Rolf Noyer: Blur avoidance in Polish

[Martha McGinnis]

Hello everyone,

I wanted to respond briefly on the question of No Blur, brought up by
Andrew Carstairs-McCarthy.  I personally do not feel that No Blur effects
argue in any way for the reality of "paradigms", pace A C-M, (although
Paradigm Economy, which I am not going to talk about here is not quite so
easily explained).

Instead, as I believe I argued (perhaps rather elliptically!) in the NELS
paper that Andrew cited earlier, No Blur can be modelled in a
non-paradigmatic grammar via two constraints.  The first asserts that stems
can be assigned at most one diacritic feature of the sort that can
condition affixal allomorphy; the second asserts that rules introducing
affixes cannot contain disjunctions of such diacritics.  Consider a
"blurred" distribution as follows, where columns 1-4 represent inflectional
"classes" and the rows represent categories with a-f representing
"realizations" of this category:

1 2 3 4
a a b b   Category A
c d e f   Category B


Assume that /a/ is the Elsewhere affix for category A. How can the
distribution of /b/ be obtained?  Assuming that stems belonging to classes
1, 2, 3, and 4 are assigned the diacritics [+1] [+2] [+3] and [+4], the
distribution of /b/ could be generated only the rule:

[+3] OR [+4] -->  /b/

This violates the second stricture I mentioned above (and a good stricture
it is, since learning disjunctions of this type is computationally much
harder).  The alternative is to group classes 3 and 4 together as a
superclass, call it "beta" and assign stems of either class 3 or class 4
the diacritic [+beta].  Then the rule would appear:

[+beta] -->  /b/

This move, however, violates the first stricture, since stems of classes 3
or 4 require two diacritics, [+3 +beta] in the case of stems of class 3,
and [+4 +beta] in the cases of stems of class 4.

If this is correct, then the only way to achieve the above distribution is
for the grammar to contain two rules:

[+3] -->  /b/
[+4] -->  /b/

This means (i.e. this is how the grammar expresses) that we are dealing
with two accidentaly homophonous instances of /b/.

Notice that the Polish case is not a problem here: the diacritics which
control the stem allomorphy ("strange" vs. "non-strange" stems, etc.) are
not subject to the same strictures, although perhaps they have as yet
undiscovered regulations on their organization: an important research
topic.

In this way, I proposed to achieve the effects of the No Blur principle
without recourse to paradigms per se, but rather through constraints on the
assignment of diacritics to stems and on the grammar which generates the
sets of forms traditionally arrayed as "paradigms."

Having said that, I'd be curious if the above works or if anyone can show
me that it does not work.  Moreoever, I should add that insofar as the
above constraints are true, they do not impugn the emprical interest of the
No Blur Effect (let us call it) -- to the extent that this effect is real I
think it does represent an important insight into the ways in which
arbitrary inflectional diacritics are manipulated by the grammar.

I have not myself studied the Polish declension, but I do believe there is
fairly good evidence for a somewhat abstract analysis for the Russian
declension, partly along the lines proposed in T. Lightner's book from the
1960s and continued in Morris Halle's article "The Russian Declension" in
the Cole & Kisseberth (eds.) volume (see DM bibliography page for a fuller
citation of the latter). The following interpretation of the facts,
however, is my own. One finds, for example, a systematic absence of velar
palatalization at morpheme boundary before front vowels in nominal
declension, hence

ruk-a  'hand (nom. sg.)'
ruk-i  'hand (nom. pl.),  not *ruC-i

An explanation for this is that underlyingly there exists a theme vowel
/a/, seen most evidently, for example, in the prep., dat. and instr pl.
/ruk-a-x, ruk-a-m, ruk-a-mi/ respectively.  This theme vowel deletes before
another vowel by a general rule of Russian phonology:

ruk-a-a -->  ruk-a
ruk-a-i -->  ruk-i

The rule of vowel-deletion applies only in derived environments (hence.
/nauk/ 'science' without deletion), and hence is "cyclic"; this rule is
robustly established in the verbal conjugation.  The rule of (primary
place) palatalization for velars is also cyclic and ordered before
vowel-deletion in a counterfeeding order.  Hence, the /i/ never gets a
chance to palatalize the /k/.  The existence of this vowel also helps to
explain certain accentual properties: if the theme /a/ is underlyingly
accented, then it follows that all the (singular) endings of the
*a-declension are also accented. (The only exception is acc sg. /u/, for
which one must postulate a special rule; in any case the number of
unaccented *a-stems which are actually exceptions to this special rule now
outnumbers the number of ones that obey it...).  A nice result from this is
that there is no longer any reason to suppose that the desinence /e/ for
prep. sg. *o-stem nouns is distinct underlyingly from the desinence /e/ for
prep. sg. *a-stem nouns, although the two appear to have quite distinct
accentual properties (the latter seems to be underlyingly accented while
the former is not: /go'rod-e/ 'city (prep. sg.)' vs. /skovorod-e'/ 'pan
(prep. sg.)').  In the latter case, the accent comes not from the desinence
itself but from the theme vowel which has been deleted:

skovorod-a'-e -->  skovorode'

Similar non-palatalization effects are seen in the imperatives, e.g.
/po-mog-i/ 'help!' ( underlyingly /po-mog-o-i/) and in the *o-stem
declension, /drug-e/  'friend (prep. sg.). (underlyingly /drug-o-e/).

Whether such an abstract analysis for Polish will clarify the declension
would be something I would look into if I were looking at this problem.

Rolf Noyer
University of Pennsylvania