LINGUIST List 12.2062

Fri Aug 17 2001

Review: Urbanczyk, Reduplication in Lushootseed

Editor for this issue: Terence Langendoen <terrylinguistlist.org>

Directory

1. Mike Maxwell, Review of Urbanczyk "Patterns of Reduplication in Lushootseed"

Urbanczyk, Suzanne (2001) Patterns of Reduplication in Lushootseed.
Garland Publishing, hardback ISBN 0-8153-4042-7, xi+254pp, $65.00
(Outstanding Dissertations in Linguistics).

Publisher's announcement at http://linguistlist.org/issues/12/12-1421.html

Mike Maxwell, Linguistic Data Consortium

This book is an account of the phonology and morphology of
three reduplicative morphemes in Lushootseed, a nearly
extinct Central Coast Salish language of Washington state
(United States). Urbanczyk's University of Massachusetts
dissertation (with some work done at the University of
British Columbia) was finished in 1996, and this book is a
substantially unrevised version of her thesis (apart from
updating references, and adding an appendix containing the
corpus of reduplicated forms). The Lushootseed data was
extracted from a dictionary, with some acoustic information
verified from archived tapes.

U's overall approach is based on Optimality Theory (OT),
specifically on a (sub-)theory of OT called 'Generalized
Template Theory' (GT), proposed by McCarthy and Prince.
Briefly, GT holds that reduplicative morphemes are
unspecified for either 'segmentism' (melodic content) or
phonological shape (syllable, foot etc.); rather, a
particular reduplicative morpheme's phonological shape is
determined by its 'morphological classification' as a root,
prefix or suffix. Reduplicative morphemes acquire a shape
based on the default shape of a morpheme of that class (and
their melodic content by copying, apart from any fixed
content they may have).

U's goal is to support GT by showing that the phonological
properties of the three Lushootseed reduplicative morphemes
are derivable from this minimal information (together with
a specification of the morpheme's alignment with respect to
its base), and from the general constraint ranking of the
language. In particular, it is not necessary to specify in
the lexicon each reduplicant morpheme's shape; nor, in the
case of infixes, is it necessary to specify in the lexical
entry where that infix appears relative to the rest of the
stem (at least in this case).

U uses the Lushootseed data to argue several further
theoretical points of OT, and GT in particular. OT uses a
family of 'Faith' constraints, which require the output
form of a morpheme to resemble some other form. For
ordinary morphemes, the Faith constraints match the output
form against the input (lexical) form of the morpheme, and
hence are collectively called Input-Output Faith (IO Faith)
constraints. But for reduplicative morphemes, the match is
between the 'base' (the form from which the affix is
copied) and the output; hence the constraints are referred
to as Base-Reduplicant Faith (BR Faith). These Faith
constraints form OT's 'Correspondence Theory', and U shows
how the Lushootseed data support this sub-theory.

A second point of GT which U argues for, is that the Faith
constraints must be split between faithfulness of roots,
and faithfulness of affixes; root faith is universally
ranked higher (more important) than affix faith. And
thirdly, U looks at a phenomenon which has become known in
the OT literature as 'emergence of the unmarked', in which
reduplicative morphemes tend to conform to the phonological
constraints of the language more than do ordinary morphemes
(which may contain phonemes which are unusual in the
language, or which violate other well-formedness
constraints of the language).

Chapter one, 'Reduplication and Optimality Theory', sets
the stage by briefly reviewing the history of analyses of
reduplication within generative linguistics, culminating in
GT. The chapter closes with a very brief overview of
Lushootseed phonology and morphology.

The second chapter, entitled simply 'Shape', shows how GT
determines the phonological shape of the three
reduplicative morphemes: a diminutive prefix; an
infix/suffix glossed as 'out of control'; and a bound root
glossed 'distributive'. Taking the diminutive first, given
that it is lexically listed as an affix, a (presumably)
universal constraint implies that it should be no more than
one syllable in length. Many morphemes which have been
called affixes in earlier studies of Lushootseed are longer
than one syllable, but U argues that most of these
exceptions are actually roots, and that the majority of
true affixes in Lushootseed are monosyllabic. (The
constraint is in any case violable.) U attributes the fact
that the diminutive prefix is CV (not CVC) to the inherent
markedness of codas, a fact embodied in a constraint.

The 'out of control' affix, on the other hand, is VC. U
attributes this to the fact that it is suffixal, rather
than prefixal. This interacts with the constraint ranking
of Lushootseed to determine the shape: it must be vowel-
initial, because if it were consonant-initial, it would add
a coda to the word (roots being predominantly consonant-
final); and it must be consonant-final, because its right
edge is "anchored" to (roughly, copied from) the right edge
of the base, which is itself consonant-final.

An interesting result here is that there are two
reduplicative affixes, both of which should--in the absence
of other constraints--have identical shapes, namely the
default shape of an affix. The difference in their shapes
(CV vs. VC) is due to the interaction of their different
alignments (prefixal vs. suffixal) with other constraints.

Accounting for the position of 'out of control' affix
is more difficult. For CVC stems, the affix acts as a
suffix. With CVCC stems, it behaves like an infix,
immediately preceding the final consonant, which U argues
results from the constraint minimizing coda consonants.
With CVCVC stems, it again behaves like an infix, usually
(but not always) attaching after the first CVC. U's
analysis in this case is that the reduplicative affix
"attempts" to match a maximal stretch of the base. At the
same time, it must be no larger than one syllable (as
discussed above), and it must be "anchored" at the right
end of the base. Since reduplication in the context of a
CVCVC base cannot simultaneously satisfy all three demands,
a lower ranking constraint must 'give'. U proposes that
what gives is the right-most anchor constraint: by infixing
after the first CVC of the base, the affix in effect breaks
the base into two separate bases, namely the initial CVC
and the final VC. By appearing after the first base, it is
only required to match that base, which means that the VC
affix is just one consonant shorter than the (left-hand)
CVC base, thereby matching against a maximal stretch of the
base. This analysis strikes me as ingenious--perhaps too
ingenious. Why should a discontinuous base count as two
bases, only one of which need match against the
reduplicant? It also implies that either children are very
clever, or there are some very peculiar provisions in
universal grammar for reduplication (more on this later).

Another oddity of U's analysis is that in effect, there is
a conspiracy of constraints which cause the affix to attach
(usually) after the first CVC: a different constraint
interaction forces this result in each case of CVC, CVCC,
and CVCVC roots. One of the arguments in favor of OT and
against traditional rule-based analyses was that different
rules conspired to cause a general effect (such as a
desired syllable structure), missing the generalization
that the language was 'aiming for' the general effect. It
seems that conspiracies have re-surfaced in OT, this time
conspiracies of conditions.

An alternative analysis might be that the 'out of control'
affix attaches after the stressed syllable, which is the
first syllable in nearly every case in the data corpus. U
rejects this analysis on two grounds: first, the initial
CVC is not necessarily a syllabic constituent, since its
second consonant may be the onset of the following
syllable. One should not be too hasty in accepting this
argument, it seems to me, until the issue of
ambisyllabicity is decided. Second, when a CVCVC stem is
stressed on the second syllable, U points out that the
infix attaches in some words after the first CVC, and in
other words after the second CVC. But the entire corpus of
CVCVC stems with second syllable stress and taking this
affix is only four words, evenly split between affixing
after the first and second CVC--two words each! This seems
like a very tiny amount of data to base such a strong claim
on. Moreover, so far as I can tell, U's analysis fairs no
better on these four words than the alternative: both
analyses get two words right, and two wrong. In summary, I
believe the alternative analysis is at least as plausible
as U's analysis. (It also avoids the conspiracy effect
mentioned above.)

The third reduplicative morpheme, glossed 'distributive',
is often used as a plural. It attaches to the left end of
the base, copying the first CVC. U argues persuasively that
this morpheme is a (bound) root, that its CVC shape is the
default shape of roots in Lushootseed, and that its
prefixal position is consistent with that of a set of
'ordinary' bound roots.

The third chapter is entitled 'Syllables, Stress, and
Syncope.' Many Lushootseed words have clusters of two
voiceless obstruents root-initially. U argues that despite
appearances, these obstruents are in separate syllables:
the first belongs to a syllable whose nucleus is a
voiceless vowel, sometimes transcribed as aspiration.

Stress, she argues, tends to fall on the most sonorous
vowel (/a/, if present); unstressed /a/s tend to delete
where possible, else to reduce to schwa. There are two
sorts of exceptions to the latter generalization: some
unstressed /i/s and /u/s delete, and some unstressed /a/s
neither reduce nor delete. U's proposal for the latter case
is that the exceptional stems which retain unstressed /a/s
have a different constraint ranking, which seems to me a
questionable move. (However, I must admit that exceptions
are difficult in any theory of phonology. Derivational
theories have a variety of ways of handling exceptions,
including strata, exception features, and various theories
of prespecification vs. underspecification, which may be
seen as an embarrassment of riches.) U mentions some
possible solutions for the unexpected deletion of some /i/s
and /u/s, but in the end leaves these cases unresolved.

Chapter four, 'Default Segmentism', concerns an unexpected
/i/ vowel that appears in the diminutive and 'out of
control' (but not the distributive) reduplicative affixes
when these are stressed. The generalization (described in
earlier work by Dawn Bates) is that the epenthetic /i/
vowel appears when copying the base's vowel fails. One
circumstance in which copying fails, is when the base vowel
is long. U proposes that "Lack of copying satisfies
transfer because if there is no corresponding vowel,
whether its length is preserved or not is a moot issue...with
an epenthetic segment in [the diminutive affix], there is
no corresponding vowel and transfer is vacuously obeyed"
(page 133). This once again strikes me as odd (although I
do not have a better explanation). Length (in the form of a
mora or a branching rhyme) is presumably part of the
prosodic structure. But prosodic structure (and
particularly syllable structure) is often not copied under
reduplication. At any rate, U's analysis does nicely
collapse the various situations where copying fails, and
the epenthetic /i/ appears. Her explanation for why the
epenthetic /i/ does not appear with the distributive
reduplicative morpheme, is that the latter is a root (as
discussed above), and roots in general are able to support
more marked structure than affixes.

U points out that a consequence of this analysis of
epenthetic /i/, is that epenthetic segments must be
assigned to a particular morpheme, even when they appear
between two morphemes. One of the arguments against
explicit morpheme boundaries in early versions of
generative phonology, was that epenthesis to the left or
right of a morpheme boundary was arbitrary. In light of U's
analysis, it may be worth re-visiting these earlier
arguments.

A word in Lushootseed can contain more than one
reduplicative morpheme; such words are the topic of chapter
five, 'Double Reduplications and the Base.' The base of the
outer reduplicative morpheme is the entire string
consisting of the root plus the inner reduplicative
morpheme. That is, the base in Lushootseed is
phonologically, not morphologically, determined.

In certain configurations, the vowel of the (outer)
distributive affix unexpectedly contains the vowel /i/,
rather than copying the schwa of the root. U argues that
this happens when the (inner) diminutive morpheme contains
the epenthetic /i/ (as described earlier), and shows how
this is accounted for by the constraint ranking. (In a
derivational theory, this might be accounted for by cyclic
rule application.)

In this same configuration (distributive-diminutive-root),
the distributive morpheme appears as a CV, rather than its
usual CVC; this is attributable to antigemination. (U
assumes that it is the distributive's final C that does not
show up, rather than the diminutive's initial C. It is not
clear to me that there is a principled basis for this
decision, or indeed that it cannot be left ambiguous.)

In the way of typos, the accent (stress) marks are
crucially missing from the polysyllabic examples of (46b)
on page 150, making it difficult to follow the argument in
the text. Occasionally footnotes are misnumbered; footnote
37 on page 97 should be 35, while footnotes 51-53 on pages
109-110 should be numbered 49-51. The second and third
sentences immediately after example (26) on page 184 appear
to be in the wrong order. There are in addition a
substantial number of misspellings and other typos in the
English text, none which should cause confusion.

While the focus of this work is on phonology, not
semantics, I would have found it helpful if each morpheme
in the Lushootseed examples had been glossed, in addition
to the word-level glosses. One can decipher how a three
morpheme word might be glossed 'gravel' ((12b) page 177),
but morpheme level glosses would have made it easier.

The index is not particularly complete; in particular, not
all constraints are indexed. (A summary listing of all the
constraints, together with their relative ranking where
that can be determined, would also have been useful.)

Overall, U has thrown light on a very interesting set of
data, and made some innovative suggestions for how that
data might be accounted for. If I am not wholly convinced
by her argumentation, it is nevertheless an interesting and
thought-provoking analysis, and one which others working on
reduplication will want to look at closely. In this
context, the last sentence of her preface seems
appropriate:

 If I contradict my thesis now or in the
 future, it simply means that my
 teachers and colleagues have done a
 good job, and I am still looking for
 explanations of the explanation.

I would like to add a few paragraphs about the general
treatment of reduplication in OT. While it is possible the
human language learning faculty is innately endowed with
constraints specific to reduplication (the Base-Reduplicant
Faith constraints), this seems (to me, at least) unlikely.

Nor does it seem likely that every child learning a
language with reduplicative morphology makes up the
constraints from scratch (since not even the most
intelligent linguists agree on the correct formulation of
such constraints). That is, both the radical innatist
position and the radical empiricist position seem flawed.

What both these positions (as I have described them) share
is the notion of a Base-Reduplicant (BR) constraint set.
Perhaps the problem is in this set of constraints. I am
reminded of Chomsky's view of parasitic gaps: it is
unlikely that there are special provisions in Universal
Grammar (UG) for parasitic gaps. Rather, parasitic gaps are
the 'accidental' result of the working together of other
aspects of the grammar. I suspect it is the same with
reduplication; there are no special provisions in UG for BR
constraints; rather the resemblance of tokens of
reduplicative affixes to their respective bases is the
result of some other constraints--constraints (or processes)
which are operative in more common situations, perhaps
including ordinary phonological assimilation.

While this review is not the place to present a fleshed out
alternative analysis dispensing with BR constraints, I will
suggest a direction which may be worth exploring. Suppose
we do eliminate BR constraints. We may retain U's (and
OT's) treatment of reduplicative morphemes as minimalist
morphemes, i.e. morphemes with no phonological content,
merely a category which determines their phonological
shape, and whose melodic content is the result of "using"
the base's melody.

The difficulty for this approach is accounting for what U
(and McCarthy and Prince in work on which U's thesis was
based) refer to as "the emergence of the unmarked", a
result which is accounted for by universally ranking IO-
Faith above BR-Faith, and by allowing phonological
constraints to be ranked between these on a language-
particular basis. But if as I have suggested BR-Faith does
not exist, then it clearly cannot be ranked differently
from IO-Faith. I would suggest that the reduced ranking of
faith for reduplicative morphemes should instead arise
automatically from the 'parasitic' nature of their relation
to the base, in the same way that segments which have
assimilated to other segments have a parasitic relationship
with features of those other segments. Putting this
differently, IO-faith corresponds to a direct relationship
between the features of some segment in the input and in
the output; but parasitic faith (both in reduplication and
in assimilation) represents the situation in which the
output features of some segment depend on the input
features of some other segment, through spreading or
copying. It is the indirect relationship of this parasitic
faithfulness that makes it somehow less 'important' (and
more susceptible to being overridden by other constraints)
than direct faithfulness.

I will also make a few final comments which do not have to
do with Urbanczyk's book directly, but rather with the
status of endangered languages. It seems clear that
reduplication offers insights into the human language
faculty that we cannot obtain in any other way, and that
different forms of reduplication have different stories to
tell us. At the same time, reduplication is at best
sporadic in most major languages. The theoretical insights
worked out by Urbanczyk would have been virtually
impossible to come by had it not been for the detailed work
of descriptive linguists like Thom Hess, Dawn Bates, and
others, not to mention the willingness of the Lushootseed
people themselves to share their language. I hope that
linguists and native speakers will be as careful to
preserve other endangered languages for future study, both
for the intrinsic value of those languages and the cultures
they represent, and for the light they can throw on the
human language capacity.


Mike Maxwell works in the morphological and phonological
grammar development for the Linguistic Data Consortium. He
has a Ph.D. in linguistics from the University of
Washington.