Thursday 25 March 2010

Poverty of Stimulus and Ecological Laws

A question about a term I’ve been using is a nice segue into an important moment in the history of the ecological approach.

Poverty of stimulus is a term that came from Chomsky, but the intuition has been underpinning theories of perception for as long as there have been theories of perception. The term describes a problem, in which the information required to achieve something is not present in the environment. In language, the argument runs
  1. Certain patterns of correct language use can only be learned with exposure to negative evidence (i.e. evidence about what counts as incorrect)
  2. Children learning languages only encounter positive evidence (i.e. evidence about what counts as correct)
  3. Children do acquire the patterns in (1).
Chomsky accepted these premises, and the conclusion is that a) the stimulus (language exposure) is not sufficient to explain learning, so b) there must be supplementary information available from another source (innate grammar). As I mentioned when posting about Skinner this was the essence of the attack on Skinner’s behaviourist account of language acquisition; if the stimulus is not sufficient, behaviourism fails.

In perception, the argument is similar, but is rooted in the physiology of the eye (on the assumption that this is the gateway into the brain for light; any bottlenecks or limitations in the eye would be reflected in the information the brain had to work with to achieve the vision of our every day experience). One specific example is 3D information – depth perception.
  1. The world has 3 spatial dimensions (length, breadth and depth).
  2. The eye can only form a 2D image on the back of the retina (length and breadth; no depth).
  3. People do see in 3D.
If you accept the premises, the conclusion is that the 2D stimulus doesn’t contain enough information to unambiguously resolve the third dimension. We do achieve this, however, so we must be supplementing (‘enriching’ is the term favoured by the Gibsons) the 2D information from another source. That source is generally some form of past experience; this gets implemented as memory, dedicated representations, etc.

The problem of depth perception has a long and noble history in philosophy of mind and psychology. The state of the art when Gibson was working was David Marr’s computational model of vision, where the primal sketch provided by the eye was supplemented by texture information into a 2.5D sketch and eventually enriched into a full 3D model of the world.

Gibson’s move was to stop thinking about the physiology of the eye. His focus on what there was to be perceived (a functionalist stance, in the fine tradition of American psychology started by William James) led him to develop ecological optics (for vision). An optic array is the description of the ever changing (flowing) visual information available to an organism, in which relations remain invariant over the transformations and thereby permit specification to work.

Skinner was ‘killed off’ by Chomsky’s review of his book. Two major names in cognitive science attempted the same thing on Gibson’s 1979 book. In 1981 Jerry Fodor and Zenon Pylyshyn published a critique of the notion of specification that is, I have to say, full of rather patronising bullshit. (A friend of mine at IU knew Fodor, who apparently still thinks he killed specification off with the paper and hasn’t thought about this in years; this is apparently pretty typical behaviour.) Unlike Chomksy’s review of Verbal Behaviour, Fodor & Pylyshyn’s paper provoked a detailed response in a paper that has effectively defined the ecological approach ever since.

The critique
The critique is this: Gibson claims that perception contains information that is specific to an object or event in the world. Perception is therefore direct. Fodor and Pylyshyn note that this simply renders the problem to be trivial (in the technical sense) because Gibson’s theory does not a priori constrain the class of things that can be detected. Without this, the ecological programme of research devolves into a tautology hunt:

All things of a certain type (e.g. shoes) have at least one property in common that is specific to them (specifically the property of being a shoe). If you wish to pick out the shoes from a line up of objects, F&P claim that Gibson could simply say ‘the perceptual system simply detects the property ‘being a shoe’’.

This is somewhat of a dick move by F&P; it ignores everything Gibson had ever said about the optic array, etc. However, if it was true, it would be fatal to the programme, and so, dick move or no, it had to be addressed.

The reply
Enter Turvey, Shaw, Reed and Mace. Their reply to F&P is simple: Gibsonian information is suitably constrained (i.e. you can’t just claim any old property can be perceived) and that these constraints are not simply ad hoc, but rooted firmly in a consideration of the physics of the world in which vision evolved. This paper is elegant and clear; in fact its elegance and clarity manage to retrospectively make F&P’s paper worthwhile because it caused Turvey et al to write this.

The constraints lie in the existence of ecological laws. A law describes a set of conditions and the necessary consequence of those conditions (if the temperature of pure water at sea level is 100°C then the water will boil). Laws have scope: the scope is laid out in the set of conditions. Within that scope, the conclusion is necessary, i.e. if the conditions obtain and the scope is correct the conclusion must be the case. Ecological laws lay out the conditions and consequences by which a pattern in (for example) an optic array can specify an object or event in the world; these conditions must relate to the physics of the situation (so ‘shoeness’ is not the kind of property we are dealing with). The programme of research from an ecological viewpoint is to elucidate these laws correctly and identify the information that specifies the relation in question. If you have yet to identify a specification relation in your research, you have yet to correctly formulate the law governing the behaviour under examination. This last point has recently become a point of contention that will be the topic of future posts.

