LINGUIST List 7.12

Wed Jan 3 1996

Sum: Unusual sound change

Editor for this issue: Anthony M. Aristar <aristartam2000.tamu.edu>


Directory

  1. benji wald, sum:unusual sound change? t>h, t>ch

Message 1: sum:unusual sound change? t>h, t>ch

Date: Tue, 02 Jan 1996 21:16:00 PST
From: benji wald <IBENAWJMVS.OAC.UCLA.EDU>
Subject: sum:unusual sound change? t>h, t>ch
As I waited for responses, I made notes for a sum. The notes 
eventually became long, but (I thought) worthwhile 
and informative, something for everybody. I gratefully 
acknowledge responses from the following:

Leslie Barrett
Andrew Carnie
Richard Sproat
Caoimhin P. ODonnaile
Alan R. King
Johannes Heinecke
Henry Rogers
John Phillips
Luciano Giannelli
Bill Fisher
John E. Koontz
Jeanine Ntihirageza
Marc Picard
Paul Foulkes

A lot of interesting issues came up one way or another.

To anticipate further discussion, I did not receive any messages to
indicate that "t > h" via the route evident for Sabaki recurs
elsewhere. However, that does not convince me that it should be
considered an unusual sound change in the manner of the arch
example of the correspondence between Armenian [erk] and IE *dw.
Incidentally, Marc Picard cautioned me to distinguish between
"correspondences" and "changes". But I didn't understand the
point until I thought of this example (he remained abstract).
I still don't know how this correspondence came about, because
even though it recurs ad nauseam in the literature, it is always
to make Marc's point, and I guess it is only explained in specialist
discussions. It is obviously not a sound change, but at best a
coincidental series of such things. In contrast, the following sum
and message is about sound changes. The main question (which I will
not answer) is: why should there be an "unusual" sound change? If a
sound change CAN occur, why doesn't it occur more often? There will
be hints about the answer to this mystery below.

Many of the messages were about the Irish and Gaelic change of t > h,
apparently via the stage theta. Strictly speaking, then, this is not
the "same" as the Bantu *t > h change, since it was clear from my
introductory discussion that the Bantu change was from post-alveolar
rather than dental "t" (because I mentioned that the intermediate
voiceless /r/ was attested). Nevertheless, I appreciated the example
,
and could not fault the respondents, since dental and alveolar t's
are both t's, as far as "t > h" is concerned.

(Here I deleted some personal ranting about how I got marked down
for what I said about t's and the biunique phonemic principle on my
intro to linguistics midterm.)

In fact, whether the *t is dental or not, the principle may be more o
r
less the same. First it weakens to something retaining some of the
buccal features, and the voicelessness, and finally it weakens to the
ultimate voiceless segment /h/. Meanwhile, the dental t also made me
wonder what happened in the course of Germanic, because Grimm's law
shows that original *t was dental, but the t that comes from IE *d ha
s
become alveolar in Germanic, for the most part. Why?

An unclarity about the Irish case, for me, is whether the intermediat
e
theta fell together with *s or not, because *s > h also occurred in
Irish -- and *s > h is a very common change in the world. For Gaelic
I was informed that theta > s occurred first. Anyway, theta directly
to h is also attested. The proto-variationist observations of Gaucha
t
and later Hermann for the Romansch spoken in Charmey came to mind. T
here
theta had started to vary with h (variable theta > h), and the variat
ion
continued over the 20 year observation period. No intermediate stage
s
were reported, and I haven't checked what the earlier source of theta
was (but Latin t seems likely -- what else could it be?) It's time
somebody went back and saw if theta has been eliminated yet (if the
population hasn't switched to Swiss-German or Swiss-French?).

The main lack of parallelism of the Celtic case and Bantu t > h is
that the Irish-Gaelic t > h is a CONDITIONED change, by intervocalic
position, so that initial t often remains, except where the
conditioned change became morphophonemicised in the famed Celtic
"aspiration" (lenition), e.g., Irish tigh "house" but
mo thigh "my house", where graphic "th" in Irish is pronounced /h/
because of the change. (And if the great Irish writer GB Shaw had
actually been literate in Irish, he wouldn't have thought English had
the most "illogical" orthography in the world, from a "phonetic" poin
t
of view -- Irish orthography is extremely etymological, because of th
e
multiplicity of sound changes postdating the orthographic conventions
,
only slightly amended, though mercifully, since the spelling reform o
f
1948.)

(NB. Grammarians always comment on the inappropriateness of the term
"aspiration" applied to Irish lenition processes, but in Irish
orthography the lenition is marked by putting an "h" after the
consonant affected, e.g., t > th, s > sh, m > mh etc., so it's
really a reference to the spelling -- probably scholastically based
on the Greek "aspirates" p*h*i, t*h*eta and c*h*i, as we spell them i
n
the Latin alphabet.)

I did speculate that the Bantu *t > h originated in intervocalic posi
tion,
but the evidence is sparse, and in both areas of occurrence it comple
tes
as an unconditioned change. Later I will indicate why paradigmatic sh
ift
of *t > h might be expected in Bantu, given the prior evolution of th
e
voiced stops, but not necessarily for Celtic. In fact, logical as
the Bantu *t > h is, I received no messages to indicate the same rout
e
in any of the other cases.

