8.1275, Review: Hung: Edge Constituents (Part 1)
The LINGUIST List
linguist at linguistlist.org
Sun Sep 7 20:26:04 UTC 1997
LINGUIST List: Vol-8-1275. Sun Sep 7 1997. ISSN: 1068-4875.
Subject: 8.1275, Review: Hung: Edge Constituents (Part 1)
Moderators: Anthony Rodrigues Aristar: Texas A&M U. <aristar at linguistlist.org>
Helen Dry: Eastern Michigan U. <hdry at linguistlist.org>
T. Daniel Seely: Eastern Michigan U. <seely at linguistlist.org>
Review Editor: Andrew Carnie <carnie at linguistlist.org>
Associate Editors: Ljuba Veselinova <ljuba at linguistlist.org>
Ann Dizdar <ann at linguistlist.org>
Assistant Editor: Martin Jacobsen <marty at linguistlist.org>
Software development: John H. Remmers <remmers at emunix.emich.edu>
Zhiping Zheng <zzheng at online.emich.edu>
Home Page: http://linguistlist.org/
Editor for this issue: Andrew Carnie <carnie at linguistlist.org>
==========================================================================
What follows is another discussion note contributed to our Book Discussion
Forum. We expect these discussions to be informal and interactive; and
the author of the book discussed is cordially invited to join in.
If you are interested in leading a book discussion, look for books
announced on LINGUIST as "available for discussion." (This means that
the publisher has sent us a review copy.) Then contact Andrew Carnie at
carnie at linguistlist.org
=================================Directory=================================
1)
Date: Sat, 30 Aug 1997 00:48:12 +0200
From: "Loren A. BILLINGS" <billings at rz.uni-leipzig.de>
Subject: Hung, H: Edge Constituents. Part 1
-------------------------------- Message 1 -------------------------------
Date: Sat, 30 Aug 1997 00:48:12 +0200
From: "Loren A. BILLINGS" <billings at rz.uni-leipzig.de>
Subject: Hung, H: Edge Constituents. Part 1
[***Editor's note. Due to the length of this review it has been posted
in two parts. The second 1/2 of the review appears in a separate message]
Hung, Henrietta J. (1995) _The rhythmic and prosodic organization of edge
constituents: An Optimality-theoretic account._ Bloomington: Indiana
University Linguistics Club Publications, vi + 172 pp., $24.
Reviewed by Loren Billings, <BILLINGS at RZ.UNI-LEIPZIG.DE>,
Universitaet Leipzig
1. Technical preface to this review:
In order to render metrical grids in e-mail format, it's essential
that this review (specifically, certain examples herein) be read using a
font which renders each character with equal horizontal spacing and with
width of at least 75 characters. Additionally, numerous types of
parenthetic symbols and the like are included and may not show up in some
reader's configurations: braces, _{_ and _}_, which Hung uses to mark the
edges of syllables; parentheses _(_ and _)_, which she uses to mark edges
of feet, abbreviated as _Ft_ below; square brackets _[_ and _]_, which she
uses to mark the edges of a prosodic word, which she and I abbreviate as
_PrWd_; and angle brackets _<_ and _>_, which are used to mark unparsed
underlying elements. I am rather sure that most of these will not be a
problem. (Linguist List, which uses the US-ASCII character set, sends all
of these.) If any of these show up so far as strange codes, you may wish
to search below and replace. I apologize for this inconvenience, but I
tried to do without these symbols and found it nearly impossible to discuss
Hung's argumentation while changing her notation drastically.
Additionally, I use _M_ where Hung uses a lower-case Greek letter mu for
"mora", and _!_ for an empty or epenthetic element where she uses an
upper-case delta. In place of a lower-case sigma I generally use _Syl_
(but I occasionally use _SYL_ and _syl_, however, as shorthand for heavy
and light syllables, respectively). I use _'_ before a vowel to indicate
(main) stress; if secondary stress is indicated, I use _`_ before the
vowel. Finally, I quote Hung's examples verbatim occasionally with her
numbering; any examples I've modified are represented by letters.
Hung generally makes a typographic distinction between the OT
constraints that she adopts/proposes, which she shows in bold face, and
constraints/rules proposed in other works, which she renders in italics
(cf. fn.*, p. 27). This use of bold-faced constraints is extremely
convenient for the reader when paging back for the most recent definition
or previous use of a particular constraint. For the purposes of this
review (again due to e-mail limitations), the constraints which Hung shows
in bold face are in all caps here, while the others are between underscores
(e.g., RHYTHM and _Nonfinality_, respectively).
2. Overall assessment of the book:
One of the original criticisms of Optimality Theory (OT), was that it
just throws together various stipulations about language in constraint form
and then re-arranges their ranking to arrive at why human languages vary.
For example, in Prince & Smolensky's (1993) seminal work, the _Nonfinality_
constraint is introduced, which merely stipulates that stress not be
word-final. Hung has taken that stipulative observation about non-final
stress and linked it to metrical rhythm in general: Just as two stresses
shouldn't be too close together, nor should the final stress and the end of
the adjacent PrWd. The idea of this work, therefore, is a very good one.
Not only is the RHYTHM constraint discussed extensively; so too are
two others (PARSE and STRICT-PARSE) which, together with RHYTHM, account
for numerous exceptional phenomena having to do with PrWd edges, often
referred to as _Extrametricality_ in the pre-OT literature.
Hung's book covers a wide breadth of the world's iambic-stress
languages to show a diverse variety of ways languages avoid having
PrWd-final stress. Had this book concentrated just on iambic systems, it
would have been quite an accomplishment. Hung also forays into two
trochaic language types to show how her RHYTHM constraint can affect the
stress even if the feet have initial stress. While her discussion of
Cairene Arabic is convincing, Hung's redefinitions of RHYTHM and
STRICT-PARSE in the section on Latin seriously degrade her own robust
definitions of these two constraints elsewhere in the book. Despite these
problems, this discussion of trochaic systems is worthwhile.
While my review is limited in many ways by its e-mail medium, Hung's
book has none of these limitations. Admittedly, finding fonts with each of
a dozen-odd languages' diacritics and special characters would be a major
task. Still, this book shies away from such fundamental diacritics as
ha^ceks and Umlauts. In some cases digraphs are used instead of the
desired letter plus superscript second letter. For example, instead of
using clear symbols throughout her discussion of Axininca Campa, to which
the better part of a chapter is devoted, Hung merely lists her makeshift
code (fn. 1, p. 57). For other languages no legend is provided at all.
These digraphs and other fudges can be very distracting to the reader
especially when asked to scan unsyllabified examples for closed and open
syllables.
Particularly distracting is the use of consecutive vowel letters (pp.
57ff) to render diphthongs (and, in places, long vowels): for example,
_ai_ (and _oo_) instead of far more transparent _aj_ or even _ay_ (and
_o:_). Agreed, "whether a second root node is referred to as a vowel or as
a glide is not really important" (p. 59) from a theoretical point of view;
a sequence like _a:tai_ "iremos" will be footed as (a:)(tai) regardless of
whether the latter foot is ({taj}) or *({ta}{i}), which are both bimoraic.
Still, from a readability standpoint it would have helped to either modify
the orthography or at least show syllable breaks throughout.
This book has numerous typographic errors. Many of the misspellings
could have been avoided with a simple spell-checker, but are not too
distracting. I point out any of the more problematic errors in turn below
in the "Typos" section at the end of this review. In one respect, the 1994
Brandeis University dissertation upon which this book is based, available
in PostScript version at <http://ruccs.rutgers.edu/roa.html>, has fewer
errors: Several mistabulations in the table of contents do not appear in
the earlier version. While this book holds together well ("perfect-bound"
according to the publisher's brochure; paper acidity not specified), and in
a very convenient size (about 23 x 15 cm), its content is little changed
over the dissertation itself, errors and all. There is no index; nor does
the bibliography appear to have been updated. (I have found a few of the
manuscripts which Hung cites and listed them in their published or
WWW-archived forms at the end of this review.) The dissertation's and this
book's footnotes and examples are numbered differently.
3. Overall organization of the book:
Chapter 1 provides a good introduction to the theoretical framework.
Chapters 2 through 5, the heart of the work, go through various
iambic-stress systems showing the interaction of Hung's RHYTHM constraints
with numerous others. The primary languages discussed here are Negev
Bedouin Arabic, Hixkaryana, Cayuga, Axininca Campa, Choctaw, Aguaruna,
Southern Paiute, Hopi, Ulwa, Yidi~n, Ojibwa, Creek, and Araucanian.
Chapter 6 discusses two trochaic systems: penultimate-stress Cairene
Arabic, and antepenultimate-stress (Classical and Pre-Classical) Latin.
Chapter 7 returns to iambic systems with a typological overview thereof.
The only language to which Hung devotes an entire chapter is Yidi~n
(chapter 4, pp. 93-117). Interestingly, glosses and sources of examples
are listed, along with explanations of special characters, at the end of
the chapter (pp. 116-117). This allows a much more complete gloss
(explaining, for example, the meaning of each inflected form).
Unfortunately, however, this led to two unglossed forms in a tableau and
footnote, shown in (Ka-b) below.
It is odd that the final chapter, "7. Conclusion: A typology of
iambic systems," appears _after_ the chapter on trochaic systems. (See
also my section below on "Foot-structure typology.")
The remainder of this review has the following organization: In the
following section I discuss Hung's definition and use of the RHYTHM
constraint. Section 5 discusses two other constraints in Hung's book which
work closely with RHYTHM: PARSE and STRICT-PARSE. In section 6 I return
to the RHYTHM constraint to further take issue with Hung's redefinition of
the RHYTHM constraint in Latin. Section 7 deals with various
insufficiently addressed issues discussed in the book. Sections 8 and 9
deal with improperly defined constraints and terms that could have been
clarified better. Section 10 addresses Hung's overall mastery of various
Optimality-theoretic issues. In section 11 Hung's choice not to discuss
primary word stress and other word-internal rhythmic phenomena is
addressed, while in section 12 I criticize Hung's having skirted the issue
of iambic-trochaic typology. Sections 13 and 14 list typos and references.
Finally, I close in section 15 with a brief mention of my own interests and
background.
4. Defining RHYTHM:
This is the definition Hung uses through most of the book (but
deviates from somewhat haphazardly in her discussion of Latin; see below).
(p. 10):
(9) RHYTHM:
Every grid mark x at level n + 1 (where n [is
greater than or equal to] 1) must be followed by a
beat of height n such that there is no beat of
height greater than n which intervenes.
She shows in the following examples how RHYTHM is violated in various grid
configurations:
(A) Grids with violations of RHYTHM:
a. (2) x x
(1) x x x x x
b. (3) x
(2) x x
(1) x x x x
c. (3) x
(2) x x
(1) x x x x x
d. (3) x* x*
(2) x x
(1) x x x x
e. (3) x* x
(2) x x x
(1) x x x x x x
f. (2) x x*
(1) x x x x
g. (3) x
(2) x* x
(1) x x x
h. (3) x*
(2) x* x*
(1) x x x
(I've added the grid-level numbers, which Hung uses occasionally for
illustrative purposes.) The book deals primarily with the arrangement of
level-2 grid marks, but does occasionally discuss three- and four-level
systems (as do I below). As the clause "where n [is greater than or equal
to] 1" clarifies, level-1 grid marks always satisfy RHYTHM. Any level-2
grid mark must be immediately followed by a column of level 1 height. The
assessment of level-3 (or higher) grid marks is a bit more complicated:
It's perhaps clearest to strip away all but the next lower level of the
grid to illustrate the assessment most clearly; the following primed
examples show the grids in (A) which have three levels (level-2 asterisks
have been erased):
b'. (3) x
(2) x x
c'. (3) x
(2) x x
d'. (3) x* x*
(2) x x
e'. (3) x* x
(2) x x x
g'. (3) x
(2) x x
h'. (3) x*
(2) x x
The spacing between columns of grid marks does not matter. All that
matters is that any level-3 mark look to the next column that is at least
level 2 in height. Thus, (Ab', c', g') are identical with respect to
assessing the level-3 grid marks . Turn next to (Ah'): The level-3 grid
mark violates RHYTHM because it is not followed by _any_ column of marks;
the same goes for the grid-final level-3 mark in (Ad'). The first level-3
grid marks in (Ad') and (Ae') are followed by a column at least level 2 in
height, but these are taller than level 2, making them bad. This
mechanism of determining good and bad grid marks with regard to RHYTHM is
relevant to the discussion below.
