thy thigh etc. (external sandhi in English)

[Robert Whiting]

On Tue, 21 Nov 2000 Douglas G Kilday <acnasvers at hotmail.com> wrote:

> Leo A. Connolly (16 Nov 2000) writes:

<snip>

>> 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.

> The loss of final consonants probably is later than the
> establishment of initial [T:D] distinction,

Not unless this distinction was firmly in place well before the
time of Chaucer.  The 'thy/thine' (and 'my/mine') allomorphic
distinction is already in evidence in very early ME, and final
[x] > 0 was accomplished by the time of Chaucer as shown by his
rhymes and spellings.  Indeed, one begins to get spellings of
'thigh' as early as 1200 which show the loss of [x] in this word.
So unless you can place the initial [T:D] distinction before
1200, this statement is unlikely to be true.

> but one should not assume that this distinction arose by the
> selective voicing of initial [T] to [D] in what Whiting loosely
> labels "function words".

First, the concept of "function words" is not a Whiting loose
label.  "Function word" and "content word" are technical terms
in grammatical theory, and hardly my invention.  You may, as you
claim, be unaware of the difference, but that does not mean that
there isn't one.  Any grammatical handbook or discussion of
grammar will explain the terms to you.  If you don't have one, a
dictionary will do.  Here are the definitions from Webster's
Collegiate Dictionary 10th edition (2000):

   function word n (1940) : a word (as a preposition, auxilliary
   verb, or conjunction) expressing primarily grammatical
   relationship

   content word n (1940) : a word that primarily expresses
   lexical meaning -- compare FUNCTION WORD

Interestingly, Webster's Encylopedic Unabridged Dictionary (1994)
has a longer definition of "function word":

   function word, a word, as a pronoun or preposition, that is
   used in a language as a substitute for another or as a marker
   of syntactic relationship; a member of a small, closed form
   class whose membership is relatively fixed.  Cf. empty word,
   full word.

but has no definition of "content word" using instead the term
"full word":

   full word, (esp. in Chinese grammar) a word that has lexical
   meaning rather than grammatical meaning; a word or morpheme
   that functions grammatically as a contentive. Cf. empty word.

I don't particularly like this last part since, although
"contentive" fairly obviously means "content word" (rather than
something that makes Mr Borden's cows give milk), the dictionary
nowhere defines either contentive or content word.

But, all things considered, "function word" and "content word"
are well-established categories that distinguish words that have
purely predictable grammatical function from words that have
(from the point of view of sentence structure) unpredictable
content.  In general, content words provide the new information
(content) in an utterance and function words tie this information
together in a particular way.  For example, simple sentence
structures can be used as a way of introducing new content words
in teaching situations by using a fill-in-the-blank approach:

   This              is              a              ___________.
   function word     function word   function word  blank for any
   demonstrative     copula          indefinite     content noun
   pronoun                           determiner     or NP

The blank can then be filled in with any noun or noun phrase:
e.g., 'house', 'screwdriver', 'mastadon', 'very sorry-looking
piece of equipment', 'cuneiform tablet written in Sumerian in
2012 BC by a scribe named Ur-Suena'.  This approach is frequently
used in phrase books for foreign languages where a functional
sentence skeleton is provided, followed by a list of content
words that can be used to fill out the sentence in various
situations.

It should be obvious from this that function words are among the
most commonly used words in a language because they are used over
and over again whenever the grammatical function they serve is
invoked in speech or writing while content words only appear when
their content needs expressing.  Any corpus analysis will confirm
this.  In fact there is a sort of Zipf's Law for this because
function words are relatively few in number but are used very
frequently while the number of content words is huge (and growing
constantly) but the individual words are much rarer in use.  This
(the presence of initial [D] in many function words) accounts for
the fact that [D] is one of the most common consonants in English
speech (ranking 6th) while its contrastive partner [T] is one of
the rarest (only more common than [Z]) despite the fact that
there are many more English words with [T] than with [D].

Basicly, content words tell what is being talked about and
function words tell what is being said about it.  You can see
this by taking a simple sentence and stripping out the functional
elements leaving the content words:

  I saw the book on a shelf in the room.
    see     book      shelf        room

A simple check on this can be had by looking at the opening lines
of Lewis Carroll's celebrated (by linguists) poem "Jabberwocky":

    'Twas brilling and the slithy toves
    Did gyre and gimble in the wabe:
    All mimsy were the borogoves,
    And the mome raths outgrabe.

One could hardly ask for a better illustration of the difference
between function words and content words.  It can immediately be
seen that there is nothing grammatically incorrect or even
inconsistent in the passage.  This is because all the function
words and other functional elements have been left intact.  The
content words, on the other hand, are all semantically empty.
Although they are all phonotactically acceptable English words
they have no meaning in the lexicon.  The result is that we know
exactly what is being said, but we have no idea what is being
talked about.  One could further note that if the function words
were removed or replaced with semantically empty expressions, the
result would be completely unintelligible gibberish.

Words can be both function words and content words depending on
how they are used, and sometimes a content word becomes a
function word (grammaticalization) through usage (e.g., French
'pas' "step" ==> "no, not"), but the concept of function words
and content words remains clear.

So people do not assume that there was a selective voicing of
[T] to [D] on the basis of any loose labels.  They assume it on
the basis of the evidence.  A succinct summary is provided by
Edward Finegan in "English" in Bernard Comrie (ed.), _The World's
Major Languages_ (London & Sydney: Croom Helm 1987), p. 91

   Significantly, initial /<edh>/ in Modern English is limited to
   the function words _the_, _this_, _that_, _these_, _those_,
   _they_ and _them_, _there_ and _then_, _thus_, _thence_,
   _though_ and _thither_, with initial voiceless /<theta>/ in
   Old English later becoming voiced by assimilation when
   unstressed, as these words often are.

This is not just an appeal to authority (although expert
authority does have its appeal).  This is to show that this is
the generally accepted position, not something that I have made
up to entertain the list.

But this was published some 13-14 years ago.  Perhaps it is no
longer the mainstream opinion.  Perhaps I missed your review of
this work where you pointed out to the misinformed multitudes
that "what [these words] share is definiteness, not some murky
'functionality'."  If so, please tell me where it is published so
that I can see what evidence you base this assertion on.

Besides, function words wasn't my label to start with.  I
originally referred to them as deictic words and pronouns.
Somebody else introduced the term function words, presumably
because this is how they are usually referred to in the
literature, and I just let it ride because, well, they *are* all
function words, and function words is easier to write than
pronouns and deictic words.  But in the final analysis, I don't
think that the voicing of initial [T] in these words relies
specifically on the fact that they are function words.

> (It escapes me how <thy> has any more "function" or less
> "content" than <through>.

No doubt.  But pronouns are quintessential function words.
Pronouns (particularly personal pronouns) have very general
reference and little inherent meaning and essentially carry only
grammatical information (person, number, gender, case).  Their
specific referents come from their immediate surroundings.

