Reversal of merger

[Larry Trask]

----------------------------Original message----------------------------
On Wed, 2 Dec 1998, Bh.Krishnamurti wrote:

> Recent postings on the above theme have not shown a genuine case of
> reversal of merger.

> 1. Most mergers create homonyms. In other words two or more words
> which contrast in a pair of phonemes will become homophonous when
> one of the phonemes merges with the other through sound change.
> After this event, no subsequent sound change will ever retrieve the
> two distinct words by recreating the lost contrast. If there are
> attested instances which appear to restore the lost contrast, that
> can happen through borrowing from an older or a different dialect
> which was not affected by the sound change in question.

[snip of Telugu example]

> The question is: Is there any language in which mergers which
> created homonyms were undone by a subsequent sound change.  Henry
> Hoenigswald in his posting of Nov 27 clarified that in the
> neogrammarian framework, 'there is no room for reversal of merger by
> 'sound-change'..

Yes, in the Neogrammarian framework this is so.  The irreversibility of
mergers by sound change has been widely accepted for generations; it was
formalized by Paul Garde in 1961, and Labov has dubbed it `Garde's
principle' -- perhaps unfortunately, since `Garde's principle' was
already in use for something quite different.

However, none of the proposals in my provisional list contradicts this
principle, since none of them proposes that a merger has been reversed
merely by sound change.  Every one of them concludes, for the particular
case(s) for which it was proposed, either that no merger ever occurred
or that a merger occurred and was then reversed by some means other than
sound change.

> 'there was either no merger, or there is more than one line of
> descent'. Here lines of descent refer to different lines of
> transmission, either inheritance or borrowing.

Yes; agreed, at least for most cases that have been discussed.  But some
of the proposals in my list explicitly conclude either that no merger
occurred or that a merger was reversed by the influence of a different
line of descent.

> 2. Reference has been made to the variability between merged and
> unmerged entities in the aftermath of a sound change. In all cases
> of lexical diffusion of a sound change, this happens normally. In
> certain Gondi dialects word initial s becomes h, and later h becomes
> 0. As we go through the affected lexical items, this two-step sound
> change is attested as s,s/h,h, h/0,0, in different words, in
> different social groups and in different areas. The variable items
> attest to the fact that the sound change is in progress. In the
> Southern dialects only 0 forms occur with the change totally
> accomplished and completely regular. This problem is treated in
> detail in a paper of mine which is published in *Language Variation
> and Change* 10:2.193-220. But please note a there is no case of
> reversal of a merger, i.e. s/h goes later to h and not to s,
> similarly, h/0 goest to 0 and not to h.

>>From this I cannot tell what instance of a potential merger is being
examined.  However, if there is a variable sound change of, say, /s/ to
/h/, then surely there remains the possibility that the variation will
eventually be resolved in favor of the original /s/, and that the
variant in /h/ will disappear.  Indeed, I believe I have seen some such
cases reported in the literature, though I can't name one with certainty
off the top of my head -- but see the Australian case below.

And, if the variable change of /s/ to /h/ threatens a merger with a
pre-existing /h/, then I see no reason why the original /s/ - /h/
contrast should not be restored in this manner -- even if that's not
what happened in Gondi.

A possibly relevant case here is the change of /h/ to zero in English.
This has been pervasive in England for centuries, and it has gone to
completion in most vernacular accents, which no longer contrast /h/
even variably with zero.  This merger of /h/ with zero is continuing to
spread in vernacular speech: for example, it has now reached areas of
East Anglia which formerly retained a phoneme /h/.  But it has not
become general in England, since prestige varieties continue to retain
many (not all) earlier instances of /h/, and hence we have a separate
line of descent to consider.

In Australia, though, things are different.  There is evidence that the
merger of /h/ with zero was widespread in 19th-century Australia.
However, no doubt as a consequence of the influence of more prestigious
varieties which had not undergone the merger, that merger has now been
completely reversed in Australian English: today all Australian
varieties exhibit a robust contrast between /h/ and zero.

> 3. Phoneme-like units in expressives, in my view, do not belong to
> the normal phonlogical system of a language.

That depends.  In some ancestral form of Basque, it is clear, the six
palatal and palato-alveolar consonants (hereafter `palatal' consonants)
never occurred in ordinary lexical items at all, but only in expressive
variants of these, or in expressive formations generally.  At this
stage, it was true that the six consonants stood somewhat outside the
ordinary phonological system.  But then things changed.

For one thing, phonemes identical in phonetic nature to some of these
six occurred in some neighboring Romance languages, where they were
ordinary phonemes with no special value.  Some Romance words containing
these segments were borrowed into Basque complete with their palatal
segments.  The resulting loan words therefore contained the expressive
segments without possessing any expressive value.  And, at this stage,
it was perhaps no longer possible to maintain that the palatal segments
in Basque still lay outside the ordinary phonological system, since they
occurred in everyday words of non-expressive nature.  At the same time,
of course, the six segments still retained their expressive value in
native words.

Still later, in Michelena's account, the historical /j/ merged with the
expressive segment [esh], notated <x>, thereby producing a state of
affairs in which some instances of <x> were expressive while others were
not.  At this point, I think, it was hardly possible to maintain that
the consonant <x> still lay outside the ordinary phonological system,
even though some instances of it still retained expressive value.

Finally, please note that in none of my postings (apart from my
endorsement of Michelena's account of the Basque case) have I been
trying to argue that mergers either can or cannot be reversed by any
means at all.  My only purpose is to compile a list of the accounts of
*apparent* reversals of merger which have been reported in the
literature, whether or not these accounts posit any genuine reversal of
merger.

By the way, I now have a seventh proposal, thanks to Laura Wright.
Simplifying somewhat, she suggests that a case of apparent merger may in
fact represent no merger at all in speech, but merely a failure of the
imperfect orthography to distinguish two phonemes which remained
distinct in speech.

Larry Trask
COGS
University of Sussex
Brighton BN1 9QH
UK

larryt at cogs.susx.ac.uk