Hung enriches her model of the grid and refines her definition of
RHYTHM in order to account for antepenultimate-stress systems such as
Latin, discussed in section 6.2 (pp. 140-151). Some trochaic systems
exhibit penultimate stress and present no complications: Each syllable is
represented by a level-1 grid mark, while the head of each foot is marked
with a level-2 mark. Because of the definition of RHYTHM above, grid marks
on level 2 or higher are the only ones that need to be followed by a column
of marks one level lower in height. (So-called penultimate systems come in
two types, reflecting how final odd-numbered strings of light syllables are
footed: [syl ('syl syl)] and [('syl syl) syl]; Fijian is an example of the
former, p. 135, and isn't discussed further, while Cairene Arabic -
specifically, "the pronunciation of Classical Arabic at two institutions in
Cairo" - is Hung's example of the latter, treated in section 6.1, pp.
136-140.) The Latin case requires an elaboration to the grid, however. In
addition to the two aforementioned grid levels, antepenultimate-stress
systems require an intermediate grid level: level 1 shows each syllable;
level 2 shows each _footed_ syllable; level 3 shows the _head_ of each foot
(= stressed syllable); main stress in the PrWd is represented by an even
higher grid level (if there are multiple feet, cf. p. 144, but Hung does
not dwell on this). This intermediate level is exemplified in (B), the
grid marks of which are the same as in (Ag) above:
(B) (3) x
(2) x* x
(1) x x x
[('syl syl) syl]
Hung's reason for this additional grid layer is to allow RHYTHM to decide
that stressing the antepenult is preferable to stressing the penult. If
there were no such footed-syllable level, then the stressed syllable would
be followed by two unstressed syllables. Hung justifies this extra grid
level as follows (fn. 5, p. 141; misspellings corrected): "It is assumed
that the head of a foot is universally stronger than a non-head. That a
footed syllable should be stronger than an unfooted syllable however is
subject to parametrization. This can be achieved by positing two
conflicting constraints, one which says that a footed syllable is stronger
than an unfooted syllable, and another which limits the number of layers in
a grid (a constraint of the *Struc variety). Only if the former dominates
the latter do we get a Latin type system." (I return to the issue of this
intermediate footed-syllable grid layer in the section entitled
"Foot-structure typology" below.) The addition of a grid level does not
constitute a redefinition of RHYTHM as such; grid marks violate this
constraint under the same conditions, as shown by the asterisk on the
left-hand level-2 grid mark of the grid in (B).
Hung also modifies considerably the definition of the RHYTHM
constraint itself (pp. 142ff). A brief background of the stress facts is
necessary to set up her redefinition of RHYTHM: Latin stresses a heavy
(i.e., _CVC_ or _CV:_) penultimate syllable; otherwise the antepenult is
stressed. Latin is so obstinately rhythmic that the only examples of
word-final stress are monosyllabic words, as in (Ca). Four types of
disyllabic words are shown in (Cb-e), each of these having initial stress:
(Cb) has two light syllables; (Cc) is light-heavy; (Cd) has two heavy
syllables; finally, (Ce) is heavy-light. (Hung gives no glosses in the
Latin section.)
(C) Mono- and disyllabic forms in Classical Latin
a. x* |b. x |c. x |d. x* |e. x*
x* | x* x* | x* x* | x | x
x | x x | x x | x x | x x
({r'e:})|({t'o}{ga})|({'a}{mo:})|({'am}){bo:}|({fl'a:}){re}
In order for (Ca) to have a stressed, footed syllable at all, stress in
monosyllables must be final; thus multiple violations of RHYTHM are forced
by a bundle of unviolated constraints referred to collectively as
minimal-word considerations. Disyllabic words, regardless of syllabic
weight, have initial stress, Hung argues, in order to satisfy RHYTHM: The
forms in (Cb-c) each begin with a light syllable. In order to be in a foot
with at least two moras, these forms must include the second syllable in
the same foot; this results in multiple violations of RHYTHM, but it is not
possible to satisfy RHYTHM in any better way without incurring minimal-word
violations (or epenthesizing, which Latin does not resort to). Both of
(Cd-e) begin with a heavy syllable; the second syllable in each is not
needed by the first to form a suitable (bimoraic) foot, so this second
syllable is left unfooted. That is, as long as each word has a bimoraic
foot, then every other constraint gives way: (Cd-e) each have a final
unfooted syllable, while (Cc-d) each have a final heavy syllable without
stress (in violation of WEIGHT-TO-STRESS: "A heavy syllable is
stressed."). Footing the second syllable in (Cd-e), or stressing the
second syllable in (Cc-d), would just add bad grid marks to the second
column in the grid. Up to this point in the book, therefore, the more bad
grid marks there are, the less rhythmic a word is. Incidentally, the grid
in (Ce) is a correction of Hung's (24a), p. 143; it is clear from Hung's
preceding discussion that she intended the grid as shown in (Ce).
In other Latin data, however, Hung modifies this definition of RHYTHM:
She considers first the grids in (Da-b):
(D) a. x b. x
x* x x x*
x x x x x x
Note that there is exactly one bad grid mark in each of (Da-b). The RHYTHM
constraint, as defined theretofore in the book, says that rhythmicity boils
down to which form has fewer bad grid marks. But Hung needs the
descending-staircase arrangement in (Da) to be the optimal one. And, in
order to do so, she resorts to the following "refinement" of this
constraint's definition (pp. 142-143, her emphasis):
Intuitively, we can see that only in [(Da)] is
nonfinality truly met, reflecting the observation
given by Mester (1994:17) that "(in final position)
avoid foot-head and avoid footing."[...] It seems
then that Latin calls for a refinement of the
definition of RHYTHM; not only should we look at the
bad grid marks, but we should also look at the good
grid marks. More specifically, we prefer the good
grid marks to be in _different_ columns.
