[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: your mail
Well, here is another way to look at it, that is mechanism independent.
Discrimination and identification can be (and generally are) tested by the
same operations. The result of the testing, assuming it is properly done
and the response set is limited to the number of stimuli the experimenter
intended to use (with noise, of course, there might be lots more) is an n
x n matrix. In the 2 x 2 case it is easy to extract independent
parameters representing the discriminability of the two, and the bias to
respond more often with one response or the other (d' and beta). In the
case of n>2, it becomes very difficult to derive such metrics for each
stimulus. We sometimes create n distinct 2X2 matrices, with the rows and
columns representing the stimulus j, and not-j (all the others). This
does give you a fairly useful measure of the discriminability of stimulus
j from the remainder of the stimulus catalog, and of the bias to use
response j. And it is general for all n (including 2). But better
mathematicians than I have convinced me, over the years, that this simple
solution to relating identification to discrimination has some unfortunate
weaknesses. I'll bet that someone out there has a better approach, by
PS Note that I didn't say a word about proclivities or capabilities.
On Mon, 12 Oct 1998, Al Bregman wrote:
> Dear List,
> I don't agree that discrimination implies underlying set size of 2, and
> that this is the only difference between identification and
> discrimination. The critical difference is that in discrimination, the
> items to be distinguished are physically present on the trial.
> Here's an illustration:
> Create a hundred figures resembling chinese characters but built out of
> short, straight, connected lines, such that the only thing that
> distinguishes them is the number, orientation, and position of the
> component lines. Confusing? You bet!
> In a discrimination experiment, on any one trial, randomly select two of
> these figures, and present them side by side for as long as the subject
> wishes. Presumably the subject will score very high AS LONG AS THE
> DIFFERENCES EXCEED THE SENSORY CAPACITY TO REGISTER DIFFERENCES (e.g., in
> size of angle). Even though the size of the underlying set of objects is
> 100, the two to be discriminated are always physically present. From this
> we see that set size is irrelevant in discrimination. One need never have
> formed a mental representation of the stimuli before the trial, if
> inspection time is unlimited.
> Now teach the subject names for the 100 stimuli. On each trial, present
> one at random, ask for its name, then give feedback about correctness.
> Treat the number of trials to a criterion (say four correct responses in a
> row for stimulus n) as a measure of the subject's ability to perform an
> identification. This will be an incredibly difficult task, even though
> the underlying set size is the same as in the discrimination task, and
> even though the subject could discriminate any member of the set from any
> other member in the discrimination task.
> The critical difference is that in discrimination, the to-be-discriminated
> items are all physically present (there need not only be two). The
> psychological mechanisms for noticing differences between them involve
> very-short-term memory as you glance back and forth from one to the other,
> but this is not the same as the long-term memory required to remember
> their names.
> Of course when you speed up the responses in a discrimination task or you
> present the stimulus very quickly, the role of long-term memory increases
> and set size becomes important.
> So every discrimination task taps (1) the limits of sensory capacity to
> register differences, (2) very-short-term memory (3) long term-memory
> representations. By changing the properties of the task, you can tap
> these processes in various combinations.
> I think to understand what's going on it is important to consider the
> underlying mental processes and not just the formal properties of the
> measurement task.
> Albert S. Bregman, Professor, Dept of Psychology, McGill University
> 1205 Docteur Penfield Avenue, Montreal, Quebec, Canada H3A 1B1.
> Phone: +1 514-398-6103 Fax: -4896 Email: firstname.lastname@example.org
> Lab Web Page: http://www.psych.mcgill.ca/labs/auditory/laboratory.html
> Email to AUDITORY should now be sent to AUDITORY@lists.mcgill.ca
> LISTSERV commands should be sent to email@example.com
> Information is available on the WEB at http://www.mcgill.ca/cc/listserv
Email to AUDITORY should now be sent to AUDITORY@lists.mcgill.ca
LISTSERV commands should be sent to firstname.lastname@example.org
Information is available on the WEB at http://www.mcgill.ca/cc/listserv