Alec Marantz: Blur avoidance in Polish

[Martha McGinnis]

[I'm a bit slower than Rolf Noyer, so I was working on the below reply when
Rolf's posting came in over the DM hot-line.  My posting covers much of the
same theoretical ground -- Rolf's has the advantage of including real
details and illustrations.  If you read only one of the postings, read
Rolf's.]

Andrew writes to the DM list in a kind of shorthand that I thought I might
expand for some of our readers.  When you read his last posting, you might
wonder what relevance the discussion of paradigms has to DM -- you might
even think that Andrew meant that evidence for paradigms was evidence
against the general approach to morphology outlined in DM.  After all, it
is written in the book of DM, "If you do morphology by itself, you will do
morphology by yourself," and direct reference to paradigms in principles is
as morphological by itselfish as it gets.

But of course Andrew is assuming that everyone knows the intertranslation
between claims about paradigms and claims about the structure of the
Vocabulary in DM.  In fact, the "No Blur" principle translates into a
stronger claim about Vocabulary Items, namely "No Curly Brackets."

Recall that "No Blur" is about rows in paradigms; actually, about single
cells defined in terms of any semantic or syntactic features.  To
translate, then, "No Blur" is about sets of VIs that share the same
substantive features and the same contextual features referring to such
features as gender on the stem.  What "No Blur" would claim about such VIs
is that they can't use curly brackets in contextual features referring to
arbitrary inflectional classes of the stems to which they attach.  That is,
VIs can't have disjunctive contexts defined in terms of arbitrary class
features.  You either have VIs that mention a single inflectional class,
thus serving to identify that class, or you have a VI that doesn't mention
a class, serving as a default for the particular set of substantive and
contextual features.

Readers of this List know that Carstairs-McCarthy's beef with DM doesn't
surround this issue of whether to describe things in terms of features on
Vocabulary Items or in terms of paradigms, since as far as I can tell, on
the empirical issues discussed in the literature, there is nothing to choose
here.  Rather, Carstairs-McCarthy brings up the interesting question of the
role of phonological zeros as VIs.  Of course here it isn't a question of
whether or not DM can translate C-M's proposals about zeros into
proposals about the Vocabulary -- DM could.  But most DMers wouldn't make
this move -- we want our zeros.  Rather, we would accept the challenge to
re-analyze apparent violations of the DM version of "No Blur" caused by the
recognition that zeros are Vocabulary Items.  No curly brackets for
phonologically zero VIs either.

The thing is, most DMers assume a much stronger version of "No Curly
Brackets" than "No Blur" as a principle of the acquisition of Vocabulary
Items.  Rather than the acquisition principle that C-M assumes is behind
"No Blur" in his _Language_ paper, "Avoid Homophony" (a.k.a. the principle
of contrast), the principle is more like that assumed by Jakobson -- "Avoid
Synonymy."  C-M concentrates on the horizontal dimension of paradigms, by
"Avoid Synonymy" operates in all dimensions and across paradigms as well.
The general principle is "Avoid Disjunction in VIs" or "Avoid Curly
Brackets."  This means that there's something wrong with an affix that is
used for both DAT and ACC unless there is a feature shared by both these
cases and the affix can be
underspecified for the features that distinguish DAT and ACC.  This also
means that if the same VIs (identified by their phonology) appear in both
verbal agreement and in pronouns, and are associated with the same person
and number features in both the verbal and pronominal domains, they must be
treated as the same VIs, underspecified for whatever distinguishes the
verbal and pronominal environments (thus, the same VIs across paradigms) --
see recent work by Karlos Arregui-Urbina on Basque.

Although "No Curly Brackets" can be understood as a violable constraint in
acquisition (no curly brackets unless you're damn sure that you need them),
one can take the principle as a formal constraint as well.  Where one might
be tempted to use curly brackets, one must instead bite the bullet and
admit two unrelated yet homophonous VIs.   This is a tricky and interesting
area for investigation.

As Rolf Noyer has pointed out and as Isabel Oltra-Massuet has worked out
for Catalan verbal morphology, inflectional classes could be hierarchically
organized, and in fact probably are.  In "No Blur" terms, this means that
apparent "blur" can be the result of two inflectional classes sharing the
same affix for some set of features  because they also share the property
that their inflectional class features belong to a broader
"meta-inflectional" class.  Oltra-Massuet shows that this isn't just a way
to avoid No Blur/No Curly Brackets - it captures significant
generalizations about the Catalan inflectional system (see MITWPL 33, which
includes a wealth of morphological material).

As for the Polish paradigm that C-M describes, one wonders whether his
proposal is what he really wants to say about facts.  Treating "with a
strange stem alternate" on par with gender features makes the prediction
that such a feature as "has a strange stem alternate" could enter into
agreement relations, as does gender (but not inflectional class).

This consideration of the implications of one's proposals outside the
domain of paradigms is of course not associated with byitselfish morphology
but is, I'm sure, along lines that C-M would approve.

--Alec Marantz
marantz at mit.edu