minimal pairs are not always there

[Robert Whiting]

On Thu, 13 Apr, Dr. John E. McLaughlin <mclasutt at brigham.net> wrote:

>[Larry Trask]
>This is *a* method [finding minimal pairs] of establishing phonemes.
>But it is not *the only* method of establishing phonemes.  If the
>distribution of two sounds cannot be stated by rule, then they can't
>be assigned to a single phoneme.

>[Robert Whiting]
>I would say that even a minimal pair is not a sufficient condition to
>establish two sounds as separate phonemes.

>[Pat Ryan] I would have to say that you are wrong.  There is no
>phoneme in any language which has not been established as a component
>of a minimal pair.

>[Me]
>This is not true, Pat, although I'm not ready to throw minimal pairs
>out with the bath water as Robert seems to be.

I must agree that what Pat says is not true, but I also must say that
I am not ready to throw minimal pairs out either.  Minimal pairs are
a useful heurisitic, but that is all they are.  While enough minimal
pairs may be convincing in themselves if they occur in a variety of
different environments, a minimal pair in itself may be evidence for,
but not necessarily proof of, a phonemic distinction.  I hope you can
see the difference.  Because of the definition of phoneme, you really
have to consider the entire system before you can declare phonemicity.

>I think that you truly have to consider teeth/teethe to be a minimal
>pair.

I think that you truly have to consider both 'teeth' and 'teethe' as
morphophonemic variants of 'tooth'.

>Historically, yes, these two forms were not (the 'e' on the end of
>teethe was a phonetic element which put the voiceless /th/ in a
>voicing environment, but synchronically, there is no distinction
>between the two except for the final voicing of th/dh (the lengthening
>of [i] in 'teethe' is due to the voicing of dh, it does not cause the
>voicing).

Historically, this is nonsense.  the lengthing of [i:] in 'teethe'
is a matter of stress.  It is a matter of vowel quantity, not vowel
quality.  Both 'teeth' and 'teethe' have [i:] and if the ending is
not stressed, both have the same vowel quality.  The [i:] in both
'teeth' and 'teethe' is the result of umlaut caused by the addition of
the plural ending (beginning with '-i') and the verbal suffix
(beginning with '-j'; exactly the same change that took place in
'doom' - 'deem'), respectively.  The fact that many speakers introduce
this additional distinction by stressing the ending of the verb
suggests that they do not consider the [th] - [dh] distinction to be
sufficient (i.e., they do not consider it phonemic).  If 'tooth' had
not preserved its umlaut plural (i.e., if 'tooth' [+ plural]  -->
*'tooths'), the question wouldn't arise.

The 'e' on the end of 'teethe' is secondary, to indicate that it is
voiced.  The form was probably 'te:thian' when the voicing took place.
The loss of final shwa in English (cf. Eng. 'help', Ger. 'hilfe')
freed <e> for such secondary functions.

But synchronically, there is a difference between [th] - [dh] pairs.
[dh] always indicates the verb.  Very rarely, there is no distinction
between verb and substantive ('tithe' [n.] / 'tithe' [v.]; 'smooth' /
'smoothe'), but this can hardly be a basis for claiming that [th] and
[dh] are different phonemes.

Allophonic splits occur when a conditioning environment causes a sound
to change in that environment but not outside it.  There are those who
claim that when the conditioning environment is subsequently lost,
then the allophones become phonemes because they are no longer
predictable from the *phonological* environment.  But I would maintain
that if the distinction between the sounds is still predictable on the
basis of the grammar, then the sounds are not fully phonemic because
they can still be predicted from the morphological/grammatical
environment.  Phonemes should not let you predict meaning.  If they
do, then phonemes become a unit of meaning (as some, like Pat, have
tried to claim), and this violates the rule of duality of patterning
(sounds express meaning, but the sounds themselves have no inherent
meaning). Phonemes should tell you nothing about the meaning of a
word, whether it is grammatical meaning or lexical meaning (there is
a large grey area here, that is not yet well researched, called sound
symbolism).  If they do, then you have Pat's Proto-Language(C).

>But there are several good minimal pairs in (at least American)
>English for th/dh--ether/either, thigh/thy, wreath/wreathe,
>sooth/soothe, etc.

'ether' [borrowed word] - 'either' [native word]
'thigh' [non-pronoun] - 'thy' [pronoun]
'wreath' [noun] - 'wreathe' [verb]
'sooth' [noun] - 'soothe' [verb]

Now these distinctions wouldn't mean much in themselves, but when you
consider that these distinctions based on the presence of [th] or
[dh] are *always* valid, then you get the impression that as minimal
pairs they aren't worth much.  If you sometimes found intervocalic
[th] in the [native word] column or initial [dh] in the [non-pronoun]
column or final [th] in the verb column, then one would be justified
in claiming a phonemic distinction between [th] and [dh].  Otherwise
we can say that all words that have [th] in a voiced environment will
be loans, any word that has [dh] in initial position will be a pronoun
or a deictic word and any word that has [dh] in final position will
be the verb part of a noun-verb pair.

Beyond that, it is interesting to note that each of these "minimal
pairs" is a special case of one sort or another.  For 'ether' -
'either' there is a dialectal variant of 'either' that clearly
distinguishes it from 'ether'.  Both 'thigh' and 'thy' have long and
distinguished phonological careers and the fact that they have ended
up with only [th] and [dh] to differentiate them phonetically is
simply a historical accident.  Were one of them not a pronoun, they
would have simply ended up as homynyms (like 'knight' and 'night' or
'right', 'rite', 'write', and 'wright').

'Wreath' and 'wreathe' are also a special case becuase the actual
verb derived from 'wreath' is 'writhe'.  'Wreathe' is a back formation
from 'wreath' influenced by a ME variant 'wrethen' of 'writhe'.
'Sooth' and 'soothe' is one that I don't have an explanation for.  By
all rights (and in parallel with 'tooth' - 'teethe') the verb should
have been *'seethe' (<-- OE 'so:dhian' [v.] <-- 'so:th') but it isn't.
Of course, there is already a verb 'seethe', but it didn't have this
form in OE ('se:odhan').  Something happened to block the umlaut in
'soothe'.  I'd be interested to learn what it might have been.

>However, because of the very complex morphophonemics of Central Numic
>and the historical changes that have further obscured them in
>Comanche, this language is full of pairs that look very much like
>minimal pairs on the surface, but are not.  For example, [papi] 'head'
>and [pavi] 'older brother' look very much like a minimal pair.
>However, they represent /pa=pi/ and /papi/ respectively.  (The = is a
>phoneme in Comanche that prevents the lenition of a following stop.
>It is fully justified on morphophonemic grounds without relying on the
>historical presence of /n/ in Panamint and Shoshoni which is cognate.)
>There are a bundle of these:  [ata] 'different' /a=ta/ versus [ara]
>'uncle' /ata/, etc.

Fascinating.  Please, sir, what is the phonetic realization of this
phoneme [=]?  Oh, I just realized -- it can't have a phonetic
realization or else [papi] and [pavi] wouldn't seem to be a minimal
pair.  It just blocks some normal phonetic change.  I'm sorry, John,
but this looks like a device to create a phonetic environment to
explain why some stops don't undergo lenition when the conditioning
environment that prevented it has been lost historically.  I'll tell
you what:  Let's assume that English has a phoneme (let's call it [=]
just for consistency) that prevents an intervocalic dental spirant
from being voiced.  Now let's insert this phoneme in a word like
'ether' which shows an unvoiced intervocalic dental spirant /i:=ther/.
Good -- now we no longer have a minimal pair 'ether' - 'either'.  Now
let's assume that English inserts this phoneme in all loanwords that
have an unvoiced dental spirant in a voiced environment.  Voila -- a
phonetic environment that explains why loanwords have unvoiced
intervocalic [th].  Now all we need is a rule that says /=th/ -->
[dh] /__ m# and all intervocalic [th] in English is accounted for by
phonological rules.  Hey, this is fun.

>On the issue of requiring minimal pairs,
<snip of lots of good stuff>
>With all the possibilities of root structure in Panamint, there just
>aren't many minimal pairs.  The phonemic inventory has had to be
>determined in other, more subtle ways, such as using permissible
>initial segments, morphophonemic alternations, etc.

I agree with this completely.  Minimal pairs are a heuristic that is
useful for finding possible phonemes.  When you don't have them, you
have to use something else.

>[Robert Whiting]
>The distribution by rule takes precedence.  Take the English minimal
>pair: 'thigh'  /  'thy'.  Most people would not insist on phonemic
>status for both [th] and [dh] in English on the basis of this minimal
>pair (although some would doubtless claim that there has been a
>phonemic split similar to what occurred with /s/ and /z/).

>[Me]
>I disagree with Robert on this one.

No problem.  I have no vested interest in any theory that either
requires or doesn't require [th] and [dh] to be separate phonemes.
I'm just looking to find out what the evidence is and how the evidence
proves it one way or the other.

>The evidence for establishing /th/ and /dh/ as separate phonemes is
>no worse than that for establishing /zh/, /ng/, and /oj/ as phonemes
>(depending on whether or not one considers diphthongs to be on the
>same footing as other phonemes in the language).

I think it is quite a bit worse.  How many of /zh/, /ng/, or /oj/
occur only in certain classes of words or only as morphophonemic
alternants?  Show me that in all the words where /zh/ occurs that /zh/
limits or restricts the meanings that it can have and I will grant you
the point.  It is said that /ng/ only occurs in word final position,
but even this is not true (compare 'finger' - 'singer').  These may be
difficult to establish as phonemes, but there is solid evidence:
places where these sounds provide the only contrast and cannot be
predicted by rule.  Show me the same for [dh] and I will grant you the
point.

>All phonemes do not have to be equally common, nor the evidence
>equally impressive. Patterns of morphophonemic, environmental, and
>unpredictability factors all point toward them being separate
>phonemes.  While the evidence for separating /th/ and /dh/ is not as
>overwhelming as the evidence separating /s/ and /z/, it is still
>enough to compel a separation on synchronic grounds.

While I agree with the first statement, the second does not follow
from it.  Patterns of the occurrences of [th] and [dh] overwhelmingly
show complementary distribution of the sounds and either lack of
contrasts or predictable morphophonemic alternation.  Morphophonemic
alternations may be the beginnings of phonmeic split, but until the
sounds are used as oppositions outside of the morphonemic environment,
I don't think it is legitimate to construe them as separate phonemes,
even if the original conditioning environment has been lost.

Synchronic grounds don't really have anything to do with it.
Synchronic rules are still rules and most will agree that synchronic
rules recapitulate diachronic rules.  They don't always have the same
basis, but by and large they have the same results.  The diachronic
rules voiced [th] in verbs because it was in a voiced environment.
Through the erosion of the infinitive ending, the voiced [dh] ended up
word final while the substantive retained word final [th].  The
synchronic rule simply says that you make a verb out of a substantive
with final [th] by voicing the [th].

The synchronic rule doesn't reduplicate the diachronic events, but it
does have the same result.  And the synchronic rule is what someone
who is learning the language has to learn.  And through it he knows
that any native English word that ends with [dh] will be a verb and
that it should have a corresponding substantive ending in [th] (with a
couple of exceptions as noted above).  Now if there were other English
words ending in [dh] that weren't verbs (as in 'cliff' vs. 'five' as
opposed to 'life' vs. 'live'), one would have a case for the
phonemicity of [dh] as one has a case for the phonemicity of [v].

But as long as final [dh] occurs only in verbs and the only way that
it contrasts with [th] is through a substantive with the same root, I
don't see a case for phonemicity.  It is just distribution by rule.

>[Pat Ryan]
>What in Heaven's name is a "voiced environment"? What is
>environmentally voiced in 'bathe' as opposed to 'bath'?

>[Me] Robert's referring to a historically "voiced environment".  This
>is not appropriate evidence for synchronic phonemicization unless the
>phonological or morphophonological rules are still productive.

Thanks for trying to explain to Pat, John, but although historically
the voicing of [th] in verbs was phonologically conditioned and
resulted in an allophonic split, loss of the conditioning environment
does not make the synchronic rule any less valid.  The synchronic rule
is still there.  One even notes that the rule for verbs has been
maintained against the devoicing of intervocalic [dh] in the ordinal
numbers once they had lost their voiced environment.  The example of
'tithe', which was once simply the same as the ordinal 'tenth' (OE
'te:odha' [n.] - 'te:odhian' [v.] shows this clearly.  If the voicing
rule for verbs were no longer valid, the verbs would end in [th] just
as the ordinal numbers do.  The rule may not be productive any longer
in the sense that new verb-noun pairs are not formed by this rule, but
that doesn't mean that the synchronic rule doesn't exist.

Now one could always invent a dummy, non-phonetic phoneme as was done
in Comanche that can be inserted to show where the conditioning
environment was lost (and graphemically, this is the purpose of the
final 'e' on the verbs in English) but this is obviously just a device
to provide a "phonological" explanation for why some intervocalic
stops didn't undergo lenition and thus maintain the allophonic status
of [p, v] without having to resort to a diachronic explanation.  I
don't see that there is really any difference in the outcome.

>[Robert Whiting]
>Thus it is not only as Larry says "If the distribution of two sounds
>cannot be stated by rule, then they can't be assigned to a single
>phoneme," but also 'If the distribution of similar sounds can be
>stated by rule, then they can't be assigned to separate phonemes.'

>[Me]
>The distribution of /th/ and /dh/ cannot be determined by the
>assignment of a PHONOLOGICAL rule.

Let me see if I have your take on this straight.  Are you saying that
allophones automatically become separate phonemes when the
phonological conditioning environment that maintains their allophonic
identity is lost?  And that they are phonemes even if they never
contrast in an environment that can't be predicted, so long as the
basis for predicting the environment is not phonological?

Now if this is what you believe, I am not saying that it is wrong.  It
is more a matter of how one sees the interaction of the various parts
of language.  If the determination of phonemes is an entirely
phonological process, then it is quite correct.  But if other areas
or language can affect phonology, then it is not necessarily so.
Again, it is a question of how much one area of language can affect
the others.  A matter of where you draw your lines and how you make
your definitions.

There are a number of issues about phonology, morphology, and lexicon
that are too lengthy to go into here, but in a summary fashion let me
say that the basic function of language requires certain relationships
between phonology and morphology.  Languages express meaning
(morphemes) through sounds (phonemes).  But the sounds, taken in
isolation have no inherent meaning of their own associated with them.
For this reason, phonemes can indicate differences in meaning, but are
not units of meaning.  Sounds that can only indicate a functional
difference (a difference in grammatical meaning) when contrasted with
a similar sound should not be considered a fully functional phonemes
in the environment in which it can only mark this difference. Such a
sound can be called a morphophoneme to indicate that it is restricted
to contrasts between morphological variants of the same word.  But if
your definition requires that any sound that contrasts with another to
produce a different meaning is a phoneme, then morphophonemes must be
considered phonemes.

>There is an archaic MORPHOPHONEMIC rule (make a noun into a verb by
>voicing a final /th/), but this is no longer productive, e.g.,
>'path'/*'pathe', 'math'/*'mathe'.

The fact that it is no longer productive does not make it any less of
a rule.  It is not an archaic rule.  It is a rule that exists in
modern English.  It just isn't a productive rule.  If it were no
longer a rule the verb from 'breath' would be 'breath'.  It is no
longer a productive rule because the infinitive is no longer formed in
the same way.  There was no archaic rule that said make a verb out of
a noun by voicing a final spirant.  The spirant was voiced because it
occured in a voiced environment in the infinitive.  When the
distinctive form of the infinitive was lost (i.e., the conditioning
environment was lost), the voiced/voiceless distinction in final
position was maintained to distinguish verb from noun and that is
where the synchronic rule came from.  But this is still a
morphophonemic distinction because this is the only place that it
occurs and it is regular and predictable.

>Even the intervocalic voicing of /th/ isn't always productive, e.g.,
>path [th] and paths [dh], but path's [th].

The (morphophonemic) voicing of spirants that occurs before the plural
marker [s, z] is regularly neutralized before the possessive singular
(it is a different rule):

     sing.         pl.           poss. sing.
   thief         thieves        thief's
   life          lives          life's
   sheaf         sheaves        sheaf's
   path [th]     paths [dhz]    path's [ths]
   wreath [th]   wreaths [dhz]  wreath's [ths]

The voiced - voiceless opposition of singular and plural is also
regularly neutralized in loan words

   chief         chiefs  (not *chieves)
   faith [th]    faiths [ths]
   math [th]     maths [ths]

and occasionally in native words

   death         deaths [ths]

and sometimes both occur

   hoof          hoofs ~ hooves

But there is no unpredictable contrast.

>These two phonemes are NOT predictable, cp. ether/either and
>thigh/thy.  No phonological or morphophonological rule can account for
>these pairs.

I just invented a phonological explanation for 'ether' - 'either' so
it is no longer a minimal pair (just like Comanche [papi] and [pavi]
aren't a minimal pair, although presumably these two words are both
native in Comanche).  'Thigh' - 'thy' is just a historical accident.
You say that synchronicity demands that [th] and [dh] be separate
phonemes and that non-productive forms cannot be used for synchronic
phonemicization and then you give me evidence based on loanwords and
obsolete forms.  Not good enough.

>Using semantic criteria ('if it's a pronoun, then') doesn't cut it in
>a theoretical sense.

But it's very good for describing the evidence.  Perhaps the theory
needs some reworking so that it accounts for the evidence better?

Bob Whiting
whiting at cc.helsinki.fi