> I would like to propose some candidates. The justifications can be
> found in the file which I just uploaded. It is an article published
> the Journal of the Inter. Quant. Ling. Association circa 1999.
> Here are the axiom(s) (axiom candidates)
> Axiom 0. Most changes are co-articulation effects of various kinds.
> Axiom 1. The changes go from less sonorant to more sonorant. These
> be seen in Figs 10-18 or so. (There are two Fig 12s. Mistake.)
[f] > [v] and [v] > [f] are both known. Word initially, [f] > [v] is
known from Dutch and English West Country dialects.
[v] > [f] seems to have occured in the development of Tai dialects
from Proto-Tai; the evidence lies in the tone development, which
indicates initial voicing. (The Thai letter is a modified Indic <b>;
the original Indic letter is now pronounced [ph] in Simaese and Lao,
[p] in the Northern dialects, and and has the same association with
the tone as the letter for *v.) It may be relevant that the Tai
change is part of a general set of changes whereby voicing contrasts
were lost. For example, <hm> and <m> now both represent [m] in most
Intervocally, [f] > [v] is unremarkable; word-finally, [v] > [f] is
> Axiom 2. The specific mechanism is (to the simplest approximation)
> nearest neighbor shift. (A similar phenomenon (!) (actually analysis
> of the phenomenon) in physics is called nearest-neighbor
> In other words, no shifts of the type t>a allowed. (Personally I
> also disallow k>s.)
We may have to be careful with our metrics here. The Latin to
Romanian shift -[kt]- > -[pt]- (also -[ks]- > -[ps]- and -[Nn]-
(<gn>) > -[mn]-) and Celtic -[pt]- > -[kt]- do not look like nearest
neighbour shifts. A plosive contrast of [p] ~ [k] ~ [t] can be
represented by 3 binary features, though obviously 2 will do. I've
seen a paper where the Romanian development was put forward as an
argument that [p] and [k] should be separated by a single feature, so
the contrast is [t] ~ ([p] ~ [k]).
The frequent change [k_w] > [p] is not a nearest neighbour shift in
articulatory terms. (I'd accept [k_p] > [p] as such, but I don't
think anyone has seriously suggests that Latin <qu> represented
[k_p].) I'm not sure what to make of the Greek development [k_w] >
[t] before a front vowel either. Potentially nasty is proto-Semitic
*w- > Hebrew [y]-. (_w&-_ 'and' is about the only native looking
word Hebrew has starting with /w/.) I'd be very dubious about [H]
(the semivowel in French _huit_ 'eight') as an intermediate stage.
> Axiom 3. The system should be hierarchical e.g. get the commonest
> changes to form a kind of a skeleton and add the more exotic ones on
> top of these. This is not much more than an approximation scheme
> is already in use in linguistics (as well as in many branches of
Is this an axiom or a principle for organising our knowledge?
> Axiom 4. When these are being used in reconstruction, make explicit
> your principle e.g. are you using some kind of an averaging?
Averaging is in general unsafe, especially in the presence of
There are also loss and gain changes. Initial [j] > [dj] and [w] >
[gw] are (in progress in Spanish) the most obvious examples of gains,
but we also have vowel prothesis, e.g. Latin st- > Western Romance
est- and consonant prothesis - almost mandatory in Slavic.