thy thigh etc.

[Robert Whiting]

On Thu, 16 Nov 2000 CONNOLLY at LATTE.MEMPHIS.EDU wrote:

>On Tuesday Robert Whiting responded at great length to my post

If you thought that was "great length," you ain't seen nothin' yet.

>on the English interdental fricatives, in which I attempted to
>draw some lessons from the German non-distinction between [x] and
>[c,] in _Kuchen_ 'cake' vs. _Kuhchen_ 'little cow', which I
>analyzed as /ku:x at n/ vs. /ku:+x at n/.  The gist of his argument
>seems to be that if it is fair to use phonological boundaries
>(which he identifies as always ultimately morphological
>boundaries), then it is also fair to use morphological and
>lexical information to account for the distribution of the
>intedental fricatives [T] and [D] in English.

A couple of points.  First, I am not trying to use morphological
and lexical information to account for the distribution of
interdental fricatives in English.  I am trying to use it to call
into question the evidence that is used for the distribution of
these sounds, particularly the alleged contrasts.  This is
admittedly a subtle point -- so subtle indeed, that many people
seem to have failed to grasp it.  My arguments are not aimed at
demonstrating that [T] and [D] are not separate phonemes in
English.  I have no particular reason to believe this.  Rather my
arguments are aimed at determining whether it is valid to accept
a contrast as being based on phonology when there may be another
basis for the contrast.  So whether or not [T] and [D] are
phonemes in English is not the issue.  The issue is whether
contrasts that can be shown to be based on some other feature
should be available for use as evidence of phonemic contrast.
Specifically it is about whether everything that looks like a
minimal pair (like [ku:c,en] and [ku:xen] should actually be
considered a minimal pair for purposes of determining phonemic
status.

Second, I do not question the importance of boundaries in
phonology.  I merely say that boundaries are not segmental
phonemes.  Nevertheless, they sometimes provide the only way
to distinguish meaning.  Pairs like 'that's tough' [Dæts't at f]
and 'that stuff' [Dæt'st at f], or 'the Trojan's trumpet' and 'the
Trojan strumpet' (Walt Kelly [creator of Pogo] had a great one
once based on the near identity of 'a tax on frogs too' and
'attacks on frog stew'), if said in an offhand manner at normal
speed, can be difficult for a listener to distinguish.  A voice
spectograph, however, will pick up the differences.  While the
only difference in segmental sounds may be the [t'] in 'tough'
and the [t] in 'stuff', stress onset will register in the
spectograph (as a stronger [s] in 'stuff', and so on) showing
that [t'] and [t] are not phonemes.  In this case, juncture will
correlate with stress onset and one will not be distinguished
from the other phonetically.  So both juncture and stress onset
are phonologically important in English.  But I think that they
should be considered qualitatively different from segmental
phonemes (and from each other).  Boundaries can be considered
"transition phonemes" (after Trager) and stress (as well as
accent and tone) when significant are commonly considered
"suprasegmental phonemes."

Third, I consider your statement "... phonological boundaries
(which he identifies as always ultimately morphological
boundaries) ..." an unwarranted generalization.  We were talking
about juncture which is a specific type of boundary where two
morphemes come together and you have extended this to
phonological boundaries in general.  Syllable boundaries can be
phonological but not necessarily morphological (in fact,
phonological syllable boundaries frequently cut across
morphological boundaries, e.g., 'Halloween', 'formation',
'telepathy', etc.); utterance boundaries are inherently both
phonological and morphological conditions (although if one stops
speaking in the middle of a word, it is not necessarily a
morphological boundary).

So if you have examples of phonological juncture (but not of
other phonological boundaries) that are not ultimately based on
morpheme boundaries, bring it out and we will talk about it (and
to avoid having to have a pointless discussion let us exclude
words that have been folk etymologized or reanalyzed so that
there is now a perceptual morpheme boundary where one may or may
not have originally existed (e.g., furbelow, mangrove, mongoose,
woodchuck, outrage).  But if you don't, then I think it is an
unfair characterization to claim that I ultimately identify all
phonological boundaries as morphological boundaries.  I only
identify those phonological boundaries that depend on the
existence of a morpheme boundary (including word boundaries, real
or perceived) for their existence as ultimately based on
morphological boundaries.

>He overlooks a significant difference between morphological ->
>phonological boundaries and other types of morphological
>information: boundaries can be located precisely between
>morphemes.  It is therefore to show them in phonological
>representations, for their position (and type) seems to matter
>for the phonetic realization.

Not so much overlooks as doesn't give it much weight.  And I
would probably agree with your last sentence if I were sure what
the missing word after "therefore" is (possible?, necessary?,
useful?).  In any case, I agree that they seem to matter for the
phonetic realization (in some cases).  But phonetic realization is
not the same as phonemic analysis.  Phonetics deals with the
actual sounds of speech.  Phonemics deals with the relations of
these sounds, and particularly with the relations of these sounds
in the perception of the speakers.  Phones are realia (capturable
with a voice spectrograph); phonemes are abstractions determined
by an analysis of the distribution of the sounds.  A voice
spectrograph will record the difference between aspirated [t']
and unaspirated [t]; but it won't tell you whether they are
phonemes or not.  [t] and [t'] should appear the same on the
voice spectrogram whether they are phonemic in the speaker's
language (e.g., Chinese) or allophonic (e.g. English).

So I am willing to stipulate that anything that leaves a
footprint in a voice spectrogram is part of the phonetic
realization of the language.  This includes various kinds of
boundaries, which, as I said before, often affect things like
pitch, duration, and pause, as well as stress patterns.
But I am not willing to stipulate that everything that appears
in a voice spectrogram is a phoneme.  Otherwise, why bother with
looking for minimal pairs to make a phonemic analysis?  Just use
a voice spectrograph and it will tell you what the phonemes are.
It doesn't work that way.  What shows up in a voice spectrogram
is the phonetic realization ([...]).  The phonemic analysis of
this phonetic realization (/.../) doesn't show up there.

Which is why I remain a little skeptical of the dictum that
things that don't show up in the phonetic realization can't
affect phonemic analysis since things that do show up in the
phonetic realization don't ipso facto determine phonemicity.

As an example let us consider the effects of devoicing of final
stops in German.  Now there is no difference in the phonetic
realization of 'Rat' "parliament" and 'Rad' "wheel," both being
[ra:t].  Most speakers of German, however, will insist that there
is a difference between the [t] of 'Rat' and the [t] of 'Rad'.
The first is a "real" [t] while the second is a variety of /d/
(occurring before a word or syllable boundary).  One can either
accept the perception of the speakers and say that 'Rat' and
'Rad' are phonemically distinct even though they have identical
pronunciations and phonetic environments, and the first is /ra:t/
and the second is /ra:d/, or we can insist that the phonetic
reality is correct and that both are phonemically /ra:t/.  Here
we are dealing with two levels of phonology, a surface level and
an underlying level.  Phonetic realization (as in a voice
spectrogram) can only recognize the surface level, but this in
not necessarily the determiner of phonemicity.

>But information such as "native/foreign" or "content/function
>word" cannot be so located and cannot be described in any real
>sense as "morphological", but rather as lexical and (for the
>latter, at least) "syntactic".

All true.  Except, of course, that different people will see the
different categories differently depending on how many levels of
linguistic structure they organize language into and what the
relationship between morphology and syntax is in their particular
model (for many people, morphological rules are just a subset of
syntactic rules).  As for whether "content/function word" goes
under syntax or not depends on where one puts word classes (parts
of speech) since this is a higher level classification of word
classes (pronouns, conjunctions, and prepositions are generally
function words while nouns, verbs [excepting auxiliary verbs and
modals], and adjectives [excepting quantifiers] and adverbs
[excepting demonstratives and intensifiers] are generally content
words).  I will agree, though, that most people would put word
classes under syntax (although many others consider them "lexical
categories").  But no matter how you divide language up to study
and analyze it, it is still a completely integrated system.  The
various parts interact with each other and in so doing, influence
each other, especially as this interaction is mediated by human
behavior, which, despite man's being the rational animal (by his
own definition), is often irrational and unpredictable.

But even if what you say is true, it doesn't prove that such
information doesn't affect phonemics.  That is to say that even
"junctures" of the sort that are freely used in all phonemic
descriptions do not necessarily (or even generally) have uniquely
or precisely identifiable reflexes in the utterance.  Therefore,
the fact that information cannot be precisely located in the
phonetic representation does not prove that it does not have
phonemic significance.

>There is no inconsistency in saying that boundaries are
>phonologically relevant while other information is not.

I would agree, so long as "phonologically" refers to "phonetics"
(i.e., what you can see in a voice spectogram).  But if you
include phonemics in phonology (as most people do, one way or
another, some exclusively), then I'm not sure that one can be
apodictic about what is relevant and what isn't.  Since a voice
spectograph can't tell what is a phoneme and what isn't, it
doesn't necessarily follow that only information that can appear
in a voice spectogram can be relevant to what is a phoneme and
what isn't.

But there is a lot of other inconsistency in what you have said.
First, I object to the inconsistency of using essentially
identical evidence to both defend and refute the same
proposition.

You said "It is therefore unlikely, though possible, that German
could coin new words in which [x] and [c,] would (appear to)
contrast in some other environment, ..." and claimed that because
this contrast is "unlikely," then [c,] and [x] are not phonemes
in German.  By implication (if not directly stated), since
English speakers are "likely" to coin words contrasting [T] and
[D], then "... the two are searate phonemes, and have ben so for
some time" despite the fact that they have failed to do so.

But "likely" and "unlikely" are not part of the original
proposition.  The original proposition was that sounds that the
speakers of a language can distinguish and could use
contrastively are phonemes in that language.  This is purely
mechanistic, while injecting "likely/unlikely" introduces a
subjective element because "likely/unlikely" are not absolutes
but are the ends of a sliding scale.  According to the original
proposition, a language can have no allophones that speakers can
distinguish and could use contrastively.  All such sounds must be
phonemes whether they are actually used contrastively or not.
Therefore according to this proposition, [c,] and [x] must be
phonemes in German because the speakers of the language can
distinguish them and could use them contrastively.  "Likely" and
"unlikely" don't come into it in the original formulation.

Indeed, had the original formulation been "sounds that the
speakers of a language can distinguish and could use
contrastively are *likely* to be phonemes," there would have been
no objection.  This creates no categorical imperative.  Such
sounds are likely to be phonemes, but, as the case of German
[c, x] shows, need not be.

So you can't use the original proposition to show that [c,] and
[x] are allophones in German and also to show that [T] and [D]
are phonemes in English.  According to the original proposition,
both pairs must be phonemes.  The fact that [c,] and [x] are
demonstrably allophones with no evidence of contrastiveness in
German clearly disproves the original proposition (if the
original proposition has a test for falsification).  Otherwise,
it has to be claimed that in some dark recess of the
German-speaking mind, not yet discovered and investigated, [c,]
and [x] are considered phonemes by German speakers because they
can distinguish the sounds and could use them contrastively.

In your second piece of inconsistent treatment, you said "...
Kuchen : Kuhchen form a subminimal pair: the vowel of the latter
is clearly (but non-distinctively) longer than the former,
indicating that it was in root-final position, before suffix
-chen" and then you said "True, in my pronunciation, the stressed
vowel of _either_ is longer than that of _ether_ -- but this is
in line with vowel length before other voices vs. voiceless
consonants."  In these two statements you have treated two very
similar phonological cases in completely different ways.

In the first, you have attributed the lengthening of the vowel
entirely to the presence of the morpheme boundary, without making
any allowance for the articulatory fact that you may get a
lengthened vowel when a palatal consonant follows a back vowel
(after all, something has to be going on while the tongue is
repositioning itself for the palatal fricative, which it doesn't
have to do [or not so much] for the velar fricative).  In the
second, you have attributed the lengthening of the vowel entirely
to the normal, purely phonetic and fully automatic, lengthening
of vowels before voiced consonants, without making any allowance
for the presence of the morpheme boundary between the vowel and
the consonant.  And yes, there is a morpheme boundary in 'either'
(OE æ:ghwæðer > æ:gðer > ME aither, either < *a+gi+hwæðer),
although most speakers probably won't recognize it today because
so much phonological activity has taken place around this
boundary.  The 'gi-' in this word is the collective prefix that
also appears in English 'enough' (< OE geno:h; cf. German
'genug').  Now it is possible that you don't realize that this
boundary is there, in which case you can't be blamed for
intentionally failing to mention it, and if you don't realize
that it is there, one can hardly expect a naive native speaker
to.

But when you make the point that the difference between 'Kuchen'
and 'Kuhchen' is that the latter has a morpheme boundary and the
former doesn't and then ignore the fact that one of the
differences between 'ether' and 'either' is that the latter has
a morpheme boundary and the former doesn't, this is inconsistent.
Now, admittedly, the morpheme boundary in 'Kuhchen' is
transparent to speakers of German ('Kuhchen' is not likely to be
a separate lexical item in German, but will always be recognized
as noun + diminutive suffix) while the morpheme boundary in
'either' is not ('either' is a separate lexical item in English
and it is doubtful if a native speaker would realize that
'either' contains 'whether' [although it is still possible that
the '-ther' might be perceived as a morpheme]).

But this just raises the question of whether morpheme boundaries
are only phonological boundaries if the speakers realize that
they are present (i.e., only clearly productive morphology
produces phonemic boundaries whereas frozen or lexicalized
morphology does not).  Another way of posing this question is
whether we are justified in including a morphological boundary in
the phonological representation just because we know that there
is a morphological boundary there, or are we only justified in
including a morphological boundary when we think we can detect
one in the phonetic realization.  This last may be true, but I
don't think it can be taken for granted.  And if it is true, then
a lot of morphological boundaries would have to come out of
phonemic representations.

Besides, in much the same way as speakers might perceive the
'-ther' of words like 'mother', 'father', 'brother' and conclude
that it is a separate morpheme identifying family members, it is
not impossible that speakers could subconsciously recognize the
'-ther' of words like 'either', 'other', 'whether', and 'nether'
as a separate morpheme (which, in fact, it is) and conclude
(correctly) that there is a morpheme boundary in 'either' (even
though it is only perceptual, not productive).  The perceptions
of speakers do not necessarily correspond to linguistic reality.
This is the very essence of folk etymology.  The human brain
looks for patterns -- in fact, it is essentially a pattern
detecting device.  But the fact remains that 'ether' is
monomorphemic and 'either' is not (even if speakers don't realize
it consciously).

So I consider it inconsistent to say that [ku:x at n] and [ku:c, at n]
are subminimal pairs because [ku:c, at n] has a morphological
boundary in it (indicated by a slight lengthening of the vowel
before this boundary which may or may not be attributable to the
boundary) and hence can be analyzed as /ku:+x at n/ while claiming
that [i:T at r] and [i:D at r] are true minimal pairs even though
[i:D at r] has a morpheme boundary in it (with a slight lengthening
of the vowel precisely before this boundary) which, while not as
obvious as the morpheme boundary in /ku:+xen/, may still be
perceivable by speakers.  So if it is possible to analyze
'Kuchen' and 'Kuhchen' as /ku:x at n/ and /ku:+x at n/ then it is also
possible to analyze 'ether' and 'either' as /i:T at r/ and /i:+T at r/.

Finally, you said "And there's no difference in _thigh_ vs.
_thy_."  As you would say, "This is simply wrong." :)  'Thy' is
part of a paradigm; 'thigh' is not.  Let's look at part of the
*synchronic* English personal pronoun system (if we are going to
consider 'thy' as synchronically part of English as everyone has
been; after all, if it isn't, then there is no contrast (even if
only apparent) between initial [T] and [D] in modern English):

        1st person sing.           2nd person sing.

  nom.        I                         thou
  obj.        me                        thee
  pos. 1      my                        thy
  pos. 2      mine                      thine

  (pos. 1 and 2 were originally allomorphs of the same form; 1 is
  now used as a pronominal determiner while 2 is used as the
  possessive pronoun; 1 still has some pronominal functions,
  however)

Now if we look at the oblique forms we can see in all cases what
looks like a pronominal base ('m' in the 1st person, 'th' in the
2nd person) followed by a set of postbases that correspond to
case or, synchronically, function.  It hardly seems illegitimate
to analyze 'thy' as consisting of a monomorphemic, unisegmental
base [D] plus a postbase morpheme [ai] in the same way we could
analyze 'my' into a morpheme [m] plus a morpheme [ai].  The base
tells us the person and number, the postbase tells us the case or
function. That this is true of the English personal pronoun
system in general can be seen from looking at some other forms
where we also find a similar unisegmental base also followed by a
postbase:

             3.m.s.          2.pl.          3.pl.

  nom.         he             you            they
  obj.         him            you            them
  pos. 1       his            your           their
  pos. 2       his            yours          theirs
               hisn*          yourn*         theirn*

  (* substandard; originally by analogy to 1. and 2. singular)

Clearly then, the majority of the English personal pronouns can
be considered to have a firm pattern of a base that indicates
person and number and postbases that indicate case or function.
These pronoun bases could thus be considered as bound morphemes
that occur only in initial position.

Going back to 'thy' vs. 'thigh', 'thigh' is monomorphemic, while
'thy' clearly can be considered to be made up of two morphemes,
[D] and [ai].  Let us further say that the pronoun base 'th' is
always realized as [D] by native speakers.  If we now compare this
with the German pair 'Kuchen' and 'Kuhchen', we see again that
'Kuchen' is monomorphemic (although allowing for the possibility
that it may represent a stem 'kuche-' and  a noun formative 'n')
while 'Kuhchen' is clearly composed of two morphemes, 'Kuh'
"cow," also an independent word, and '-chen', a diminutive
suffix.  German speakers know that the diminutive suffix is
always realized with [c,] regardless of the environment of this
sound, and thus will not consider the [c,] of this suffix a
phoneme even when it appears to contrast with [x].  English
speakers may or may not know that the pronoun base 'th-' is
always realized as [D] and that therefore initial [D] is not
being used as a phoneme even when it appears to contrast with [T],
but it really is begging the question to claim that "there's no
difference in _thigh_ vs. _thy_."

The difference between 'Kuhchen' and 'Kuchen' is that there is a
morpheme boundary in 'Kuhchen' but not in 'Kuchen'.  The
difference between 'thy' and 'thigh' is that there is a
(perceivable if not actual) morpheme boundary in 'thy' but not in
'thigh'.  Now admittedly, the '-chen' morpheme is productive in
German and can be freely used to create new forms while the
morphemes in 'thy' or 'my' are not productive but exist in frozen
forms.  But I can't see that it would take much intuitive power
for an English speaker to look at the forms 'my' and 'thy' and
'me' and 'thee' and figure out, at least subconsciously, which
part of these words carries the pronominal information and which
the case or function information, especially if they know the
rest of the personal pronoun system.  There is reason to believe
then, that native speakers of English may be somehow aware of a
difference between the pronominal bases in [D] and other words in
English with initial [T].  So if it is possible to analyze
'Kuchen' and 'Kuhchen' as /ku:x at n/ and /ku:+x at n/ then it is also
possible to analyze 'thigh' and 'thy' as /Tai/ and /T+ai/.

Now a similar analysis could be done with the deictic base 'th-':

  far deixis 'th-'   near deixis 'h-'   relative 'wh-' (/hw/ > /w/)

      there               here              where
      then               *[OE henne]        when
      thence              hence             whence
      thither             hither            whither
      that                                  what

Again, there is a clear pattern of a unisegmental base carrying
the deictic information and a postbase carrying the function
information, and it is not stretching credulity to point out that
native speakers could notice, even if only subconsciously, this
pattern.  A similar pattern would have been strongly visible in
the early ME forms of 'the' (of which 'that' is simply the neuter
singular) when it was still declined for gender, number, and
case.

It doesn't take too much imagination to see how speakers of
English might make a connection of some kind, even if only
subconscious, between the nature of the pronoun bases and the
deictic bases.  Allowing for a possible case of analogy and one
of dissimilation to smooth up the rough edges, it is easy to see
how the speakers of English could very well consider the pronoun
bases in 'th-' and the deictic base in 'th-' as somehow, at least
functionally, related and qualitatively different from the
initial 'th-' in other words where it functions differently.  If
we want to follow the evidence, then it would seem that native
speakers of English consider that initial [D] is reserved for
these bases (in much the same way as German speakers realize that
the 'ch' of '-chen' is always [c,] even when following a back
vowel) and only [T] should appear in initial position in other
words that do not contain one of these bases.  In short, just
as German speakers realize that [c,] after a back vowel belongs
to the morpheme '-chen', so English speakers realize that initial
[D] belongs to the morpheme(s) 'th-'.

This is not to claim that this analysis represents linguistic
reality or that this is how these words should in fact be
analyzed (linguistically it would be better to analyze the
vocalic nucleus as the pronoun/deictic base and treat the 'th-',
etc. as prebases; this would account for 0 prebases in some forms
like 'I', 'us', 'it' rather than having to claim a 0 pronoun
base).  These apparent morphemes may just be chimeras (although
the very fact that initial [D] predicts meaning already suggests
strongly that it is a morpheme) and since the pronoun system is a
relic from an earlier stage of the language, speakers don't have
to deal with the forms on any productive level.  But the
potential for speakers to perceive them in this way is clearly
there.  Perceptions of reality can't change reality.  But
perceptions of reality can affect reactions to reality.  If
everyone believes that the earth is flat, it still isn't possible
to walk to the edge and jump off.  But if everyone believes that
the earth is flat, it may make people cautious about the
possibility of getting too close to the edge, and a good con man,
or even a well-intentioned do-gooder, could probably raise money
to put up a guard rail around the edge.

In short, the morpheme boundary of /ku:+x at n/ is productive while
the morpheme boundary of /T+ai/ is only perceptual.  But as
Sommerstein says (Modern Phonology, p. 100):

   ... both productive and perceptual factors are capable of
   being used by speakers to make intuitive generalizations, and
   of determining or conditioning linguistic change.

How, or even when, this perception may have come about, I'm not
sure; I am only saying that the evidence indicates that this is a
possible *synchronic* explanation of how initial [T] and [D] are
used by native speakers of English and, if this is so, that is
why there are no new coinings in which [T] and [D] contrast in
initial position.  The sound [D] is restricted to the pronoun
and deictic bases and these bases are not productive (in the
sense that they are not used to create new forms, but then there
are seldom new coinings of function words) in English.  The
contrast between initial [T] and [D] is simply frozen in the
contrast between the pronoun/deictic bases and other words with
initial [T] and hence any new coining in English will have
initial [T] unless it happens to be a pronoun or deictic word
because the [D] in these forms is a morpheme.  It seems to me
that this perception must be at least partially involved in the
observation that 'the' has invariant [D], where it presumably
exists because it is a lenition of [T] in a usually unstressed
form, while 'theology', which invariably has an unstressed [T],
has only [T].

As Herb Stahlke pointed out on 21 Apr 2000, the explanation of
initial [D] as lenition of [T] in function words is less than
satisfying because:

  ... there was no initial contrast until the function words,
  largely deictics and largely unstressed, laxed the initial /th/
  to /dh/.  However, there was a sizable set of content words,
  like "theology" that had initial unstressed syllables beginning
  with /th/, and none of these voiced.

This is unsatisfying because the main difference between function
words like 'the' and content words like 'theology' is that the
former are much more heavily used than the latter ('the' is the
most common lemma in the English language with 6,187,267 tokens
in the 200 million word BNC while 'theology' slips in at 5237
with 1098 tokens).  One must thus rely on frequency of use to
account for the lenition of one and not the other.  But if the
distinction is based, not on function words per se, but on the
perception of a 'th-' morpheme in some words (which because of
the nature of the 'th-' morpheme(s) just happen to be function
words), then this difficulty disappears.  One can then express
the shift of initial [T] to [D] with the rule

    [T] > [D] / # ___ +

which, since # (word boundary) and + (morpheme or formative
boundary) are (as we all agree) phonological (even if not always
phonetic) conditions, does not violate anybody's phonological
rules.  In this case, frequency of use can be used as a
motivation for the rule, but the trigger comes from the morpheme
(or formative) boundary.  Objections that the morpheme boundary
may not be real or that it doesn't show up in the phonetic
representation are groundless if phonology is based on the
perceptions of the speakers of a language and not mechanistically
on the phonetic realization.

If the relationship between initial [T] and [D] is as expressed
in this rule, it turns out that it is not based specifically on a
difference between function words and content words (once again
demonstrating that correlation does not prove causality) which
accounts for the fact that 'through', which is frequently used as
a function word, does not have initial [D].  What it lacks (at
least in the perception of speakers) is the deictic morpheme
'th-'.  The fact that the lenition is not a specific property of
function words also accounts for the fact that function words
like 'for' (rank 11 in the BNC) and 'so' (rank 58) do not show
initial voicing in English.  Finally, since the laxing of [T] in
these words is only motivated by frequency of use, not triggered
by it, the extension of [D] to stressed forms of these words does
not have to be accounted for by analogy.

This analysis also can be used as a basis for explaining the
anomaly that Herb Stahlke discussed in his posting as follows:

   An oddity is that the function word "thither", which is rare
   in contemporary ModE, has initial /th/ for all AmE speakers
   I've consulted.  American dictionaries regularly show /dh-/ as
   a second pronunciation, and British dictionaries I've checked
   either give only /dh-/ or give /th-/ as a second choice.
   Apparently Americans who know the word generally don't treat
   it as a function word.

Again, if the analysis of the distribution if initial [T] and [D]
in English given above is correct, the difference between
American [TID at r] and British [DID at r] is not based on the
perception of 'thither' as a function word in one dialect and not
as one in  the other.  Nor does the word lack the 'th-' deictic
morpheme since it is clearly part of the deixis pattern (thither,
hither, whither).  All that is needed to account for the pattern
is a rule in one dialect that blocks the shift of [T] to [D] if
the word already contains a [D].  Of course, it might also be a
dissimilation rule in the other dialect that operates after the
shift.  [Incidentally, my pronunciation is the American one, but
the word to me is archaic (to the point of obsolescence) and only
used by me in fixed expressions like 'hither and thither' or
'hither, thither and yon', and I only use these when I'm feeling
archaic, generally replacing them in everyday speech with 'here
and there' or 'here, there and everywhere'.  Typically, however,
I still use the compound 'hitherto' productively (i.e., as part
of my speech production).]

Once again, let me stress that this is not the same thing as
saying that [T] and [D] are not phonemes in English.  They simply
aren't contrastive in initial position.  The shift rule has been
presented as a historical rule.  If [T] and [D] are not phonemes
(or even possibly if they are) then it is also a synchronic rule.
That is all it takes to eliminate contrast between initial [T] and
[D] in English.  While [T] and [D] may well be phonemes in English,
what I am saying is that in the same way that [ku:c,en] and [ku:xen]
should not be considered a minimal pair in German, so [Tai] and
[Dai] should not be considered as evidence of the phonemicity of [T]
and [D] in English.  In other words, one should not insist on the
phonemicity of [T] and [D] on the basis of 'thigh' and 'thy'.  That
is what I have said from the start.  The distribution of initial [T]
and [D] in English can be accounted for by (phonological!) rule.

Now there are those who say that if you just consider this a
phonemic distinction, then you don't have to do all this analysis
and this evidence can safely be ignored.  They also say that some
patterns don't need explanations, that they are just there and
that trying to find explanations for patterns just complicates an
analysis that would be much simpler if the patterns were ignored.
But then they can't account for the distribution of initial [D]
and [T] between pronouns/deictics and other words and they can't
account for the lack of new coinings with initial [D] except by
saying that it is just a statistical anomaly.  So all in all, I
think it is better to account for the evidence rather than just
sweep it under the carpet.  I must say that I was surprised to
see you buy into the "any sounds that the speakers of a language
can distinguish are phonemes" bit, but I am glad to see that you
haven't jumped on the "ignore the contradictory evidence"
bandwagon so far.  At least you are presenting facts and
interpreting evidence as you see it.

It is of course possible to claim convincingly that there is no
phonetic difference between 'thy' and 'thigh' except for [T] and
[D], but it can be claimed equally convincingly that there is a
considerable difference between their function and internal
structure.  Now when it can be shown that the words that have
this same internal structure all have [D] to the exclusion of
[T] (that is, if one of these words has an initial dental
fricative it will always be [D], not [T]), it is possible that
the perception of the speakers about the distribution of initial
[T] and [D] is based on something other than phonemic
distinction.  Now there is no doubt that 'thy' and 'thigh' are
different words, just as there is no doubt that 'Kuhchen' and
'Kuchen' or 'Tauchen' and 'tauchen' are different words.  But
speakers of German do not distinguish 'Kuhchen' from 'Kuchen' or
'Tauchen' from 'tauchen' on the basis of a phonemic contrast
between [c,] and [x] but on the perception of a different
morpheme, '-chen' (diminutive; always with [c,]) in one word but
not in the other.  In the same way, in my view, the speakers of
English do not distinguish 'thy' from 'thigh' on the basis of a
phonemic distinction, but on the perception of a different
morpheme 'th-' (pronoun base; always with [D]) in one word but
not in the other.

My point is that this perception is not necessarily based solely
on phonetics.  The perceptions of speakers can sometimes be
perverse:  sometimes speakers will insist that two quite distinct
sounds are the same sound to them (e.g., German [c, x]); at other
times they will insist that two phonetically identical sounds are
different in their perception (e.g. German [t] in 'Rat' and
'Rad'; even more compelling was Sapir's example from Athapaskan).
As far as I know, there is no mechanistic way to determine the
phonemes of a language.  Determination of phonemes has to be done
by linguistic analysis based on what sounds contrast and under
what circumstances.  Different analysts may arrive at different
conclusions about which sounds are phonemes and which not.
Phonemes are psychological entities and their identification is
often subjective, depending on how much weight different analysts
give to which parts of the evidence.

>(I do not mean to imply that boundaries are *always* relevant
>while other information *never* is, only that there is no
>inconsistency in treating them differently.)

A wise precaution, since there always seem to be borderline cases
in which a clear distinction is not possible from the data, even
if, in the vast majority of cases, it is ("All grammars leak").
But, as I say, the inconsistency lies not in treating boundaries
differently from other morphological information, but in treating
some morphological boundaries differently from other
morphological boundaries.  If you only introduce morphological
boundaries into the phonology when you want two sounds not to be
phonemes, but ignore syntactically justifiable morphological
boundaries when you want two sounds to be phonemes, you run the
risk of being inconsistent.

>A few other points: How did _thy_ and _thigh_ differ before the
>former developed a voiced fricative?  Easy: the Old English forms
>were _thi:n_ and _the:oh_, where the final -h represented a velar
>fricative.  The loss of the final consonants in these words seems
>to be much later than the initial voicing.

Here we have a major difference of opinion.  To my knowledge (not
to my sure and certain knowledge, but from what I have gathered
about it), the voicing of initial [T] did not take place before
the 16th century, possibly even later (although, as far as I
know, there is nothing that prevents it from having happened
earlier).

Second, the loss of final [n] in 'thi:n' is not the normal
erosion of final [n] affixes that took place later in, for
example, the verbal infinitive, but rather was a morphological
shift whereby the forms with and without [n] became allomorphs,
with the former being used before words beginning with vowels and
the latter being used before other words.  This use is exactly
paralleled by the use of the forms 'a/an' and 'my/mine'.
Although the 'a/an' usage persists, 'thy/thine' and 'my/mine'
eventually split into two distinct forms, one becoming the
possessive pronoun and the other becoming a possessive pronominal
determiner.  The earlier usage is still found, however, in some
frozen forms (e.g., 'mine host') or in poetic language (e.g.,
'mine eyes have seen ...'). The forms without [n] are already
attested in very early Middle English, thus clearly antedating
the normal erosion of final inflectional [n].

Finally, the loss of the velar fricative was presumably complete
in the London dialect by the time of Chaucer since he rhymes
'high' (written <hye>) with both 'pie' (written <pye>) and 'fly'
(<flye>) and also rhymes 'light' (written <lite>; this, like
'the:oh' also had the form 'le:oht' in Old English;) with
'mite'(<myte>).  In 'thigh', there are writings as early as 1200
that indicate the loss of the velar in this word (<the>, <thee>,
in contrast with <thi> 'thy').  Of course, all of these words
would have been pronounced with [i:] at that time rather than
with [ai] as today after the great vowel shift and
diphthongization of long vowels, but they still all had the same
final sound.

So we can comfortably say that the forms [Ti:] and [Ti:n] were in
use by early ME (11th-12th century) and that 'thigh' would have
had the pronunciation [Ti:] at least by the end of the 14th
century (or possibly as early as 1200).  In order for what you
say to be true ("The loss of the final consonants in these words
seems to be much later than the initial voicing") then this
voicing had to have taken place well before the end of the 14th
century (and possibly before the end of the 12th century).  As I
say, I don't have a firm date for this event, but it has always
seemed to me that it was later rather than earlier (although I
can think of several good reasons why it may have been earlier,
but, even so, I can't put it much before about 1300).  Herb
Stahlke, in his posting of 21 April 2000, said that it didn't
happen until the 18th century, but this seems much too late to
me.  If anyone has solid information, based on evidence, perhaps
we can resolve at least this particular side issue.

>As the possibility of a three-way /ph : p : b/ contrast:  While
>I agree that English speakers tend to hear unaspirated [p] as /b/
>(at least, if the stop has a lenis pronunciation; the fortis
>stops of the Romance languages are normally perceived as /p/),
>systems with this contrast are by no means rare.  It was found in
>ancient Greek and a variety of modern languages.  It could not
>*on typological grounds* be ruled out for some future version of
>English.

I think my point was that it would be difficult to make this
three-way distinction with English /b/ as we presently know it.
The lax unaspirated [p] of English /spin/ is virtually
indistinguishable from the partially voiced /b/ of English /bin/
(that is, if you taped an English speaker saying /spin/ and
saying /bin/ and then erased the part of the tape with the /s/ on
it and ran both through a voice spectograph the two sequences
would appear much the same).  The relevant feature is voicing
onset, which is not an n-ary distinction, but is a continuum.
Thus what distinguishes [b p ph] is where voicing begins with
respect to the releasing of the stop.  If voicing begins before
the stop is released you have [b].  If voicing onset corresponds
to the release of the stop you have unaspirated [p].  If the
beginning of voicing is delayed after the release of the stop,
you have [ph].  The amount of delay determines whether you have a
weakly aspirated stop or a more strongly aspirated one.

Typologically such three-way contrasts do exist, but they have to
be based on stronger distinctions than presently exist among the
current English sounds.  In English, the voicing of [b] starts
very close to the release of the stop making it difficult to
distinguish from lax unaspirated [p].  It is only the phonotactics
of the language that prevents confusion between the two.
Furthermore, even when so based, such contrasts seem to be
unstable, with a tendency for one or more of them to shift out of
the matrix.  Look at what has happened in Greek:  both /ph/ and
/b/ have shifted away from the phonetic center (bilabial stop) of
the matrix to become fricatives.

So, I too would not rule out such a three-way distinction for
some future version of English, but I don't think it could happen
with the English sounds as they are presently realized.

>And finally, I have read Chomsky & Halle.

I rather thought you had.

>Seems to me they thought that had proved we didn't need phonemes
>to do phonology. and that various problems (of the type to which
>the thy thigh controversy belongs) could be resolved if we
>accepted "minor rules".  Well, that's very nice, but that's not
>the framework in which we have been arguing, so I don't see why
>you invoked them.

I invoked them (sounds rather like a pagan ritual :>) because
you said with reference to orthography "... we ignore all sorts
of other distinctions, and it doesn't bother us."  The part that
I was suggesting that you read (but I didn't have the specific
reference at hand) or re-read is on p. 49 where they say:

  The fundamental principle of orthography is that phonetic
  variation is not indicated where it is predictable by general
  rule.  Thus stress placement and regular vowel and consonant
  alternations are generally not reflected.  Orthography is a
  system designed for readers who know the language, who
  understand sentences and know the surface structure of
  sentences.  Such readers can produce the correct phonetic
  forms, given the orthographic representation and the surface
  structure, by means of the rules that they employ in producing
  and interpreting speech.  It would be quite pointless for the
  orthography to indicate these predictable variants.

I'm not sure that they believe that you don't need phonemes to do
phonology (well, I guess they do).  What they don't believe is that
there is a "phonemic" level of abstraction between the abstract
underlying (lexical) representation and the surface
representation that represents any "psychological reality."  What
they proved is that they don't like to talk about phonemes.  But
what they call phonological representations usually look
suspiciously like what others would call phonemic transcriptions
with the morphophonemics thrown in.  It is rather that they like
to lump phonemes and morphophonemes together without referring to
either one and account for morphophonemic alternations by what
they call "readjustment rules" in the difference between what
they call surface phonology and deep phonological structure.  But
they certainly don't deny that there must be some abstract phonic
structure beneath the surface phonetic level, whatever it may be
called.  But even if one does not subscribe to their concept of
phonology, I believe that there are still truths to be found
among their pages.

I brought them up because you mentioned some of the orthographic
peculiarities of English that they make less "peculiar" through
some of their "minor rules" (or "readjustment rules").  Although
I agree that much of what they say is overstated, especially as
they steadfastly refuse to discuss 'details that they consider
irrelevant', as ad hoc explanations for the peculiarities of
English orthography, they make it easier to systematize.  As they
state, again on p. 49, but it is a recurring theme throughout,
"... English orthography, despite its often cited
inconsistencies, comes remarkably close to being an optimal
orthographic system for English."  Despite its rather sweeping
nature, I don't think this is overstated (well, not much anyway;
"not really all that bad" might be better than "close to being
optimal").  Their real contribution, however, lies in
demonstrating that it is possible to evaluate intrinsic
complexity through the concept of markedness.

And incidentally, we don't really need phonemes to do phonology.
If one leaves mentalism (i.e., the perception of the speakers)
out of it, phonemic analysis is an abstraction used by linguists
to make talking about morphemes from the phonetic point of view
easier.  Gábor Sándi has stated (26 Apr 2000) that "the main
purpose of phonemic analysis is to provide for an unambiguous way
to describe the pronunciation of every utterance in a language."
This is rather the opposite of what phonemic analysis is
(although, actually I only object to one word of this
definition).  Phonemic analysis is in this way a reductionist
strategy that linguists use so that they can talk about the
morphemes of a language without getting bogged down in the actual
pronunciation of words and other morphemes.  Whether phonemes
have any "psychological reality" or not is still a matter of
discussion (although one, or at least this one, tends to think
that they do).  But since phonemes are an abstraction, if you do
your abstraction in a different way, then you don't need
phonemes.  The trick is to find a different abstraction that is
consistent across all the levels of language that phonology
affects and still can claim some degree of "psychological
reality."

What phonemic analysis provides us is a unique way to represent
every utterance in a language without worrying about its actual
pronunciation.  In this view, phonemic transcription represents
the limit of how broad a phonetic transcription can be without
violating any of the functional distinctions made by the
language.  Here, phonemic analysis just collects those similar
sounds that don't contrast in a language into a single entity
called a phoneme.  The phoneme then can be used to stand for any
(or all) of its members (allophones) in discussions.  Phonemic
transcription is used to free us from the cumbersome restrictions
of narrow phonetic transcription.  Phonemic transcription in this
sense is just another type of phonetic transcription from which
all the predictable, redundant, non-contrastive elements have
been stripped mechanically.

But a phonemic transcription doesn't describe the pronunciation
of words unambiguously.  It may describe the words unambiguously
because of the property of contrastiveness which all phonemes
must have regardless of whose definition you use, but not their
pronunciation.  One can determine the actual pronunciation of
words from a phonemic transcription only if one has a key that
gives both the rules for the distribution of the allophones of
the phonemes in the transcription and the speech habits of the
speakers of the language.  For example the phonemic transcription
of English /spin/ doesn't tell you that the /p/ here is lax and
unaspirated.  A description of how the word in question is
actually pronounced (a narrow phonetic transcription) would be
more like [sbIn] (allowing for the limitations of the 7-bit
ASCII character set).  An even better example was provided by
Gábor Sándi himself when he said "[Japanese] /huzimori/ is
pronounced [phujimori]."  Clearly, a phonemic transcription is
fairly useless for pronunciation purposes without the key.

The point of this longish digression in an already overly longish
message is not really that we don't need phonemes to do
phonology.  It may be true, but I am not arguing in favor of it.
Yes we could do phonology with phones plus a hideous mess of
morphophonemic and allophonic rules, or with a different kind of
abstraction, in much the same way as we don't need arabic
numerals (or place value notation in general) to do arithmetic,
but could use roman numerals (sign value notation) instead.  But
using phonemes in linguistics and arabic numerals (place value
notation) in arithmetic just makes things a hell of a lot easier.
But the real point of this is actually about the relationship
between phones and phonemes and the relationship between phones,
phonemes, and orthography.

If we look at phonemic transcriptions, we will find that they are
often fairly close to the normal orthography (at least for
English [if we except the vowels and throw in some
morphophonemics], which is principally what we are talking
about).  By comparison, phonetic transcriptions will usually be
widely divergent from the orthography (especially if one gets
beyond a 7-bit ASCII medium).  This is because alphabetic scripts
are based on essentially the same principle as phonemic analysis.
Those who introduced alphabets to language must have used a very
similar analysis, providing a symbol for each non-predictable
(contrastive) sound and leaving the non-contrastive (predictable)
sounds to be sorted out by the speakers (readers).  Alphabetic
systems then will tend to have one symbol (or sequence of
symbols) for each phoneme.  Indeed, the fact that those who
introduced alphabets to writing language used a phonemic-type
analysis is one of the things that indicates that phonemes are
such a linguistically useful concept and may have some
"psychological reality" apart from a mere linguistic abstraction.

So, with this in mind, let's take another look at what you said
about English orthography:

On Sun, 05 Nov 2000 CONNOLLY at LATTE.MEMPHIS.EDU wrote:

>>... we ignore all sorts of other distinctions, and it doesn't
>>bother us.  <s> stands for /s z S Z/ (the latter two as in
>>_pressure_ and _pleasure_ -- in fact, we don't evan have an
>>official way to spell /Z/, though <zh> would seem obvious.  And
>>don't even ask about the vowels: treating <y w> as doublets of
>><i u>, we have effective five vowel symbols for a very large
>>phonemic inventory.  I'd be glad to see ,dh. in use, but it
>>really has nothing to do with whether [T D] are separate
>>phonemes.

You said that "<s> stands for /s z S Z/" but these are all
(except /s/ most of the time) morphophonemic spellings or
allophonic variations.  <s> only represents /S Z/ before a
palatal.  Therefore, I would not expect <s> to represent /S/ or
/Z/ except before a palatal <u> or before the sequence <ion> or
<ial>.  Whether it represents /S/ or /Z/ in these environments
will depend on the form of the basic, underlying word.  Thus the
realization of <s> will be predictable (with, as usual, a very
small number of exceptions).  When the distinction is phonemic
(i.e., not predictable; e.g., 'sea':'she', 'sell':'shell/,
'sore':'shore'/) you can't use <s> for /S/.  Similarly, we can
always write the plural morpheme /s/ with <s> and a native
speaker will know automatically when it is to be pronounced [s]
and when [z], but we can't write /zip/ with <s> because here it
constrasts with [s] in /sip/.  So it is not that "we ignore all
sorts of other distinctions, and it doesn't bother us"; rather we
ignore *predictable* distinctions and it doesn't bother us.

The fact that <s> can be used for /z S Z/ just tells us that
there must be some rule by which English speakers derive /z/,
/S/, or /Z/ from an underlying /s/ whenever this orthography is
used.

You said "in fact, we don't evan have an official way to spell
/Z/, though <zh> would seem obvious."  Part of this stems from
the fact that /Z/ became phonemic after the orthographic rules
were established.  But I can't think of any other way to spell
foreign names like 'Zhukov' or 'Zhdanov' or loanwords like
'muzhik' -- do you have an alternative suggestion?  As with <s>
and <z> and <sh>, when the distinction isn't predictable, you
have to be explicit in the spelling.

I agree that one shouldn't even ask about the vowels.  As far as
I know, English vowels are still shifting on a continuing basis.
I doubt that the writing system has any chance to catch up until
this stops.  Certainly there is precious little phonetic
predictability in English vowel orthography.  On the other hand,
there is a great deal of morpheme identity encoded in English
vowel orthography (e.g., 'pair', 'pare', 'pear'; 'die', 'dye';
'meat', 'meet', etc., etc.).  If these homophonous vowel sounds
were replaced with standardized orthographies, this information
would be lost.  Basically, English vowel orthography has traded
off phonetically specific information for morphologically or
lexically specific information through its myriad variant
orthographies for identical phonetic sequences.  But there are
still regular vowel alternations that are not explicit in the
orthography but are still predictable from the grammatical rules
(divine/divinity, extreme/extremity, etc., etc.).  Finally,
English vowel orthography has advantages in a situation with
continually shifting vowels or numerous dialects with varying
vowel realizations.  'Caught' and 'cot' are still
orthographically recognizable whether they have the same vowel
sound or not.

And to your last statement about the fact that [D T] are always
written only with <th> having nothing to do with whether they are
separate phonemes or not, I still say "maybe not."  But
personally, I think it has a lot to do with it.  But it certainly
has a great deal to do with the contrastive load of [D T] in the
language.  If one accepts the idea that only distinctions that
are not predictable from the rules of grammar need to be marked
in the orthography, then I don't see how the fact that the
difference between [T] and [D] is never marked in the orthography
can indicate anything other than the existence of rules that
predict the presence of [T] or [D] (with the usual small number
of exceptions).  In short, the contrastive load of [D T] is very
nearly zero.  I suppose it depends on whether one considers
phonemes to be just different sounds in a language or whether one
considers them to be sounds that can't be derived from some other
phoneme by a grammatical rule.

Since, in your view, the fact that [D T] are always written
indiscriminately with <th> says nothing about whether they are
phonemes or not, then you should be able to come up with lots of
examples of phonetic (alphabetic or syllabic) writing systems
from around the world where two distinct phonemes are invariably
written with the same symbol (not occasionally, but invariably).
I'll be waiting to see this list (note: logographic and
logo-syllabic systems are excluded since, in general, they don't
represent phonemes directly).

But if the fact that [D T] are always written with <th> says
nothing about whether they are phonemes or not, then so also does
the fact that German [c, x] are always written with <ch> have
nothing to do with whether they are separate phonemes or not.
But guess what? -- they aren't.  The basic principle of
alphabetic systems remains:  one symbol (or fixed sequence of
symbols) per phoneme.  Allophonic or morphophonemic alternations
don't have to be expressed in writing because they are
predictable to native speakers (just as allophonic alternations
don't have to be expressed in phonemic transcription).  In fact,
anything that can be predicted by native speakers from context
(like the vowels in most Semitic langugages or the many English
words that are spelled the same but pronounced differently)
doesn't have to be expressed in writing.  But when you have
contrastive sounds that can't be predicted from context
(phonetic, grammatical, or semantic), the writing system has to
differentiate them or else it fails.

Bob Whiting
whiting at cc.helsinki.fi