breathtakingly simple, and there's reams of neuroscientific evidence suggesting that it captures (real) dopamine response patterns in learning.

      If you're interested, Ramscar actually has an exegesis on Wittgenstein on concepts. It's funny, because a lot of scholars have taken Wittgenstein to be advocating a schema model of concepts -- but reading Ramscar (and then the Investigations), it becomes abundantly clear that that's just not the case.

  • Jason W. says:

    Melody, I like the development of this idea, and of your extension, Sabrina.

    [I'm a master's student just starting to think about language, so please forgive the naivete of what follows.]

    Sabrina, you mention that we paint ourselves into a corner with concepts--that, because we assign things discrete linguistic labels, we find it difficult to conceive of concepts as anything but discrete representations.

    How true is it, though, that we assign things discrete linguistic labels? Doesn't that assume the same static mental environment that you seem to be arguing against? I'm not convinced that we make any such concrete assignments.

    As you and Melody have noted, the meaning of labels, what "comes to mind" when we employ or perceive language, shifts with context. Additionally, it is a trivial observation that the linguistic labels I _employ_ in my language acts are hardly static. I may shift my accent, tonal patterns, pace, lexical and word order choices, and so on, to match my interlocutor and the conversational context. Similarly, my articulation of sounds changes dramatically given the coarticulatory context.

    [It's probably easy to dismiss these latter points as a difference between production and storage/representation, similar to the competence/performance distinction, I guess. But I wonder if such dismissals aren't begging the question, and assuming the underlying static representations at debate here--as opposed to language or representation as process similar to speech production, or indeed a tennis swing.]

    That said, this seems to be something of a restatement of the distinction between phonetics and phonemics, no? Melody and Sabrina, you seem to be arguing that we gain little in moving away from "conceptual phonetics," that "conceptual phonemics" introduces unneccesary abstractions that, to restate Sabrina, may have some theoretical or formal interest, but which may reflect very little in the way of psychological reality.

    • melodye says:

      So, of course you're right to say that a label isn't absolutely discrete, but it's a useful heuristic to use in modeling. Also -- a label is certainly *relatively* discrete as compared to the things in the world that are informative about that label. (E.g., The word "dog," pronounced in various ways, is still quite discrete compared to a flesh and blood dog, in terms of its featural properties).

      Does that make sense?

      If you're interested, we talk about this at length in The Effects of Feature-Label-Order and their implications for Symbolic Learning (Ramscar et al., 2010) -- which, by the way, argues against referential models of language. See Section 4 (Verbal Labels lack Cue Structure), pp. 917-8. We've also done some research on how we make a continuous sound space relatively discrete (i.e., how we learn phonemic categories), which is currently under review at Science -- I can send you a draft, if you send me an email 🙂

  • Avery Andrews says:

    I don't think you can get rid of the phonemes, although there's more than them going on. In English, for example, you can say 'bananas' in different ways to imply different things about your attitude towards the situation containing them ('yay, my favorite thing!' vs 'bloody hell not them again'), but there won't any difference in the information provided concerning the situation that the utterance is about ('referential meaning', one might say), as opposed to abotu the context of utterance, in particular your 'sentiment' as the Nat Lang Proc people might say.

  • Mark says:

    Hi Melody,

    Has the Ramscar paper on Wittgenstein and concepts been published? Is it going to be? Read it this morning and would like to make use of it.