Neo-Empiricism

Barsalou, Glenberg, Prinz, Damasio and other neo-empiricists have theoretically and experimentally challenged the once dominant view that representations in higher cognitive processes are amodal (eg, Fodor, Pylyshyn). They have renewed a  century-old perspective on the mind, according to which representations in perceptual processes and representations in higher cognitive processes are, if not identical, at least qualitatively similar. Additionally, they have proposed empirical evidence for their views.

In my mind, neo-empiricism is a positive development, if only because it invites skeptics and proponents of amodal theories of the mind to specify their views and to provide evidence for them.

However, I have criticized at length neo-empiricism in two recent papers (also available on my site):

Machery. 2006 Two dogmas of neo-empiricism. Philosophy Compass 1:398-412.
Machery. In press Neo-empiricism: A methodological critique. Cognition (Barsalou is supposed to reply to this paper).

I think that there are at least three problems with neo-empiricism:

  • First, there is a theoretical question of distinguishing amodal representations from modal representations.
  • Second, there are several methodological issues in neo-empiricists’ experimental work (see particularly my Cognition paper). I have identified three main issues. In brief,
    1. It is impossible to sort out the predictions made by neo-empiricism in general and the predictions made by amodal theories in general; Rather, specific amodal and specific neo-empiricist models of cognitive processes make specific predictions;
    2. It is necessary to find some tasks that are not obviously best solved by using perceptual representations; Neo-empiricist psychologists have too often focused on tasks that are maybe best solved by means of perceptual representations;
    3. Psychologists must be cautious in generalizing some findings about some processes in some domains to other processes and other domains; the issue might not be whether representations are amodal or perceptual, but which processes in which domains in which contexts use amodal/perceptual representations.
  • Third, there is a large body of evidence for amodal representations used in higher cognitive processes (eg, the representations of cardinality).

I think that issues of this kind are fruitfully dealt with by philosophers. Philosophers who are interested in these theoretical and methodological issues can have a real impact on the development of psychology.

11 Comments

  1. Your point number 3 is especially good (about the need for caution in generalizing from one particular task).

    Operationally, is an amodal representation different from a multimodal representation? To the point, could purported evidence for amodal representations be interpreted as supporting a multimodal representation?

  2. gualtiero piccinini

    I’d be curious to know more about the motivation for neo-empiricism. As I understand things, classical empiricism was motivated by epistemological considerations: establishing the foundations of knowledge in experience. But by now, I suppose this is no longer a motivation for empiricism in psychology (e.g., because we have abandoned the myth of the given and believe that knowledge may be justified in experience even if not all representations are perceptual, etc.). So what’s the motivation for resurrecting empiricism in psychology? Empirical evidence?

    I ask because having being raised on a Chomskian diet, I find empiricism prima facie empirically implausible.

  3. edouard machery

    Eric writes:
    “Operationally, is an amodal representation different from a multimodal representation?”

    I suppose that by “multimodal representation”, you have in mind those neurons or brain areas that are activated by the inputs belonging to different modalities (e.g., the left inferior parietal sulcus).

    To counter the idea that the existence of these multimodal neurons or brain areas is evidence for amodal representations, Jesse Prinz has argued that amodal representations should be distinguished from multimodal representations. Multimodal representations are a kind of modal representations. (Claim 1)

    At the same time, Jesse characterizes modal representations as being those representations that belong to a perceptual system. Additionally, he proposes a definition of a perceptual system that does not hang on the notion of a modal representation (Claim 2). (Barsalou has another definition, based on the contrast between linguistic representations and analog representations. I have argued that Barsalou’s way of drawing the distinction between amodal and modal representations won’t do.)

    I might be that Jesse can’t have it both ways. Suppose that Jesse sticks to claim 1. Then, the location of these multimodal neurons and brain areas is such that it is not the case that representations are modal by virtue of beloning to a perceptual system, given Jesse’s definition of a modal system. But if Jesse sticks to claim 2, then the multimodal representations are not modal after all, given their location.

    I note that Dan Weiskopf discusses Prinz’ reply in a short, unpublished paper.

    As far as I am concerned, I think that being multimodal is evidence for amodality, though not an operational definition.

    Gualtiero wonders about the *motivation* behind neo-empiricism. I suspect that this varies from psychologist to psychologists and from philosopher to philosopher.

    A motivation is that at least some psychologists (e.g., Glenberg) believe that neo-empiricism can solve what is called “the grounding problem”–which is, in fact, the problem of providing a semantics for mental representations. I think that this approach leads to a naive empiricist semantics. (I note that this is not Jesse’s view.)

    Another motivation is a growing body of evidence that prima facie suggests that our perceptual systems are used in numerous tasks that were believed to involve amodal representations. Barsalou’s empirical work illustrates this approach.  My discussion of Barsalou targets this motivation. I tend to believe that most of his (very interesting) findings are predicted by some amodal theories of representations together with some models of the processes defined over these representations (see the Cognition paper).

  4. gualtiero piccinini

    Another naive question: what do these neo-empiricists say about language learning and linguistic cognition, not to mention abstract concepts (such as justice and truth) and mathematics? What do they say about the poverty of the stimulus arguments?

  5. edouard machery

    Not all neo-empiricist  endorse the anti-nativism of their forerunners. Barsalou is very happy to endorse innate perceptual representations. Prinz, on the other hand, is a staunch anti-nativist.

    What makes them empiricist is their commitment (1) to the view that conceptual representations and perceptual representations are not qualitatively distinct and (2) to the view that cognitive processes are defined over perceptual representations.

    Now, it remains unclear how perceptual representations could be sufficient to account for syntactic understanding, if anything like the Comskyan account (use of a tacitly know grammar to parse sentences) is correct. Neo-empiricists usually have little to say about this.

    They have more to say about abstract concepts. They recognize that it is a challenge for their views. They typically propose that entertaining an abstract concept consists in entertaining a complex, dynamic set of perceptual representations. It is however striking that most of their examples of such sets of perceptual representations are much too thick-grained to individuate abstract concepts. Typically, their examples could correspond to numerous abstract concepts (see my Two dogmas of neo-empiricism).

  6. Concept empiricists don’t say much about language acquisition, although Jesse does gesture hopefully towards recent data suggesting that children are exposed to a wider array of constructions than nativists have traditionally supposed, as well as towards statistical (e.g., connectionist) models of language acquisition. I don’t see that a concept empiricist needs to believe an empiricist language-learning story, though. It’s a strange feature of neo-empiricism that, if language is a perceptual system (as many maintain), there can be lots of ‘abstract-looking’ representations available to form thoughts with, since concepts can potentially be composed from any perceptual vehicles.

    On abstract concepts, there are a range of strategies:

    (i) For logical concepts: these might be operations, not representations; so, e.g., a P v Q thought might be a P thought and a Q thought that stand in a certain functional relation to one another (and to other thoughts), rather than containing an abstract logical constituent.

    (ii) For mathematical concepts: these might be representations of concrete numerals, or abstract ‘quantity’ mechanisms (e.g., accumulators), following Dehaene’s _Number Sense_ view.

    (iii) For ‘lofty’ concepts: these are either images that are associated with their referents (e.g., polling booths for DEMOCRACY), or linguistic vehicles (e.g., the word ‘democracy’).

    A major difference between neo-empiricists and classical ones is in the theory of content. Informational semantics allows us to refer to abstract properties no matter what sort of vehicles we use to do so. So all empiricists usually do is find some reasonably plausible candidate vehicle (a word, a numeral, an image of an associated perceivable scene), and claim that it realizes the abstract concept.

    One problem with this approach is that while it shows how possibly these concepts might be perceptual, there often isn’t evidence that these images and words are actually used in thinking about these categories. (This is in contrast to the excellent work on concrete object concepts, where there is such evidence.) So insofar as it just shows ‘how possibly’, it’s not as compelling.

    Also, I’d add that the linguistic vehicles strategy strikes me as pretty implausible. It’s not enough to use the _noise_ ‘democracy’ to think about democracy. That noise could stand for lots of things (compare genuinely ambiguous terms like ‘bank’). You might represent democracy as something _called_ ‘democracy’, but this involves the concept of being called something. That’s an intentional notion, hence another abstract concept. Looking at the details, it’s pretty hard to see how to make this linguistic strategy work within the confines of empiricism.

  7. edouard machery

    Dan,

    why do you think that (ii) is of any help for neo-empiricists? Dehaene and colleagues think of the representation of numerosity as amodal, though analog, representations . The main reason is that these representations are activated by the inputs to different modalities.

    It is true that Jesse does not endorse the semantics developed by the traditional empiricists such as Hume. However, many psychologists do, eg. Glenberg. Barsalou is unclear on this issue.

  8. Daniel Weiskopf

    Edouard:

    I don’t think that Dehaene’s account is really a boon to empiricists. I think Jesse’s presentation of it fudges a bit. He notes that there is some analog mechanism for representing quantity non-linguistically, but proposes that this can be thought of as a _visual_ representation, e.g., a number line (presumably an unmarked one). Jesse suggests that number lines give analogical ways of understanding quantity. Of course, as you point out, this is different from Dehaene’s interpretation of the results, on which the accumulator is amodal.

  9. Allan Paivio

    For over 30 years I have published empirically supported arguments for a multimodal dual coding theory of cognition as compared to abstract, amodal (single code) theories like Pylyshyn’s. As early as 1983 I summarized over 60 empirical findings that were consistent with predictions from dual coding but not single code theories. There are now many more such findings, especially ones from brain studies showing functional dissociations between modality specific perceptual and memory representations. Comprehensive reviews of the phenomenal domains covered by the empirical-theoretical dual coding approach are available in several books, the most recent being:
    A. Paivio (2007). Mind and its evolution: A dual coding theoretical approach. Mahwah, NJ: Lawrence Erlbaum Associates.

Comments are closed.