probability & sound change (nee rhotacism from R.H.)

[Robert R. Ratcliffe]

----------------------------Original message----------------------------
H.M.Hubey wrote:
 
> Robert R. Ratcliffe wrote:
> >
> >The proposal that given enough time X>Y>X is
> > plausible is based on a hypothesis that the directionality of sound
> > change is random. It isn't.  There are clearly preferred (frequently
>
> This is something that used to bother me a lot and I used to think
> that
> there must be some universality. I am still not sure that there is not
>
> but I am more sure now that some of what we see is not really
> universal but rather due to some specific local phenomena.
 
This is not a matter of opinion, or something that can be decided simply
by thinking about it.  It is a matter of evidence, and that is what I
meant by 'empirical'. Labov's conclusion, based on case studies of some
twenty languages, is: ".. there are no directions of vowel shifting that
are forbidden to speakers of human languages, but ... some directions
are taken far more than others" (Principles of Linguistic Change 1994,
p.116).  Specifically he proposes that in chain shifts, long vowels
rise, short vowels fall, and back vowels move to the front.
 
> > attested) and disprefered (rarely or not attested) directions for
> sound
> > change. The only systematic work on this I know of is in Ch. 5 of
> > Labov's 1994 "Principles of Language", which only deals with vowels.
 
> This has to be from a specific time period and a specific set of
> languages.
 
Labov's sample (mistake in title, should be Prinicples of Linguistic
Change, sorry) is small by the standards of typology, and most of the
languages are European.  Further work in the diachronic typology of
phonological systems is certainly desirable. Nonetheless, as empirical
scientists we have to prefer a theory consistent with some evidence
(directional sound change) over a theory (random sound change) which has
not been shown to apply to any evidence.  It is certainly possible that
directionality effects are part of a short term cycle which evens out to
zero in the long run, but I haven't seen any evidence which would
support this view.
 
> There is a similar problem in probability theory, that of stationarity
> of a signal. It's impossible to prove. No matter how many languages
> you
> look into you can only look at a specific time interval. It's hard to
> say if the sound changes are due to universals or to a particular
> combination of sounds and phonotactics.
 
This passage is a bit cryptic. I haven't talked about universals, but
only about emprically observed statistical tendencies, and haven't
discussed explanatory factors at all. It would certainly be meaningless
to say that something is "due to" a statistical tendency.  The length of
the time interval is only limited by the historical record, but even so,
it isn't necessarily a problem.  If directional effects are only part of
a long term cycle X>Y>X (a cycle longer than the historical record),
this should still be demonstrable, based on the fact that not all
languages should be at the same point in the cycle.  In some langues X>Y
should be the trend in the attested period, in others Y>X.  Breadth of
the sample can substitute for depth.
 
> However, what I wrote above referred, in general, to any two sounds.
> It
> may be that for some specific X and some specific Y, the sound change
> X > Y for some specific language (i.e. specific set of phonemes and
> phonotactics) may be irreversible. But in general I do not see any
> reason to assume that no sound change is reversible given enough time.
 
You're right, of course. Directional trends relate only to what is
probable in sound change, not to what is possible or impossible. On the
other hand it occurs to me that there is one type of sound change which
IS irreversible, that is X>0, as found in mergers, phoneme loss, or
assimilation in clusters.
 
> Right. However, it is strange on the other hand to see those consonant
>
> clusters and lack of vowels in languages like Abaza, Georgian, or
> Khoisan and its clicks. The foremost question is this: if there is
> such a
> universal trend (say toward lack of cases, or toward voicing, or from
> stops to fricatives, or approximants) how then did the language (any
> language) get those stops in the first place? Or how did some language
> get consonant clusters at all?
 
Directionality does not mean teleology. To say that sound change moves
in a non-random direction is not to say that it is going anywhere in
particular.  One might propose (although I didn't) that over time
phoneme inventories get smaller, or that 'marked' phonemes tend to be
lost in favor of unmarked.  I have yet to see a teleological proposal of
this type which can survive the test of the evidence. The directionality
trends discovered by Labov appear ultimately due to physiological
constraints both articulatory and auditory. In other words the direction
of sound change is constrained by the present not by the future.  I
suppose this is what you mean by talking about local versus universal
(i.e. short term vs. long term).   There are no long term directional
trends (as far as the evidence now goes).  But we cannot conclude from
this that the sum of short term directional trends adds up to zero or no
directionality or random directionality.
    The prequisite to a realistic probalistic model of language change
is a systematic research program in diachronic typology.  We have to
establish the probable direction of change on the basis of statistical
data from observed changes-- not on the basis of a priori reasoning. I
suspect the ultimate model will look something like those typhoon maps
we see here (don't know if you have them in other countries): There is a
circle showing where the storm is and from the outer edges of the circle
there is a wedge shaped projection showing the area into which the storm
is likely to move. At the big end of the wedge there is another bigger
circle showing where the storm is likely to be at the end of a certain
period (24 hrs, e.g.).  From this big circle another wedge is projected,
with a bigger mouth showing where the storm should go in the next
period, and so on.
 
 
> The same applies to linguistics changes.There are many scales at which
> changes occur, and if we mix up these levels we create opposing ideas.
 
A priori we don't know what models or scales apply to linguistic
changes. We can't know until we have tried to apply them, that is test
them against the data.
 
 
+++++++++++++++++++++++++++++++++++++++++++++++++
Robert R. Ratcliffe
Senior Lecturer, Arabic and Linguistics,
Dept. of Linguistics and Information Science
Tokyo University of Foreign Studies
Nishigahara 4-51-21, Kita-ku
Tokyo 114 Japan