Summary
Poverty of stimulus describes an epistemological claim; the information required for us to do something we clearly do is not contained in the task environment, and so must be provided by an alternative source. For Gibson (and especially Turvey et al) this scenario simply tells you that your research has so far failed to correctly describe the task environment. The ecological methodology therefore assumes that poverty of stimulus is not an actual state of affairs in the world but a clue that your research is incomplete. Foder & Pylyshyn attempted to show that this assumption is under-constrained and therefore trivial; Turvey et al then laid out precisely how and why the assumption is entirely scientific and justifiable. The assumption made by the ecological approach might be wrong, but this is a coherent and answerable empirical question, and it is the focus of the ecological research programme.

References
Fodor, J., Pylyshyn, Z. W (1981). How direct is visual perception?: Some reflections on Gibson's “ecological approach” Cognition, 9 (2), 139-196 DOI: 10.1016/0010-0277(81)90009-3

Turvey, M. T., Shaw, R. E., Reed, E. S., Mace W. M. (1981). Ecological laws of perceiving and acting: In reply to Fodor and Pylyshyn (1981) Cognition, 9 (3), 237-304 DOI: 10.1016/0010-0277(81)90002-0

9 comments:

  1. Is there a good defense of the Skinner point of view from the Chomsky critique of language acquisition? Is there some ecological analogue, rather than the innate grammar black box?

    ReplyDelete
  2. Thanks, kiwi.
    Now THAT was an explanation.

    Question about the 3D vision: Isn't that depth mostly comes from the combination of our binocular vision?
    or is this exact point what Gibson though was misguided?

    I can't find anything in your argument that I would think is over the top.
    Actually right now I have some experiments using "error signals", which is ERP (event-related potentials) specialty.
    I was led in to the assumption that this is the way you have to do brain research with EEG because it's the way everyone else does it. You're the first one to tell me this is not the case.
    It's interesting to know that there could be an alternative.

    Has anyone compared a similar stimulus with both types of methodology?

    I wonder what my research would look like in an ecological model...

    ReplyDelete
  3. @Gavin - as I understand it language is still the great behaviourism crusher. The ecological methodological point is 'well, you just haven't found the right information' and there is a lot of information available to support language perception and production (from motor systems; see the Haskins Lab approach - and from perception, eg McGurk effects). No slam dunks, but proof of principle anyway.

    ReplyDelete
  4. @Eduardo: binocular vision is indeed a requirement to access lots of information about depth. Gibson has no problem with this: he just points out that the eyes are where they are in the heads of people to provide access to certain information. There is a lot of monocular information for depth (if you close one eye, the world doesn't suddenly appear to be flat) but it's less stable, in general.

    Marr's 3D sketch is the result of a computation on the 2D information from the eyes.

    ReplyDelete
  5. I guess much of this whole ecological debate is not really to find slam dunks. Rather, just to point out that there could be another way that the problem is solved. And, unless wonderfully clever experiments can be designed to genuinely determine how the problem is being solved by the visual system (or whatever), researchers of the Gibsonian viewpoint are generally content to perform experiments showing that it is possible (and, in some cases, simpler) for the 'problem' to be solved in an ecological way. Then it's up to the reader to decide which explanation is more parsimonious, or less computationally demanding (whatever that really means...).

    ReplyDelete
  6. Yeah, that is (at least partly) true. There is just a desire to demonstrate that that assumption of mental representation and computation is not necessary, and the number of tasks for which this has been demonstrated is growing. In one very important sense, this is all that's required just now: do good science, go after the problems one at a time and keep talking about the results to cognitive types.

    Of course, by demonstrating that there are other ways to solve the problem, and that this other way is aligned with evolution/physics etc, we do want to eventually convert people :)

    ReplyDelete
  7. Thanks for the answer.

    If you do want to convert people, you should post some books to read about the subject, specially for the non-psychologists.

    Or to put it in other words: Gimme, gimme.

    I would like to see if this theory convinces me enough to apply it to pain studies.

    ReplyDelete
  8. Your wish is my command: http://psychsciencenotes.blogspot.com/2010/03/reading-group-gibson-1979-and-heft-2001.html

    ReplyDelete
  9. I guess much of this whole ecological debate is not really to find slam dunks. Rather, just to point out that there could be another way that the problem is solved. And, unless wonderfully clever experiments can be designed to genuinely determine how the problem is being solved by the visual system (or whatever), researchers of the Gibsonian viewpoint are generally content to perform experiments showing that it is possible (and, in some cases, simpler) for the 'problem' to be solved in an ecological way. Then it's up to the reader to decide which explanation is more parsimonious, or less computationally demanding (whatever that really means...).
    I feel the urge to reply to this. My PhD advisor has been visiting and he gave a talk about some work on perception of metric shape. The result of (finally) working out what was required to allow such perception to occur (after years of no one being able to do it) did, in fact, entail a slam dunk. When you provide the correct information: people perceive and use metric shape. When you don't, they don't. It's not gradual, it's not a correlation: they either do it, or they don't, and the only error in the data comes from minor, completely normal random variations in the reach kinematics.

    Turvey is known for advocating the idea that 'if all you have is an account for some of it, you haven't found the information - keep looking'. A few people are beginning to freak out about this, but only because they can't find the information. But this made me remember that actually, slam dunks are precisely what we need (and are accumulating).

    ReplyDelete