What Hung appears to have in mind in the preceding passage is the
following: (a) _fewer_ bad grid marks represents better satisfaction of
RHYTHM; (b) if there is an _equal_ number of bad marks, then the
arrangement of the good and bad grid marks is what decides which grid is
more rhythmic. Hung returns to this general idea in her discussion of
Pre-Classical Latin. She is discussing so-called iambic shortening, the
shortening of the second syllable's vowel in underlyingly light-heavy
disyllables. The following three forms are assessed [= her (39a-c), p.
148; asterisks added]:
(E) a. x b. x c. x*
x* x* x* x* x*
x x x x x x
('a mo)M *('a mo:) *a (mo:)
The second vowel of (Ea) has been shortened; (Eb) is identical to the
Classical Latin form above in (Cc). The grids in (Ea-b) are identical. In
(Ec) only the second syllable is footed; presumably, Hung intended to mark
stress on the second syllable here. Note that these grid shapes have
exactly two bad grid marks each. Hung states that "it is due to RHYTHM"
that (Ea) is better than (Ec) without any explanation. Her
good-grid-marks-in-different-columns proposal in the preceding excerpt
doesn't appear to apply here. Here the bad grid marks are in the same
column; is this what Hung has in mind? She goes on to discuss so-called
cretic shortening, in which words ending in heavy-light-heavy sequences of
syllables become ...heavy-light-light, as (Fa-d) show [= her (40a-d), p.
149; asterisks added]:
(F) a. x (4) b.
x* x (3) x
x* x* x* (2) x* x
x x x (1) x x x
(d'i:)(ci to)M (d'i: ci) to M
c. (4) d.
x* (3) x* x*
x (2) x x*
x x x (1) x x x
(di:) ci to M *(d'i:) ci (to:)
She assumes the output in (Fa), although the footings in (Fb-c) would also
result in the attested form (_d'i:cito_, with a final short vowel). Hung
refers, correctly, to (Fb) as a more rhythmic than (Fa). She then refers
to (Fc-d) as _less_ rhythmic (p. 148): "Both are less rhythmic because
both contain fewer 'good' grid marks" than (Fa). In this discussion Hung
has apparently abandoned assessing the bad grid marks; (Fa), with one bad
mark, and (Fd), with three, each have fewer bad grid marks than (Fa), with
four. Hung is instead using _good_ grid marks to assess rhythmicity.
Thus, her definition has gone from dubious ("we prefer the good grid marks
to be in _different_ columns") to farcical. Below, following my discussion
of Hung's PARSE and STRICT-PARSE constraints, I show that these
redefinitions of RHYTHM are an attempt to subsume too much of the grammar
under a single constraint.
5. The complexities of PARSE and STRICT-PARSE:
As shown above, the RHYTHM constraint eliminates final stress. In an
iambic system a foot is head-final (= stresses the second of two syllables
in the foot). A word-final foot therefore entails word-final stress; but
RHYTHM requires the word not to have final stress. Hung therefore comes up
with an extremely intricate mechanism for allowing final syllables to be
stressless. In some cases RHYTHM forces the last syllable not to be part
of any final foot, while in others the word-final foot contains no stressed
syllable. (In yet other cases the final foot changes shape, becoming a
trochee.) Another problem arises with final heavy syllables, which
generally must be stressed. Failing to associate the weight-causing coda
consonant or second vocalic mora with the syllable node gets around this
problem. All in all, Hung's use of two kinds of parsing - strict and weak
- makes for quite a capable mechanism for explaining nonfinality effects.
Until, that is, she attempts an eleventh-hour redefinition of this
constraint (again for Latin), as I show below.
Hung also discusses the various constituents involved in the prosodic
structure (p. 18). Following McCarthy & Prince (1993b), she concludes that
while the syllable, foot and word are prosodic "constituents", root nodes
and moras are not. This leads her to propose a split definition of PARSE
in which "skeletal-level units" (Rt, M) must be sub-syllabic, while parts
of the prosodic hierarchy (Syl, Ft) must be within some PrWd:
(12) PARSE (Final Statement)
Subsyllabic constituents must be part of the
syllable; Prosodic constituents must be part
of the Prosodic Word.
This definition of PARSE allows, effectively, for only two ways to attach
each of these four constituents into the prosodic structure: Each can be
the immediate daughter of the next highest category, as (Ga-d) show.
(These structures are intended to show that while the syllable and foot
must be immediate daughters of foot and PrWd in (Gc-d), respectively, in
(Ga-b) the root and mora are both required to be merely within the lowest,
unadjoined-to syllable node.)
(G) a. Syl b. Syl c. Ft d. PrWd
/\ /\ (/) \ (/) \
... Rt ... M ... Syl ... Ft
Alternatively, each constituent type can also skip an immediately higher
node, and attach to some higher constituent. In order to satisfy PARSE,
however, an element must remain within a particular category (i.e., any
root note or mora must remain within some syllable, while any syllable or
foot must be within some PrWd). Thus, for example, if a syllable node is
immediately dominated by a PrWd node, then the PARSE constraint is
satisfied:
(H) PrWd
/ \
... Syl
This notion of satisfying PARSE, in (Ga-d), while violating STRICT-PARSE,
in (H), is a key part of Hung's approach.
The following restatement of the definition is arrived at quite
ingeniously (p. 19):
(17) STRICT-PARSE (Final Statement)
A parsed node must be parsed by a higher
node, such that the number of levels
skipped is minimal."
Hung points out four important points with regard to this constraint's
definition:
First, this constraint must be assessed in terms of the daughter (=
the parsed element), not the mother node. A node with improper daughters
does not violate PARSE or STRICT-PARSE at all (but some other constraint,
such as Ito & Mester's 1992:12 _Proper Headedness_).
Second, if some constituent violates PARSE, it vacuously satisfies
STRICT-PARSE. That is, STRICT-PARSE merely assesses those nodes which
already satisfy PARSE (as reflected in her wording: "A parsed node must
...").
