> From: "Harrison, Richard" <RJH93PY@psy.soton.ac.uk>
> Date: Fri, 24 Nov 1995 17:04:31 GMT
> Two examples, first, the claim that participants take longer to scan
> longer distances across images, and second, subjectively smaller images
> are more difficult to examine (i.e. as subjectively smaller objects are
> more difficult to percieve visually, they are also more difficult to
> The second is weaker than the first and can be accommodated in
> Pylyshyn's propositional model (e.g. smaller parts involve 'moving
> through' more networks) but the first appears to be more challenging.
> Kosslyn suggests such observations indicate "that images do represent
> metric distance and that this property affects real-time processing
> of images." Critics maintain the effects are due to experimental
> demands (participants giving the answers they think experimenters
> want to hear). However, even if such an effect doesn't explain the
> findings I don't think they need necesserily imply what Kosslyn
> concludes. For example, why can't the manipulation of propositions
> take different times depending what the proposition are?
It can, but why should it? The relation between the timing effects and
the spatial properties is fairly straightforward. To get the same effect
with propositions requires more ad hoc factors. As I wrote in my comment
on Denise's summary, however, I agree that subjects might be talking to
themselves here, or going through steps consciously because of
expectations, etc., and that these might also account for the timing.
If Kosslyn had focused only on the task, and not the introspections,
and had shown the time effect across a larger variety of tasks, and
shown them for repeated presentations too, and speeded ones, where you
would expect the conscious frills to wear off, with the subject just
doing the minimal stuff it takes to get it right -- then finding the
same timing profile would, I think, favour analog processing over
symbolic (propositional) quite neatly. And a model that did the same
thing, and did it better than a symbolic one, would also help.
> A major problem with claiming that mental imagery can be an explanatory
> construct is that we cannot know when it is being used by participants
> and when it isn't (as well as if they are at all). Also, people do not
> report mental representations (as in the computer model) but report
> what the objects look like, there seems to be a distinction that is
> involved in preventing imagery's explanatory role.
Hence it is better to focus on analog processing, and what it is and is
not good for, and forget about whether or not it is conscious.
> I didn't find this article very convicing as many of the points in
> Pylyshyn's critique of imagery were not addressed. For example, the
> point that images must be processed before being stored otherwise we
> must have an unlimited storage capacity and our retrieval
> characteristics would be different (when something is missing it is
> usually an integral thing or a relationship, e.g. remembering who was
> at the party but not where they were standing, not arbitary segments
> like a torn photograph) were not accounted for in Kosslyn's paper.
But if he has a machine model that can do the same thing the subjects
can do, and does it by manipulating sensorimotor anaolgs rather than
symbolic descriptions, does that not get around Pylyshyn's objection
about the danger of infinite regress, the need for undue storage
capacity, prior interpretation, etc.?
This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:57 GMT