What is ASA and who cares? (Al Bregman )

Subject: What is ASA and who cares?
From:    Al Bregman  <bregman(at)hebb.psych.mcgill.ca>
Date:    Fri, 23 May 1997 17:02:52 -0400

Dear List,

I'd like to reply to recent postings by Pierre Divenyi and Alain
DE Cheveigne.

Pierre wrote:
       "Talker A says "X", while
        Talker B says "Y".
If the (male) listener happens to perceive the structure * (asterisk as
in linguistics)
      * Talker A says "Y", while
      * Talker B says "X",
is he considered as having segregated the two events?"


Alain then replied:
"I'd distinguish the following levels of "segregation":
     (0) unaware of multiple sources
     (1) aware of more than one source (say, two)
     (2) able to recognize both sources
     (3) able to associate each source with other attributes
         (in Pierre's example, correctly attribute "come here,
         love" to wife rather than rival)."


To which, Pierre responded:
"However, in my interpretation, and I would like
to ask Al Bregman whether or not he concurs, auditory scene analysis (ASA)
implies that both the source and the message have to be identified and
correctly associated with each other. In other words, ASA is not complete
unless the attribute and its value are accurately perceived for all
concurrent events.


My reply is this:

It all depends on whether you want the term ASA to label
processes or accomplishments.  If it labels processes, ASA has
occurred because it has bound features into perceived events.
If it labels accomplishments, ASA only occurs if the problem is
solved correctly.

Consider failures of binding.  Pierre's example is one (if it is
an actual observation).  Another is the incorrect binding in
Diana Deutsch's octave illusion, when at a particular instant,
you might experience a low tone at the left when being actually
presented with a high tone at the left and a low tone at the
right.  In both examples, the percepts are internally well-formed
(using the grammatical perspective espoused by Yoshitaka
Nakajima), but do not correctly map actual physical signals.  Has
ASA taken place?  It depends on the definition.

Pierre's example is different from Diana's: everything except the
word meanings seem to be allocated correctly.  Certainly it is a
failure of correct binding, but is it a failure of ASA?  It's certainly a
failure of perception, but it all depends on how far you want to
push the notion of ASA.  If you want to make it co-extensive with
the solving of the cocktail-party problem (now to be relabeled as
the "potential adultery problem that occurs after you leave
the cocktail party") then it is a failure of ASA.  I would expect
however, that the failure in the potential adultery case is due to
top-down processes (e.g., wishful thinking) rather than bottom-up ones.

In the end, of course, such terms as ASA mean what we want them to mean,
and need only be as specific as the requirements of the problems at hand.
My own non-positivistic belief is that precise definitions of theoretical
concepts need to wait until we have a good understanding of the processes
being referred to.  Prior to that, initially rough definitions, gradually
refined, will be sufficient to point us toward the correct problems.

In reference to Alain's points, I think of ASA as the
contribution of bottom-up constraints acting to favor correct
recognition; however, even when these constraints don't lead to
the correct recognition of the lexical identity of a word (e.g.,
that it is the French word, cheval), listeners might still be
able to bind together all its basic auditory properties:
location, loudness, timbre, pitch, and modulations in these, and
hear it clearly as a distinct sound, even a voice, without
knowing that it was the French word, "cheval".

Alain distinguished different degrees of segregation according to levels
of awareness.  Of course, attention is involved in awareness, and
there is some reason to believe that auditory organization doesn't
require attention. But leaving attention aside, my view is that
the bottom-up constraints involve a binding of features, but not
necessarily a specific recognition.  So in his terms, it involves
level 3, but skips level 2.

Does this make anything any clearer?  Probably not.

- Al