Even more restricted than Irish *t > h is Arabic *t > h. Henry Roger
s
reminded me of it. It only takes place for the feminine suffix -at-,
as in ma:lik-at- (ruler-ess-) "queen", in pause position. That is,
ma:lik-at > ma:lik-ah. In contrast to the other cases, it doesn't
take place "intervocalically", which in Arabic would be in a
genitive construction (called the "construct state" in Semitic), as i
n
ma:lik-at an-na:s (ruler-fem.sfx the-people) "queen of the people".
It is also reconstructed for Hebrew and Syriac and deduced for Punic,
in all of which the result in orthography is -a(:)
 *-at > -ah > -a: >...
Was there an intermediate stage between *-at and -ah?

In Hebrew it remains -et in the construct state. Interestingly, in
Arabic the feminine suffix is written by a special letter which is
a cross between Arabic "t" and "h", called "ta' marbu:t.a" (as if a
"t"/"h" ligature, cf. Irish "th" -- but presumably pronounced as
theta in Irish before that changed to "h").

It's hard to say what the actual conditions on this sound change are,
or even if it's really a "sound change" since it only applies to one
particular morpheme. Some Semiticists have pointed out that a suffix
-a: occurs independently as a feminine suffix, as in the feminine
form of Arabic colour terms, e.g., bayd.-a: "white f.", s.afr-a:
"yellow f.", etc. (the masc forms are a-byad, a-s.far, etc), where
there is no evidence for the origin of this -a: in -at, and it is not
reflected in the Arabic spelling. By the way, the Semitic *t is
dental, not alveolar. (Northern Swahili borrows Arabic words with
a dental "t" but English words with a post-alveolar "t"; more about
this distinction later.)

Luciano Giannelli reminded me of the Florentine change in which t > h
in some verb endings -Vto (past participle) and -Vte (2p present indi
c).
Change t > h, also a few lexical items (but not all?) with the same
environment, e.g., patata > pataha "potato". This is attested from t
he
19th c onward, and follows on the heels of the general Tuscan lenitio
n
process which started with k > h (via x, hence called "gorgia toscana
",
once held to be an Etruscan substratum by unjustifiable myth, an exam
ple
of why substratal theories of sound change had a bad reputation in
much of the 20th c). In a more restricted area, including Florence,
t > theta followed, and finally p > phi. The t > h is peculiar and
implies, phonemically at least, t > k for the restricted contexts I g
ave,
which are mainly grammatical suffixes. I wonder about the lexical it
em
patata > pataha because of the two successive t's. It reminds me of 
the
Spanish alternation in the diminutive suffix between -ito and -ico, a
s in
chiquit-ito/ico "real little (and probably cute for that)". In any
case, lenition remains conditioned and the status of Tuscan t > h as
a Neogrammarian-type sound change is highly suspect. Seems to be an
abrupt change in "underlying spelling" from /t/ > /k/ for specific
lexical items in a specific phonological environment.

Bill Fisher informed me of a Yucatec Mayan morphophonemic change
(unglottalised) stop + same stop > h + stop, e.g., tt > ht.
There is always a morpheme boundary between the two occurrences of
the same stop. We can probably assume a sound change, though Bill
stopped short of that. He also clarified for me that it's optional,
so it may also be a fairly recent change, which would be sufficient
to indicate that there are no intermediate stages. In characterising
the nature of the change, I would opt for something like reduction of
a geminate to a single "long" segment and then delay in occlusion of
that segment (because of the "internal" morpheme boundary?), causing
the initial /h/ impression because devoicing precedes the occlusion.
If my interpretation of the process of change is right, it's caused b
y
changes in the coordination of timing mechanisms for articulation
(something that happens, for example, in all the classic assimilation
cases, but in a different way here, and there are no intermediate
stages worth mentioning. It shares with other t (and any other
voiceless stop) > h the reduction of a segment (or "half" a segment)
to voicelessness alone (oh, and egressive pulmonary air flow of cours
e).

Another example of what I would interpet as changes in the coordinati
on
of timing mechanisms is offered in a later message I received from
Jeanine Ntihirageza. She reminded me that in KiNyarwanda and KiRundi
(interior Northeast Bantu) the aspirated voiceless prenasalised stops
(often?) lose the stop component. The way she put it was:

" In Kirundi and Kinyarwanda (more strongly in Kinyarwanda than in
Kirundi) rapid speech, /p/, /t/, and /k/ are so strongly aspirated
after nasal sounds (in which case they are normally prenasalized) to
the point that they become /h/, e.g./u-mu-ntu/ ----> [umunhu] 'a pers
on';
/mpa/ --->[mha]'give me; /i-n-ka/ ---> [i-velar nasal-ha] 'a cow'."

This looks like a conditioned variant of t > h (also p, k > h followi
ng
the previously homorganic nasal). The thing that most interested me
here was the part about "rapid speech". My own observations about
Nyarwanda at least is that the change is complete at all tempos.
In Shambaa, a language much closer to Sabaki, there is still a great
deal of variation, so that sometimes I transcribed the voiceless stop
as a superscript mun[t]hu, etc. I have long been of the opinion that
the route of change is one of timing mechanisms involving aspiration,
cf. my suggestion for the Mayan observation above. That is, the
aspiration ("caused" by the homorganic nasal) is anticipated to such
an extent that the distinct "explosiveness" of the stop is overwhelme
d.
Nothing intervenes between the released occlusion of the nasal and th
e
aspiration of the (former) stop. In Shambaa, the superscript for the
stop intended that it was very "subtle", sometimes difficult to hear
,
and in the case of (n)t(h) even "flappy" (in its brevity). Sometimes,
often in fact, in Shambaa the stop was clearly there. In Nyarwanda
it just wasn't there at all (but is clear to speakers because of the
morphophonemics, as in the example Jeanine gave, gu-ha < *ku-pa 'to-g
ive'
but m-ha <*n(i)-pa via m-pha 'me-give'; where unconditioned *p > h is
part of the same lenition process as in Sabaki, but *mph > mh is a
totally different and unrelated change). The change is also common
in a large area of Tanzania, so that, for example, the language that
used to be referred to in print as Nkwele is now sometimes spelled
Ng'hwere (one of numerous examples of alternate spellings among
African languages designed to confuse casual readers).

To complete the story on this change, I should explain that in a
large part of the Bantu area, the prenasalised voiceless stops,
i.e. *mp *nt, *nk, are aspirated on release, as mph, nth, nkh. I
have to pause to ask here if this also counts as an UNUSUAL change?
 *NC > NCh where C is a voiceless stop & N homo-nasal
At least it can't be any more usual than prenasalised stops to begin
with -- so that even illiterates syllabify simba 'lion' as si-*mb*a,
etc NEVER si*m-b*a, as English speakers would. In Swahili, for exampl
e,
the English syllabification is not pronounceable. Swahili orthography
does not distinguish mbaya 'bad person' *mu-B > m-b, with syllabic m
and implosive 'b', from mbaya 'bad' *N-b > mb, homorganic nasal and
explosive voiced stop, the only context for explosive voiced stops.
There will be reason to further discuss aspiration and prenasalised
stops later. One point will be that prenasalised stops in Bantu
are single segments, so, of course, the stop component cannot be
pronounced independently, e.g., for syllabification.

Moving on to theory-inspired messages, Paul Foulkes (p.c.) particular
ly
appreciated the point about intermediate stages in my original postin
g.
He had written a dissertation of p > h and observed that there is alw
ays
an intermediate stage (he probably means in the unconditioned and
intervocalic cases). This was important to him because he proposed
that the lenition of p and the reduction of lenited "p" to /h/ had
different motivations. So, he proposed that lenition of p has an
articulatory motivation, as I understand him, but the further reducti
on
to /h/ has an acoustic motivation. Projecting this thesis onto the
*t > h change I had mentioned, he began with the question:

"Of course, a major problem is 'what is [h]'? The label [h] covers so
many things"

As he developed this theme, it became clear that he was suspicious
that in some of the languages I mentioned what was described as h
was actually at some intermediate stage, presumably with some additio
nal
features from its origin. It seemed that he assumed I had gotten my
information from secondary sources which were not trustworthy, rather
than from direct observation. All I could answer is that the /h/ is
like /h/ in English, especially British English. It's voiceless in
deliberate speech and citation of words (translated from Swahili), bu
t
sometimes it's voiced intervocalically, and in connected speech it's
often omitted with or without an effect on syllabification of the
vowels it separates when it's there. I should have added that
other than that, there are no features that would betray its origin
in *p or *t or anything else.

Similarly with interest in the intermediate stages, which are indeed
of central importance to understanding *t > h (and any other "unusual
"
change), Marc Picard (p.c.) pointed out that stages of the Bantu chan
ge
occur independently elsewhere. Thus, he recognised the tapped Americ
an
English /t/ as akin to the flap that I described for the voiceless /r
/
intermediate stage (or more similar to the occasional flapped "t" in
Mombasa Swahili). Then he went on to point out that European
Portuguese has the rolled r, but (Northern) Brazilian Portuguese has
an /h/ reflex of /r/, to which I added that the Puerto Rican Spanish
rolled ("double") rr is often pronounced as a velar like in French et
c.
The implication is that Brazilian may have passed through this stage
in the evolution of r > h. This might seem to support what I
understood to be his thesis that the Machame R (Parisian voiceless R)
should be counted as a stage in *t > h even in Sabaki. I had suggest
ed
this for Siha (and will back off that below), but I disagreed for Sab
aki
since no R stage is attested. I don't see why there had to be one. T
hat
is, a velar or uvular does not seem to be a necessary waystation to /
h/.
That doesn't happen with s > h, and obviously wasn't happening in the
Charmey theta-h variation, so why should it be necessary with hr > h?
In essence, my objection is that Occham's razor should not be applie
d to
THEORIES about possible paths, but to EMPIRICAL EVIDENCE, there being
 no
empirical evidence for a R intermediate stage between hr and h in Sab
aki.
Yes, I realise that that's a strong statement for people doing
historical reconstruction, but all the better for us sound change in
progress devotees.

I may have misunderstood, but I thought that Marc's method insisted o
n
putting in all the stages for all languages, if any language anywhere
had such a stage, to create one invariant universal path, whether the
re
was evidence for them all or not in some other case. Just on general
principle, that method, whether it is Marc's or not, should not go
unchallenged, and I touch on counter-evidence in some discussion
below -- but it's not my main concern.

So far then, we don't have anOTHER example of unconditioned t > h.
Promising was John (Koontz)'s posting on Siouan. It was somewhat
tentative. Something perhaps unpleasant to contemplate is that
Siouan *th > Dakotan h should be analysed as t+h > h, as in Seneca --
with subsequent t > 0 /_h (from Pat Crowe's posting). The point of
Pat's posting was that what might look like t > h in Seneca is actual
ly
th > h, "simple" loss of a segment. The only way I could interpret t
his
observation, interesting in its own right, as relevant to discussion 
to
*t > h, was that it was a gentle caution to researchers not to jump t
o
hasty conclusions in searching for further examples of *t > h. (Like 
if
someone were to misconstrue the older English alternation "them/hem",
with later "hem" > 'em, as evidence for *th > h > 0, where the /th/, 
in
fact, would be an earlier theta descending from IE *t, hence *t > h -
-
I think not) Anyway, Pat's caveat was not a reaction to John's Sioua
n
observations since they were posted at the same time.

To quote from John's posting:

"The mechanics of the process of *Ch to h in Siouan are not clear.
> There are instances of regular shift of *p (not *ph or *hp) to w (i
n
> Winnebago and Chiwere), and it is possible, but not at all certain,
> that *W and *R (not *w and *r) originated in PMV as lenitions of *p
> and *t (not *ph and *th). A more likely scenario than lenition in 
the
> Siouan context seems to be *Ch to *CH, where H is some sort of back
> voiceless fricative, with subsequent reduction of *CH to *H and mer
ger"

John's final suggestion above is not far from the Seneca case. The
*W and *R as lenitions of *p and *t would be typical lenitions a la
Sabaki, but they do not culminate in h, as far as I understand.

Moving on, now, to further discussion of the elements involved in
understanding "unusual" sound changes like *t > h, first, for those
who want a published source to refer to, instead of, or in addition t
o,
my postings, the following discusses the various *t changes in Sabaki
.

 Derek Nurse & Thomas J. Hinnebusch (1993) Swahili and Sabaki:
 A Linguistic History. University of California Press.
 Vol. 121. 67-71.

My observations are from my own fieldwork (unpublished because my
focus of interest is on grammatical rather than phonological change).
Some of the things I'll say are more detailed, or just different from
N&H, maybe because they are more interested in reconstruction and
genetic classification than in the processes which resulted in the
current reflexes. For example, H (who did the phonological chapters)
represents only the voiceless bilabial fricative as the Pokomo reflex
of *p. The various dialects I encountered often had affricate reflex
es.
Since lenitions of *p are more widespread in the area than those for 
*t,
it is apparent that it is the older sound change in the area as a who
le,
as H deduces. However, with respect to Pokomo, as I suggested in my
original message, *t > h, which only affects the Southern dialects of
Pokomo, has occurred FASTER than *p > h (assuming, faultlessly, that
Pokomo is headed in that direction). Northern Pokomo has the intermed
iate
stage of voiceless flap or tap (often a voiceless ROLLED /r/). At th
e
same time ALL Pokomo has the bilabial affricate and fricative reflexe
s of
*p, older, but closer to the original stop sound than what has happen
ed
to *t in Southern Pokomo, whither the change must have spread from th
e
Miji Kenda languages to its south. *t > h occurs in all Miji Kenda
languages, alike those in the North, which have *p > h, and those in 
the
South, which have *p > beta.

The interesting point for theory is that the area indicates that
 stage and age
are quite different phenomena. *p > "whatever" may be older
throughout the area, but *t > h is MORE ADVANCED along the discernibl
e
path of change in Southern Pokomo and (Southern) Miji Kenda, because
*t has lost ALL its buccal features but *p has not. Alternatively, i
n
Pokomo as a whole reflexes of *p are so close to their origin that it
is hard for me to shake the idea that it is more recent but has sprea
d
throughout Pokomo FASTER than *t (> voiceless r) > h. The alternativ
es
are a problem of interpretation for which current theory gives no
guidance, as far as I know. *So file that as an ISSUE.*

To give some data for the sake of concretising, N&H offer reflexes
of *tapika "vomit" (still so pronounced in urban Swahili), which has
the virtue of containing the three original principal voiceless stops
of Bantu. I'll use hr for voiceless apical flap, F for voiceless
bilabial fricative (phi) and V for voiced bilabial fricative (beta).

*Sabaki * t a p ika (unchanged in urban Swahili)
Northern Pokomo hr a (p)F ika (I add the affricate element /p/ here
)
Southern Pokomo h a (p)F ika (ditto)
Northern MijiK h a h ika
Southern " h a V ika
Vumba r a (p)V ika (I add the affricate element /p/ here
)

Vumba is a rural Swahili dialect, on the periphery of the Southern
Miji Kenda area. So the vertical alignment approximates areal
contiguity, and you can draw isoglosses which put Northern MijiK
at the core (and the origin?) of the most advanced changes. South
of that *p and for Vumba also *t deviate from the path of h-hood.

The Vumba (p)V is an attempt to represent the "tenseness"
of the bilabial along with the indifference to voicing which
often accompanies it, but it never sounds like a "cardinal"
*voiceless* bilabial fricative or affricate. That is not the dimensi
on
along which it is varying. It is much closer to the Southern Miji
Kenda reflex than to the Pokomo reflex. In pronouncing the sound
myself it seems to me that the tension is in the lips, where a real
voiceless released stop /p/ generally also has more lip tension than 
a
bilabial (or labio-dental) af/fricative. I think this is also true of
English, and certainly is of urban Mombasa Swahili, where there is no
distracting "aspiration" (delay in following vowel onset). Mombasa h
as
a distinct aspirated /ph/, cf. phaa "antelope" ( < *mpala) vs. paa "r
oof"
( < *(li-)pala). The other Sabaki languages also once had the contra
st.
The aspirated voiceless stops are not subject to the change. Thus,
the contrast is reflected in Vumba as phaa vs. (p)Vaa, and so on for
the other Sabaki languages (e.g., No.MK phala vs. hala).

In Bajuni (a Northern Swahili set of dialects on the periphery of the
Southern Pokomo area, N&H call it T'ikuu) the reflex of *tapika is ch
apika,
where /ch/ represents the voiceless palatal affricate, as in English.
I'll discuss the Bajuni unconditioned *t > ch change later, since I t
hink
it is most unusual and significant.

I also mentioned that *t > h recurs in the Siha dialect of Chagga.
Derek Nurse discusses dialectal sound changes in his dissertation on
Chagga, but I don't have the reference at arm's length (the published
version is referenced in the N&H volume). Again, I observed these
things directly in my own research. Some examples from my field note
s:

 *moto "fire" *teta "speak"
 (unchanged in urban Swahili) (unchanged but means "bicker"
 in urban Swahili)
Bosho mudo DeDa
Vunjo mOhrO zhezha
Machame moRo ReRa
Siha muho heha

I'm approximating my narrow transcription here. O for "open o".
D for "implosive" alveolar stop (the Swahili "d") -- doesn't
seem to be on the path to t > h, but another way for *t to go in
Chagga. (Some may question this, but (pre)voicing derails from the
track to /h/.) Vunjo /zh/ for the voiced palatal fricative, but
sometimes it sounds like the Czech palatalised r (with hachek). In
Vunjo there is general voicing variation for /(h)r/, so I don't think
it's on the track to /h/, but rather to voiced /r/. In Machame, R is
my feeble attempt to represent the VOICELESS velar (uvular?) liquid
that I said sounds like Parisian /r/ (as in"t*r*ois"). It's a very
noisy sound, maybe like Northern Dutch "g". Non-linguists would call
it "guttural". (I left out the tone markings for Chagga words as
irrelevant to the issue here.)

Although I suggested a connection between Machame R and Siha /h/,
one piece of data I found in my notes makes me want to reconsider.
This is the Chagga word *tende for "leg/foot" (Swahili mguu).

Bosho DEnde Vunjo rEndE Machame (I didn't get it, should be REnd
E)

BUT Siha hrende with a beautiful trill on the "r" (cf. Vunjo word
for "fire" with voiceless "r", but voiced "r" for "leg" = voicing
variation for Vunjo "r", not distinctive.)

Lexical diffusion of hr > h? In any case, it suggests to me that
the path in Siha may have been the same as the Sabaki path *t > hr >
 h,
without the Machame detour hr > R.

Incidentally, in ALL Chagga *p > h, and then as elsewhere, e.g., in
Sabaki, in connected speech intervocalic h of any origin is voiced
and/or omitted. Thus, there is no problem in Chagga to see the
lenition process starting with *p and then spreading to *t.

Again incidentally, lenition processes including *t > r occur
sporadically throughout Bantu, e.g., in Mozambique and South Africa,
and in Cameroon (where the result may actually be more l than r-like)
.
I think it is the continuation of the lenition process that over
most of the Bantu area first affected the "voiced stops", so that
 *b > beta, w, 0
 *d > l (sometimes varies with r), 0 (in some areas)
 *g > gamma, 0
(In Kamba, for example, adjacent to Sabaki, all the voiced stops were
lost, resulting in many words with successive vowels, e.g., *gaba > a
a
'distribute', *deda > ea 'cry', and the delightful n-o-o-e-iw-E <
*n-o-god-ed-iw-E Foc-he-buy-for-Passive-Perf 'he was bought it'; the
only shift in voiceless stops is *p > beta, as in Southern MijiKenda)
.

There is no doubt that these sounds descend from stops, since the
non-apical ones remain sporadically in some areas, but there is much
doubt, at least in my mind, that voicing was the primary dimension
that distinguished them from the *p *t *k series (as opposed to the
prevoiced implosives, for example, "voiced" (or "gulped") correlates
of ejectives/glottalised consonants, cf. IE glottalic theory. These
sounds occur almost everywhere as voiced stops when prenasalised,
cf. Swahili n-dimi (tongues)/u-limi (tongue).

Essentially, I would say that in the Bantu consonantal paradigm
lenition of "voiced" stops is UNCONDITIONAL. The prenasalised voiced
stops are another series in the consonantal paradigm, i.e.,
*mb, *nd, *ng, and enter into morphophonemic alternation with the
"voiced" stops for clear historical reasons. Lenition of the voiceles
s
stops, where it occurs, shows the same UNCONDITIONAL shift (and recal
l
that the prenasalised voiceless stop series becomes aspirated in Saba
ki
and a much larger Bantu area.)

BUT, while *t > r is not unusual in Bantu, and while, unsurprisingly
crosslinguistically, *p > h is quite widespread, *t > h (by whatever
path) seems to be confined to Siha and Sabaki. Thus, it seems
that *hr (< *t) most commonly voices in Bantu and becomes a second
liquid taking a place in opposition to *l (which remains in most
Bantu as /l/, the reflex of deeper reconstructed *d).

The rarity of *t > h, as opposed to *p > h, leaves questions for a
theory that sees *p > h as a catalyst for *t > h, rather than the
more common *t > hr > r (once it voices, forget about h). Such a
catalyst may have INITIATED *t > h in Sabaki (cf. Northern Miji
Kenda example *tapika > hahika) and independently in Siha, but then
*t > h took on a life of its own, as in the rest of the Sabaki *t > h
area, where *p > h hasn't happened (cf. *tapika > Southern Miji Kenda
haVika, Southern Pokomo > ha(p)Fika.)

Thus, it seems necessary to distinguish the origin of a sound change
(in one area) from its spread still as a perfectly acceptable
Neogrammarian sound change (to adjacent areas), not a difficult conce
pt
to accept, but one that I haven't seen emphasized in the literature o
n
sound change (and which also applies to all other forms of linguistic
change, so that "linguistic", as opposed to "social", motivation
for a change will not always be evident unless you can locate the
*original* area of the change, an exercise in dialect geography.)

*File this as another ISSUE, not generally attended to in the
literature on structural motivations for change, but somewhat
recognised in the sociolinguistic literature on change -- at least
to the extent that sociolinguistic factors are quite capable of
frustrating structural factors in linguistic change.*

[cf. From Alexis Manaster Ramer's Summary: Relative Chronology of
Sound Changes ... Lauren Sagart (responded) "In Gan dialects of
Chinese, One change (call it change a) turns th- into h- over a
certain area ... Then new th-'s are recreated by a subsequent change
in certain parts of that same area AND BEYOND (ie, where th- had not
been destroyed): that change (call it change b) creates new t-'s and
th-'s out of original post-alveolar affricates tS-'s and tSh-'s ... M
y
interpretation is that (b) is a response to (a), that (b) arose in th
e
area where (a) had taken place, but that later on (b) continued
spreading *without reference to its original motivation*, spreading t
o
areas where (a) had not taken place."]

My emphasis with the *'s. "original motivation" = "linguistic motivat
ion".
By the way, notice *th > h (= a), with no question that the initial *
th
is a single segment, and *t(h)S > t(h) (= b).

The issue of *p > h as the catalyst for *t > h, as the *original*
motivation for *t > h (as opposed to the SPREAD of the latter), is
not something I'm proposing seriously, but simply for the sake of
making a point. The point has to do with what's lurking behind the
entire discussion of "unusual" sound changes. The question is: if
it CAN happen, why doesn't it happen more often? I guess we'll just
have to see what the unusual sound changes turn out to be, and
what linguistic motivations can be adduced for them, before we
can decide whether this is even a legitimate question -- any more
than "explaining" why subject-object order (whatever that means) is
more common than object-subject order. Maybe it's just historical
accident (from a linguistic as opposed to a sociopolitical point of
view).

In this context, Labov's (1994) book on paths of changes in
vowel systems implicitly dismisses such a question by illustrating
the various routes that vocalic chain shifts can take, e.g., whether
the highest back vowel fronts or whether it diphthongises and heads
for /aw/. No explanation for alternative paths is given, and the
principles given imply that asking why one path occurs more commonly
than another is devoid of linguistic interest. This is not yet
obvious for "unusual sound changes", but these seem to be changes
in consonants, not vowels, so the implications of Labov's study may
be more restricted. We'll see (or we won't, as the case may be).

Anyway, what about Bajuni
 t > ch
Completely regular Neogrammarian sound change. Any other examples
of UNCONDITIONED t > ch in the world?

Koontz (p.c.) proposed one to me:

"Winnebago (...they prefer their native name Hochank ...) has c^
everywhere for *ht and *t#, j^ everywhere for other *t and *th. It d
oes
retain t? for *t?, and there is a [d], written t, from *R (not *r). 
One
presumes that this began as a normal shift of *t to *c^ before front 
vowels,
and then was extended to back vowels, too. There is an obvious analo
gical
pathway for this in Siouan, as many verbs end in e in some contexts (
like
the singular) and a in others (like the plural). There is a tendency
 for
the plurals in -ta to switch to -ca to reduce root allomorphy. This
tendency is observed in, e.g., the Dhegiha languages Osage and Kansa.
"

So, for understandable reasons given our textbooks' first example of
an explicable sound change (palatalisation), he seems to equivocate o
n
this sound change by speaking of an "analogical" (Neogrammarians
contrasted this with "real" sound change) pathway in Siouan. But
the real attempt to reconcile the change with what we already think
we know is the part about "one presumes" it began with front vowels.
If we always assume a conditioning factor to make a sound change look
like one that we already know about, we won't get very much further
than what we already know -- but this is a typical problem for
historical linguists to make their reconstructions look plausible.
Next we have no control on our understanding of generalisation, if we
allow the conditioning factor to simply disappear, to end up with an
unconditioned sound change. Such a device will, of course, allow us 
to
go from any sound to any other sound, something which we can probably
 do
anyway, but which, in my opinion, is not very interesting except as a
curiosity to fascinate non-linguists if we can keep their attention t
hat
long. All this is not to say that John's explanation cannot be right
for the Siouan case, but it has nothing in common with the Bajuni cas
e,
if it is. In further communication, when I suggested the preceding
criticism to John, he said that he was aware of it and agreed. He ad
ded
that the extension of *t (add appropriate features) > c/j to /_ back 
vowel
might have been facilitated by retroflexion of the back vowels, somet
hing
he had personally observed in some other Southeast Amerind languages,
e.g., Quapaw, and is an areal feature. It will be seen that mention
of "retroflexion", I'll speak of "retraction", moves us closer to the
Bajuni *t > c change, but not because of any influence of adjacent
vowels. So, let's examine that case, starting with some previous
changes that set the motivational stage.

First, Northern Swahili (including Bajuni) is differentiated from
Southern and standard Swahili by the shift *ch > T, where I'm using
T for dental "t", e.g., *mucho > mTo "pillow/headrest" (with syllabi
c
initial "m" < mu). This contrasts with mto < *muto 'river', where
post-alveolar *t remained unchanged.

In contrast to the lenition processes, i.e., changes in *manner* of
articulation, this shift is essentially one of *point* of articulatio
n.
Thus, it also affects the prenasalised palatals, e.g., *nchupa > Thup
a
(Southern Swahili ch'upa) "bottle, gourd". (It's irrelevant whether 
the
sequence was *nchupa > nTupa > nThupa > Thupa, or *nchupa > nch'upa
> nThupa > ..., or even ch'upa > Thupa)

A parallel shift occurs in *nj > nD, e.g., *njoo (still Southern) > n
Doo
'come!' (vs. ndoo'pail/bucket'). Thus, the change is sensitive to
point of articulation, not manner of articulation, and is distinct
from lenition processes.

This change just seems to be extreme apical fronting.
Intermediate stages are reflected in Miji Kenda /ts/ and /n(d)z/. Th
en
I suppose the affricate dentalised to Tth(eta), and finally lost the
slow release, as in nonstandard English Tthinker > Tinker "thinker"
(which is different in articulation from "tinker".) [cf. *t(h)S > t(h
)
mentioned above in the Gan dialects of Chinese shows similar
deaffrication.] Some affrication is still audible in Bajuni (because
the dentalisation is so extreme), but in Mombasa it is simply a denta
l
stop. Fronting to dental is attested elsewhere; something like that
happened in Castillian, giving theta from some less fronted affricate
s.
And in Sabaki, the language Elwana simply fronted *t > T, otherwise n
ot
found in Sabaki, and giving rise to substratal theories of its origin
(from Cushitic). (Dentalisation of fricatives, not stops, is common
in Kenya, but not in Sabaki except for Bajuni, where there is a
transparent conspiracy to embarass theories that posit "s" as the
"first" fricative in a language)

The end result is a contrast between dental and post-alveolar stops.
Interestingly, a generalisation of this shift to the palatal nasal,
*ny, is more restricted to the extreme North, e.g., Bajuni *nyumba
(still so pronounced elsewhere) > Numba 'house' etc. (later a new
*ny arises from *my, but we're not interested in the chain shifts set
in motion here.)

Next, there is a new source of ch, from *ky. That happens throughout
urban Swahili, North and South, but some rural dialects like Vumba
still have /ky/, e.g., *kyakula 'food' (still Vumba) > chakula.

Meanwhile, some principle like maximal differentiation seems to be
at work in Northern Swahili, to increase the distance between t and
T. Elsewhere I am only familiar with this principle in discussions
of vowel systems, which may have to do with the unusualness
of the t > ch change, or be a scholarly oversight. Frankly, in
Southern Swahili where *t remains opposed to *ch, it is also pretty
POST-alveolar, maybe no less so than in those Northern dialects
where nothing subsequent happened. Of course, I can invoke the
principle that it started in the North where I can give the motivatio
n
for it and then spread to the South -- but I'll control myself.

Anyway, in Mombasa (a Northern urban dialect), it is easy to tell the
dental T from the post-alveolar t when listening through headphones
to excellent recordings, because the dental T has the dull
release expected of a dental, while the t has a white noise release
that sounds s-like to me. It is more in the nature of an overtone
to the release than an affrication, not at all like the Miji Kenda
clear /ts/ reflex of *ch, so I don't entertain any thoughts about an
incipient *t > ts in Mombasa. (By the way, immigrants learning
Swahili, even in Mombasa, after the "critical age" -- 12? -- don't
pick up the difference. They think that mto "river" and mTo "pillow"
are homophonous. Similarly, Northern loanwords in Southern Swahili
convert the dentals into post-alveolars. mTo > mto 'pillow' is an
example so it is homophonous with 'river'. This wouldn't happen
in learning Bajuni, at any age, where nTo 'pillow' is clearly
distinct from ncho 'river' for any listener, though a Southern
Swahili speaker would post-alveolarise nTo to nto in imitating it.
Perversely, of course, *mcho should be the Southern word for "pillow"
,
while it is the Bajuni word for "river" -- with nasal assimilation.
The assimilated nasal remains syllabic so that aspiration is not
induced.)

Hinnebusch in his discussion of *t > ch in Bajuni claims to hear an
occasional /tr/-like sound in Mombasa for *t, but I have reservations
about that, although I have an idea about why he hears that, as I'll
explain (beside the t > hr change in other Sabaki languages). The
same differentiation occurs for nD (Southern nj) and nd. But here,
especially in Bajuni, the impression of the retracted postalveolar
nd is definitely /ndr/ with a slightly rolled release to the "d" (a
few rapid taps at best). So a parallel /tr/-like sound is definitely
plausible as an intermediate step toward *t > ch in Bajuni. The
rest of the change you can probably imagine for yourselves at this po
int,
but I'll try to put it into words. These sounds, which have an r-lik
e
auditory effect, are actually a laxity in the tongue behind the
RETRACTED post-alveolar point of occlusion. The apex does not resist
the egressive air flow. Instead, the apex flaps as air escapes over
it -- by an aerodynamic principle. We could imagine this as
a step toward lenition to "hr", if apical articulation became even
laxer. The Neogrammarian concept of sound change would then posit
 t > tr (as a single segment) > hr
but insist on a completion of t > tr BEFORE tr > hr. Variationist
observations of sound change in progress might lead one to expect,
instead, a variation between t and tr (which H hears for Mombasa),
and then a variation between tr and hr (not attested), or even a
simultaneous variation among all three stages (not attested).

In any case, *tr > hr did not happen in Bajuni. Maximal differenti-
ation (from T) is about point of articulation, not manner, and keeps
the focus on the *point* of occlusion. If *tr > hr had occurred, it
would have separated the fate of *t from *nt (probably already *nth
or even *th) and *nd, where the latter two are not candidates for
lenition according to its Bantu phonological logic.

Instead, in the evolution of Bajuni *t, the occlusion held against
areodynamicity, but at some point the retraction must have made a
noisy release which was quite affricated. In fact, being etymolo-
gically aware, I was not sure at first that the Bajuni reflex of *t
was EXACTLY the same as the Bajuni ch reflex of *ky. But it is
(psychologically and even spectrographically). The apical stop and
noisy release, somewhere in the really POST-alveolar region, fell
together with ch (< *ky). That is, ch < *ky was a catalyst, and serv
ed
as a landing site for completion of the change.

The retraction of original *nd, so that its release sounds r-like, is
the best support for the notion that maximal differentiation was at
work, since it is clearly retracted beyond Southern (or even Mombasa)
reflexes of *nd. Laxity accounts for the flap, but the "r" impressio
n
of the flap is a result of the retraction (can be several taps, not
just one like the English flapped "d").

In sum, I think (1) maximal differentiation from the dental T actuall
y
caused further retraction of the post-alveolar t, (2) a laxity in
the tongue behind the apical occlusion in the post-alveolar position
caused some kind of white noisy release as retraction continued
(3) the noisy release got sucked into palatal position, not a great
distance to travel, dragging the point of occlusion with it.

Meanwhile, as Hinnebusch observed, Bajunis do not hear their pronunci
ation
of nd as having an "r" release. They just don't identify the release
with the distinct phoneme /r/ which exists in Bajuni. Neither do
Mombasans, who think the Bajuni accent is hysterically funny, and
can do good imitations of it with the substitution of /ch/ for *t
and even the little flappy release for /nd/, which doesn't normally
occur in Mombasa. Of course, the most humorous thing about Bajuni
speech to Mombasans is that they lisp. That is, in many Bajuni
varieties there is no /s/ at all. A new wave of dentalisation has
converted all /s/'s into thetas (and /z/'s into edhs).

In sum, Bajuni post-alveolar *nd has not gone as far as post-alveolar
*t in retracting to palatal position. Instead, we hear the retractio
n
of Bajuni /nd/ as something with an r or tap-like release that doesn'
t
suggest a future metamorphosis into /nj/. (And note for the catalyst
theory that Bajuni does not have phonetic nj in its inventory, though
Bajunis are quite capable of pronouncing it in speaking other forms o
f
Swahili -- hole in the pattern? -- more so than in changing their ch'
s
into t's where appropriate.)

The catalyst theory, then, proposes that when an "unstable" sound is
produced by a sound change it MAY be attracted to an articulatorily
(very) similar sound that already exists in the language, merging wit
h
it. "unstable" is the controversial concept in this proposal. What
makes a sound unstable? Foulkes might suggest that its auditory
distinction from a very similar pre-existent (?) sound is to blame, b
ut
I think that the dynamicity of the bases of articulation already
established in the language is at least as likely a culprit.

(Bases of articulation!? -- In that case we obviously want to know wh
y
in Mayan tt > ht by *delay* in occlusion of the "segment", but in Ban
tu
nth > nh by *anticipation* of aspiration. These seem to be connected
with different syllabification trends. tt spans two syllables in
Mayan, but *nt(h) is a syllable initial in Bantu. With regard to "sim
ilar"
sounds, we also want to know why the various waves of palatalisation 
in
Slavic did not result in more mergers than they did, e.g., in Russian
.
In this context it is significant that in Slavic a process reminiscen
t of
maximal differentiation tends to oppose palatalised consonants to
"velarised" ones (the so-called soft vs. hard consonants), where, for
example, products of earlier palatalisation like /zh/ end up sounding
velarised in contrast to later palatalisations, and the peculiar --
for Indo-European -- effect of "velarisation" on the
vowel *i in East Slavic, for example. Originally, this sound comes
into Slavic from IE *u:, but probably as a front high rounded vowel o
r
diphthong /wi/, whence the pre-glide eventually velarised the nucleus
.
The various waves of palatalisation made palatals out of IE (labio)ve
lars
before original and evolving front vowels, leaving the surviving vela
rs
"hard" -- should I say "velarised"? Needless to say, the genesis of 
this
differentiation is historically more complex than Bajuni dental:post-
alve-
olar differentiation. With regard to bases of articulation, the hard
:soft
distinction in Russian provides some soft /zh/s, either onomatopoetic
 or
in loans from French -- hole in the pattern? -- but even so, the loan
zhurnal 'newspaper' has a "hard" initial, as I've heard it, and sound
s
very unFrench. I think only the French /zh/ followed by a front high
rounded vowel comes into Russian as a "soft" /zh/.)

The question remaining about the Bajuni *(n)t > (n)ch change is: woul
d
it have happened if there wasn't already a ch (< *ky) in the language
.
In other words, would retraction of *(n)t stopped/reached there, if
there wasn't already a rest station there? That is, of course,
unanswerable, unless we find other examples of unconditioned *t > ch
in the world, and can reconstruct its path and relevant aspects of th
e
consonantal system in which the change completed. (For that matter,
it is not an incontestable fact that *ky > ch occurred BEFORE *t > ch
in Bajuni. It is only most likely because North of Bajuni, Mwini
has the *ky > ch change, just like urban Swahili to the South of
Bajuni, but does not have *t > ch. In fact, Mwini has *ki > chi
generally, unlike any form of urban Swahili.)

In sum, I would suggest that unusual sound changes are either the
result of long sequences of usual sound changes (not interesting in
itself but the issue was probably raised in the hopes of justifying
reconstructions or correspondences at tremendous time depths) or
indications of the effect of historically produced idiosyncrasies in
particular phonological systems on paths of change for individual
sounds (after all, what two languages/dialects have exactly the same
phonological systems? -- though the differences don't matter for
"usual" sound changes). The latter is inherently interesting from a
linguistic point of view because it involves the notion of linguistic
"system", not simply the pre-Saussurian Neogrammarian notion of "blin
d
sound laws" (as in, blind to what else occurs in the language's
phonological system -- and on this basis correspondences and
reconstructions may either rise or fall when we know more about how
these things work. -- BW
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue