minimal pairs are not always there

[Robert Whiting]

On Tue, 07 Nov 2000 Stanley Friesen <sarima at friesen.net> wrote:

>At 05:40 PM 11/6/00 +0200, Robert Whiting wrote:

>>On Wed, 26 Apr 2000 Gábor Sándi <g_sandi at hotmail.com> wrote:

[RW]
>>yourself a question:  If the phoneme /b/ occurs in an English
>>word, what does that tell you about the meaning of that word?
>>The answer should be:  "absolutely nothing."

>Seems reasonable, though perhaps a bit overstated.  There are
>plenty of words with some sort of sound-symbolism involving
>perfectly good phonemes to make this too strong a statement.
>(And I do not just mean onomatopoeia).

Yes, I grant you the point about sound-symbolism.  I always try
to allow for this, but it is a very tenuous thing (at the present
state of research).  To be convincing, there needs to be evidence
of fundamental links between sound and meaning that are
consistent over a large number of languages.  There would have to
be the same pattern of sound and meaning in languages of diverse
types, and finally, these patterns would have to be confirmed by
experiments linking these sounds and meanings to speakers'
intuitions about relationships between sounds and meanings.  In
the meantime, we can't just say that there is nothing to it,
because the data to establish it has not yet been collected and
analyzed in detail.  But there are patterns of sound symbolism, so
I suspect that there must be rules about such patterns.

But what kind of sound symbolism are you proposing that accounts
for the fact that pronouns and deictic words in English have [D]
while other words have [T]?  You can't just make the original
point go away by some vague reference to sound symbolism without
showing some connection to the present discussion.  For, if, as
you agree, when sound symbolism isn't involved, the presence of a
particular phoneme in a particular word shouldn't tell you
anything about the meaning of that word, then how is it
possible that the phoneme /D/ in initial position belongs
exclusively to pronouns and deictic words?  When you can provide
a convincing answer for this, then you will have made your point.
Until then, I will continue to maintain that the presence of a
particular phoneme in a particular word should tell you nothing
about the meaning of that word (leaving aside the possibility of
sound symbolism).

>>In the pair ether/either, 'ether' belongs to a class of words
>>where the pronunciation of <th> as [th] is required (by rule
>>based on a clearly established pattern [Greek loans originally
>>containing theta]);

>I disagree. I do not respond to any such rule in my own
>listening.  No such rule governs any pattern in my speech
>production.  As far as I am aware, I just memorized the word
>'ether' as such.

Well I can't argue with your perceptions, because they are in
your head, not mine.  But this means that you have memorized
the pronunciation of all other English words with <th> as well
and when you see a new word with <th> you don't have a clue how
to pronounce it.

>>  'either' belongs to a class of words where
>>the pronunciation of <th> as [dh] is required (by rule based on a
>>clearly established pattern [native words with an intervocalic
>>spirant where the original voiceless spirant was voiced by
>>rule]).

>Ditto.

Ditto.

>>   Since the respective pronunciations of these words are
>>required by rule, this cannot be a phonemic distinction because
>>the choice of these sounds is not arbitrary.

>It is in MY speech.  And I doubt many other English speakers
>could be shown to have even a subliminal awareness of any such
>rules.

You still seem to be operating on the assumption that it is
necessary to know how a system works in order to be able to use
it.  By and large, native speakers do not know what the rules are
in any conscious way.  They just know what sounds right.  And it
is fully automatic.  The "rules" that we talk about are for the
most part simply a model that linguists use to account for the
patterns of grammatically acceptable usages that they find in a
language.

>>Greek word with theta doesn't make any difference.  There must be
>>some synchronic rule that forces this [th] pronunciation or else
>>the pattern wouldn't be there in the language synchronically.

>Not at all - there is such a thing as a *relictual* pattern. one
>that is left over from prior history, but which has no *current*
>linguistic significance.

The entire language is a relictual pattern left over from prior
history.  Otherwise you have a creole.  There are only relictual
patterns that are no longer fully productive and ones that are.
The latter are used to create new forms in the language by its
speakers; the former are not.  Here is a good example of a
relictual pattern that is no longer productive:  the English
personal pronoun system.  Now tell me that this pattern has no
significance for the *current* language or that native speakers
aren't aware of it because it is a "relictual pattern left over
from prior history."

Native speakers aren't born with their language already in their
brains.  They are born with a language acquisition device that
makes it possible to acquire the language that they hear spoken
around them in a manner that we don't fully understand.  They do
this before they are capable of memorization, and indeed, before
significant long-term memory capabilities are developed.  So
language is passed from generation to generation.  Native
speakers learn the language of their parents (or care-givers) as
a historical relic, which as far as they are concerned has just
been invented.  So each generation reinvents the language.
Sometimes they get things wrong and introduce change (or at least
alternation) into the language.  But anything that gets passed on
is part of the language.  And it is part of the language
*synchronically* because it is known by the new batch of native
speakers and they don't care when it came into the language.  As
far as they are concerned, it came into the language when they
learned it because they have no prior knowledge of the language.
If it does not get passed on, then it becomes part of history and
disappears from the currently spoken language.  But anything that
new native speakers learn is part of the language
*synchronically* and has *current* linguistic significance.

>The only evidence I would accept for a synchronic rule is
>evidence from the speech behavior of native speakers.

Okay, so you won't accept the list of words and pronunciations
that I provided based on a contemporary corpus and dictionary.
Fine.  Take the same list and ask a couple of hundred native
speakers how they pronounce these words.  If you find significant
differences from what I have provided, then you will have made
your point.  But I think that you will find that this is the way
that these words are pronounced by native speakers today.  We're
not talking about words that were pronounced one way in the 15th
century and are pronounced differently now.  We're talking about
the way these words are pronounced at this very moment as we sit
here in front of our computers.

>The mere existence of a pattern is not, by itself, sufficient to
>establish a rule.

The existence of patterns is the only basis for establishing
rules.  That is what science is all about.  Grammatical analysis
is virtually entirely devoted to trying to deduce rules that
explain the patterns of grammatically acceptable forms.  That is
why linguists spend so much time looking at languages trying to
find patterns.  Now when you have a pattern that you can't
account for, you are justified in saying that you can't account
for the pattern, but you aren't justified in saying that the
pattern has no significance because you can't account for it.
But above a certain level of coincidence or randomness, patterns
not created by rules are ruled out by the second law of
thermodynamics.

>The very fact that having characters names Lith and
>Lihthe(/lidh/) is *comprehensible* in English shows that English
>speakers *perceive* the two sounds as being semantically
>arbitrary.  If they did not, then someone hearing such a pair of
>names would necessarily attempt to interpret them according to
>the supposed rules, and thus have a difficult time hearing them
>as a pair of names.  Instead they would hear two words in
>different categories, as required by the rules.

This seems to be a cart/horse problem to me.  If you present your
audience with two words that differ in only one segment and tell
them (or let them figure out for themselves) that the two words
are arbitrarily semantically different, then you have a phonemic
contrast.  That's what a phonemic contrast is.  But when you say
that if you created such words (in the absence of such words
already existing) and speakers could tell them apart that means
the sounds are already phonemes, that is a logical fallacy.

You are trying to convert a necessary condition into a sufficient
condition.  "Distinguishable sounds" is necessary condition for
phonemes.  Therefore "all phonemes are distinguishable sounds" is
a valid statement.  But "distinguishable sounds" is not a
sufficient condition for phonemes.  Therefore "all
distinguishable sounds are phonemes" is not a valid a priori
statement.

The simple converse of a universal statement is not a valid
inference.  Thus "all S are P" does not imply "all P are S".
Your argument is based on the fact that all phonemes must be
distinguishable sounds.  This is an a priori valid statement.  In
this formulation, S is "phonemes" and P is "distinguishable
sounds."  When you then say all distinguishable sounds must be
phonemes, you are trying to tell me that "all P are S" because
"all S are P."  It is exactly the same as telling me "all birds
have wings, therefore anything that has wings is a bird."  Or if
you have a mathematical bent it is like saying that "all prime
numbers greater than 2 are odd numbers" (a valid statement) and
then claiming that "21 is an odd number greater than 2, therefore
21 is a prime number."  The simple converse of a universal
statement is valid if and only if S and P are an identity (i.e.,
S and P are identical sets).  If there is a member of P that is
not also a member of S then the simple converse of "all S are P"
is not a valid inference.

Therefore the fact that there exist two distinguishable sounds in
a language (e.g., [c,] and [x] in German) that are not phonemes
in that language proves that "phonemes" and "distinguishable
sounds" are not an identity and therefore the converse of "all
phonemes must be distinguishable sounds" (i.e., "all
distinguishable sounds must be phonemes") is not a valid
inference as you claim it is.

The fact that two sounds are not distinguishable by the speakers
of a language will prove that the sounds are not phonemes in that
language.  But the fact that two sounds are distinguishable by the
speakers of a language does not prove that they are phonemes.  It
shows that they could be phonemes if they were used as phonemes,
but "could be" and "are" are not the same thing.  It is like
proving paternity with blood groups.  If the blood groups don't
match that proves that there is no paternity.  If the blood
groups do match, it shows that paternity is possible, but it
doesn't prove paternity between two individuals.

This entire discussion grew out of a stated belief that "minimal
pairs" were a sine qua non for determining phonemes.  Many people
pointed out that this was not true.  I pointed out that even
apparent "minimal pairs" don't necessarily guarantee the presence
of different phonemes.  Now you and Mr. White are trying to tell
us that you don't even need minimal pairs at all in a language to
determine phonemes.  You can make up your own and if you judge
that the speakers of a language would be able to recognize these
as different words then the sounds in question are phonemes (and
have been for a long time).  I say that this is not true.  Native
speakers of English could probably hear the difference between
French 'beau' and French 'bon'.  Does this mean that nasal vowels
are phonemes in English?  No, it does not.  But you (and Mr.
White) are trying to tell us that it does.  And I for one, do not
believe it.

>>to be pronounced with [dh].  There is nothing arbitrary about the
>>choice of [th] and [dh]; these pronunciations are required by
>>rule, therefore the contrast is not phonemic.

>The pronunciations are "required" by *historical* rules that are
>no longer active.

The rules are active because the words are pronounced according
to them.  If the rules were no longer active (i.e., had been
replaced by a rule that was active) the words wouldn't be
pronounced that way.

Here is an example:  At the time of Chaucer 'bite' was pronounced
with [i:].  By the late 18th century it was pronounced with [ai].
The pronunciation with [i:] is a *historical* rule that is no
longer active.  The pronunciation with [ai] is a *historical*
rule that is still active.

>In my speech I have merely learned the two words by rote.  I
>apply no rules to determine their pronunciation.

And this is precisely why naive native speakers are not a
reliable source of grammatical information.  Such speakers can
tell you infallibly whether something is grammatically correct
(acceptable usage) or not, but if asked why it is correct or why
some other form is incorrect can usually only say "I don't know,
it just is."  In exactly the same way you can say "I know of no
rules about the way these words are pronounced, that's just the
way they are pronounced."  Consider for example the use of the
suffix '-ist' meaning "one who does ...".  Why is a person who
does physics called a 'physicist' and not a '*physicsist' or why
is a person who does science called a 'scientist' and not a
'*scienceist'?  Are there rules that determine this?  Well, yes.
Is a naive native speaker likely to know these rules?  Well, no.
Is a naive native speaker likely to know the correct forms?
Well, yes.  If you ask a naive native speaker why 'scientist' is
correct and '*scienceist' is incorrect, what answer are you likely
to get? --  "I don't know, it just is," or perhaps "I don't know
of any rules, that's just the way I learned it."

>smuggled into phonology.  So it is not a matter of "no
>nonphonetic conditioning factors," it is a matter of how much
>morphological conditioning you are willing to allow into
>phonology.

>As little as possible to still produce a coherent theory of a
>given language.

A sound rule, but who gets to decide what is a coherent theory?
And what happens to the evidence that isn't included and who gets
to decide what evidence can be safely excluded from the theory.
This sounds awfully like the way Ruhlen goes about creating
Proto-World.  Take as little phonological and grammatical
information as possible to produce a coherent theory and don't be
specific about how much or what kind of evidence is excluded.

>I.e., apply Occam's Razor here.  Do not postulate an entity - in
>this case a particular type of morphological conditioning -
>unless it is *necessary* to explain the facts. You might also
>call this the KISS principle.

Call it whatever you like, but Occam's Razor doesn't tell you
what evidence to exclude.  Occam's Razor just tells you that the
simplest solution *that accounts for all the facts* is to be
preferred.  Occam's Razor does NOT tell you that if you can get a
simpler solution by ignoring some of the evidence then that is
what you should do.

>>My own answer to this tends to be "as much as you can."

>Then here is the fundamental disagreement.  I prefer to minimize
>special handling, not maximize it.

Again, a sound rule.  I could better express my position as "as
much as you need, but don't arbitrarily exclude rules that tell
you how words are pronounced from phonology."  Or if you do
exclude them from phonology, then at least account for them
somewhere.

>>> Take two of your examples above:

>>> /ng/ - not a phoneme. We can analyze it as the allophone of /n/
>>> before /k/ and /g/. /g/ is then dropped ("zero allophone" - why
>>> not?)

>Because it is *simpler* to treat it as a phoneme?

Well, this is what I said (see below).  You are arguing against
what Gábor Sándi said here (although he obviously presented it as
a straw man argument that he didn't really believe in), not
against what I said.  You are in fact agreeing with what I said.

>[Though in this case there is actually a reasonable argument for
>allophonic status, given its *very* limited distribution - and
>the fact that trying to use /ng/ to coin arbitrary new words
>results in words that do not sound like English].

I don't know that words like 'zing' or 'humdinger' sound
particularly un-English.  Expressive or imitative, yes; obviously
not ancient, yes -- but un-English, not particularly.  The
limited distribution is more of a problem.  English also has what
could be considered a palatal allophone of n [ny] whose
distribution is even more limited (doesn't occur in final
position).  It could be said to be present initially in 'nyah'
(usually reduplicated 'nyah nyah'; a cry of derision, not to be
confused with the Indian school of philosophy 'Nyaya'), but
otherwise only in loans (e.g, 'gnocchi', usually with Anglicized
pronunciation [n], but pronounced [ny] by the cognoscenti).  In
medial position it is still not common ('opinion', 'pinion'
'senior', etc.; also in more recent loans from French [<gn>],
Italian [<gn>], Spanish [written with enye, n with tilde] and
Portuguese [<nh>]), but it does contrast with /n/ ('cannon' or
'canon' vs. 'canyon').  Despite this contrast, I have never heard
anyone suggest that [ny] is an English phoneme and usually it is
not even treated as an allophone of /n/.  The palatal is simply
analyzed as a separate phoneme /Y/.

>>before the agentive suffix '-er'.  Stem-final /g/ goes to zero
>>immediately and (nearly) all derivative suffixes (agentive '-er',
>>'-ing', '-ish', '-ly', '-y', etc.) are added to this base.  All
>>that is needed then is a rule that triggers the restoration of
>>/g/ before the comparative '-er' (and for completeness, ...

>I would prefer to reorder the rules so that the comparative
>suffix is added prior to the application of the rule that deletes
>the /g/.  I do not like rules that restore.

But you only need the rule if you treat /ng/ as an allophone.
Otherwise, it is easier to write the /g/ into the rule for the
comparative and superlative as an allomorph (i.e., add -er, -est
[-ger, -gest after /ng/] to the adjective stem).  Another reason
for treating /ng/ as a phoneme.

>>All things considered then, it is probably simpler to operate
>>with /n/ and /ng/ as different phonemes because this does not
>>require the zero allophone of /g/ to maintain their allophonic
>>status *and* there is no indication of a morphophonemic
>>alternation between [n] and [ng]

>The main problem with that is that this supposed phoneme cannot
>be used arbitrarily.  A made-up 'ngib' is NOT a possible English
>word contrasting with 'nib'.

Now you are arguing against what I said as well as what you said
just above.  But this is a phonotactic rule that excludes /ng/
from initial position and has nothing to do with the
contrastiveness of /n/ and /ng/.  English just doesn't allow /ng/
in initial position.  Of course, phontactical rules are relaxed
for foreign loans.  Thus there are loans from languages like
Maori with initial /ng/ (such as the name of the famous mystery
writer, Ngaio Marsh) (there are probably a number from Vietnamese
as well now, but I don't know any; Ross Clark could probably
explain this in detail).  So a foreign borrowing 'ngib' could
contrast with initial /n/.  But limited distribution doesn't
disqualify a sound from being a phoneme.  Many languages strongly
restrict the allowable final consonants (Finnish for example
allows only /n/, /s/, and /t/ as final consonants).  This does
not affect the phonemic status of the disallowed consonants.  You
couldn't have a made-up '*sfit' to contrast with made-up 'snit'
in English either (although you have a borrowed 'sphere' to
contrast with 'sneer') but this doesn't show that /f/ and /n/
aren't phonemes.  It just shows that phonotactic rules have
higher precedence.

>[Interestingly, when playing with this "word" I tended to "hear"
>it as 'gib' (hard 'g') - which supports a 'silent g'
>interpretation].

I can't see why.  If there is a [g] sound that has come from a
zero allophone of /g/, that means that the silent g is restored
when [ng] is in initial position and that can't be right because
you don't like rules that restore.  More likely, you simply are
unpracticed at using initial [ng] and are getting some voicing
of the velar from the following vowel.

Bob Whiting
whiting at cc.helsinki.fi