reversal of merger

[bwald]

I appreciated Larry Trask's enumeration of distinct cases which resemble
reversal of merger.  Among the cases he mentioned, the following is
particularly interesting for the meet:meat:mate case:
 
(4) The merger occurred, but only variably, and speakers retained both
merged and unmerged pronunciations, though they may have reported only
the merged one (Milroy).
 
More examples come to light almost daily, and they tend to be just on the
other side of an isogloss of an innovative merger.  Speakers report the
reference dialect, not their own.
 
According to the Milroys (if I remember correctly -- and I think it is
mentioned in Labov 1994 as well), Belfast maintains a separate "meat"
class, distinct from both "meet" and "mate" in that the "meat" class varies
between both the "meet" and "mate" pronunciations (stressed vowel, that is)
while the "meet" and "mate" classes are invariably distinct.  If I got this
straight -- and I think I have the gist right if not all the details, and
it exemplifies Larry's (4) above (though it shows a class which merges with
two distinct other classes (or vowels) and has no unique vowel of its own
except that it varies between two distinct vowels), then the "residue" in
most current varieties of English may have arisen through collapse of the
variable class into the two invariant classes.
 
That is,  FIRST, Class/Vowel 1 = i: as in "meet", Class/Vowel 2 = e: as in
"mate" and the problematic Class/"Vowel" 3 = i: OR e: as in "meat", THEN,
Class/"Vowel" 3 breaks up into Classes 1 and 2 but eventually no words
remain in both.  This is not neogrammarian sound change, but reassignment
of the words of a distinct (variable) word class to two distinct word
classes.  The problem remains why most of the "ea" words were reassigned to
/i:/, but much fewer of "ear$" words were.  Again, there seems to be
phonological conditioning of a familiar type in English here, such that the
centralising tendency of a following tautosyllabic r inhibited reassignment
of "ear$" words to /i:/ (rather a more mid vowel).  How often in practice
can we distinguish phonologically conditioned lexical partition of a
variable phoneme from sound change?
 
Of course, the Belfast situation may not reflect the situation of the
earlier reported London merger of "meat" and "mate" and then unmerger and
merger with "meet", but, in the absence of compelling evidence to the
contrary, it remains a possibility to the extent that such a linguistic
possibility as the one described for Belfast is attested.  It also goes
without saying that it is quite interesting to contemplate how such a
variable phoneme could arise in the first place X -- collapse of two
"dialects" into one?  (more on this later.)
 
I also remember that when I read Labov's (1994) account of the Belfast
situation I was struck that he did not report phonetic measurements (of
vowel formant structure) for the three word classes at issue, so even if my
characterisation of the Milroy's remarks on the Belfast variable class is
accurate, it is still not clear that there is indeed absolute merger with
the two other distinct word classes, rather than a near-merger (for the
mid-level "meat" and "mate").   This problem is aggravated for
extrapolating to historical London by the problem of whether or not /ey/ as
in "hair", "pain", etc. had merged with /e:/ as in "hare", "pane" etc., in
the relevant time period -- it did later, by the end of the 18th c., and
they are still distinct in many British areas.  That's why I preferred the
"meat"-"mate" example to the "sea" - "say", since "say" descends from a
diphthong and may have never become a monophthong -- again the "reversal"
problem of whether [ay] > [ey] OR [ay] > [a:] > [e:] > [ey] (the "reversal"
if the diphthong ever became a monophthong and then diphthongised again).
 
With regard to Belfast, I do not know if the Milroys ever performed
acoustic measurements to determine whether Class 3 shows mergers with
Classes 1 and 2, or whether they rely simply on ear and self-report
(neither of which is sufficient for near-merger).  I wonder if I should
assume that they did do measurements, given that Labov (1994) does not
discuss Class 3 as a possible case of near-merger.
 
With regard to the whole concept of a "variable" phoneme (one that is
phonemically but not phonetically distinct from one (or more?) other
phonemes), I was struck in the mid 1970s by "unmerger" in second language
phonology.  The particular case I examined was the distinction between /ch/
and /sh/ in the fluent English of L1 Spanish (Mexican) speakers.  Although
/ch/ and /sh/ had the same range of allophones (namely, [ch] and [sh] and
something in-between), the patterns of realisation of English /ch/ and /sh/
words was different, such that /ch/ words were more often realised as [ch]
for the same speaker and vice-versa for /sh/ -- with the same phonetic
conditioning for both (the sound [ch] was more favored word initially for
both classes and [sh] more favored elsewhere -- but still affected in
frequency by the English word class).  Further work has been done on such
L2 processes since then.  Certainly, investigators generally speak of
"developmental" processes in such cases, and L1 acquisition is somewhat
similar, .but the speakers I looked at were adults, some quite old, and had
spoken English for decades (since their early 20s), and their systems
seemed to be stable -- as if they had distinct phonemes /ch/ and /sh/, and
a single pronunciation rule which variably merged them, with different
frequencies in different phonetic positions.
 
Trask's case (2) piqued my curiosity:
 
>(2) The merger genuinely occurred, but just one of the merged segments
possessed a distinctive phonological role in the language, allowing
speakers later to separate out instances of the merged phoneme which had
>this role from those which did not (Michelena).
 
I am not familiar with the Michelena example.  Is the general case the one
that, say, Kiparsky used to explain the general case of Yiddish final
voiced stops.  That is, for verbs and nouns the lexical morpheme may appear
finally or intervocalically due to inflection, e.g., hob-en 'have' but
'hop'  '(I) have.  The phonological rule is: devoice all final stops.
Intervocalic position preserves the word class distinct from the word class
of final voiceless stops, which remain voiceless in intervocalic position.
Generally, Yiddish has revoiced the final stops when they are voiced in
intervocalic position, i.e., Yiddish has withdrawn the rule: devoice final
stops.  Kiparsky proposed that the former rule can be detected where there
was no alternation, e.g., with the adverb 'avek' ("away", historically from
'a-veg').  This kind of "unmerger" is apparently what Halle was hoping he
could establish for the shift in "ea" words, but it turned out to be an
inappropriate example.  But, now, what is the Michelena example?
 
Last but, most emphatically, not least, I want to apologise to Alan King if
I misrepresented the thrust of his remarks on the complexity of linguistic,
even phonological, even phonetic, change.  As I've said before, I am
grateful to Alan for his frequent insightful examples and comments.  The
Welsh problem deserves to become a classic, and I do not doubt that lasting
fame will come to the linguist (or group of linguists) who manage to make
the linguistic community at large aware of the general theoretical
significance and challenge posed by the problem (and to who/mever "solves"
the Welsh problem).