Third, this definition relies, after a fashion, on the OT principle of
gradient violation: When comparing two candidate forms with regard to a
particular constraint, the form with incrementally more violation of this
constraint is suboptimal. That is, when a syllable is not the immediate
daughter of a foot, as in (Ia),but of a PrWd, as in (Ib), then this
structure violates STRICT-PARSE slightly less than if the same syllable is
daughter of neither a foot nor the smallest PrWd domain, but rather
adjoined to a larger PrWd domain, as in (Ic). The following are my
modifications of Hung's exx. (18a-c), p. 20. (See my discussion of
adjoined structures below.)
(I) a. ... b. ... c. PrWd'
| | / \
PrWd PrWd PrWd \
| / \ / \ \
Ft Ft \ /...\ \
(/) \ / \ \ \
(...) Syl /...\ Syl Syl
In (Ia) both PARSE and STRICT-PARSE are satisfied. However, while both of
(Ib-c) satisfy PARSE (because Syl is within some PrWd node), neither (Ib)
or (Ic) satisfies STRICT-PARSE. The syllable node in (Ib) fails to be the
immediate daughter of some foot node. In (Ic) the syllable node fails as
well to be within the (unadjoined-to) PrWd node. This means that the
structure in (Ib) incurs _some_ violation, while (Ic) incurs _more_
violation, of STRICT-PARSE.
Fourth and finally, Hung observes that structures like (Ic) are never
likely to be optimal; if violation of STRICT-PARSE is forced by some
dominant constraint, a structure like (Ib) would nonetheless be preferable
to (Ic). That is, no other constraint appears to prefer (Ic) over (Ib).
This entails that a structure like (Ic) will never be the optimal candidate
(and therefore is never attested). This leads Hung to generalize that
whereas STRICT-PARSE is a gradient constraint, it is never the case that
candidates like (Ic), which violate STRICT-PARSE more than minimally, are
ever attested. Hung stops short, however, of claiming that forms such as
(Ic) are not part of the candidate set (i.e., not produced by _Gen_, the
black box that produces the various candidates in an OT grammar).
Therefore, alongside the candidates in (Ga-d), which satisfy STRICT-PARSE,
are those in the following four structures, which _minimally_ violate
STRICT-PARSE .
(G') a. Syl' b. Syl' c. PrWd d. PrWd'
/ \ / \ / \ / \
Syl \ Syl \ Ft \ PrWd \
/ \ \ / \ \ / \ \ /\ \
/...\ Rt /...\ M /...\ Syl /__\ Ft
The structures in (Ga-d) and (G'a-d) are, according to Hung, the only way
that elements within the PrWd can be part of the prosodic structure. The
structures in (G'a-d) are said to be "weakly parsed" (i.e., satisfying
PARSE but violating STRICT-PARSE). The weakly parsed element is uttered
overtly, but fails to be counted in the "skipped" category's prosodic
prominence. In (G'a-b) sub-syllabic segments are not part of the
unadjoined-to syllable. They are thus uttered (satisfying PARSE) without
making that syllable prosodically heavy. Hung calls these "weightless"
consonants and moras. The "loose" syllable in (G'c) fails to be part of
the foot. Finally, the foot in (G'd) is not part of the inner PrWd, this
"headless" foot has no stress. Each of these mechanisms can prevent
word-final stress: Some languages require heavy syllables to be stressed
(WEIGHT-TO-STRESS, see below). If a syllable has a long vowel or is closed
by a consonant as in (G'a-b), then such a syllable can be uttered with the
long vowel or final consonant but nonetheless escape the WEIGHT-TO-STRESS
constraint that requires a heavy syllable to bear stress. In languages
with iambic feet, the mechanism in (G'c) can keep a syllable from being
within any foot, thus allowing that (final) syllable not to bear stress.
Finally, the headless foot in (G'd) allows for a final foot with no stress,
thus also avoiding final stress.
Hung's use of adjoined prosodic structures is also worth commenting on
in several respects:
First, in her notation an apostrophe indicates a re-instantiation of a
particular prosodic category; cf. _PrWd'_ above _PrWd_ in (G'd) and _Syl'_
above _Syl_ in (G'a-b).
Second, in order for sub-syllabic constituents to be within some
syllable and still be outside of the lowest syllable, Hung proposes that
the weightless mora or consonant is "adjoined to" Syl. Skipping to (G'd),
for a foot to satisfy PARSE it must be within a PrWd, yet in order violate
STRICT-PARSE it must skip a PrWd node. The answer, as in (G'a-b), is for
Ft to be adjoined to PrWd. Note, however, that the loose syllable in (G'c)
is not adjoined to anything, but instead skips the foot node and is
daughter of PrWd. The syllable is unique in having two prosodic categories
above it to choose from - Ft and PrWd - while still satisfying PARSE. Syl
could therefore either adjoin to Ft (not shown) or be the immediate
daughter of PrWd, as in (G'c). Though Hung doesn't address this, the
latter allows for a loose syllable without requiring adjunction and
therefore seems preferable to the presumed expense of an adjoined
structure.
Third, another point which Hung doesn't discuss is whether the lower
Syl nodes in (G'a-b) themselves satisfy STRICT-PARSE; her wording of this
constraint (see above) allows for more intervening structure (the addition
or Syl' below Ft) as long a no actual _skipping_ of nodes takes place.
Fourth, Hung proposes that adjunction to a prosodic category is "by
definition" allowed only at the edges of a PrWd (p. 19). This entails that
the only kind of word-internal violation of STRICT-PARSE is a weak
syllable, shown in (G'c), because no adjunction is involved in this kind of
weak parsing. This makes sense with regard to weakly parsing a foot: How
would a foot adjoin to a PrWd between immediate daughters of this PrWd
without crossing lines in the prosodic tree? Hung admits, however (fn. 10,
p. 23), that it's unclear why adjunction of sub-syllabic elements to Syl
should be restricted to PrWd edges.
Fifth, Hung makes clear that a syllable can be weakly parsed anywhere
in the PrWd, while Rt, M, and Ft nodes can adjoin only at word edges (p.
80). This only part of the story, however: Because weight-bearing
sub-syllabic elements are syllable-final, and (by the preceding
stipulation) adjunction is only at PrWd edges, adjunction to Syl can only
be at the _final_ edge of a PrWd. Thus, Syl can be weakly parsed anywhere,
Ft can do so at either PrWd edge, but Rt and M can do so only at the right
end of the PrWd. Additionally, while a root node can be weakly parsed to
any final syllable, a mora can be weakly parsed only to a vowel-final
syllable (fn. 4, p. 101). I've found two apparent deviations from this
array of allowed adjunction sites: In Cairene Arabic adjunction to Syl is
apparently only allowed phrase-finally (p. 140); this is not so much an
exception as a further restriction (which is not explained). Hung also
points out an apparent case of word-internal adjunction of a mora to Syl in
Pre-Classical Latin verse (which I discuss below at the end of this
section). The fact that a loose syllable can be anywhere in the word is
nonetheless effectively restrained by RHYTHM, which needs the loose
syllable to be PrWd-final in order to avoid word-final stress (p. 61).
Sixth, a distinct advantage of Hung's adjunction model is that a lone
foot cannot adjoin to PrWd because there must be at least one "properly
mothered" (= non-adjoined) foot in order for there to be a PrWd in the
first place (p. 39). It's impossible to adjoin to nothing. This explains
why single-foot words don't have the option of lacking a stressed syllable.
This, in turn, explains why many iambic languages resort to a trochee just
in case the word is disyllabic (to avoid final stress yet allow the lone
foot to satisfy STRICT-PARSE).
Seventh, I should add that Hung stops short of formalizing the
difference between Syl and Syl' (respectively, PrWd and PrWd') in the
constraints themselves. For example, she adopts the following constraint
(p.79):
(79) ALIGN-LEFT:
Any [PrWd is aligned with a (Ft.
(_PrWd_ and _Ft_ are subscripted. Read: "Any left edge of a prosodic word
is aligned with a left edge of a foot.") She then shows the following
schematic of various numbers of initial light syllables (modified in the
list of typos below and further modified here with _syl_ and _SYL_ instead
of _L_ and _H_, respectively):
(J) a. [(syl SYL) ...
b. [(syl syl) [(SYL) ...
c. [(syl syl) (syl SYL) ...
d. [(syl syl) syl (syl SYL) ...
e. [(syl syl) (syl syl)(syl SYL) ...
Hung clarifies (p. 80) that this constraint requires every instance of
PrWd, including _PrWd'_, to be left-aligned with some foot. This is
illustrated in (Jb), which has two PrWds; the leftmost square bracket
denotes the adjoined-to _PrWd'_ node, while the inner square bracket
corresponds to the unadjoined-to _PrWd_ node. Crucially, Hung's use of
_PrWd_ and _PrWd'_ as labels is not a formal property of the two instances
of this category. Other linguists propose constraints which specify a
particular adjoined level: Legendre (1997) proposes an OT constraint
requiring a particular clitic's left edge to align with the right edge of a
zero-level PrWd (literally, _PrWd_ with a _0_ superscript). L"ohken (1996)
proposes numerous constraints requiring stress to align with particular
PrWd projections: _PrWd'_, _PrWd''_, etc. (I should point out that I
consulted an earlier manuscript of L"ohken's work.) If constraints such as
Legendre's and L"ohken's are taken seriously, then they formalize a
difference between, say, PrWd and PrWd'; that is, these prosodic categories
may have similar names (differing only in the apostrophe), but they have
different formal properties. Thus they're really different categories;
_PrWd'_ really isn't a prosodic word but some sort of intermediate category
between PrWd and MiP (minor phrase). Hung's STRICT-PARSE, on the other
hand, merely rules out skipping any Syl or PrWd node, whether or not there
is an apostrophe. The fact that weakly parsed Rt, M or Ft nodes end up
adjoining, while Syl ends up in an non-adjoined structure shows that Hung's
constraint makes no reference to some particular level of adjunction.
Eighth (and finally), yet another interaction of PARSE, STRICT-PARSE
and an ALIGN-type constraint emerges in chapter 4, on Yidi~n. This
particular interaction requires a bit of explanation: In Yidi~n
STRICT-PARSE is not very prominent in the hierarchy. Hung proposes the
following constraint (p. 104; her emphasis):
(35) ALIGN-R:
The right edge of a |Stem is aligned with
a }Syl*, _where Syl* is a footed syllable._
(Read: "Align the right edge of stem with the right edge of a footed
syllable.") The second proviso is extremely suspect from the start and, as
I will show below, is also inaccurate. I'll summarize Hung's main
arguments with regard to each type of data: "The relevant property of this
constraint is that it treats an unparsed vowel and a [weakly parsed] full
syllable [adjoined to PrWd/L.A.B.] in the same way: both are part of the
morphological stem, yet neither is footed." Following Kirchner (1993),
Hung proposes another constraint (p. 105):
(38) END-RULE:
The rightmost footed syllable is heavy.
This appears to be a way of marking the head foot of the PrWd with special
prominence. And, since "heavy" syllable in Yidi~n is defined only as a
syllable with a long vowel, END-RULE entails that vowel-lengthening be the
way of achieving such PrWd prominence. This constraint is dominated by
ALIGN-R, which essentially says that END-RULE may apply to any PrWd where
ALIGN-R is already violated. ALIGN-R is further dominated by a constraint
requiring each foot to be disyllabic (not just bimoraic). The data are
divided into odd- and even-numbered syllables with further subdivisions
according to final-syllable/-foot shape:
1. Odd-parity words:
Because each foot must be disyllabic, a PrWd with an odd number of
syllables must somehow fail to parse one of these syllables into a foot.
(As in Latin, discussed above, epenthesis is not an option.)
1a. Loose syllable:
One possibility is to weakly parse one syllable; RHYTHM, although a
relatively low constraint in the hierarchy, emerges to require that it be
the final syllable which is weakly parsed. For example, in a
three-light-syllable word like /gudaga/ "dog"-absolutive, the optimal parse
is [({gu}{d'a!}){ga}], where _!_ represents the epenthetic mora. That is,
because this word has an odd number of syllables, and every foot must be
disyllabic, {ga} must be weakly parsed (= adjoined to PrWd), forcing a
violation of ALIGN-R (because no _footed_ syllable coincides with the end
of the stem). Thus, because ALIGN-R is "already" violated as it were,
there is no harm in violating ALIGN-R in some other way.
1b. Unparsed segment:
Another possibility for certain stems with an odd number of
underlyingsyllables is to just eliminate some final segment(s) as the
following examples show:
(K)
a. /gudani/ b. /mabi-Ngu/ c. /guygal-ni/
({gu}{{da!}n})<i> ({ma}{{bi!}N})<gu> ({guy}{{ga!}l}){ni}
no gloss no gloss "bandicoot"-genitive
[ex.(51), p. 108] [fn. 9, p. 112] [fn. 9, p. 112]
Hung deals only with data like (Ka), calling the phenomenon "vowel
deletion". As (Kb) shows, however, certain final consonant-vowel sequences
can be deleted; (Kc) shows that other final consonant-vowel sequences do
not drop, and instead weakly parse the final syllable (as in /gudaga/
above), if a morpheme boundary appears in a particular position, etc.; I
will not expound on the conditions here. The prosodic organization on the
second line of (Ka-c) is my interpretation of Hung's intentions.
Crucially, all three of these involve either deletion or a weakly parsed
syllable. Suffice it to say that in odd-syllabic-parity words either
weakly parse the final syllable or delete some final material. Both
constitute violations of ALIGN-R. Because ALIGN-R is already violated,
epenthesizing a mora is no worse.
2. Even-parity words:
Any PrWd with an even number of syllables will be exhaustively footed
(i.e., ALIGN-R is met). In no such cases is mora-epenthesis attested,
because this would trigger an unnecessary violation of ALIGN-R. Three
cases illustrate this:
2a. Final heavy syllable:
If the final foot is an iamb with an underlyingly heavy (PrWd-final,
foot-head) syllable, then END-RULE is satisfied vacuously. Mora-epenthesis
is unnecessary.
2b. Final light, open syllable:
An even-syllabic-parity PrWd can, however, also have a short PrWd-final
vowel. This can happen because either the final foot is a trochee,
[...('Syl syl)], or the final foot is iamb and the final stressed syllable
is not inherently bimoraic, [...(syl 'syl)]. Inserting a mora in such
forms violates ALIGN-R as /waRa:buga/ "white apple tree" shows: The
inherent length of a preceding even-parity syllable requires a trochaic
parse: [({wa}{R'a:})({bu}{g'a})]. If an epenthetic mora were added to the
last footed _Syl*_, then the form would end in *...{g'a!})]. This _Syl*_
ends in the epenthetic mora, while the Stem's right edge is _a_; this
constitutes a violation of ALIGN-R, albeit a meticulously defined one.
Here END-RULE is violated due to ALIGN-R.
2c. Final light, closed syllable:
An even-syllabic-parity word ending in a short vowel + consonant such as
/yadi:riNal/ "walk about"-going-transitiviser-present [= Hung's (9b), p.
97], is footed and stressed as [({ya}{d'i:})({ri}{N'al})], without
lengthening the final vowel. Here Hung evokes an extremely involved (and
erroneous) argument for the lack of vowel length in the last syllable: The
preceding attested form violates END-RULE in order to satisfy ALIGN-R, but
in a roundabout way. Inserting a mora, so the argument goes, would force
the final syllable to be superheavy, which Hung analyzes as having the
following internal structure: *...{{N'a!}l})], with a weakly parsed final
consonant. (Supposedly *{N'a!l} would have to be trimoraic, and therefore
disallowed, although there are PrWd-internal CV:CC sequences that require
special explanation, pp. 106-107.) Hung labels the internal syllabic
braces in ...{{N'a!}l})] with a _Syl*_ subscript, and the outer braces as
_Syl'_ (which she defines elsewhere as an adjoined-to syllable). Thus, her
argument goes, if a mora is epenthesized, this forces a misalignment of the
right edges of _Syl*_ and the stem. The fallacy in Hung's argument is that
the syllable demarcated by the _outer_ syllabic braces, despite being
labeled _Syl'_, is nonetheless fully footed! It seems that instead of
requiring a footed "Syl*" and stem to share right edges, it's preferable to
a try some other approaches. Hung mentions (fn. 6, p. 104) that "Green
(forthcoming) argues for the alignment between the edge of a word and the
edge of a foot-head." It's unclear why this approach isn't pursued;
aligning the stem with a stressed syllable is no more stipulative than
aligning to a footed syllable. Unfortunately, if such a constraint were
ranked above END-RULE, it would require a odd-syllabic-parity word to
weakly parse a syllable other than the last one. Hence Hung resorts to
alignment with a footed syllable. Another approach is evident in McCarthy
& Prince's (1993b) discussion of ALIGN-R effects (p. 105):
(37) Consistency of Exponence:
No changes in the exponence of a
phonologically-specified morpheme are
permitted.
Following the Consistency idea, specifically with regard to the stem's
right edge, the various data types discussed above in 1a-b and 2a-c that
any violation of PARSE or STRICT-PARSE (all of the odd-syllabic-parity data
in 1a-b) result in an epenthetic, stressed-syllable mora. Those which
achieve perfect parsing (namely, the even-syllabic-parity data in 2a-c)
never have an epenthetic mora. Unfortunately for Hung's definition of
STRICT-PARSE, a violation of PARSE by definition constitutes satisfaction
of STRICT-PARSE. Thus, she can't just define ALIGN-R as "strictly parse
the right edge of the stem." This would fail to capture the unparsed data
in (Ka-b) above, since they would vacuously satisfy STRICT-PARSE. What
seems to be needed is a definition like "The right edge of the stem is
satisfies both PARSE and STRICT-PARSE." This would do the trick (but could
be objectionable for other reasons).
To summarize so far, the notion of weak parsing allows a constituent
to prevent the PrWd from having final stress (and other seemingly very
different yet theoretically unified phenomena) without actual deletion.
Each constituent can be uttered (satisfying PARSE) while being attached to
the prosodic tree one level higher than it should (violating STRICT-PARSE),
which may be driven by a dominant constraint prohibiting final stress
(i.e., RHYTHM).
A serious departure from STRICT-PARSE: I show above how the violation
of STRICT-PARSE, so called weak parsing, allows for phonetically overt
material to elude certain constraints, namely RHYTHM. Hung is very
specific about this (p. 20, my emphasis): "While unparsed elements are
segmentally realized, and strictly parsed elements are both segmentally and
rhythmically or quantitatively present, weakly parsed elements are
_segmentally present_, but rhythmically absent." It is, therefore, quite
difficult to understand the following about-face in the preceding
definition, specifically with regard to weakly parsed moras (p. 23): "The
theory here gives two ways to represent vowel shortening. Either a mora is
unparsed, violating PARSE, or it is weakly parsed, violating STRICT-PARSE."
If pronouncing an underlyingly long vowel as short involves failing to
utter a portion of this vowel, then the second mora is completely unparsed,
not just weakly parsed. The entire notion of weak parsing depends on overt
utterance - in this case, of the duration of the vowel - while ignoring its
existence when assessing RHYTHM. Here Hung is saying that weakly parsing
the mora not just ignores the weight; this weight is absent altogether.
This dubious reversal of weak parsing's very definition is applied quite
problematically to Pre-Classical Latin (sec. 6.2.2, pp. 146-151), which
(unlike the later classical period) undergoes so-called iambo-cretic
shortening. Whereas both periods have identical placement of stress, in
the pre-classical period "there are in addition a couple of optional
shortening rules." Iambic shortening turns underlyingly light-heavy
disyllables into two light syllables: /amo:/ --> _'amo_; cf. (Cc) above.
Cretic shortening turns words ending in heavy-light-heavy sequences into
...heavy-light-light ones: /di:cito:/ --> _di:cito_ (unlike Classical
Latin _di:cito:_). This analysis, aside from radically altering
STRICT-PARSE, raises several problems/questions: First, exactly how is
iambo-cretic shortening "optional"? Second, while Hung's discussion deals
with shortening long vowels, it is not very specific whether this applies
to syllable-final consonants. Additionally, Hung discusses word-internal
iambic shortening (fn. 9, p. 147): "words like vo.lup.t'aa.tees.
ve.ree.b'aa.mi.mii are scanned as LLH..." How is the syllable {lup} in the
first word "scanned as" light? Is /p/ dropped? Third, in connection with
the preceding passage, because these data are strictly textual, what sort
of evidence is there about whether such scansion actually resulted in
shortening in the pre-classical period? Fourth, and finally, discussing
the same two word-internal shortened iambs, Hung concedes that this
constitutes word-internal weak-parsing of a mora, which is only possible at
the end of a PrWd (same fn.): "I assume that in these cases there may be
two PrWd structures assigned to the lexical word, with weak parsing at the
right edge of the first PrWd." If the vowel length is not overt, then it
is unparsed, not merely weakly so; completely unparsed elements are allowed
PrWd-internally. These problems, similarly to her reformulation of RHYTHM
discussed at the end of the preceding section, place Hung's analysis of
Latin in serious doubt.
6. More on Hung's "refinement" of RHYTHM:
Having introduced the PARSE and STRICT-PARSE constraints, I now return
to some of the structures at the end of my discussion of RHYTHM. I wish to
show here that Hung is trying to force the RHYTHM constraint to perform too
much of the grammaticality determination. I begin by repeating the
Pre-Classical Latin examples above in (Ea-c) [= Hung's (39a-c)] as (E'a-c);
I have re-drawn the bottom line of these examples, based on my objections
to Hung's use of STRICT-PARSE to delete moras:
(E') a. x | b. x | c. x*
x* x* | x* x* | x*
x x | x x | x x
[({'a}{mo})]<M> | *[({'a}{mo:})] | *[a({mo:})]
Recall that these represent the pre-classical period's so-called iambic
shortening. Hung eliminates (E'b) because of an imbalanced foot (violating
the F=MM constraint, discussed below). Hung argues that the grid in (E'a)
is more rhythmic than in (E'c), even though the grids have exactly two bad
marks each. It isn't necessary to resort to RHYTHM here: These two grids
tie with respect to RHYTHM but can be differentiated using STRICT-PARSE;
(E'c) has a weakly parsed initial syllable. (At the end of the RHYTHM
section above, I disagree with Hung's analysis that the dropped mora in
(E'a) should be interpreted as weakly parsed. With the final mora in (E'a)
analyzed as unparsed (i.e., a violation of PARSE and not of STRICT-PARSE),
then the choice of (E'a) over (E'c) is merely a matter of STRICT-PARSE
dominating PARSE in the constraint hierarchy.
I also repeat the grids above in (Da-b) as (D'a-b); I've added (D'c):
(D') x | x | x*
a. x* x |b. x x* |c. x
x x x | x x x | x x x
({d'i:}{ci}){to:}|({d'i:}){ci}({to:})|({d'i:}){ci}{to:}]
Hung compares only the grids in (D'a-b), with no actual examples shown. I
have added the examples and footings underneath. While (D'a) is Hung's
analysis for the classical period, either of (D'b-c) would yield the
attested form as well (with initial stress and initial and final vowel
length). The footing in (D'b) is the only possible one that could
correspond to the grid above it. Hung attempts to show that while (D'a-b)
share the same number of bad grid marks (one each), (D'a) is more rhythmic
because the good grid marks are in "_different_ columns". Why not simply
observe that while these are equally rhythmic, other constraints
differentiate them? For example, while (D'a-b) each have an unfooted
syllable (in violation of STRICT-PARSE), and the two forms each have a
violation of WEIGHT-TO-STRESS (the final long syllable in each being
unstressed), only (D'b) has a headless foot (also in violation of
STRICT-PARSE). Thus, it is the STRICT-PARSE constraint, not RHYTHM, which
differentiates these two.
Comparing (D'c) with either of (D'a-b) likewise reveals the same
number of bad grid marks: one each. All three of (D'a-c) violate
WEIGHT-TO-STRESS. As for STRICT-PARSE, (Db-c) each have two violations:
(D'b) has a headless foot and a loose syllable, while (D'c) has two loose
syllables. Thus, while Hung does not consider a grid like (D'c), it can
likewise be ruled out using STRICT-PARSE.
******* THIS IS THE END OF PART ONE OF THIS REVIEW. PART TWO IS
******* POSTED IN A SEPARATE ISSUE OF LINGUIST
---------------------------------------------------------------------------
LINGUIST List: Vol-8-1275
More information about the LINGUIST
mailing list