NW vs. E Gmc

[Sean Crist]

On Thu, 27 Jan 2000, Herb Stahlke wrote:

(I had previously written that *z > r can be shown to be an independent
change in NGmc and WGmc.)

> As one only generally familiar with linguistic issues of early
> Germanic, I'm puzzled by Sean's clause "after the two dialects had
> separated".  I've rather gotten the impression from sources like Orrin
> Robinson's Old English and its Closest Relatives and Hans Frede
> Nielsen's The Germanic Languages: Origins and Early Dialectal
> Interrelations that Sean's presupposition is something of an
> oversimplification.  Rather, I've gathered, early Germanic, even after
> the Gothic migration, had very much the characteristics of a dialect
> continuum. Certainly, one can distinguish EGmc as a separate group
> earlier than N or W, but N and W don't become clearly distinct
> branches, rather than extremes on a continuum, until quite late,
> perhaps after the continental period.  Or is Sean simply using a
> convenient shorthand that I'm being overly critical of?

Okay, here's the details.  There are three rules, one in NGmc and two in
WGmc, which are sensitive to the *z/*r contrast, which therefore must not
have collapsed until after some of the specifically NGmc and WGmc sound
changes had occurred.

In NGmc, but not in WGmc, the following rule occurred:

	*ai > a: / _ h, r (but not *z)

Hence Old Icelandic *sair > sa'r "wound", but *maizan > meiri
"more" (Noreen, 1904, p. 75).

In WGmc, there is a sporadic rule which deletes *z before *n, *d with
compensatory lengthening.  The details are quite messy, because you get
deletion in some words in some WGmc languages but not in others, and
sometimes not in any:

Goth	OHG	OE		OFris	OS
mizdo	me:ta	me:d/meord	me:de	me:da	"reward"
--	lerne:n	leornian	lerne:n	li:no:n "learn" (*z > r)
huzd	hort	hord		--	hord	"treasure"
razda	rarta	reord		--	--	"language, speech"
gazds	gart	gierd		--	gard	"sting, switch, goad"

(There are alternate spellings in some cases which, for clarity, I haven't
listed, but which have the same outcomes of *z as those listed above.)

This is a mess, but the point is that while the deletion applies variably
to *z, the deletion _never_ applies to original *r; so WGmc was sensitive
to the the *z/*r contrast.

Another such case occurred in Ingvaeonic, where word-final *z in stressed
monosyllables deletes with compensatory lengthening (word-final *-z in
unstressed syllables had already gone):

OE	OS		OHG	Goth.	PGmc
me:	mi:		mir	mis	*miz	"me (dat.)"
we:	wi:, we		wi(:)r	weis	*wi:z	"we"
ma:	me:r		me:r	mais	*maiz	"more"

(The -r was restored in OS me:r, and later in OE, by analogy with the
comparative adjectives.  Notice that rhotacism later applied in OHG.)

Notice that this deletion never happens to original *r, as in *he:r > he:r
"here".

So we have one rule in North Germanic which is sensitive to the *r/*z
distinction, but which is clearly NGmc only, hence postdating the
WGmc/NGmc split.  Thus, the merger cannot yet have happened when NGmc and
WGmc separated.

We likewise have two rules within WGmc which are sensitive to the
contrast; same conclusion.  (As I noted in a previous post, Ingv. is
securely WGmc; it undergoes none of the NGmc sound changes, but it
undergoes some seven different sound changes shared by all the WGmc
languages before starting to innovate on its own, as with the Ingv. rule I
listed above.  Thus, the *z/*r contrast must have survived in WGmc until
WGmc itself was already beginning to break into dialects.)

There's one further bit of evidence: runic inscriptions clearly postdating
the breakup of Germanic by centuries maintains a clear distinction between
*r and *z.  The inscribers don't get them mixed up, even in a period where
NGmc and WGmc must have already separated, based on external historical
evidence.

If we didn't have such clear evidence of a *z/*r contrast within NGmc and
within WGmc, we'd say that *z > r is an innovation which occurred in
Proto-NWGmc before the split between NGmc and WGmc.  But we clearly can't
say this; the evidence is just too strong that this is not so.  So the
question is, what do we make of this?

One perfectly reasonable possibility is that rhotacism is a purely
independent parallel development in the two branches; rhotacism rules are
not uncommon (one occurred in Greek, for example).  Another possibility is
that *z had already come to be somewhat phonetically rhotacized in
Proto-NWGmc (without actually collapsing the contrast yet), so that the
merger was phonetically likely, but didn't actually occur until later.  A
third possibility is that rhotacism was an innovation which spread across
dialect boundaries after NGmc and WGmc had become clearly differentiated
from one another but were still in relatively close contact.  It can't be
an _inherited_ common innovation, however.

I should probably mention what I'm working on in my in-progress
dissertation, which is about conspiracies of historical sound changes;
it's relevant to this discussion. The elimination of *z in West Germanic
is a complicated picture; there are actually several other rules which
eliminate it in various ways, with rhotacism taking care of the residue.
Traditionally, these processes eliminating *z have been treated as
separate, unconnected rules; there is no way under traditional rule-based
to conflate them, even tho they all mysteriously have the effect of
eliminating *z within a fairly short period of time.

What I'm saying is that conspiracies of this sort are no accident.  I'm
working in a recent formal model of phonology known as Optimality Theory,
which makes use of ranked constraints rather than conventional rewrite
rules.  Previous authors working in OT have suggested that historical
phonological change is to be modeled as a change in the relative ranking
of the constraints.  I accept this, and the point of my dissertation is to
discuss a previous unexplored consequence of this idea: namely, that a
change in the ranking of constraints can sometimes result in clusterings
of seemingly connected surface phenomena for which traditional rule-based
accounts can give no unified formal account.

One of my chapters looks at the WGmc *z elimination conspiracy in close
detail; I'm saying that all of the historical rules eliminating *z are
actually the outcroppings of a single change in the grammar; namely, the
rise in ranking of a constraint prohibiting *z, with the details of the
individual rules (deletion with or without compensatory lengthening,
rhotacism, etc.) falling out automatically from other, independently
motivated aspects of the grammar.

  \/ __ __    _\_     --Sean Crist  (kurisuto at unagi.cis.upenn.edu)
 ---  |  |    \ /     http://www.ling.upenn.edu/~kurisuto/
  _| ,| ,|   -----
  _| ,| ,|    [_]
   |  |  |    [_]