Back to the test

[From Rick Marken (931005.1600)]

Martin Taylor (931005 16:00) --

This is all getting too esoteric for me. Let's get back to where I feel
most comfortable -- in the world of data. Bruce said that one detects
"contrasts" using the pair test which is a version of "the test for the
controlled variable". I agreed. All I did was point would that the
"contrasts" being discovered by this test are simply perceptions that
differ from reference specifications. So "contrasts" reveal something about
the perceptions being controlled for in speech. When a person tries to
say "spin" they apparently don't care whether the sound after the "s" is
aspirated or not; this is not part of the perception being controlled.
They DO care about aspirating when they are trying to say "pin". If they
fail to aspirate the result will be treated as an "error" (becuase it
sounds like another word was said -- bin). I am sure that aspiration
is controlled when saying pin and not spin because sbin does not
happen to be an english word (yet). So the speaker doesn't need to control
aspiration when saying spin. Bruce seems to be saying that the fact that
the "pin"/"bin" distinction (this kind of distinction being one of Bruce's
meanings of "contrast", I think) exists in english is the reason why
we hear the p/b distinction when we say pin vs bin but not when we say
spin vs sbin. I think this is just a confusing way to look at it. I'm
sure we can hear the p/b distinction in both cases if we want to; we just
don't need to notice it when we are hearing (or saying) spin/sbin becuase
we don't need to CONTROL it (in order to be understood -- ie. control for
others understanding us).

What "the pair test" should reveal (and as I believe it does) is that people
do not control for mispronunciations of speech sounds that don't make a
difference (that are non-contrastive in the linguistic sense that Bruce
seems to use quite frequently). The "pair test" is a good start at testing
for the variables controlled in speech; I'm sure other tests would show
that in many cases these variables (like aspiration) are controlled
in order to make sure that the word said "contrasts" with other possible
(and similar) words.

By the way, I think I am learning quite a bit about speech from these
discussions with Bruce; it is obvious that Bruce knows quite a bit more
about speech than I know about PCT. Thanks Bruce.



[Martin Taylor 931015 19:10]
(Rick Marken 931005.1600)

About "contrast" vs "perceptual difference, Rick says

This is all getting too esoteric for me.

I don't see why. All I'm trying to get you to see is that the levels in
the hierarchy of perceptual control behave differently. At the category
level it makes no difference whether an acoustic signal is made from
"spin" or "sbin" because they are in the same word category. At the
continuum configuration-transition-event levels it does make a difference.

When you are pronouncing a word, you are controlling perceptions at all
these levels, so it is not true that

When a person tries to
say "spin" they apparently don't care whether the sound after the "s" is
aspirated or not; this is not part of the perception being controlled.

Yes it is. If it weren't, the word would sometimes be spoken with an
aspirated p sometimes without and sometimes with something in between.
But in fact, it is always spoken without aspiration unless the talker
chooses to speak it with, which is hard to do for an English speaker
untrained in foreign phonetics.

But you are rarely conscious of well-controlled perceptions, so when
you speak or listen, you rarely hear the difference without some training.


[From Oded Maler (930907)]

Re: Rick looking for tests.

In those dark ages when I was preparing my first Ph.D. proposal,
I was interested in Student-Modelling. This is a sub-topic
in computer-aided instruction, and the general setting is like this:
there is a "correct" way to do some excercise (e.g. addition).
The student is observed to some errors. So he presumably (modulo
some noise) employs a wrong program, and the goal of student modeling
is to keep on asking him more excercises in order to diagnose him
and see exactly what is his "program".

The names I recall were Burton amd Brown in SRI at the beginning of the
80's. They made a classification of virtually all types of systematic
errors (=erronous algorithms) in the domain of decimal integer
subtraction. One flaw of this work, from your point of view, that
it is a sequential symbolic task belonging to higher level, and I'm
nut sure how it will translate into C of P framework (e.g.,
"and now control for the difference between the sum of the two
leftmost digits, etc."). On the other hand it is a nice, but different
example of trying to identify a black-box from its external behavior.




Oded Maler, VERIMAG, Miniparc ZIRST, 38330 Montbonnot, France
Phone: 76909635 Fax: 76413620 e-mail: