On Welsh voicing

[bwald]

----------------------------Original message----------------------------
Joe Eska writes:
 
> In Welsh, broadly
>speaking, phonemic /p b/, for example, are [p^h p] phonetically, i.e.,
>the contrast is one of phonetic aspiration, rather than voicing.
>There is no process of intervocalic devoicing in Welsh, but there is
>one of intervocalic de-aspiration for voiceless stops, which, in many
>dialects, is accompanied by partial voicing.
 
I feel I am not risking too much to read into this message that he is
suggesting that Welsh fits into "natural" sound change with respect to
voicing in intervocalic position, i.e., that the Welsh varieties at issue
are variably indulging in an EXPECTED sound change: [-asp] -> <[+voi]> /
V_V
(if this is indeed restricted to intervocalic position).  Perhaps there are
Welsh varieties for which the sound change is completed as well, but
Patrick Honeybone's message warns against taking transcriptional practices
or changes in orthography at face value (with regard to culling the
literature on Welsh without listening to the sounds themselves).  A deeper
issue, then, implicates the situation in Proto-Celtic (whether there were
true intervocalic voiced stops), and perhaps even ultimately whether I-E
"glottalic" theory is significantly reflected in Celtic (or Proto-Celtic).
That is, either voiced stops existed at some point intervocalically BEFORE
Welsh emerged, in which case there was devoicing under some conditions (not
made clear), or, as far back as we can go into the ancestry of Welsh there
were NEVER (consistent) intervocalic voiced stops.
 
Alan King's message is more complex, and contains many interesting ideas,
some of which intersect with Honeybone's, where Patrick emphasises the
issue of distinctive or "primary" vs. secondary features.  To take
Patrick's point first, the issue of the nature of sound change remains.  As
far as I know, the idea that [+stop] > [-voiced] /V_V (and ONLY in V_V)
would be an "unnatural" change, REGARDLESS of whether [voiced] was a
distinctive feature of Welsh or not.  That seems to be because the common
change, [+stop] > [+voiced] / V_V has a PHONETIC explanation which leads to
unexpectedness of the reverse change on PHONETIC grounds alone.  It is not
clear to me to what extent distinctive vs. "redundant" features has ever
been admitted among practicing historical linguists as a mitigation of the
phonetic explanation.  It's clearly an issue of basic and general interest.
Can phonology (patterns of phonetic organisation in language) mitigate
phonetically regular (and Neogrammarianly "blind" phonetic) changes?
Without pausing to think of examples, I suppose so, but I don't know about
such a case as intervocalic DEvoicing, with respect to this issue.
 
Now, for Alan's comments and suggestions.
 
>Actually if we're looking at the northern Welsh model, the [+voiced] bit is
strictly unnecessary.  The change can be represented simply as: [+stop] >
[+geminate] / V __ V, which I should think is intuitively more plausible.
Of course, this occurred under the circumstance that voiceless stops in the
specified position were already geminate (for diachronic reasons that we've
>already been into).
 
This feeds into the issue of phonological mitigation of "blind" or strictly
phonetically motivated change.  It did give me pause in my last message,
but I accepted it as a fact (provisionally).
 
Next, Alan writes:
 
>Again, relevant to the Welsh phenomenon is the fact that the gemination
probably only occurs when the preceding vowel is stressed, so another
>modification is necessary: [+stop] > [+geminate] / [V, +stress] __ V.
 
Off the top of my head, the +stress condition may be important, since the
following vowel is then less prominent, and hence its features, such as
voicing, may play less of a role in inhibiting a devoicing motivated by
other factors (such as "pattern pressure").  That is, devoicing following a
stressed vowel may be more "natural" than preceding a stressed vowel (due
to anticipation of its maximally prominent voicing).  Still, I don't know
that the thesis of "naturalness" of intervocalic voicing of stops has ever
considered such a distinction in offering its explanation in terms of
strictly phonetic motivation.
 
Alan again:
 
>I would suggest there are two ways of "understanding" this change.  One is
just in the terms implied by the foregoing notation: northern Welsh
speakers "decided" to geminated all stops between a preceding stressed and
>a following (unstressed) vowel.  "For fun", so to speak.
 
I don't get the point of the personification.  I'll ignore it.  Alan continues:
 
> The other view
might be that since voiceless stops in that position were already always
geminate, they "decided" to *simplify* their phonological system by
extending the "gemination habit" to all stops, regardless of voicing.
 
That seems more plausible to me.  The change is [stop] > [+geminate].  NO
CONDITIONING.
For maintaining distinctions, however, that seems to necessitate accepting
that there were previously voiced stops, which goes against what Esko
seemed to be suggesting.  Actually, it doesn't if the distinction affects
only an inventory of [+/- aspirate] stops. The aspirate dimension remains
after the change, e.g., -tth- vs. -tt-.  The further change -tt- >  -dd- (>
-d-), or whatever, can be seen as maximal differentiation of +asp and -asp,
but intervocalic position might be unnecessary and remains problematic (so
far) as a CONDITIONING factor.
 
Next Alan writes:
 
>Actually they went further than that.  Gemination in the said position was
extended to *most* consonants; notably including /n/, /l/ and /r/.  At this
point we might say that they were really getting "carried away", since
length in these particular consonants (only) had until that time been
phonemic, as it still is in southern Welsh (although phonetically most of
the work to distinguish short and long liquids is probably performed by the
>preceding stressed vowel through a compensatory length contrast).
 
Generalisation to resonants (or whatever you call them) detracts from the
"maximal differentiation" suggestion, UNLESS there were previously resonant
aspirates, which do exist in some languages I'm familiar with, and
therefore is not a priori implausible.  (I suppose Welsh /ll/ might even
count as a liquid "aspirate' -- I'm unfamiliar with its historical sources.
It only now occurs to me that "devoicing" (aspiration?) of Spanish /rr/ in
some varieties as opposed to /r/ may be another case of maximal
differentiation, but I'm accustomed to viewing this tendency, where it
exists, as part of the general Spanish tendency to devoice "fricatives",
cf. Buenos Aires variable /zh/ > /sh/, significantly from /y/ (?and /ly/?),
which becomes /dzh/ in much of Spanish, where prior existence of /ch/ might
play a role in keeping devoicing of the voiced affricate at bay -- there's
also common Spanish /ch/ > /sh/, in many areas, but that's another matter)
 
Alan again:
 
>So far from doing such a strange thing as geminating intervocalic voiced
stops for no apparent reason, we could say that they pretty much had a
"gemination party", geminating nearly everything they found in the position
>/ [V, +stress] __ V.
 
I need to be assured that the conditioning factor is indeed necessary.  I
had previously suggested (on the basis of incomplete data) that the
gemination was unconditioned but, perhaps, has only been recognised in
intervocalic position, i.e., it reflects a more general paradigmatic shift
among stop series.
 
Alan's next comments are most interesting:
 
>Why?  I don't know if it's a motivation or merely an effect, but the
outcome of this development is a very characteristic *staccato, almost
syncopated* rhythm to northern Welsh speech, since most stressed syllables
are pronounced with a short, rather energetic vowel followed by a long and
also energetically articulated consonant which seems to go
>implosion-interruption-explosion.
 
Implicit in this suggestion is a METRIC explanation for the change in terms
of syllabic rhythm in words.  This is certainly worthy of further serious
discussion.  There's a chicken-and-egg problem involved.  It also somewhat
(but not completely) overlaps with my earlier suggestion that Welsh
stressed vowels may be much more prominent and influential on (preceding)
consonants than unstressed vowels, relevantly with respect to voicing.  It
would indeed be interesting (and relevant) if Welsh unstressed vowels had
any tendency to devoice.  I've never heard anything along those lines for
Welsh.  Variable devoicing is a well known feature of Swahili (and some
adjacent languages) POST-stress vowels (which are necessarily word-FINAL,
and voicing alone is not a distinctive feature of Swahili consonants -- the
stops written as voiced in Latin or Arabic letters are imploded in
first-language Swahili, truly voiced stops are prenasalised -- some other
languages in the vicinity borrow words with voiced stops by prenasalising
them, e.g., English "soda" > Gikuyu "sonda", Swahili just implodes the "d"
in such words. Gikuyu doesn't have imploded or simple "voiced" stops.
Meanwhile Gikuyu has variable denasalisation of the prenasals, so the
actual pronunciation can be "so(n)da" without reference to the English
original, or the faintly possible Swahili intermediary.)
 
Alan continues:
 
>Interestingly, in some speakers at
least, the intensity of the stressed and the posttonic unstressed vowels
doesn't seem to contrast as much in this "articulatory style" as in, say,
English or even Spanish (so that foreign ears may have difficulty
interpreting which is the stressed syllable), and it now occurs to me that
this may be explained if we assume that the "staccato-ey" features I've
described have taken over the function of identifying the position of the
stress.  In particular it is common for the pitch of the posttonic syllable
>to be higher than that of stressed syllable.
 
That does not support what I just had in mind concerning the difference
between stressed and post-tonic vowels, but continues the theme of
metricity as a crucial feature of the relevant Welsh sound changes.
 
Alan again:
 
>As I already said previously, stops in general tend to be strictly
voiceless in pronunciation, with other features such as aspiration doing
most of the work of differentiating the voiced and voiceless series.  It
seems to me that in this form of speech intervocalic geminate stops are
*particularly* voiceless: i.e. voicing is interrupted along with everything
else in their intervocalic articulation.  I believe that the voiced and
voiceless series nonetheless remain phonologically fully differentiated in
this pronunciation (although some foreign speakers might be excused, again,
for getting the signals wrong).  But *phonetically* one might argue that
you already have intervocalic devoicing here, at least as a subsidiary
effect - assuming of course that some voicing is there in the voiced stop
>series to begin with.
 
Again, Alan is presupposing an earlier change FROM voiced stops.  For the
moment this has to weighed against the contrary implication of Esko's
remarks, which seem to leave open the possibility that where there is no
voicing in Welsh there historically never was.
 
Alan again:
 
And if that is the case, it appears to me that we might describe the
phenomenon as one of dissimilation.
 
Exactly.  But in the way I described it above as "maximal differentiation"
[+asp] = [-voice]
vs. [-asp] > [+voice].  I continued to question the conditioning factor,
but Alan's comments on metricity leave me hesitant as to what is a more
likely (and even more "natural") account.
 
Alana again:
 
>There are surely enough precedents for
that is phonology?  When intervocalic stops get voiced, that is
assimilation to the voicing of the surrounding vowels, I take it; why then,
when the opposite happens it is surely dissimilation.  The motivation might
be to heighten the contrast between adjacent segments: vowels voiced,
>consonants unvoiced.
 
The whole issue of dissimilation is complex and requires a lot of thought
(I vaguely remember that a paper or a few have addressed this subject at
length in general).  To be sure, if the PHONETIC explanation of
intervocalic voicing is (voicing) ASSIMILATION, then it is most expected
that any attempt at explanation of a reverse phenomenon would begin with
the notion of DISSIMILATION.  I have my reservations, but the  "maximal
differentiation" notion fits into a general notion of dissimilation on the
paradigmatic level.
 
"Maximal differentiation" is a most well-known concept in accounting for
the tendencies of VOWEL systems, but I know at least a few examples where
this can be posited as an underlying motivation for changes in certain
CONSONANTAL systems (even beyond dissimilation of coarticulation effects
for consonants).  EG, among "Swahili" varieties Bajuni exhibits the change
t > ch and nd > ndr, where "ndr" is what English ears hear (and have
commented on), but is simply an effect of the release of the retracted
consonant and is not heard as an "r" release by Bajunis.  The t > ch is a
bona fide palatal affricate, but arose from the same retraction .
Dissimilation is involved to the extent that these two sounds have become
more distinct from the opposing dental stops t. and nd. (which, ironically,
are reflexes of earlier *ch and *nj, still maintained in Southern and
standard Swahili).
 
Finally, "dissimilation" alone is not sufficient explanation for
(problematic) intervocalic devoicing, but I hasten to add that Alan went a
lot further than to suggest that it could be.