From HubeyH at Mail.Montclair.edu  Sat Nov 27 17:27:28 1999
From: HubeyH at Mail.Montclair.edu (H. Mark Hubey)
Date: Sat, 27 Nov 1999 12:27:28 -0500
Subject: [language] What exactly are allophones?
Message-ID: <SAT.27.NOV.1999.122728.0500.>

<><><><><><><><><><><><>--This is the Language List--<><><><><><><><><><><><><>


I sent this to the author. It is just as well putting it here.

I have followed the exchange on phones, allophones and phonemes with
interest. Of course, I have to disagree with almost everyone :-).
(Just kidding.)

The reasons are that there are at least three separate issues;
(1) "theoretical" or what "ought-to-be" (deontology)
(2) what really exists in various forms
(3) some kind of a distillation of (2) which we may call
"practical" or "what-is" (ontology)


What I see in the literature, ranging from acoustics, to phonology,
and from articulatory to phonemics, is that

1. we have some acoustic signal which we can record electrically
and display. For concreteness, let us call this some kind of an <i>.
(I purposefully do not use [.] or /./]. We may record other
samples of this <i> so that this is a stochastic process and it should be
treated as such. Here there is no reason to invent a new name. Instead
we should use the words that already exist. This would be called a
sample function, first because it is a sample (which is written above
also) and second because it is a function of time. IT should be noted
here that this should also have an identifying subscript so that it
should be written something like <i>n because there will be many of
them. So <i>n is the nth sample function which is a real acoustic
signal not a set.

As it stands this signal is practically useless. We compute some derived
statistics from this signal. After massaging this signal then we may think
of putting this sample function into some set. It is here that problems begin.

2. Suppose we extract N statistics from this signal. Like other signals
of this type, this sample function then will be representable as a
point in N-dimensional space. A cloud of such points in this N-space
will represent a hypervolume. Equivalently we may think of it as a set
of points in this N-space. We can then think of cutting up this
hyperspace into hypervolumes according to some rules. It is here that
allophones, phones, and phonemes start to get confused. Some authors
immediately would write this and refer to it as "phone [i]". What
are they referring to? Here we have two choices; (a) how it should be
done and (b) how it is done. We can generate (or pretend that we have
generated) many many of these sample functions and we make many many
people listen to them, and they ask them to categorize them somehow
into groups. But they will categorize them according to the "sounds"
of their language. If we pretend that we have conducted an experiment
in such a way that we have had hundreds of thousands of people speaking
thousands of languages listen to these sounds and have categorized
them (according to the "sounds" of their languages which is naturally
what we'd expect), then what we have is that we can cut up this
hypervolume into smaller hypervolumes in many different ways, at least
as many ways as there are languages. These hypervolume divisions are
then (more or less) the phonemes of various languages. In practice
it might be slightly more complicated. We cannot do this and have not
done it but instead pretend that what would have happened if such a
thing were to be attempted has already been figured out by some
people who are presumably experienced and competent to do this. So the
division is really made by linguists some of whom have conducted
experiments, some have read of such experiments, some know a little
about acoustics, some a little about articulation etc. These pass off
as the equivalent of this hypothetical experiment. Even if we were to
perform such an experiment the judgements of the subjects would be
colored by their education. So these would be denoted by a symbol
like /i/n. (The n is a subscript so that to be clear and exact we
would have to specify which language it is a phoneme of. In practice
most of the time the subscript is implicit.)

3) We have yet another way to divide up this hyperspace. We can divide
it up into smaller pieces such that any of the language-specific
hyperspaces could be expressed in terms of the union (i.e. set theoretic
union which will be like concatenation of these spaces) of these smaller
hyperspaces. In principle, then we would never have to create any smaller
objects than these. We should denote these as [i]. Note that both
[i] and /i/ refers to a whole set of phones. But in many books
the word "phone" refers both to a set like [i] but also to a specific
realization (i.e. a sample function) which should be denoted by <i>n and
which should also be tagged with a subscript because there are many
of these. So then since this is not with respect to any language
(theoretically at least) these then are some kind of absolute
quantities (not relative). IT should then be possible to describe any
phoneme of any language in terms of unions of these phones.

Now when someone says "allophone" which phone is meant? If the phone
is in the sense of [i], then it means something different than if it
is meant in the sense of <i>n (please note the subscript, i.e.
<i> is not a set but a sample realization, a token or instantiation
in the terminology of computer science). The literature has both
meanings but usually what is meant (presumably) is that these allophones
are really different phones in the sense of [.] not <.>. In other
words (as explained in 3) a phoneme /&/ in some language might
actually consist of the concatenation of the hyperspace for two
phones like [&] and [#]. This is how some authors and many linguists
mean it. Here one starts getting into difficulties about what a
phoneme is or should be.


--
Sincerely,
M. Hubey
hubeyh at mail.montclair.edu             http://www.csam.montclair.edu/~hubey

---<><><><><><><><><><><><><><>----Language----<><><><><><><><><><><><><><><>
Copyrights and "Fair Use":     http://www.templetions.com/brad//copyright.html
"This means that if you are doing things like comment on a copyrighted work, mak
ing fun of it,
teaching about it or researching it, you can make some limited use of the work w
ithout permission.
For example you can quote excerpts to show how poor the writing quality is. You
can teach a
course about T.S. Eliot and quote lines from his poems to the class to do so. So
me people think
fair use is a wholesale licence to copy if you don't charge or if you are in  ed
ucation, and it isn't.
If you want to republish other stuff without permission and think you have  a fa
ir use defence, you
should read the more detailed discussions of the subject you will find through t
he links above."


You are currently subscribed to language as: language at listserv.linguistlist.org
To unsubscribe send a blank email to leave-language-4283Y at csam-lists.montclair.edu


From HubeyH at Mail.Montclair.edu  Sat Nov 27 17:27:28 1999
From: HubeyH at Mail.Montclair.edu (H. Mark Hubey)
Date: Sat, 27 Nov 1999 12:27:28 -0500
Subject: [language] What exactly are allophones?
Message-ID: <SAT.27.NOV.1999.122728.0500.>

<><><><><><><><><><><><>--This is the Language List--<><><><><><><><><><><><><>


I sent this to the author. It is just as well putting it here.

I have followed the exchange on phones, allophones and phonemes with
interest. Of course, I have to disagree with almost everyone :-).
(Just kidding.)

The reasons are that there are at least three separate issues;
(1) "theoretical" or what "ought-to-be" (deontology)
(2) what really exists in various forms
(3) some kind of a distillation of (2) which we may call
"practical" or "what-is" (ontology)


What I see in the literature, ranging from acoustics, to phonology,
and from articulatory to phonemics, is that

1. we have some acoustic signal which we can record electrically
and display. For concreteness, let us call this some kind of an <i>.
(I purposefully do not use [.] or /./]. We may record other
samples of this <i> so that this is a stochastic process and it should be
treated as such. Here there is no reason to invent a new name. Instead
we should use the words that already exist. This would be called a
sample function, first because it is a sample (which is written above
also) and second because it is a function of time. IT should be noted
here that this should also have an identifying subscript so that it
should be written something like <i>n because there will be many of
them. So <i>n is the nth sample function which is a real acoustic
signal not a set.

As it stands this signal is practically useless. We compute some derived
statistics from this signal. After massaging this signal then we may think
of putting this sample function into some set. It is here that problems begin.

2. Suppose we extract N statistics from this signal. Like other signals
of this type, this sample function then will be representable as a
point in N-dimensional space. A cloud of such points in this N-space
will represent a hypervolume. Equivalently we may think of it as a set
of points in this N-space. We can then think of cutting up this
hyperspace into hypervolumes according to some rules. It is here that
allophones, phones, and phonemes start to get confused. Some authors
immediately would write this and refer to it as "phone [i]". What
are they referring to? Here we have two choices; (a) how it should be
done and (b) how it is done. We can generate (or pretend that we have
generated) many many of these sample functions and we make many many
people listen to them, and they ask them to categorize them somehow
into groups. But they will categorize them according to the "sounds"
of their language. If we pretend that we have conducted an experiment
in such a way that we have had hundreds of thousands of people speaking
thousands of languages listen to these sounds and have categorized
them (according to the "sounds" of their languages which is naturally
what we'd expect), then what we have is that we can cut up this
hypervolume into smaller hypervolumes in many different ways, at least
as many ways as there are languages. These hypervolume divisions are
then (more or less) the phonemes of various languages. In practice
it might be slightly more complicated. We cannot do this and have not
done it but instead pretend that what would have happened if such a
thing were to be attempted has already been figured out by some
people who are presumably experienced and competent to do this. So the
division is really made by linguists some of whom have conducted
experiments, some have read of such experiments, some know a little
about acoustics, some a little about articulation etc. These pass off
as the equivalent of this hypothetical experiment. Even if we were to
perform such an experiment the judgements of the subjects would be
colored by their education. So these would be denoted by a symbol
like /i/n. (The n is a subscript so that to be clear and exact we
would have to specify which language it is a phoneme of. In practice
most of the time the subscript is implicit.)

3) We have yet another way to divide up this hyperspace. We can divide
it up into smaller pieces such that any of the language-specific
hyperspaces could be expressed in terms of the union (i.e. set theoretic
union which will be like concatenation of these spaces) of these smaller
hyperspaces. In principle, then we would never have to create any smaller
objects than these. We should denote these as [i]. Note that both
[i] and /i/ refers to a whole set of phones. But in many books
the word "phone" refers both to a set like [i] but also to a specific
realization (i.e. a sample function) which should be denoted by <i>n and
which should also be tagged with a subscript because there are many
of these. So then since this is not with respect to any language
(theoretically at least) these then are some kind of absolute
quantities (not relative). IT should then be possible to describe any
phoneme of any language in terms of unions of these phones.

Now when someone says "allophone" which phone is meant? If the phone
is in the sense of [i], then it means something different than if it
is meant in the sense of <i>n (please note the subscript, i.e.
<i> is not a set but a sample realization, a token or instantiation
in the terminology of computer science). The literature has both
meanings but usually what is meant (presumably) is that these allophones
are really different phones in the sense of [.] not <.>. In other
words (as explained in 3) a phoneme /&/ in some language might
actually consist of the concatenation of the hyperspace for two
phones like [&] and [#]. This is how some authors and many linguists
mean it. Here one starts getting into difficulties about what a
phoneme is or should be.


--
Sincerely,
M. Hubey
hubeyh at mail.montclair.edu             http://www.csam.montclair.edu/~hubey

---<><><><><><><><><><><><><><>----Language----<><><><><><><><><><><><><><><>
Copyrights and "Fair Use":     http://www.templetions.com/brad//copyright.html
"This means that if you are doing things like comment on a copyrighted work, mak
ing fun of it,
teaching about it or researching it, you can make some limited use of the work w
ithout permission.
For example you can quote excerpts to show how poor the writing quality is. You
can teach a
course about T.S. Eliot and quote lines from his poems to the class to do so. So
me people think
fair use is a wholesale licence to copy if you don't charge or if you are in  ed
ucation, and it isn't.
If you want to republish other stuff without permission and think you have  a fa
ir use defence, you
should read the more detailed discussions of the subject you will find through t
he links above."


You are currently subscribed to language as: language at listserv.linguistlist.org
To unsubscribe send a blank email to leave-language-4283Y at csam-lists.montclair.edu