'Through' is also often a function word, although not invariably
since it is also used as an adverb and an adjective.
Furthermore, it has a lexicalized stressed variant 'thorough'
(always an adjective), something that is characteristic of
function words.  I suspect that what keeps 'through' out of the
group of function words with initial [D] is that it doesn't
appear to be a word with a pronominal/deictic 'th-' base as the
other members of the group do.

> What English words with initial [D] share is definiteness, not
> some murky "functionality".)

Definiteness is a functional (grammatical) category.  But where
is the definiteness in 'though'?  As a conjunction, 'though' is a
function word, but is no more definite than 'but' (with which it
is sometimes synonymous).  It can be adversative, disjunctive, or
conditional, but not definite.  But, even though it lacks any
overt "definiteness," it can still be considered a deictic or a
demonstrative.  While modern 'though' is a Scandinavian loan, it
replaced an OE word that was homonymous with part of the 'the'
demonstrative family, and was probably perceived as part of this
family despite lacking definiteness.  Finally, in some northern
dialects of British English it is pronounced with [T], clearly
indicating that it is a borderline case.

But I will agree partially with your premise.  These words *are*
all function words, but the fact that they have initial [D] is
not specifically because they are function words, but because
they share some other feature.  So you were on the right track;
you have just taken a disastrously wrong turn.

> Old English clearly did not consider [T:D], [f:v], and [s:z]
> to be phonemic oppositions.

Good.  Then we can dispense with the notion that any sounds that
the speakers of a language can distinguish are phonemes.

> The characters "thorn" and "edh" are used indiscriminately in
> the manuscripts, and no special signs for "v" or "z" are found.

The writing of [v] in foreign words (like 'fers/vers' "verse")
was done with the <u/v> graph, as later.  You will find both
spellings (<f> and <u/v>) in OE sources (although writings with
<f> are more common).  You will also find writings like
<Uespatianus>, <Uenus>, <Uesoges>, <Nouembris>, and, of course,
<Eue> (Adams bryd).  But you won't find *<Fespatianus>, etc.  In
native words, you will find 'over' written both <ofer>
(historical spelling) and <ouer> (pronunciation spelling) already
in OE.

The writing of [z] is found almost exclusively in foreign names,
particularly in the Bible:  <Azariah>, <Baldazar> (for later
Balshazzar or Balthazar, etc.), <Zabulon>, and so forth.
According to the grammar, z (named zede) was used for writing
Greek words.  Hence <zodiacus> and <zefferus>.

The Anglo-Saxon latin alphabet consisted of the 23 letters of the
then current latin alphabet (lacking j, v [u and v were simply
graphic variants at this time and counted as a single letter],
and w) plus two necessary signs left over from the runic alphabet
(thorn and wen or wyn) and one unnecessary sign (eth or edh)
added by the Irish monks who introduced the latin alphabet to
England.  There were also vowel ligatures notably <æ> (ae) that
represented sounds that had had signs in the runic alphabet but
these were not counted as separate letters although this sign had
a name ("ash") inherited from the name of the runic character.

So it is not true that "no special signs for 'v' or 'z' are
found" in OE.  They are there, but they aren't used frequently
and generally only in foreign words (except that 'u' is sometimes
substituted for wen or for <f> in intervocalic position in native
words).  What you can say without fear of contradiction is that
they are not used to indicate contrasts between [f] and [v] or
[s] and [z] because there aren't any in OE.

> In the usual treatment the OE fricatives are regarded as unvoiced
> when doubled, when adjacent to unvoiced stops, and when occurring
> initially or finally in words or compound-elements. The first two
> conditions are solidly established, but the third may be doubted:
> did morphemic boundaries in OE necessarily make adjacent
> fricatives unvoiced in connected speech?

Probably.  Otherwise words like 'nothing' (OE na:thing < na:n +
thing) and 'anything' (< æ:nig + thing) would doubtless have [D]
rather than [T] since these compounds came into existence in OE.
Since the environments around the boundaries in these words are
invariant, surely the sandhi-type rule that you propose below
would have been more likely to operate here than in the more
variable environments of the boundaries between words.  I can't
see any alternative:  if there was a sandhi rule operating across
morpheme boundaries it would have fixed 'nothing' and 'anything'
with [D] already in OE.  Similarly, compounds like 'within' and
'without' would have been fixed with [D] at an early date by such
a rule.  Since none of this happened, it speaks against the
existence of such a rule.

> Some light is shed on the problem by the "Cuckoo Song"
> manuscript of circa 1240, which contains an Early Middle English
> folk-song in parallel with a Medieval Latin hymn. The English of
> the song shows no significant influence from Norman French or
> Latin. The scribe uses typical ML orthography with "u" and "v"
> indiscriminately representing both the vowel [u] and the voiced
> fricative [v], which creates no ambiguity in practice. For the
> English text, the scribe requires "w" and "thorn" to convey the
> non-ML sounds. More importantly, he carries the ML graphemic
> distinction "f:u/v" into his English transcription. We find "f"
> in <after>, but "u" in <nauer>, <calue>, and most significantly
> <bucke uerteth> '(the) buck farts'. Here the scribe has preserved
> initial [v], conditioned by a final vowel in the previous word.
> This verb appears in later ME as <farten>; it is unattested in OE
> but is a transparent cognate of Greek <perdomai> and Sanskrit
> <pardate>.

By your own description, the scribe has not "preserved" initial
[v] in this word but has shifted initial [f] to [v] since the
cognates clearly show that the sound is etymologically [f] (<
PIE *p per Grimm's law).  The [v] can have been "preserved" here
only if there was some kind of sandhi rule in operation for which
this writing then becomes the only evidence.  Using the word
"preserved" to describe this writing is begging the question.  It
is building what you are trying to prove into the argument

> The manuscript does not distinguish [s:z] and [T:D] because
> these oppositions are unknown in Medieval Latin.

A contrast between [s] and [z] may not have existed in Medieval
Latin, but there were Medieval Latin words that historically had
inherent [z] (e.g., 'zodiacus') borrowed from Greek.  In fact,
the letter <z> was added to the latin alphabet specifically to
write the large influx of Greek loan words after the Roman
conquest of Greece.  That's why it's tucked in at the end of the
alphabet.

There certainly was no contrast between [T] and [D] in Latin (of
any stripe) since [D], certainly, and [T], probably (although
possibly in late periods), never existed in Latin.  Graphic <th>
in classical Latin represents [th] (aspirated rather than
spirantized) and was not native to Latin, but entered through
Greek influence both from loanwords (e.g., 'bibliotheca') and
from the prestige of Greek literature and learning which caused
aspirated pronunciations (and eventually spellings) to be
introduced into native Latin words after the conquest of Greece.
Before that time Greek words with [th] were borrowed into Latin
with [t].

A similar phenomenon occurred in English.  When the <th> was
restored in words like 'throne' (ME 'trone') and 'authentic' (ME
'autentik' based on the prestige of classical literature and
learning, there was probably a "knock-on" effect so that <th>
found its way into some words of classical origin that had never
had theta, such as 'anthem' and 'author/authority'.

> In my opinion the Early ME dialect recorded here most likely
> made no phonemic distinction between voiced and unvoiced
> fricatives, but maintained an allophonic contrast on the basis of
> adjacent sounds regardless of morphemic boundaries, and this
> scheme was probably inherited from OE. For the composer of the
> "Cuckoo Song" the words <thin> and <thih> would have begun with
> [T] or [D] according to the unvoiced or voiced nature of the
> preceding sound.

"Composer" or "copyist"?  Do you have proof that these sounds
were in the original, or only that they are in this particular
copy?  Since, as far as I know, there is only one manuscript, the
question is rhetorical.  But the thing about folk songs is that
they often (perhaps even usually) don't have a composer.  That's
what makes them folk songs.

Sounds like the writer/copyist was Irish or Welsh.  I must say
that I have never heard of this sandhi rule for OE.  Can you give
me a reference to a publication of it so I can see if there is
any other evidence besides the _hapax phenomenon_ <bucke
uerteth>?  Surely if such a sandhi rule could be proved for OE,
those who claim massive Celtic influence on English would be all
over it like a shot even though it is only similar, not identical,
to Celtic lenition (mutation).

But if <bucke uerteth> is the only evidence, this is certainly
too flimsy a foundation on which to build such an elaborate
structure.  There are simply too many other ways to account for
this single piece of evidence.  For example:

  a) The writer/copyist might have been a speaker of a language
     that had a similar rule (e.g., Irish or Welsh), which
     influenced his writing.

  b) The dialect of the writer/copyist might have shifted (or
     been in the process of shifting) initial [f] to [v].  Some
     English dialects did do this.  In fact the only native
     (inherited) modern English words with initial [v] ('vat'
     [cf. G 'Faß'], 'vixen' [cf. G 'Füchsin'], 'vane' [cf. G
     'Fahne']) are borrowings into the standard dialect from just
     such a dialect.

  c) The distinction between [f] and [v] was not considered
     phonemic by the writer/copyist and, although he had
     different signs available to make the distinction, he felt
     no compulsion to indicate it consistently in the writing of
     English (in much the same way as thorn and edh and <s> and
     <z> were used).

  d) The use of the voiced member of a voiced/voiceless allophone
     pair may have been sound-symbolic, used perhaps as a jeu
     d'esprit in a word where the writer/copyist considered a
     voiced sound more appropriate to the meaning.

Other explanations are also possible (like the simplest one that
the writer/copyist meant to write <f> and wrote <u> instead; this
is usually called "scribal error").

And all this assumes that the writing <uerteth> actually
represents the verb 'fart', which is not universally accepted.
Admittedly, OED takes it this way, but others have taken it as a
form of a verb based on _vert_ 'green' (Cambridge History of
American and English Literature).  More convincing in my view is
a loan based on Latin _vertere_, French _vertir_ 'turn'.  There
are quite a number of borrowed English compounds based on this
root ('avert', 'divert', 'convert', 'revert', 'pervert', invert,
etc.), but the bare root seems unknown as a verb (although common
in other words:  'version', 'vertigo', 'vertex', 'verse',
'versus', etc.).  The native cognate is found in the suffix
'-ward(s)'.  Perhaps this hapax legomenon is a borrowing that
never became established or even an aphetic form of some compound
used to preserve the meter (this is, after all, a song).

Although there is nothing philologically unsound about a
derivation from 'fart', it seems out of place in the context of
the song.  No one has ever been able to explain to me why a
farting buck is a transparent metaphor of spring (from eating all
the newly sprouted greenery?).  Admittedly, I have lost touch
with 13th century England, and it is possible that in early
spring the air was redolent with buck farts, and your average
churl or thane took a deep breath and said "Ah, spring -- the
buck is farting."  Or perhaps the metaphor is auditory rather
than olofactory and buck farts were so sonorant that in spring
they resounded from every hillside and could be heard for miles
around, and your average ch. or th. said "Hark, sumer is icumen
in -- the buck farteth."  But I have a hard time seeing it this
way.  In keeping with with the rest of the song and the
parallelism within this couplet, some more visible physical
activity rather than an auditory or olofactory one would be more
appropriate.  Thus "the bullock starts, the buck turns/twists
(about)" is better suited to the context and the structure of the
song.  The main advantage to translating "the bullock starts, the
buck farts" is that it preserves the rhyme and scan of the
original (bulloc sterteth/bucke verteth).

As I say, this is just my own impression, based on a solid urban
tradition.  So if some of our more bucolic colleagues have
knowledge of why there may be more to this metaphor than meets
the eye (ear, nose), and why a farting buck heralds the coming of
spring/summer, please share it.

But in the absence of corroborating evidence, calling <bucke
uerteth> unequivocal evidence for a sandhi rule involving
voiced/voiceless fricatives in OE and ME (when in fact it is
neither the most likely nor the most plausible explanation) and
then using this rule to account for the present distribution of
these sounds in modern English (when in fact it doesn't) seems a
bit on the fanciful side.

As you say, this is your opinion, but in my opinion, statements
like "most likely ... maintained an allophonic contrast on the
basis of adjacent sounds regardless of morphemic boundaries, and
this scheme was probably inherited from OE", are considerably
stronger than the evidence warrants.  "May have" rather than
"most likely" and "could have been" rather than "was probably"
are more appropriate to a theory built on a single piece of
evidence without corroboration, especially in the face of
conflicting evidence.

> Turning now to Chaucer we find the "f:v" distinction
> consistently made, initially and medially, even in native words
> like <yeven> 'to give' from OE <giefan>. The character "z" is
> used only in foreign words, mostly from French and Arabic:
> <zeles>, <zodiak>, <azimut>, etc. Chaucer retains "s" for medial
> [z] in native words: <hasel>, <hasard>, <bosard>, etc. Printed
> editions give no information about the dental fricatives, using
> "th" everywhere.

The history of the orthography of the English fricatives, given
both here and below, is quite correct (except that ME 'give' is
generally considered a Scandinavian borrowing rather than derived
from OE 'giefan', 'gifan' and except for the loose labelling of
<zeles>, <zodiak>, and <azimut> as "from French and Arabic"; all
of these words come into English from French or Late Latin; all
are borrowings into French (or Latin), the first two from Greek,
the last from Arabic; the one word that comes into French (or
Late Latin) from Arabic (via Spanish) is actually based on a
Latin borrowing into Arabic), but basically it just shows what I
have been saying all along:  there is no need to distinguish
sounds in writing until they are used phonemically.  As long as
the distinction is allophonic or morphophonemic or
non-contrastive, there is no need for a graphic distinction.

But the picture given here, while factually correct, implies
quite a bit more regularity than is actually found in Chaucer's
spelling.  While he does use <z> only in foreign words (as was
done in OE), he also uses <s> for [z] in foreign words (as was
also done in OE) and he also frequently writes <z> for the plural
morpheme in foreign words (but this could be a borrowing from
French).

For example in two consecutive sections (Astrol. I 19-20) he
spells 'azimuths' as

    Azymuthz
    A3imutz (3 for yogh; probably a miswritten or miscopied <z>)
    azymutz
    azimutes

Furthermore in these same sections he writes 'zenith' as both
<senyth> and <cenyth>, but not with <z>; he also writes 'horizon'
as both <Orisonte> and <orizonte>.  So his spelling is pretty
much all over the place as far as indicating [z] with both <s>
and <z>, in foreign words.

> Chaucer's language has a large component of recent loanwords
> in <v-> sometimes forming minimal pairs with older words in <f->:
> <vailen> 'to have value' ~ <failen> 'to fail', <vers> 'verse' ~
> <fers> 'chess-queen'.

The thing is, all four of these words are foreign to English.
And _fers_ 'chess-queen' is not older in English than 'verse'
which occurs already in OE (written both <uers> and <fers> but
without any indication of a sandhi rule governing which is used;
they are simply in free variation).  There were plenty of loan
words with initial [v] as well as with intervocalic [f] (e.g.,
'coffer') in English before the 14th century.  It is just that
before this time English speakers didn't consider [f] and [v]
different sounds.  So long as they didn't contrast this was
possible.

Now, while it is true that there was an earlier borrowing of
'verse' in OE with a variable initial <f>/<u/v>, there is no
indication of a sandhi rule affecting which way it is written.
Since <f/uers> was neuter in OE, it is often preceded by <thaet>
as the correct form of the definite article, and one finds both
<thaet fers> and <thaet uers> -- simply free variation.  This is
not to deny that some speakers may have had such a rule.  There
is just no indication that it was the norm.  Similarly, there is
no rule involved in the OE variation in the writing of <ofer> and
<ouer>.  This last word was always pronounced with [v] in OE,
just as <f/vers> was probably always pronounced with [v] despite
the fact that it was much more frequently written as <fers>.  In
OE there was no contrast between [f] and [v] and hence no
consistency in the way they were written was required, although
speakers could probably tell the difference between the sounds.

One could legitimately question the extent to which OE written
materials actually reflect the spoken language of the time and
to what extent written OE was influenced by Latin with its
[f]/[v] contrast.  Most people who knew how to write OE (mostly
monks and/or bureaucrats) were familiar with, if not fluent in,
Latin.  Did OE monks and bureaucrats know Latin?  Well, does the
Pope know Latin?  Did the writer/copyist of the Cuckoo Song know
Latin?  The answer is right there in the manuscript.  Your saying
that "The English of the song shows no significant influence from
Norman French or Latin" is again begging the question.  These
things being so (as Caesar was wont to say), it is difficult to
ascribe <f>/<u/v> orthographic variants in OE or even in early ME
to the single cause of a sandhi rule that voiced initial
voiceless fricatives in a voiced environment across morpheme
boundaries (especially when such a rule is not in evidence in
recent compounds).

> It appears that this large influx of words beginning with [v]
> which does not alternate allophonically with [f] has forced the
> older words beginning with [f/v] to abandon the voiced allophone
> and become words beginning with invariant [f].  That is, this
> influx of loanwords has created a new phonemic distinction /f:v/
> in initial position which the orthography of Chaucer's time (late
> 14th cent.) regards as fully established.

Rather, it was the French (Latin) contrast between [f] and [v]
that was imported in pairs like 'failen' and 'vailen' and
'coffer' and 'cover' or 'coffin' and 'coven'.  English now had to
deal with words with inherent [f] and inherent [v] in contrastive
positions.  'Vers' "verse" probably was reborrowed from French in
ME along with 'fers' "chess-queen".  But this time there is an
[f]/[v] contrast so the two sounds can't be treated as variants.
Furthermore, the fact that OE <f/uers> was not fixed with initial
[f] but with initial [v] runs counter to your assertion.  So the
[f]/[v] contrast was just imported along with the words.

But this is true whether there was a sandhi rule operating or
not.  Speakers now have to take account of words beginning with
inherent initial [v] (borrowed) as well as with inherent initial
[f] (native or borrowed) as well as words with inherent
intervocalic [f] and inherent intervocalic [v].  If there are a
lot of borrowed words with initial [v], especially when some of
them contrast with words with initial [f], either borrowed or
native, and there are a lot of borrowed words with intervocalic
[f] then speakers either have to nativize the borrowed words
with initial [v] to [f] and nativize the borrowed words with
intervocalic [f] to [v] (and lose the contrast in both cases) or
they have to accept the [f]/[v] contrast as phonemic.  Obviously
they did the latter.  But this certainly doesn't require a sandhi
rule for its implementation.

In fact, the observation that native words were fixed with
invariant [f] rather speaks against such a sandhi rule.
Otherwise, one might have expected at least some native words to
have been fixed with initial [v].  After all, if what you say is
true, English no longer had words with inherent initial [f], but
only a sandhi-governed initial [f]/[v] alternation.  Since you
claim that this situation had existed since OE, native speakers
could have had no memory of words with inherent initial [f] by
the 14th century.  Unless, of course, you want to argue that
'vat', 'vane', and 'vixen' are not borrowings from the southern
dialects, but are in actuality such words.  But then you have to
account for the distribution of [v] in these particular words:
e.g., why in 'vixen' but not in 'fox'?

If definiteness is the criterion, then surely we would expect
'four' and 'five' to be written *<vour> and *<vive>.  While 'for'
and 'fore' might have flopped back and vorth between initial [f]
and [v] for centuries, surely invariable compounds like 'before'
and 'afore' would have been fixed as *<bevore> and *<avore>
within a week if there was a sandhi rule operating that voiced
fricatives in a voiced environment across morpheme boundaries.
So phonemicization of [v] doesn't require a sandhi rule for its
implementation.  In fact, it is easier to explain without one.
Otherwise you have to say that native words in initial [f] and
borrowed words in initial [v] is just another huge statistical
anomaly.

This is a choice that is faced whenever there is a massive influx
of loanwords that use a segment in a position where it doesn't
occur in the borrowing language.  Speakers will either nativize
the segment or leave it as it is.  In the latter case, if it
doesn't contrast (or at least not heavily) with native segments
then it may simply be retained as a foreign phone(me) and will
ever be a marker of a foreign word (e.g. [z] in Latin words of
Greek origin).  If it does contrast heavily, or even if speakers
just happen to like the sound, it will become a new phoneme, and
will eventually find its way into new environments and contrasts
and new (native) words, perhaps even ousting some native phoneme
in certain words.

> For initial [s:z] Chaucer has only the sub-minimal pair <zeles>
> 'zeals' ~ <seled> 'sealed', and since "z" is otherwise used to
> denote loanwords this provides no direct evidence for a phonemic
> opposition.

Hallelujah -- alhamdulillah -- barak hashem.  At last someone who
realizes (and is willing to admit) that distinctions that can
have other explanations do not necessarily provide evidence of
phonemic contrast.

> The status of initial [T:D] in Chaucer is equivocal.
> Contractions like <artow> 'art thou' and <seistow> 'sayest thou'
> suggest that <thou> already had invariant [D], since the old rule
> would require [T] here and [tT] usually contracts to [T], which
> would yield *<arthow> etc. However, this argument is extremely
> weak. Statistical analysis of Chaucer's alliterations in "th"
> might resolve the issue, but I am not aware of such a study.

Extremely weak is perhaps an understatement.  Actually, in
addition to contractions, <t> in 'thou' (and the other cases of
this pronoun) is often found after [s], [t], and [d].  Thus you
find things like 'bi-hold tou'.  Any conclusions about whether
initial <th> in these words was voiced, either invariantly or
environmentally are very iffy.  But I will agree that the
possiblility that it was already voiced is not excluded.

> At any rate it is clear that the English /f:v/ distinction
> resulted from loanwords and was well established by Chaucer's
> time. The /s:z/ distinction, if not already made by Chaucer,
> followed shortly. It can be attributed partly to loanwords,
> partly to the earlier establishment of /f:v/ as phonemic.

There are other factors, which you seem anxious to ignore,
involved in the phonemicization of the voiced fricatives as well.
One of these is the collapse of the English short vowel system
and the loss of final inflectional '-n' followed by loss of final
[@] (schwa).  This uncovered some of the former intervocalically
voiced fricatives (particularly in infinitives) and left them
bare at the ends of words, which removed the phonetic
conditioning environment that both created and maintained the
identity of the allophones.  Since they couldn't remain
allophones without the conditioning environment, something had to
happen to the voiced fricatives that were now in word final
position.

Another factor was the loss of intervocalic long fricatives
[ff TT ss] which were simplified into [f T s] producing
intervocalic unvoiced fricatives which could contrast with the
intervocalic voiced fricatives [v D z] which had previously
been simply allophones of [f T s] in intervocalic position.
Since it was no longer possible to predict [f T s] or [v D z]
based on intervocalic position, something had to happen to the
intervocalic voiceless/voiced fricatives that were no longer
predictable from their environment.  For instance, modern 'since'
comes from a contraction of ME 'sithens' (with adverbial genitive
-s) which in turn came from OE 'siththan' (itself a compound of
'si:th' "after" and 'tham' or 'thon' [dative or instrumental of
"that"]).

So the loss of the conditioning environments that identified the
voiced fricatives as allophones of the voiceless fricatives may
have had as much (or more) effect on phonemicization of voiced
fricatives as the massive influx of French loanwords in the
13-14th century.  Certainly the two events reinforced each other.
And the large number of loans with inherent initial [v] and the
smaller number with inherent [z] quickly established these two
sounds as phonemes in English.  New coinings with /z/ and
contrasts between /s/ and /z/ follow very soon (e.g., 'fuss',
'fuzz', 'buzz') because of the suitability of [z] for expressive
and imitative words.  Now that /z/ is a phoneme instead of just a
sound in the language it is possible to have words with inherent
[z] in them in any position (so long as phonotactic constraints
are observed).

But [T] and [D] did not have any such crutches to aid in their
phonemicization.  English was not in contact with any languages
with a [T]/[D] phonemic opposition.  In fact, ME did not have any
source of words with [D] at all (except Celtic where it existed
only as a lenition of [d]) and [T] was only available through the
medium of classical Greek or Latin (or more rarely, through
Celtic).  Even French had lost the [T] pronunciation and hence a
number of words borrowed from French lacked <th> (e.g., ME
'trone' "throne").  As a result, the same thing that happened
with the [f]/[v] and the [s]/[z] oppositions just did not happen
with [T]/[D].  There simply was no massive import of words with
[D] or with a [T]/[D] contrast to make it happen.

> The [T:D] opposition was left as an orphaned allophonic pair,
> very difficult for speakers to maintain, especially in initial
> position. They would have been obliged to enunciate some words
> with invariant initial [f], [v], [s], [z] and others with
> variable [T] or [D] according to the preceding sound. Under these
> circumstances any mechanism tending to fix [T] or [D] in
> particular words would have operated rather strongly.

Now it's my turn to say that I don't see why.  According to you
this sandhi rule had been operating for over 500 years and now,
with the loss of two of the allophone pairs to which it applied,
speakers are suddenly anxious to get rid of it.  One of the
principles of sound change rules is that they operate until they
no longer have anything to operate on and then they disappear.
It should have been no more difficult to maintain this alleged
sandhi rule for [T]/[D] than it was to maintain it for the other
fricatives in the face of other fixed voiceless/voiced
oppositions (as in the stops).

But I must say that you paint a vivid picture of the difficulties
faced by 14-15th century speakers of English.  One can almost see
the churls and thanes tossing and turning on their straw pallets
or feather beds, driven to sleeplessness by the difficulty of
maintaining the allophonic alternation of initial [T] and [D] now
that [f] and [v] and [s] and [z] have become phonemic contrasts.
Much of the energy of the English-speaking peoples during this
time must have gone into finding a solution to this problem.  It
is easy to see now that the Hundred Years' War wasn't about
English territorial claims in France, but was just a fit of pique
because the French hadn't fixed initial [T] and [D] with their
loanwords at the same time that they fixed [f]/[v] and [s]/[z].
And Richard III wasn't pleading for a "horse" at Bosworth Field,
but for a "source" (obviously for fixing the runaway allophonic
variation of initial [T] and [D]).  Shakespeare just got it
wrong -- but what can you expect from someone who was secure with
his invariant initial [T] and [D] and who never gave a thought to
the struggles that his forebears went through to bring him that
security.

> In my opinion this probably began with the statistics of
> utterances: demonstratives like <the> and <this> commonly follow
> prepositions, which in a majority of cases end in voiced sounds.

And you call functionality murky.  Do you have any other examples
of syntactically conditioned phonetic change?  I have seen a lot
of ontological ingenuity expended by classical phonologists to
devise a purely phonetic environment for some obviously necessary
sound change, but this is a new one.

If the shift to [D] is syntactically statistical in nature (i.e.,
it occurs when a variable sound is pronounced more often one way
than the other), why shouldn't it have affected all nouns
beginning with [T].  After all, nouns are commonly preceded by an
article, either definite or indefinite and these always end in a
voiced sound.  I rather suspect that nouns are preceded by
articles more often than demonstratives are preceded by
prepositions (especially prepositions that end precisely in
voiced sounds).  About the only thing that would make this
scenario play is the fact that demonstratives, being function
words, occur very frequently in spoken language -- much more
frequently than nouns, which are content words.  So why not say
that function words, which are used much more heavily than
content words, were subject to this effect?  Oops, I almost
forgot -- you can't say this because you don't know what function
words are.  It escapes you how 'the' has any more "function" or
less "content" than 'theology'.

Or, if the shift to [D] is syntactically statistical in nature,
why shouldn't it have affected the verb 'think' which is among
the most common lexical verbs in English (rank 64 in the BNC with
153,881 tokens, right behind the deictic 'then' at 63).  After
all, this verb, which requires an animate subject, is most often
preceded by a personal pronoun, all of which (except for the
neuter, which can't properly be a subject of this verb) end in a
voiced sound.  And when it isn't preceded by a pronoun it is most
often preceded by either a modal verb or an adverb ending in a
voiced sound.  If it is because 'think' is not a demonstrative,
then your environment for the sound change is still semantically
or syntactically conditioned, not phonetically conditioned.

Or, if the shift to [D] is syntactically statistical in nature,
why shouldn't it have happened already in OE where, before the
loss of final inflectional n and the eventual loss of all final
short vowels, many more words ended in vocalic sounds than in ME.

> Once the invariant [D] was fixed in the most common
> demonstratives, it would have spread by analogy to the other
> "th"-words of pronominal origin which carry definiteness.

Are there any "th"- pronouns that are indefinite?

Besides, this doesn't account for the spread of this feature to
the conjunction 'though' which does not have any of the
characteristics you deem necessary for this change; i.e., it
lacks definiteness, it never follows a preposition (or even an
article).

This is one reason why the mainstream theory that [D] in these
words is a lenition of [T] in unstressed forms of function words
is easier to believe.  Rather than basing the distinction on
"definiteness," categorizing these words as function words
accounts for all of them (which "definiteness doesn't do [cf.
'though']).  Besides, 'that' is not infrequently used in an
indefinite sense ('That's just the way it is').  So if you want a
"one size fits all" covering term for all the English words with
initial [D] it has to be "function word" not "definiteness,"
despite the fact that definiteness is the most obvious
characteristic of things like the definite article.  Ultimately,
though, I don't think that the phenomenon can be completely
ascribed to the fact that these are all function words.

Lenition in function words is a normal development because
function words are usually unstressed.  It is content words that
tend to be stressed because, well, they carry the content of an
utterance.  Function words are mostly predictable from the
grammatical structure.  I realize that this all sounds like
mumbo-jumbo if you don't know the difference between function
words and content words, but I suppose that at some point you
will just have to learn.

Function words often have stressed and unstressed forms.
Sometimes these are even lexicalized ('of' and 'off', 'through'
and 'thorough', and although not (yet) lexicalized, any native
speaker will know the difference between 'please' and
'puh-leeze', which have similar but quite distinct meanings [one
indicates a polite request, the other is a warning]).  Lenition
of pronominal forms would also be a natural thing, as is shown by
the fact that all the "h"-pronouns have a lenited form without
the "h" ('e, 'er, 'im, 'is, 'ers).  In fact, we even have a
nursery rhyme to imprint on young minds the fact that unstressed
'beside her' rhymes with 'spider'.

These lenited forms of the "h"-pronouns are obviously very old,
because, despite the fact that the third person plural pronouns
with "th" in English are Scandinavian loans, there is still a
native lenited form ('em) from the old "h"-pronoun plurals used
as an unstressed form of 'them' ('Give 'em hell, Harry', 'stick
'em up').  So the lenition of the "th"-pronouns would have been
a completely natural event considering the example of the
"h"-pronouns, which must have already existed before the
replacement of the "h"-pronoun plurals with "th"-pronoun plurals.
The structural similarity of the "th" pronouns and the "th"
deictics provides a strong basis for similar phonetic treatment,
whereas other words with initial [T] lack this structure and are
not affected.  The fact that these lenitions occur in pronouns
may also be significant because other very frequent function
words like 'for' and 'so' do not show any tendency for lenition.

> The situation would then have resembled the pre-Chaucerian stage
> of ME in which one class (recent loanwords) had invariant initial
> [v] and another class (older words) had variable initial [f/v].
> In that case the older class was forced to fix its initial sound
> as [f], establishing /f:v/ as a phonemic distinction. In the case
> under discussion the non-pronominal, non-definite words beginning
> with [T/D] were forced to fix their initial sounds as [T], which
> yielded the fixed opposition in initial [T:D] that is found
> today. I would guess that the process was complete (in East
> Midland at least) by 1500, if indeed it was not by Chaucer's
> time.

An entertaining story, and every bit as plausible as how the
elephant got its trunk or how the camel got its hump, but I'm
afraid that that's all that it is.  It depends on too many
suppositions that are not borne out, and even contradicted, by
the evidence.  In summary, it specifically requires:

   a) That there was a sandhi rule that voiced initial unvoiced
      fricatives in voiced environments across morpheme
      boundaries.  This is a sine qua non for your theory.  The
      only evidence for this is <bucke uerteth> in the Cuckoo
      Song and that requires accepting this interpretation out of
      a wide range of other possible interpretations.  Opposing
      this is the fact that compounds created in OE do not show
      any evidence of this sandhi rule and that OE writes loans
      with initial [v] with either <f> or <u/v> but such
      variation is not with regard to such a sandhi rule but
      simply free variation.

   b) That the importation of loans with initial [v] forced the
      abandonment of the alleged sandhi rule by requiring words
      with variable initial [f]/[v] to be fixed with [f].  This
      process is not illustrated by any of the evidence that you
      have provided.  In fact, the importation of 'verse' with
      initial [v] caused the earlier loan of the same word with
      variable initial <f>/<v> to be fixed as [v].  Rather, the
      evidence suggests that the [f]/[v] distinction has been
      imported along with the words since all the contrastive
      pairs that you have advanced are loan words.

   c) That by analogy to this unproved (and contradicted) fixing
      of words with initial [f]/[v] alternation with initial [f],
      words with "definiteness" with an alleged initial [T]/[D]
      alternation were fixed with initial [D] because
      "demonstratives like <the> and <this> commonly follow
      prepositions, which in a majority of cases end in voiced
      sounds."  No evidence is provided that this is what
      happened, or even that this can happen; this is simply
      presented as a "just-so story."  Since this contrasts with
      the currently accepted theory, surely some evidence is
      needed.  The mainstream view is that the words with initial
      [D] in English are unstressed forms of function words with
      a lenited initial consonant.  Lenition arises not from a
      sandhi rule but because the unstressed form facilitates
      assimilation of the initial consonant to the vocalic
      nucleus.  The mainstream view is supported by the facts
      that:  The forms with initial [D] are all function words;
      function words frequently have stressed and unstressed
      forms; lenition of initial consonants is found in
      unstressed forms of some function words.  The only counter
      to this that you offer is that you don't know what function
      words are or you can't understand how function words work.

   d) That this fixing of initial [D] spread from common
      demonstratives that "commonly follow prepositions, which in
      a majority of cases end in voiced sounds" to certain other
      words with [T], the criterion for the spread being
      "definiteness."  This ignores the fact that words that lack
      "definiteness" have this feature ('though') while words
      that have "definiteness" ('three', 'thirty', 'thirteen',
      'thrice') lack it.

   e) That the fixing of words with "definiteness" with initial
      [D] forced all other words in the language with an initial
      dental fricative to be fixed with [T].  The only evidence
      for this is that all the other words that don't fall into
      this small group with initial [D] do have initial [T].  But
      they might always have had invariant initial [T] and the
      situation would be exactly the same.  [BEEP - BEEP - BEEP
      -- Warning, Occam's Razor violation -- two entities
      proposed (sandhi rule, fixing rule) where only one
      (lenition rule) is required.]

   f) That because this process of fixing initial [T] and [D] is
      analogous to the alleged fixing of initial [f] and [v] and
      of initial [s] and [z] and since these last pairs are now
      phonemic distinctions, the contrast between [T] and [D] is
      also a phonemic distinction.  This ignores the fact that
      there is no predictable difference in meaning between words
      like 'fers' and 'vers' that can be based on the fact that
      one has initial [f] and the other has initial [v], whereas,
      in sharp contrast, all words with initial [D] share,
      according to you, a feature called "definiteness" (however
      you may want to define this so that it covers all these
      words, as long as the definition is independent of the
      sound [D]) and words with initial [T] lack this
      "definiteness."

      In other words, according to your scenario initial [D] in
      English is a marker of "definiteness" and initial [T] is a
      marker of "non-definiteness."  Thus the occurrence of
      initial [T] or initial [D] in a word can be predicted on
      the basis of some inherent semantic or lexical quality of
      the word and vice versa.  So, if what you say about words
      with initial [D] sharing "definiteness" is true, then the
      distinction between initial [T] and [D] is not phonemic
      because the occurrence of one or the other predicts
      something about the meaning of the word in which it occurs.
      Phonemes are units of sound, not of meaning.  Morphemes are
      units of meaning.  If a sound predicts meaning, then it is
      not (just) a phoneme.  If the only contrast that you have
      involves a prediction of meaning, then you can't use that
      contrast to prove a phonemic distinction.

> Hence one should not say that <thy> or its antecedent
> "developed" a voiced fricative. What happened, one way or
> another, is that the pronunciation with the unvoiced fricative,
> occurring after unvoiced sounds, was lost in this word.

Or, it was lost when the (comparatively rare) stressed forms of
these words were pronounced with [D] by analogy to the unstressed
forms.

> So is the initial [T:D] distinction phonemic in present-day
> English? Absolutely; it can and should be written /T:D/.

This is one of the problems that one often encounters with
hypotheses that are based on different interpretations of the
same evidence.  Even when the interpretation is put forward with
due diffidence and properly qualified with appropriate
conditionals (although, admittedly, this was not done in this
case), by the time the conclusions are reached, they are
absolute.  The hypothesizer seems to forget that other
interpretations are possible and the conditionals and modals have
disappeared.  There is a gradual promotion of "possible" (p > 0)
to "probable" (p > .5) culminating in conclusions that are
"absolute" (p = 1).

This is not the dispute the truth value of your conclusion.  It
may very well be true.  All I say is that it doesn't follow
necessarily from the argument presented.  It is possible to have
a false premise and an invalid argument and still reach a valid
conclusion.

> It was phonemic as soon as the process of fixing invariant [T]
> and [D] on particular words was completed, even though minimal
> pairs for these phones probably did not exist at that time.

But the "fixing" of invariant [T] and [D] on particular words is
just a supposition, and an unnecessary one at that, based almost
entirely on the highly questionable <bucke uerteth> as an
indication that there was a systematic allophonic variation in
initial fricatives in English.  Without this alleged variation,
there is no "fixing."  The problem here is that, even if the
supposition is correct, there is still a conditioning environment
involved in the "fixing" of initial [D].  According to the
scenario, only words with "definiteness" have initial [D].  Since
initial [D] has meaning associated with it, it can't be
considered a phoneme on the basis of a contrast with another word
that doesn't have "definiteness".  Phonemes are not units of
meaning.  Initial [D] is a unit of meaning, even in your
scenario.  If you start with salami, no matter how you slice it,
it's still salami.

> Minimal pairs are luxury items. They are nice to have, but one
> can (and in many situations must) establish phonemic oppositions
> without them.

Quite true.  But the availability of minimal pairs is still more
often the rule rather than the exception.  A lack of minimal
pairs is often an indication of fairly recent phonemicization.
It is also a measure of the functional load of the contrast in
the language.  If [T] and [D] are phonemes, then their functional
load in English is very nearly zero.  Obviously all (or even
most) of the phonemes of a language cannot have a nearly zero
functional load, so minimal pairs for phonemes will be much more
common than not.  But then I have never said anything different.
What I have said is that even an apparent (purely phonetic)
minimal pair (like German [ku:c,en] and [ku:xen], or Comanche
[papi] and [pavi], or even English [Tai] and [Dai]) should not be
accepted uncritically as evidence of phonemicity.  There is no
mechanistic way to determine phonemes, and that includes minimal
pairs taken in isolation.

But my point is not different from yours.  If [T] and [D] are to
be established as phonemes, it must be done without using the
opposition of 'thigh' and 'thy' as evidence.

> When distinct phones occur in contrastive distribution in
> corresponding morphemic positions, they are either allophones of
> one phoneme or represent distinct phonemes.

Sometimes they are both, although not in the same situation.
Consider the [T] in 'path'.  Now consider the plural and the
possessive singular of this word.  The first has [Dz] and the
second has [Ts].  Is the [D] of the plural a different phoneme
from the [T] of the possessive singular?  If it is, then the
plural form is a suppletion while the possessive singular is not.
But suppletions are completely unpredictable (i.e., the
suppletive form cannot be derived from the base form by any kind
of generalized rule: e.g., 'am', 'be', 'was'; 'fero', 'tuli,
'latus'), whereas the [D] of the plural is predictable from a
final spirant voicing rule that operates in the environment
[PLURAL].  Now consider the [T] in 'faith'.  The plural of this
word has [T].  Is the [T] of 'faith' the same phoneme as the [T]
of 'path'?  There is no reason to say that they are not.  Is the
[T] of 'faiths' the same phoneme as the [D] of 'paths'?
Classical phonology says that they can't be. Now consider the [T]
in 'bath'.  This word has both forms of the plural, with [Ts] and
with [Dz] (i.e., a speaker can use either one, and the hearer,
even if he has the other pronunciation, will not take the
received form as a different word).  Is the [T] of one
pronunciation a different phoneme from the [D] of the other? If
they are, then the two forms of the plural must be different
words.  But instead, the two forms are simply variants of the
same word, so the two sounds are in free variation.

Similarly, in the morphemes of the plural, possessive, and third
person singular of the verb, the variations of the
(morpho)phoneme //s// ([s z Iz]) are predictable from the
phonetic environment.  But final [s] and [z] are also phonemes
('fuss':'fuzz').  Thus /s/ and /z/ are contrastive in 'fuss' and
'fuzz' but are distributed allophonically in the plural morpheme.
To avoid the confusion and complication of conflicting
definitions resulting from this situation we use a reductionist
strategy known as "once a phoneme, always a phoneme" so that we
don't have to deal the complications of sounds that are sometimes
allophones or morphophonemic alternations of one another and
sometimes phonemes.

While phonemes/allophones may a mutually exclusive
relationship in a particular situation, it is not mutually
exclusive across the entire language.  Thus [k]:[s] in
'kill':'sill' or 'cat':'sat' is phonemic while [k]~[s] in
'public' ~ 'publicity' ~ 'publication' ~ 'publicize' is
allophonic or morphophonemic.  We simply make it mutually
exclusive across the entire language with the axiomatic dictum
"once a phoneme, always a phoneme."  This is another way of
saying that just because two sounds can be shown to be
allophonically distributed in some situation does not mean that
they are not phonemes if their phonemicity can be demonstrated
elsewhere.

If sounds exhibit only non-contrastive (complementary,
predictable, or free variation) distribution across the language
we label them allophones, but if they exhibit contrastive
(non-predictable) distribution somewhere in the language we label
them phonemes.  The problem arises when they exhibit contrastive
but predictable distribution.  Then we can't call them allophones
because they contrast.  But we shouldn't call them phonemes
because they are predictable.  If the prediction is made on the
basis of a grammatical environment, they are usually called
morphophonemes, and if they can be otherwise shown to be
phonemes, then we can invoke "once a phoneme ..."  But in order
to invoke "once a phoneme ..." you have to have that "once a
phoneme."  If you don't have it, you shouldn't use the
morphophonemic alternation to establish phonemicity, because it's
not the same thing.  Phonemic contrasts are arbitrary.
Morphophonemic alternations are based on some predictable
difference in meaning, function, or phonetic environment.
Phonemic contrasts should tell you nothing about the meaning of a
word except that it is different from some other word with
different phonemes.  Phonemes effect meaning; they don't affect
it.

This is where [T] and [D] are now.  They should be phonemes
because [f]/[v] and [s]/[z] are phonemes and the [T]/[D]
alternation was created at the same time, by the same historical
change.  If the others have become phonemes, then so should [T]
and [D] have become phonemes.  But as you have gone to lengths to
point out, [D] did not have the same driving force for
phonemicization behind it that [v] and [z] did, lacking the
massive influx of Romance and classical loanwords that were so
instrumental in the phonemicization of especially [v] and to a
lesser extent [z].

Another reason why [T] and [D] should be phonemes is because
otherwise there is a hole in the pattern, since the distinction
between [T]/[D] is voiceless/voiced which is normally phonemic in
English.  But there are other holes in the pattern of English
phonemes as there are holes in the phonemic patterns of many, if
not most, languages (in fact, Hockett considered "gaps,
asymmetries and 'configurational pressures'" to be a universal of
phonological systems), so this is not sufficient, by itself, to
declare these sounds phonemes.

But in any case you shouldn't use a contrast where [D] indicates
that a word has "definiteness" (in your view) or is a pronoun or
deictic (in my view) and [T] indicates that a word lacks these
qualities as evidence of phonemicity.  Even though we expect [T]
and [D] to be phonemes we shouldn't accept questionable evidence
just because it confirms our expectations.  Once we have evidence
that [T] and [D] are phonemes, then we can consider 'thigh' and
'thy' as a phonemic contrast (even though it isn't) because of
"once a phoneme ...".  But you can't establish the phonemicity of
[T] and [D] on the basis of such a contrast because it really
isn't phonemic.

> Allophones are regularly distributed according to phonologic
> environment.

Yes, but phonologic != (does not equal) phonetic.  While the
greatest part of phonology does involve phonetics, it is not the
only factor that affects phonology.  Phonetics is about speech
sounds.  Phonetics studies speech sounds independently from the
rest of language.  It treats speech sounds as if they have no
other purpose but to be speech sounds.  But linguistically,
speech sounds do not exist to be speech sounds; they exist to
convey semantic information (meaning) between speakers of a
language.  Phonemics is about the relations among the speech
sounds of a particular language.  These relations are
abstractions and exist in the perception of the speakers of the
language.  They are not necessarily detectable from a phonetic
transcription of the spoken language.  The fact that phonemes
cannot be deduced from a phonetic transcription implies that
there are factors other than purely phonetic ones that affect
phonemicity.

> No conceivable phonologic rule could predict the distribution of
> [T] versus [D] in modern English words.

Fine.  Show me a nice fat (or even a thin) content noun or verb
or adjective in English with initial [D] and I will believe it.
It doesn't even have to contrast with a word with initial [T].
The existence of an English word that begins with [D] that isn't
a pronoun or deictic (or that doesn't have "definiteness") would
be sufficient to show that you can't predict which modern English
words begin with [D] and which with [T].  Just one.  So where is
it?

And don't say that phonologic == phonetic.  Phonologic refers to
the relations of the speech sounds of a language in the
perception of its speakers.  Saying that phonologic is the same
as phonetic implies that speakers know nothing about their
language except its speech sounds.  In order to say that "No
conceivable phonologic rule could predict the distribution of
[T] versus [D] in modern English words," you have to equate
phonologic with phonetic.  So if you mean phonetic, say phonetic.
The distribution of initial [T] and [D] is quite predictable in
English.  I can quite easily conceive of a phonologic rule that
predicts their distribution.  Saying that there isn't one is like
putting on a blindfold and then saying "I can't see."

Your claim that classical phonology can't predict the distribution
of initial [T] and [D] in English is simply a limitation of
classical phonology, not a reason for denying the obvious truth
of the predictability of initial [T] and [D] in English.

> They are indeed distinct phonemes.

This may be true, but it isn't established by your argument.

Even if everything you say is true, the difference between words
with initial [D] and initial [T] in English is not arbitrary but
is based on "definiteness."  This is not the basis of a phonemic
distinction.  It's still salami, just sliced differently.
Besides, whether [T] and [D] are distinct phonemes or not is not
the issue.

Bob Whiting
whiting at cc.helsinki.fi