Publishing Partner: Cambridge University Press CUP Extra Publisher Login
amazon logo
More Info

New from Oxford University Press!


How Traditions Live and Die

By Olivier Morin

This book brings together cognitive science and quantitative cultural history to look into the causes of cultural survival.

New from Cambridge University Press!


The Acquisition of Heritage Languages

By Silvina Montrul

"This work centres on the grammatical development of the heritage language and the language learning trajectory of heritage speakers, synthesizing recent experimental research."

Review of  The Rhythmic and Prosodic Organization of Edge Constituents: An Optimality-Theoretic Account

Book Title: The Rhythmic and Prosodic Organization of Edge Constituents: An Optimality-Theoretic Account
Book Author: Henrietta J. Hung
Publisher: IULC Publications
Linguistic Field(s): Phonetics
Issue Number: 8.1276

Discuss this Review
Help on Posting

[****Editor's note: Due to the length of this review, it has
been posted in two parts. What follows is the second 1/2 of
the review]

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@RZ.UNI-LEIPZIG.DE>

(continued from previous issue)

7. Loose ends:

In this section I discuss other issues that arise in Hung's treatment
of the various languages. Following brief discussions of Cayuga and
Choctaw, I discuss Aguaruna in some detail. I then return to Yidi~n,
followed by a discussion of Ojibwa.

Section 2.3 (pp. 54-56), on Cayuga, "is modeled closely after the
analysis given in Kager (1993:422-425)" appears to be a late appendage to
chapter 2 too large to be a footnote. (The book is sprinkled elsewhere
with footnotes referring to Kager's alternative approach: fn. 4, p. 11;
fn. 4, p. 33; fn. 4, p. 122.) Hung admits that this analysis is "extremely
sketchy". One inconsistency is presented by the following passage (pp.
55-56), specifically Hung's example (6) and the following paragraph:

It seems like most words which are disyllabic
have an initial long vowel, as the following
examples, from Dyck (1993) indicate.

(5) a. th'o:hah "almost"
b. 'o:neh "now"

[...] Mithun and Henry (1982) gives a form in
which there is a prothetic vowel which is lengthened.

(6) /s-h-e?/ 'i:she? "he is going back"

[...] Cayuga imposes strict conditions on
lengthening, resulting in cases where the initial
syllable of a disyllabic word may end up unstressed
and unlengthened. Examples of such overriding
conditions on lengthening are as follows: (i) do not
lengthen when the syllable ends with an [h], [?], or
an [s] if the [s] is followed by another C (although
it is okay to lengthen if [s] is intervocalic [...]

The initial (prothetic) vowel in Hung's (6) should not lengthen if the
first syllable ends in [s]. Unfortunately, no syllable breaks are shown.
Nor does Hung provide a reference for "Mithun and Henry (1982)" in her
bibliography. I contacted Marianne Mithun, who provided the reference at
the end of this review; this word is listed as _ishe?_ in Mithun & Henry
(1982:539), which likewise doesn't show syllable breaks. Mithun confirms
that while the _i_ in Hung's example (6) is prothetic, it should be
[is.he?], with neither stress nor length on the first syllable. It
appears, therefore, the example as shown in (6) contradicts the paragraph
after it.

Choctaw (treated in sec. 3.2.1, pp. 85-86), shows no stress, only
length, on every second (non-final) light syllable. If the word consists
of an even number of light syllables, then the final syllable is not
lengthened; instead, the final foot is headless (= weakly parsed, adjoined
to PrWd). A word with only two light syllables likewise shows no length.
Because this foot can't adjoin, it's improbable that this foot is headless.
By the process of elimination Hung shows that such a foot is trochaic
(though abstractly so). Hung analyses PrWd-final consonants, which never
cause a syllable to be heavy in Choctaw, as weakly parsed (= weightless,
adjoined to Syl). This then begs the question: Are there /CVC/ or /CVCVC/
words in Choctaw? If /CVC/ is attested, the final consonant should count
because of minimality considerations: [({CVC})]; alternatively, such words
might all require the vowel to lengthen, keeping the final consonant weakly
parsed: [({{CV:}C})]. Similar issues could be clarified by /CVCVC/ words
as well; if such forms exist, do they have vowel length? Alas, these
questions are unanswered in Hung's treatment of Choctaw.

Aguaruna (sec., pp. 87-88) is not unlike other languages Hung
discusses in being iambic unless there are just two syllables, in which
case a trochee (SYL. syl) results, forced by more highly ranked RHYTHM.
Only the initial foot of a PrWd bears discernible stress. Aguaruna's
uniqueness lies in the following two properties having to do with the
elision (in some dialects, mere devoicing) of certain vowels: First, final
vowels are elided in all but disyllabic PrWds. Hung explains the lack of
final elision in disyllables as resulting from minimality considerations.
(All of Hung's Aguaruna data are underlyingly vowel-final and there is no
mention of whether monosyllabic words exist.) Second, elision of a
non-initial foot's non-head syllable also takes place. Presumably the
stressed foot's non-head syllable does not undergo this because of a
strongest-foot effect of some sort. (Stressed vowels are indicated below
with a preceding apostrophe. I would like to have shown syllable breaks,
but where these are located is unclear from Hung's presentation; i.e., is
_Nk_ a pre-nasalized stop or a cluster? The syllabic affiliation of
consonants preceding elided vowels is also unclear.) The data in (L) are
arranged by syllable length (from two to eight syllables):

(L) Underlying form Surface form Gloss
--------------- ------------ -----
2 kasa k'asa thief
3 namaka nam'ak<a> river
4 CaNkina-na CaNk'inan<a> basket+af
5 iCinaka-na iC'in<a>kan<a> clay pot+af
6 CaNkina-Nu-mi-na CaNk'in<a>Numin<a> basket+afs
7 iCinaka-Nu-mi-na iC'in<a>kaN<u>min<a> clay pot+afs
8 iCinaka-Nu-mi-na-ki iC'in<a>kaN<u>minak<i> clay pot+afs

Hung shows analyses the seven- and eight-syllable examples in (Ma-b). A
non-head syllable of a non-initial foot (labeled _w_ for "weak") is elided,
except for the penultimate syllable of the eight-syllable word (labeled
_w*_). Every other use of angle brackets in the book signifies complete
non-parsing; I take this, along with the coda-syllabification of the
preceding nasal, to mean that final <a> here is unparsed, but other "weak"
syllables are parsed.

(M) a. /iCinakaNumina/
w s w s w s w
(i Ci) (na ka) (Nu min) <a>

b. /iCinakaNuminaki/
w s w s w s w* w
(i Ci) (na ka) (Nu mi) (na ka) [sic.]

Hung appears to mistakenly show final _a_ instead of _i_ in the bottom line
of (Mb). I assume the following form to be her intention:

(Mb') w s w s w s w* w
(i Ci) (na ka) (Nu mi) (na ki)

Additionally, she does not show elision of this vowel using angle brackets.
That is, Hung appears to place final vowels only in odd-numbered numbered
syllables within angle brackets; cf. <a> in the seven-syllable example,
(Ma). She does not provide such an analysis of shorter words. This
notation - namely, two weak syllables in the last foot - reflects Hung
analysis of the extrametricality effect: The final foot in a PrWd with an
even number of (and more than two) syllables is adjoined to PrWd. It is
unclear what Hung's angle-brackets notation of the seven-syllable example
(Ma) is intended to mean; the general convention is to use angle brackets
to refer to (completely) unparsed material. Hung surely does not intend a
completely unparsed final mora or syllable because this would leave the
head of the final foot in final position, in violation of RHYTHM. Nor
could she mean a totally unparsed vowel/mora (for the same rhythmic
reasons). In order for some semblance of a syllable to remain final in the
odd-parity PrWds, it is necessary for a syllable to be somehow parsed in
the structure. The final syllable of a PrWd with an odd number of
syllables is "loose" (= the immediate daughter of PrWd). In four-, six-
and eight-syllable PrWds the final foot is "headless" (= adjoined to PrWd).
Such final adjunction is not possible with two-syllable words, thus causing
trochaic footing. To be clear, the following are my interpretation of what
Hung apparently has in mind:

(N) a. 2 [('SYL syl)]
b. 3 [(syl 'SYL) syl]
c. 4 [[(syl 'SYL)] (syl syl)]
d. 5 [(syl 'SYL) (syl SYL) syl]
e. 6 [[(syl 'SYL) (syl SYL)] (syl syl)]
f. 7 [(syl 'SYL) (syl SYL) (syl SYL) syl]
g. 8 [[(syl 'SYL) (syl SYL) (syl SYL)] (syl syl)]

Incidentally, Hung conjectures (p. 88) that the non-elision of the
penultimate vowel of (Mb), marked _w*_, is to keep the _n_ and _k_ from
forming a PrWd-final cluster: "The resistance of this vowel to elision
must be accounted for by other means, such as syllable structure
restrictions." The preceding conjecture assumes that resyllabification
takes place even when afinal vowel in a word with an even number of
syllables is elided. This statement is at odds with the differing
notations of the final vowels in (Ma) and (Mb), corrected as (Mb').
Furthermore, the mere devoicing of some dialects' weak vowels suggests that
the syllable somehow remains even if its vowel is affected. More
importantly, the absence of a seventh syllable in (Ma) entails that RHYTHM
is violated, because of the stressed sixth and final syllable. I assume,
therefore, that just the vowel in a weak syllable is affected (possibly a
completely unparsed mora therein). Since each of Hung's examples is
underlyingly vowel-final, it is hard to determine conclusively whether
final elision is the same as weak-vowel elision. Assuming, with Hung, that
these two phenomena are the same, it is tempting to analyze the final foot
in (Nc, e, g) as being trochaic. Hung is loath to do this, because (in her
original definition of RHYTHM) PrWd-internal, consecutive foot-head
syllables are just as bad as final stress. The grids in (Oa-b) are based
on Hung's (21b, a), p. 63, showing a 4-syllable word; asterisks show
RHYTHM-violating grid marks.

(O) Possible grids (of a 4-syllable PrWd)

a. x* x (2)
x x x x (1) (initial iamb, final trochee)

b. x x* (2) [= (Af) above]
x x x x (1) (initial iamb, final iamb)

Hence Hung's rather concerted effort to analyze such a final foot as
headless, as in (P). While the grid has no RHYTHM-violating marks, it
violates STRICT-PARSE (by adjoining the final foot to PrWd):

(P) x (2)
x x x x (1) Hung's analysis of (Nc)

I can think of two alternative analyses which might be preferable: The
first of these relies (contra Hung) on internal elision being a distinct
phenomenon from final elision. This idea is still in keeping with (Na-g),
however. A final vowel is either completely elided, resulting in its
would-be onset consonant resyllabifying, or reduced (in some dialects),
resulting in a weak final vowel. Internal elision affects all non-head
syllables of a foot. Crucially for this idea, a weakly parsed foot cannot
have a "weak" syllable; for there to be a "weak" there must be a "strong",
as it were. (I return to this idea in my discussion of Ojibwa below.) The
second alternative analysis assumes, with Hung, that final and internal
elision are the same phenomenon. It's clear that the initial foot is head
of the PrWd. Although Hung does not deal extensively with PrWd-headedness
(but see p. 144), main word stress is formalized in the literature using a
higher grid level with a mark over the stressed syllable, as in (Q), (R)
and (S); in the (b) examples only the final foot is trochaic, while the (a)
examples are parsed entirely as iambs:

(Q) a. x (3) b. x
x* x (2) x x*
x x x x (1) x x x x

The level-1 and -2 grid marks in (Qa-b) are identical to (Oa-b). The new
level-3 marks each satisfy RHYTHM (because the next level-2 grid mark is
exactly level 2 in height). In light of Hung's attempted "refinement" of
the RHYTHM constraint's definition, the grids in (Qa-b) are identical in
the relevant respects to those in (Da-b), where she argues that the
descending-staircase configuration in (a) should be better than the one in
(b), because the good grid marks are "in _different_ columns" (p. 143).
The same different-columns argument, which I criticize above, would not be
possible, of course, for longer words, as (R) and (S) show:

(R) a. x (3) b. x
x x* x (2) x x x*
x x x x x x (1) x x x x x x

(S) a. x (3) b. x
x x x* x (2) x x x x*
x x x x x x x x (1) x x x x x x x x

In all, I agree with both points of Hung's conclusion regarding this
language's footing (p. 88): "While the case for Aguaruna having trochaic
solitary feet within a fundamentally iambic system is quite convincing, the
case for headless terminal feet may be less so."

Hung's discussion of Yidi~n (pp. 93ff) begins by showing that a heavy
syllable must be the head of a foot and that the only heavy syllables are
ones with _CV:_ shape; she then lists intervocalic sequences of _mb_ (p.
94). Only much later (p. 107) does Hung mention that nasal-stop sequences
have been analyzed as prenasalized-stop onsets. Contrary to this, however,
she shows the nasal in a separate syllable (and foot) from the stop:
(gul'am)(bar'a:) "march fly" (p. 97). This choice of syllabiification and
footing seems unnecessary, since nothing in her discussion of Yidi~n
prohibits placing _mb_ together: (gul'a)(mbar'a:). That is, far more
complex onsets (consisting of liquid, nasal and stop) are argued for (p.
107). (In many other instances in this chapter the _y_ of the
lamino-alveopalatal _dy_, _ny_ fails to be superscripted.)

Similarly, Yidi~n requires that in a PrWd with an odd number of
syllables all syllables but the last be footed (p. 109). In such cases
either the last syllable is weakly parsed or final segments are altogether
unparsed ; cf. (Kc) and (Ka-b), respectively, above. Hung also discusses
how a long vowel in an odd-numbered syllable in PrWd with an odd number of
syllables must shorten. She adds (fn. 8, p. 109) that this type of
shortening is PrWd-internal. This then begs the question: What happens to
_CV:C_ or _CV:_ syllables at the end of words with and odd number of
syllables? In one of the examples illustrating the preceding phenomenon,
the word /barganda:dyinyu/ "pass by"-antipassive-purposive is discussed.
(Both instances of _y_ should be superscripted.) The attested form is
_barg'anda<:>dy'i!ny<u>_; the final vowel and the internal mora are
unparsed. In her tableau (62), p. 110, the second foot is compared. The
two candidate forms are ...(dady'i:)... and *...(d'a:dyi)...; whichever
vowel is long is also stressed. There is one empirical problem with these
forms: Because the final vowel is unparsed, the final consonant _ny_
should also be in this foot, making the candidates ...(dady'i:ny)... and
*...(d'a:dyiny)..., respectively. (Following Hung's discussion of CV:C
syllables earlier in chapter 4, this final consonant is adjoined to
syllable: ...{{dy'i:}ny}...; whether the consonant is adjoined, however,
does not bear on the tableau.) More importantly, Hung might have indicated
that the final mora of the first candidate is epenthetic (to satisfy
END-RULE) and made use of angle brackets around the unparsed underlying
mora: ...(da<:>dy'i!ny)... and *...(d'a:dyiny)...; this would clarify
which mora is underlying and which is epenthetic.

Ojibwa (sec. 5.1, pp. 119-129) exhibits a phenomenon that potentially
challenges Hung's main proposal in the book: This unyieldingly iambic
language ranks RHYTHM relatively low in the hierarchy, resulting in final
stress if every parsed syllable is footed. That is, if no syllable is
weakly parsed, and each foot is an iamb, then there must be a final stress.
Strangely, if there is a (final) weakly parsed syllable, then it too
appears to bear (secondary) stress. Moreover, such a weakly parsed
syllable is invariably immediately preceded by a stressed syllable, as
exemplified in (Ta-d):

(T) a. [ ({mi}{z'i}) ({na}{h`i}) {g`an} ] "book"
b. [ ({n'i:}) ({ni}{m`i}) {z`i} ] "he is weak"
c. [ ({ni}{n'i}) {b`a} ] "I sleep"
d. [ ({bi}{m'o}) ({s`e:}) ] "we walk"

A long vowel makes a syllable heavy, as in the first syllable-foot of (Tb)
or the last syllable-foot of (Td). Consonants, with one exception
discussed below, do not contribute to syllable weight. Hung concentrates on
"the rather unexpected pattern of adjacent stresses in words which end in
an odd-numbered string of light syllables." This is exemplified by (Ta-c).
In addition, (Td) shows that an even-numbered sequence of light syllables
before a heavy syllable likewise produces consecutive stresses. Hung
observes that this phenomenon "goes directly against the very essence of
rhythmicity: not only does it create clash [= consecutive stressed
syllables/L.A.B.], but it also violates final stresslessness." Examples
(Ta-c) each also appear to have a stressed, unfooted light syllable - also
ruled out by the (nearly) universal requirement that feet be at least
bimoraic. The alternative analysis wouldn't be more favorable, resulting
in a final degenerate (= monomoraic) foot in both of (Tb-c). Hung proposes
that not all secondary stresses are really "stress" as such. Instead,
Ojibwa marks main stress and reduces the vowel in each non-head syllable of
a (disyllabic) foot. Evidence in support of this analysis is that it is
possible to reduce each vowel without a preceding stress mark (_'_ or _`_)
to schwa or zero; cf. (Wa-b) below. This reduction/apocope does not happen
in weakly parsed syllables. Thus, Hung argues, "a transcribed stress does
not necessarily imply that the vowel is the head of a foot, but rather that
it is simply not subject to reduction, for whatever reason." (I continue
to show "secondary stress" below with the understanding that these probably
represent the absence of reduction.) Final secondary stresses in examples
like (Ta-c) mean merely that "the vowel is full and not reducible for
independent reasons."

The "independent reasons" which Hung proposes are quite suspect and
challenge another of her main constraints. This time it's STRICT-PARSE
that is called into question. What I haven't mentioned is that the final
mora of a word's underlying representation is unparsed (in all but
disyllabic words). This phenomenon differs from the aforementioned
reduction of vowels in being categorical and non-gradient. For example, a
final short vowel or glide is dropped, as shown in (Ua-b), while a final
long vowel is shortened, as in (Uc).

(U) a. /ni-nagamo/ nin'ag`am "I sing"
b. /ni:nimizi-w/ n'i:nim`iz`i "he is weak"
c. /ni-niba:/ nin'ib`a "I sleep"

Hung's book shows many phenomena which serve to shorten words to allow a
particular optimal footing; this one is strange in that it takes a word
with potentially ideal parsing and reduce the syllable count. For example,
(Ua) could ideally form two iambs *[({ni}{n'a})({ga}{m`o}] but instead a
trisyllabic surface form is attested with a weakly parsed final syllable:
[({ni}{n'a}){g`am}]<o>. With (Uc) the footing could also be potentially
ideal: *[({ni}{m'i})({b`a:})]; instead, the unparsed final mora likewise
forces a weakly parsed syllable: [({ni}{m'i}){b`a}]<:>. (This phenomenon
also shortens some would-be weakly parsed forms into strictly parsed words.
For example, /bimose:-w/ "we walk" results in two feet from elision:
[({bi}{m'o})({s`e:})]<:> [= her (8b), p. 121].) To account for this
unusual phenomenon, Hung slightly modifies Prince & Smolensky's (1993:101)
solution to a similar phenomenon in Lardil, where certain words must not be
vowel-final (p. 122):

(27) _Free-Mora_:
Word-final moras must not be parsed.

Hung admits (fn. 3, p. 121) that this explanation for final deletion is
lacking in motivation. (Prince & Smolensky, incidentally, also mull over
this issue. They do show, however, for Lardil that their Free-V constraint
interacts with other constraints and is not just an unviolated stipulation.
Hung's FREE-MORA likewise interacts; it is violated if the resulting word
would become too small.) I do not wish to criticize this constraint,
however. Rather, I take issue with Hung's rationale for why the vowel in
the last syllable of (Ta-c) doesn't reduce: In (Ua) the surface final
consonant is not "linked" to a following syllable onset; a non-linked coda
preserves the vowel's quality. This argument is very questionable. Hung's
explanation of (Ub) is as follows (p. 125):

Let us suppose that the mysterious /w/ [in (Ub)] is
like a vowel in that it is underlyingly moraic, so
that word-finally, it provides the mora for the
apocope rule. On the other hand, it is like a
consonant [as in (Ua)] in that it protects a final
vowel from weakening. [...] Thus despite the fact
that the final syllable is phonetically open, it
appears that it is phonologically closed.

