Rolf Noyer: Noun Compounding Question (reply to Carson Schutze)

[Martha McGinnis]

Hello everyone,

I think Carson is right that Lieber's 1980 analysis attributes most of the
post-root material in German nouns to the stem allomorph and little to an
actual desinence's spell-out.  For this reason, perhaps, her analysis met
with some criticism.  But the actual analysis is a bit complex so it may be
useful to review it.

First, I do not think it is correct that the compound stem form is always
the same as the plural unless it is linking -s-. The example that I am
familiar with is Stern-en-schein  'starshine' but Stern-e 'stars'.

Second, according to Lieber, a stem such as Geist/Geist-er is listed in the
lexicon in a static morpholexical relation; it is part of a system of
morpholexical stem relations of the form X ~ Xr; any particular stem
participates in one (or perhaps more than one) static morpholexical
relation, but the number of such relations is not limitless and so although
the exact relation(s) that a stem participates in is (are) to some extent
arbitrary, on the other hand it is not quite as arbitrary as simply listing
a suppletive stem, for example.  If it turns out (and I think this is
mostly true) that the choice of compound stem form (aside from "linking"
-s-) is  the same as the stem used in the plural, this is not necessarily a
problem, it simply means that the "second" stem that a root has is
polyfunctional.

My reading (of many years ago, now -- my copy has disappeared in the hands
of someone who, if s/he is reading this, please return it!) of Lieber 1980
has the following table of affixes for German nouns:

        sg      pl
nom     -0      -e/-0
acc     -0      -e/-0
gen     -s (m.) -e/-0
dat     -0      -n

The alternation -e/-0 in the plural is conditioned phonologically: -0
occurs following an unstressed -e plus a sonorant.  Note that this sonorant
can be part of a stem extension itself.

Stems can be in the following morpholexical relations:

I:  Non-umlauted/umlauted
II: X ~ Xs
III: X ~ Xr
IV: X ~ Xn
V:  X ~ Xe

It seems that the plural suffix -e/-0, as a property of that suffix,
attaches to the second stem (i.e. the umlauted or r-suffixed or n-suffixed
stem) *if it is available* for a given root.

I'm not entirely sure what happens in Hund/Hund-e 'dogs', whether we have
Hunde+e --> Hund+e by phonology or whether we have Hund+e to begin with.
We clearly need Hund+e as a stem allomorph to get Hund+e+müde 'dog-tired'.
If the former, then one must say that the plural suffix always chooses the
second stem in the morpholexical relation (how to actually "name" or "point
to" the stems is a difficult issue as well, but I'll leave that aside for
now).  Also I don't recall how multiple relations are established, i.e. how
one deals with Korn, Körn-er  'seed', whether by "stacking" the
morpholexical relations or by some other means.

So, if we are dealing with  Geist-er 'spirit (pl.)' we have in fact
Geist+r+0/e  where the 0 plural allomorph is chosen after a sonorant
(epenthesis is assumed before /r/).

A prediction that Lieber does make -- or at least my extension of these
ideas to the problem of the Verbindungsvokal, although I don't recall if
she explicitly states it -- is that the dative plural /n/ could never be
inside a compound.  So we predict that we will never find

Root+er+n+Root ...
Umlauted Root + n + Root ...

In other words, one doesn't see something like:

Geist-er-n-seher  cf.  Geist-er-seher  'visionary seer'

In Stern-en-schein, however, the /n/ is not the dative plural /n/ but is
the stem-extension /n/.

I have nothing to add on the point of learnability except to say that this
issue is not confined to German but arises wherever stem allomorphy
involves affixal material.  For example, in Modern Irish there exist stem
extensions /t/ /r/ and /@x/, /@n/ among others which may attach to a root
in forming its plural stem.  Thus we have:

/kar at g'/            'rock'
/kar at g'-r'- at x-i:/   'rocks'

/stal'/              'horse, stallion'
/stal'-t'-r'- at x-i:/  'id. pl.'

where the plural suffix is /i:/.  Historically these stem extensions were
originally part of the root of a set of consonantal-ending stems; later the
extensions were lost except in certain cases and were reinterpreted as,
e.g., genitive singular desinences, and finally, in the modern language
they are mostly idiosyncratic stem-extensions whose distribution has spread
into new parts of the vocabulary (i.e. stems which didn't have these
extensions historically).  So the /r/ extension in 'rock' occurs in Cois
=46hairrge Irish but not in "standard" Irish, where the plural is
/kar at g'- at x-i:/.  Likewise, the standard Irish plural for 'horse' is
/stal'- at n-i:/ with a different stem-extension entirely.

So on the whole my suspicion is that yes, indeed, learnability is
problematic in these instances and we see historical reanalysis happening
all over the place owing to re-segmentation.   I do agree that it is a
vexed problem to argue for a clear-cut distinction between suffixal
material that is introduced by readjustment rules (or, equivalently, is
part of a stem "allomorph") and suffixal material that is in fact the
"spell-out" of a particular morpheme.  However, this distinction is not
unlike that between phonological material which "discharges" morphological
features and that which is merely conditioned by morphological features, a
distinction I tried to argue for in my dissertation.  To the extent that
this distinction exists, so might we also expect there to be a distinction
between stem-extensions and affixes.

Another point to be made is that many of Morris Halle's analyses of other
Indo-European languages' nominal morphology, such as Latvian, Russian and
Latin, assume the existence of a nominal "theme" vowel whose function is
not unlike that of Harris' word-marker for Spanish.  Carson seems to be
assuming that the child comes to the data with the expectation that there
is a simply a root plus a number suffix.  However, this may not be correct.
Rather, it may be that the default expectation is a category-neutral root
plus a category-making suffix (=3D theme) plus, potentially, further affixal
material on top of this.  If this is even remotely correct, then the stem
allomorphs in German need not be readjustment-rule-added; rather one might
(?) construe at least /r/ /n/ and /e/ as theme vowels which provide
morphological category to a root.  One would have to assume a 0-theme as
well, of course, for bare root nouns.

This analysis makes perfect sense for a language like classical Greek,
where compounds typically involve Root + Theme + Root ....  where the Theme
acts as "glue" between the two roots.  Interestingly, the Theme in
compounds is not always the same as the theme in nominal inflection.  For
example, we have /ikhthuu-s/ 'fish' -- where it's not at all clear if there
is an overt theme at all -- but /ikthu-o-pool-ee-s/ 'fish-monger'.  There
are plenty of such cases to be found in Smyth's grammar, pp. 247ff.

Another useful analogy may be to French adverbs which are traditionally
thought to be formed from the "feminine" form, e.g. _certain-e-ment_
'certainly.'  It is, I think, evident that there is nothing "feminine"
about adverbs and that no plausible "rule of agreement" could supply the
property "feminine" in such cases.  Logically one is led to the conclusion
that the "feminine" form is a stem allomorph and that the -e- here does not
"spell" "feminine" per se.

Rolf