To account for (Uc), Hung suggests that "the vowel is linked to two
moras" despite the final mora being phonetically absent. What Hung appears
to be proposing for (Uc) is some sort of abstract bimoraicity: The mora is
abstractly present but is segmentally absent. This is the flipside of weak
parsing: mora segmentally present but absent for abstract prosodic
purposes. (Hung's account is reminiscent of Kiparsky 1992, which posits
that some un-segmented material can still be considered in the prosody.
Hung discusses Kiparsky's approach elsewhere - fn. 2, p. 7.) Suffice it to
say, Hung's explanation for (Ua-c) is dubious; I suggest the following
approach instead: Following-up an idea I mentioned above in connection
with Aguaruna, it might be that a vowel is "weak" only if it's the non-head
of a foot. This results in the apparent secondary stressing of a final
weakly parsed syllable. A weakly parsed syllable - like the initial
syllable of a weakly parsed foot in Aguaruna - is neither weak nor strong.
Because there is no perceptible secondary stress as such, such a syllable
has apparent secondary stress. My suggestion also accounts for the
consecutive stressed syllables in (Td).

Regardless of my counterproposal, there appears to be one bit of
evidence against the weakly-parsed-syllable analysis exemplified in (Ta-c)
above: Hung mentions briefly (fn. 5, p. 126) that main stress is
apparently assigned to the antepenultimate unreduced vowel; if the word is
too short, then the first of these bears main stress. Hung's data include
the following types of examples with at least three unreduced vowels (_syl_
and _SYL_ indicate a syllable with a light and heavy vowel, respectively):

(V) a. [(syl 'syl) (syl `syl) (syl `syl)]
b. [('SYL) (`SYL) (`SYL)]
c. [(syl `syl) (syl 'syl) (`SYL) `syl]
d. [(syl `SYL) ('SYL) (syl `syl) `syl]

The three-foot examples in (Va-b) both stress the initial foot, while
(Vc-d) each have an additional weakly parsed syllable and stress the second
foot. Hung might have pursued this issue further, especially because it
casts doubt on her weak-parsing analysis of Ojibwa, but also because RHYTHM
does not appear to be the factor deciding between first- and second-foot
main stress in these examples.

One more unexplained feature of vowel-reduction in Hung's treatment of
Ojibwa is the non-uniform occurrence of reduction. She lists forms which
fail to reduce the non-head syllable in every foot:

(W) a. b.
Unreduced nig'i:nam`ad`ap nig'i:nam`adab`im`in
Schwa-reduced n@g'i:n@m`ad`ap n@g'i:n@m`ad@b`im`in
Deleted n_g'i:__m`ad`ap n_g'i:n_m`ad_b`im`in

No glosses are given for these.[ = her (11b-c), p. 122]. I've added
underscores to show where segments are missing in the "Deleted" row of (W).
The unexplained factor is that whereas every consonant in (Wb) is
maintained in all three columns, in (Wa) the _n_ after _i:_ is dropped.
Hung lists other examples as well with unexplained deletion of one weak
vowel but non-reduction of another: /tatanisi-w/ _ttanisi_ "he stays for a
while" [= her (14), p. 122]. Assuming this word is footed into two iambs,
[({ta}{t'a})({ni}{s`i})]<w>, then why isn't the third vowel reduced?
Hung's only hint (fn. 4, p. 122): "Kager (1993:427) argues that one must
resist the temptation of characterizing this weakening effect as a purely
iambic phenomenon."

Despite being riddled with violations, RHYTHM emerges in two ways in
Ojibwa: First, as (Ta-c) and (Vc-d) show, when some syllable must be
weakly parsed, it is the last syllable. This is especially apparent in a
word with three light syllables: [({min}{d'i}){d`o}] "he is big". As far
as STRICT-PARSE is concerned, the first syllable could be weakly parsed;
this would result, unnecessarily, in PrWd-final stress:
*[{m`in}({di}{d'o})]. Another, more unique, way in which RHYTHM emerges is
in words with /CVNCV/ shape (where N = nasal): Whereas /CVCV/ words are
invariably parsed as [({CV}{C'V})] iambs, forms in /CVNCV/ have initial
main stress and no reduced second vowel. Hung proposes that, whereas
consonants aren't usually moraic, nasals can violate Ito & Mester's (1993)
_Mora Sonority Threshold_ in order to achieve non-final stress:
[({g'on}){d`a}] "these"-animate, instead of *[({gon}{d'a})]. Showing how a
constraint functions even a little bit despite frequent violation is one
sign of a good OT work and a credit of this framework overall.

8. Constraints not properly defined:

Hung is quite rigorous about defining her OT constraints, often
repeatedly (with clear labels such as "first statement" and "final
statement") and with ample citations to adopted/adapted constraints'
sources. In a few exceptions, however, this is not the case:

Aside from glides, Axininca Campa allows certain nasals to appear in
the coda of a syllable (p. 60). Hung adopts the following constraint:

(12) _Coda-Constraint_ (McCarthy & Prince 1993a:27)
A coda consonant is a nasal homoorganic to a
following stop or affricate.

Hung observes that the lack of a following consonant rules out PrWd-final
nasals. Then, in section 3.1.3, she goes through each type of heavy
syllable: CVC, CVV, CVG (where G = "glide"). In the CVC subsection she
repeats this constraint but does not really address PrWd-final nasals.
That is, a final long vowel shortens, a final glide remains unchanged, and
a final obstruent picks up a following epenthetic vowel (forming a new
syllable with it), but what actually happens with a final nasal (or do such
cases ever arise)?

Two new constraints are used in tableaux (74) and (75), p. 114:
PARSE-F (where F = "features"), is discussed briefly; FILL-ROOT is not
defined at all. (Incidentally, in section 4.4, pp. 112-116, on vowel-zero
alternations in Yidi~n, Hung uses the word "root" several times with two
different meanings: "an unaffixed word morpheme" and "root node".)
Presumably FILL-ROOT means "don't epenthesize a root node (= segment)."

Another constraint similar to the more common requirement (that feet
be at least bimoraic) is one in trochaic languages which requires feet to
be "balanced" (pp. 137ff). Here again, this constraint is discussed in
passing in a few places; a proper definition can only be pieced together
from several different mentions of it: "the notion of the left-headed
'balanced' foot" (p. 137), "F=MM, the requirement that trochaic feet be
even" (p. 138), "the importance of the bimoraic (= moraic = even) trochee"
(p. 140), "demands that a trochee be even, is referred to by Prince and
Smolensky as _Rhythmic Harmony_ (1993:590[)]; its effect there is to rule
out HL trochees" (p. 147). From these excerpts it appears that this
constraint allows either a heavy syllable or two light ones, ruling out
heavy-light feet. It also appears from this constraint's use in one
tableau that this constraint also rules out light-heavy trochees; cf. my
discussion of (E'b) above.

While discussing footing in Classical Latin (p. 145), Hung introduces
the EDGEMOST constraint with neither definition nor references to other
uses of this constraint in the literature:

"That we get antepenultimate stress at all indicates that RHYTHM is
also obviously more important than EDGEMOST.

|(29) | | RHYTHM || EDGEMOST |
|--> | (sp'atu)la | || * |
| | spa(t'ula) | *! || |

(As I point out in the next section, the RHYTHM cells of this tableau
should have one and two asterisks, not zero and one, respectively.)
Although EDGEMOST is a relatively widely used constraint in the OT
literature, it should still be defined clearly. Another problem with this
constraint's introduction is that while this tableau establishes the
dominance of RHYTHM over EDGEMOST, the data used in this tableau do not
actually justify the use of EDGEMOST in the first place. RHYTHM alone
justifies the attested footing, with the descending-staircase metrical grid
in (B) above. Hung should have used data like (Xa-b) to make this point.
As these grids show, these footings violate RHYTHM equally (one bad grid
mark each); the only constraint ruling out (Xb) is EDGEMOST, which places
the foot further from the right edge of the PrWd than in (Xb).

(X) Possible footings and grids of /patricia/:

a. x (3) b. x
x* x (2) x* x
x x x x (1) x x x x
[{pa}({tr'i}{ci}){a}] *[({p'a}{tri}){ci}{a}]

c. x (3)
x* x* (2)
x x x x (1)

d. x (3) e. x
x* x* x* x* (2) x* x* x* x*
x x x x (1) x x x x
*[({p'a}{tri})({ci}{a})] *[({pa}{tri})({c'i}{a})]

f. x (4) g. x*
x x (3) x x
x* x* x* x* (2) x* x* x* x*
x x x x (1) x x x x
*[({p'a}{tri})({ci}{a})] *[({pa}{tri})({c'i}{a})]

The other possible footings of this word, in (Xc-g), illustrate another
point which Hung does not emphasize [(Xa) = her (21a); (Xf) = her (27b)]:
The fewer feet are formed, the fewer violations there are of RHYTHM. As
long as there is only one stressed foot, then minimal-word requirements are
met. Any exhaustive footing of this word - be it with two stressed feet,
(Xf-g), or with one headless and one stressed foot, (Xd-e) - results in at
least four bad grid marks. Among the structures with just a single foot,
(Xb-c), it is RHYTHM which forces the foot to be non-final; (Xa-b) each
have fewer bad grid marks than (Xc) has. Thus, the nearly absolute
dominance of the RHYTHM constraint in Latin prevents any more than one foot
from being formed. (Unfortunately, Hung does not discuss whether Latin has
secondary stress; two pre-classical examples, _vo.lup.t'aa.tees_ and
_ve.ree.b'aa.mi.ni_ (fn. 9, p. 147), the only words long enough to have
secondary stress, are shown only with main stress. I know, for example,
that in Macedonian, which also has antepenultimate stress, there is no
perceptible preceding secondary stress.) Additionally, in Hung's (27a),
repeated below as-is, should be corrected as in (Xg) above. That is,
neither level-3 grid mark in Hung's grid should have an asterisk, but the
level-4 grid mark should have an asterisk. This error does not, therefore,
alter the total number of bad marks in this grid.

(27) x
x* x
x* x* x* x*
x x x x
a. (pa tri) (c'i a)

9. Other terms that could have been clarified better:

It could have been made clearer that adjunction involves skipping a
potential mother node in the prosodic tree (pp. 19ff).

Hung also uses "stress" and "head" (of a foot) interchangeably (pp.
35ff). This is not immediately clear. A foot that is weakly parsed
(violating STRICT-PARSE) is one with no stressed syllable (= with no head).
Such a foot is segmentally present but prosodically out of consideration.

In her discussion of Yidi~n (pp. 98ff) Hung makes very clear the
distinction between violations of PARSE (segmentally absent) and of
STRICT-PARSE (segmentally present, but not counting prosodically). In the
subsequent diagrams, however, some of the data can be confusing as to
whether material is completely unparsed (deleted) or merely weakly parsed.
In some places she uses only foot boundaries, _(_ and _)_, while in others
she uses all three prosodic-category-edge markers, as in /gudaga/
*[({gu}{dag})]a "dog"-absolutive [= her (28), p. 101]. In this example it
is possible to determine that the final vowel is completely unparsed only
by knowing that sub-syllabic items must adjoin to a syllable; it would be
far clearer to just place the vowel in angle brackets: *[({gu}{dag})]<a>.
Hung uses this notation elsewhere in the chapter, but not consistently.
Every use of angle brackets indicates completely non-realized (=
unpronounced) material.

Hung also oversimplifies the following tableau, on Classical Latin (p.

|(28) | | RHYTHM || STRICT-PARSE |
|--> | pa(tr'ici)a | || * |
| | (patri)(c'ia) | *! || |

The two candidates correspond to the metrical grids below in (Xa, e),
respectively. As those grids show, these two forms actually have one and
four violations of RHYTHM, respectively. The same oversimplification takes
place in the STRICT-PARSE column of asterisks as well; the upper (attested)
candidate should show two violations (corresponding to each unfooted
syllable), while the lower candidate should show one violation (the
headless foot). Hung is apparently using single asterisks to mean "more
violations" rather than showing the specific number thereof. With this
tableau she's establishing that RHYTHM dominates STRICT-PARSE, as indicated
by the double line between the constraints' columns. (She continues this
representation in the RHYTHM column of her tableau (29), repeated below my
re-discussion of EDGEMOST, as well as in all four of the tableaux for
Pre-Classical Latin at the end of chapter 6.) Such simplification of
asterisks is potentially confusing to readers unfamiliar with this
technique, especially since it is adopted without explanation. As I show
as well immediately in the preceding section with regard to Hung's
treatment of Pre-Classical Latin, this simplification technique can also
lead to sloppy argumentation.

10. Command of overall issues of OT:

Hung takes the reader through many of the issues in the framework
itself. For example, she argues that constraint-family dispersion - making
many little constraints instead of using one (p. 23); for example,
PARSE-Vowel, PARSE-Consonant, PARSE-Syllable, etc. - is to be avoided.
Hung sticks to this philosophy rather carefully in chapter 4 (on Yidi~n),
positing special PARSE-C and PARSE-F[eatures] and separate STRICT-PARSE
versions for consonants and syllables only when unavoidable.
Unfortunately, Hung is not so careful with FILL-type constraints; she
proposes the following very specialized constraint (p. 105):

(39) FILL-NUC:
No empty moras.

She also uses FILL-ROOT (undefined, p.114; presumably: No empty root
nodes). Why not just "FILL: No empty material" (of any kind) in either of
these cases? Incidentally, Hung claims that PARSE-Consonant is unviolated
in Yidi~n; see a counterexample in (Kb) above.

Hung devotes less than a page to Ulwa (sec., p. 91).
Interestingly, words consisting of two light syllables can be either iambic
or trochaic. She doesn't pursue this issue further. Is such an option
subject to whether the next word has initial stress (as in English
_nineTEEN_, *_NINEteen_, but _NINEteen NINEty_)? Another apparent
optionality is in Araucanian (sec. 5.3, pp. 131-133), where "disyllabic
content words ending in a vowel may be stressed on either syllable":
_ruk'a_ and _r'uka_ "house". If these are completely optional phenomena,
then there may be equally-ranked constraints at work. This aspect of OT,
variation, is avoided in Hung's book. Tied constraints have been a problem
for OT and avoided by many of the framework's adherents. (This issue,
dubbed "variation", has recently been discussed on the OT e-mail list

11. Limiting the focus of the work:

Hung is very clear about what her book is intended to cover (p. 11):
"the focus here is on the nature of lexical stress, rather than phrasal
stress." By "lexical" Hung appears to mean "stress assigned to a lexical
word" and not "stress lexically encoded on a particular syllable of a
word" (which she also deals with in places, but doesn't focus on). Much of
the books theoretical punch comes from uniting word-internal and -final
effects. Heretofore the internal effects were known as Rhythm, Clash
Avoidance, Perfect Grid; at the end of the word the phenomenon was known as
Extrametricality, Extraprosodicity and Nonfinality. Hung's RHYTHM unifies
the avoidance of stress clash (consecutive syllables with stress) and the
nonfinality of stress.

It is understandable that Hung does not deal with other, word-internal
issues of the perfect-grid, such as constraints to keep rhythmic beats from
being too far apart. (Hung does, however, devote a subsection (3.1.4, pp.
77-80) to internal rhythmic effects within her larger investigation of
Axininca Campa.) Failing, however, to pursue the relevant constraints
thoroughly beyond edge phenomena to the inside the word - short of a few
brief mentions (pp. 12, 23) of the by now classic alternation of Cayuvava,
which stresses every third syllable - is a missed opportunity. The
following is Hung's longest discussion of this relatively unique phenomenon
(fn. 7, pp. 142-143):

Ternary parsing effects like Cayuvava (Hayes 1991,
Hammond 1992) may also be viewed as having a three-
layer rhythmic grid. [R]hythmicity is maintained at
the expense of STRICT-PARSE throughout the word, so
it is optimal to have weakly parsed syllables
between feet. What happens inside the Cayuvava word
then, is no different from what happens at the end
of the Latin word; this is also Hammond's view.

This idea is intriguing, but not followed-up with even the simplest of
examples. True, Cayuvava behaves identically to Latin (if there is a final
sequence of light syllables): [(SYL.syl)syl], where _SYL_ just means
"stressed (light) syllable". Unlike Latin, however, Cayuvava sets up other
stressed feet prior to the antepenult in ternary alternation:
[(SYL.syl)syl(SYL.syl)syl]. As I show above in my discussion of RHYTHM in
Latin, if RHYTHM truly dominates STRICT-PARSE, then (assuming that words
need at least one stressed foot due to minimal-word considerations) there
should be no more than a single foot, no matter how long or short the word
is. Recall that, in Hung's three-tier model of antepenultimate-stress
grids, _every_ foot has at least one bad grid mark; cf. the Latin grids in
(C), (E), (F), (D') and (X) above. In a string of, say, six or seven
(light) syllables, it's better to just weakly parse all but the footed
antepenult and penult: [syl.syl.syl('syl.syl)syl],
[syl.syl.syl.syl('syl.syl)syl]. In fact, no mechanism in Hung's constraint
system can prevent this. If STRICT-PARSE dominates RHYTHM, however, then
the result would just be exhaustive footing if possible,
[('syl.syl)('syl.syl)('syl.syl)], with loose syllables perhaps only if
there's one syllable left over, where RHYTHM would presumably emerge:
[('syl.syl)('syl.syl)('syl.syl)syl]. That is, I'm afraid that Hung's
constraints alone are insufficient for Cayuvava.

While main word stress occasionally arises in the discussion (see,
e.g., pp. 67ff, p. 144, 149), there is no real attempt to apply the RHYTHM
constraint to determination of main stress. The only thing considered is
the location of the final stress and any possible clashes with an
immediately preceding stress. While the book's title makes clear that it
is the _edge_ that is being investigated, it might have been worthwhile to
see how Hung's RHYTHM constraint affects placement of main stress. (See
also my discussion of Hung's treatment of Ojibwa secondary vs. primary
stress, above.)

12. Foot-structure typology

Citing Prince (1990) and others, Hung lists the types of foot attested
in human languages (p. 29):

(Y) a. Syllabic trochee ('Syl Syl)
b. Moraic trochee ('syl syl), ('SYL)
c. Iambic foot (syl 'syl), (syl 'SYL), ('SYL)

Note that (Ya) is not sensitive to syllabic weight, while (Yb-c) are
weight-sensitive. Hung deals primarily with iambic systems, as in (Yc).
In quite a few of her case studies, otherwise-iambic languages can result
in trochaic feet (in which the first or only syllable in the foot is
stressed) in order to avoid having PrWd-final stress. In one instance (fn.
1, p. 135) Hung refers to a trochaic language, Western Aranda, that results
in an iamb under certain conditions (apparently not related to finality of
stress): whether or not the would-be head syllable lacks an onset.
(Unfortunately, Hung does not cite any sources for this in that footnote;
apparently she has Davis 1988 in mind. I have uncovered three more sources
that appear to discuss this phenomenon: Davis 1985a-b and Endo 1987.)
Disappointingly, Hung stops short of proposing an OT mechanism for why some
languages are trochaic and others are iambic. She adopts constraints which
imply - but stop short of claiming overtly - that, other things being
equal, languages should be iambic (respectively: pp. 30, 30, 46):

A heavy syllable is stressed.

If there is a head, it is on the right.

In a rhythmic unit (W S), S is heavy.

(In Hung's notation "stressed" = "head of a foot"; W = weak, S = strong.)
It would seem that Hung's main proposal would be strengthened greatly if it
could be shown that these three constraints are overridden by RHYTHM (and
possibly other constraints), resulting in a language with invariably
trochaic footing. That is, while FOOT-FORM requires every language to be
iambic, it is RHYTHM (or possibly other constraints) which make the
language trochaic. Hung implies that the presumption is iambic unless
there is positive trochaic evidence in the following discussion of Yidi~n
(p. 93): "while the question Hayes [1992] and Kirchner [1993] ask is, why
aren't feet always trochaic, the question I ask is, why aren't feet always
iambic?" Hung returns to this issue again at length (p. 95):

Regarding headedness, there are basically two views:
(1) feet are iambic unless X compels trochaicity,
and (2) feet are trochaic unless Y compels
iambicity. The first view is the one presented
here [...]. For me, the X which compels trochaicity
is RHYTHM. [...] For Kirchner (p. 17), the Y which
compels iambicity is the principle of WEIGHT-TO-
STRESS. Hayes [...] proposes that _Trochaic Defult_
applies when it does because it is not contradicted
by _Iambic Uniformity_; more specifically, _Trochaic
Default_ applies when there is no canonical [light-
heavy] iamb [...]. Thus for Hayes (1992) the X
which compels iambicity is the presence of the
canonical iamb.

(There seems to be a typo; Hung probably means "... the _Y_ which compels
..." in the last sentence.) Hung then makes the following conclusion,
which strongly suggests that she allows for two separate headedness
constraints in the universal inventory, but a language can choose only one
or the other (p. 96): "I assume therefore that in Yidi~n, the statement of
FT-FORM [... Fn. 1:] Unlike Kirchner, I assume there is only one
statement of FT-FORM, not two. Either the statement makes reference to the
right or the left, but not both. The choice may ultimately be a matter of
parametrization." Hung returns to this thread in the opening page of
chapter 6, on trochaic languages (p. 135): "under the appropriate
conditions, RHYTHM can interact with _Iambic_ FT-FORM. We have seen many
cases where a trochaic parse has been deemed optimal given the rhythmic
situation. This is not true of _Trochaic_ FT-FORM." Her use of italic
_Iambic_ and _Trochaic_ seems to be a reference back to p. 96, fn. 1.
Strangely, in none of the tableaux or argumentation does it appear
necessary to use the trochaic version of FT-FORM. Why not just come out
and claim that iambs are produced universally unless overridden by RHYTHM?

The Yidi~n language apparently requires each foot to be disyllabic,
not just bimoraic (p. 94), as is the case with many other languages
discussed in the book. It seems worth pointing out that this type of foot
binarity is typical of trochaic systems. This fact potentially weakens
Hung's claim that this language is essentially iambic and strengthens
Hayes's (1992) and Kirchner's (1993) view about Yidi~n's underlying

One final detail having to do with foot-structure type is Hung's
mechanism for accounting for antepenultimate-stress trochaic systems.
There is an intermediate grid level for _footed_ syllables, shown in (B)
above. Hung adds that such a mechanism has no rhythmic advantage in iambic
systems (fn. 2, p. 136): "Since the iambic foot-head is on the right, it
will never be followed by a non-head" as in (B). I agree that there is no
_rhythmic_ advantage. Still, in at least one other place in the book Hung
points out that the initial syllable of the main-stressed iamb is exempt
from various weakening effects: As I mention above, Aguaruna reduces
vowels in non-head syllables of every foot except the foot bearing main
stress; see (L) above. Hung's constraint (p. 141, fn. 5) "which says that
a footed syllab[l]e is stronger than an unfooted syllable" might be at work
here. Recall that it is the first (iambic) foot that bears stress in
Aguaruna; the unreduced syllable is thus always word-initial. The
rhythmically expensive use of an intermediate grid level might be motivated
by some (initial) END-RULE constraint; cf. pp. 105, 124 for discussion of
such a mechanism. Other feet don't get this extra grid level, because they
don't help (initial) END-RULE in any way.

13. Typos:

Aside from the typographic errors mentioned above so far, I point out
several others, with possible corrections:

"Hammond (1990)" apparently should be Hammond (1992) in p. 12.
In tableau (55), p. 72, the second and fourth candidates should begin
with _(_ and not _{_.
In tableau (63), p. 74, the constraint labels appear to have been
switched (i.e., the column before the double line should have the label
FILL-NUC, while the column after the double line should be marked CODA-C.
Hung is describing the footing of various numbers of consecutive light
syllables in Axininca Campa (80): "it appears that ALIGN-LEFT is
unviolated, even in the case where the initial foot is adjoined."
Unfortunately, in her schematic representation (of a heavy syllable
preceded by various syllables), Hung omits the first square bracket at the
left edge of the second row. The left-hand column in (Z) is Hung's (86),
while the one on the right is my correction thereof:

(Z) a. [(LH)... [(LH)...
b. (LL)[(H)... [(LL)[(H)...
c. [(LL)(LH)... [(LL)(LH)...
d. [(LL)L(LH)... [(LL)L(LH)...
e. [(LL)(LL)(LH)... [(LL)(LL)(LH)...

A subsequent example, _(kima)(n'iNta)(k`ina)_ "estoy lleno" (fn. 16, p.
83), fails to show PrWd square brackets. In keeping with the preceding
schematic, this word's initial headless foot is likely to have been
adjoined to PrWd: [(kima)[(n'iNta)(k`ina)]].
In fn. 14, p. 80, there's a pronoun _he_ with two potential
antecedents: "J. Payne 1990" and "Hayes 1991". Hung is apparently
referring to [Bruce] Hayes, not to J[udy] Payne. (Thanks to Sandy Peavy
for helping to clarify this.)
In the following example, meaning "dog" in Yidi~n, the penult syllable
was changed from _da_ to *_ga_ in the lower line (p. 108):

(52) CV [-son] V# ex. /gudaga/ gud'a!ga!}Syl* ga|Stem

The second row of this example should read as follows: "...da!}Syl* ga|Stem".
Just a page after stating that in Ojibwa "the vowel in the final
syllable of the _surface_ representation is _always_ stressed" (p. 120, her
emphasis), the following apparently inaccurate example appears: /minogi-w/
min'ogi "he is growing well". Based on discussion elsewhere in section 5.1
it appears the final vowel should bear "secondary" stress.
Hung briefly discusses Prince's (1983) grid-based analysis of Hawaiian
(p. 124). One of the steps in that analysis is "Perfect Grid (RL)", where
_RL_ means "right to left". In her example (17), however, this step is
mistakenly labeled as "PG (LR)"; this should be "PG (RL)".
On p. 131 Hung refers to example "(4)"; the intended reference appears
to be "(38)".

14. References:

The following includes any works mentioned above as well as the
following: certain works that have appeared since the version cited in
Hung's bibliography (shown with Hung's identifier, if it differs, in square
brackets). Many of the manuscripts (and even published items, including
Hung's 1994 dissertation itself) cited in the book's bibliography are
available on the Rutgers Optimality Archive
<>; those titles are shown below with
their ROA numbers. Additionally, rather than list the numerous OT works on
rhythmic and related phenomena, I refer the interested reader to the
convenient "search" mode on the ROA web site.

Burzio, Luigi ([in press, a]1994) _Principles of English Stress._
Cambridge, England: Cambridge University Press.
Crowhurst, Megan & Mark Hewitt (1995) "Prosodic Overlay and Headless
Feet in Yidi~n." _Phonology_ 12:1, 39-84.
Davis, Stuart (1985a) "Syllable weight in some Australian languages."
In M. Niepokuj, M. VanClay, V. Nikiforidou, & D. Feder (eds.).
_Proceedings of the eleventh annual meeting of the Berkeley Linguistics
Society: February 16-18, 1985._ Berkeley: Berkeley Linguistics Society,
Davis, Stuart (1985b) _Topics in syllable geometry._ University of
Arizona dissertation.
Davis, Stuart (1988) "Syllable onsets as a factor in stress rules."
_Phonology_ 5, 1-19.
Endo, Yuichi (1987) "A metrical analysis of onset-sensitive stress
languages: Concerning syllable internal structure." _Journal of the
English Institute_ [Aoba-ku, Sendai, Japan] 16, 57-88.
Dyck, Carrie (1993) "On the prosodic definition of 'word' in
polysynthetic languages." Paper presented at the CLA, Carleton University,
May, 1993.
Green, Thomas (forthcoming) _Configurational constraints in a
generalized theory of phonological representation._ Massachusetts
Institute of Technology dissertation.
Hammond ([1992]1995) "Deriving ternarity." In C.M. Fitzgerald & A.
Heiberg (eds.). _Coyote Papers. Working Papers in Linguistics from A-Z_
9, 39-58.
Hayes, Bruce (1982a) "Extrametricality and English stress."
_Linguistic inquiry_ 13:2, 227-276.
Hayes, Bruce (1992) "Iambs in Yidi~n." Ms., University of
California, Los Angeles.
Hayes, Bruce ([1991]1995) _Metrical Stress Theory: Principles and
Case Studies._ Chicago : University of Chicago Press.
Hung, Henrietta J. ([1993b]1994a) "Iambicity, rhythm, and
non-parsing." ROA-9.
Hung, Henrietta J. (1994b) "Iambicity, rhythm, and weak parsing."
_Cahiers Linguistiques d'Ottawa_ 21, 78-102.
Ito, Junko & Armin Mester (1992) "Weak layering and word binarity."
Ms., University of California, Santa Cruz.
Ito, Junko & Armin Mester (1993) "Licensed segments and safe paths."
_Canadian journal of linguistics_ 38, 127-214. (= C. Paradis & D. Charit'e
(eds.). _Constraints, violations and repairs in phonology.)
Ito, Junko, Armin Mester, & Jaye Padgett ([1993]1994) "NC: Licencing
and underspecification in Optimality Theory." ROA-38.
Kager, Ren'e (1993) "Alternatives to the Iambic-Trochaic Law."
_Natural language and linguistic theory_ 11, 381-432.
Kiparsky, Paul (1992) "Catalexis." Ms., Stanford University and
Wissenschaftskolleg zu Berlin.
Kirchner, Robert (1993) "Optimizing Yidi~n phonology." Ms.,
University of California, Los Angeles.
Legendre, G'eraldine (1997) "Clitics, verb (non-)movement, and
Optimality in Bulgarian." ROA-165.
L"ohken, Sylvia C. (1996) _Deutsche Wortprosodie: Abw"achungs- und
Tilgungsvorg"ange._ Technische Universit"at zu Berlin dissertation;
published by Stauffenburg-Verlag, T"ubingen. (= Studien zur deutschen
Grammatik, 56.)
McCarthy, John & Alan Prince (1993a) _Prosodic morphology I:
Constraint interaction and prosodic satisfaction." Ms., University of
Massachusetts, Amherst, and Rutgers University, New Brunswick. (= TR-3
Rutgers Center for Cognitive Science, Piscataway, New Jersey.) To appear
from MIT Press.
McCarthy, John & Alan Prince ([1993b]1994) "Generalized alignment."
In G. Booij & J. van Marle (eds.). _Yearbook of morphology, 1993._
Dordrecht: Kluwer, 79-153. [Also available as ROA-7.]
Mester, Armin (1994) "The quantitative trochee in Latin." _Natuaral
language and linguistic theory_ 12, 1-62.
Mithun, Marianne & Reginald Henry (1982) _Watewayestanih: A Cayuga
Teaching Grammar. Brantford, Ontario, Canada: Woodland Indian Cultural
Educational Centre.
Payne, J. (1990) "Asheninca stress patterns." In Doris Payne (ed.)
_Amazonian linguistics: Studies in lowland South American languages._
Austin: University of Texas Press, 185-209.
Prince, Alan (1983) "Relating to the grid." _Linguistic inquiry_
14:1, 19-100.
Prince, Alan (1990) "Quanitative consequences of rhythmic
organization." In M. Ziolkowski, M. Noske, & K. Deaton (eds.) _The
parasession on the syllable in phonetics and phonology._ (= Papers from
the 26th regional meeting, 2.) Chicago: Chicago Linguistic Society,
Prince, Alan & Paul Smolensky (1993) _Optimality Theory: Constraint
interaction in generative grammar._ Ms., Rutgers University, New
Brunswick, and Universiy of Colorado, Boulder. (= TR-2 Rutgers Center for
Cognitive Science, Piscataway, New Jersey.) To appear from MIT Press,
Cambridge, MA.

15. Short biography of the reviewer:

I am interested primarily the interfaces of syntax with other grammar
components (phonology, morphology, animacy). I find Optimality Theory to
be an ideal model for assessing such interfaces, because the components
interact, as it were, only inasmuch as optimal forms are selected that
favor certain grammar components over others. My training and continuing
interest is in the Slavic languages. My dissertation is titled
_Approximation in Russian and the single-word constraint_ (Princeton Univ.,
1995). I'm currently interested in Macedonian, which exhibits
antepenultimate stress (hence my interest in Hung's topic) and, with
Bulgarian, has extensive nominal- and verbal-clitic systems.

(I'm grateful to Matthew Baerman for comments on an earlier draft of
this review. Any errors that remain are my own doing.)


Amazon Store: