Benua, Laura (2000) Phonological Relations Between Words.
Garland Publishing, hardback ISBN: 0-8153-3810-4, x+271 pp.,
Outstanding Dissertations in Linguistics, US$52.50, GBP35.00
Antony Dubach Green, University of Potsdam
The publisher's announcement of this book can be found in
http://linguistlist.org/issues/11/11-1144.html#1; however,
the price and page numbers are different from those listed
above. The prices listed above are from the website of
Routledge, the distributor. The book is a revised version
of the author's 1997 UMass-Amherst dissertation,
"Transderivational Identity: Phonological Relations Between
Words".
According to the abstract, the main hypothesis of Benua's
dissertation is that "morphologically related words are
required to be phonologically identical by ranked and
violable constraints" (p. ix). This hypothesis is tested
by examples of phonological under- and overapplication,
i.e. by instances where the requirement for morphologically
related forms to be identical requires phonological
processes to apply in places where they are not expected
(overapplication), or to fail to apply in places where they
are expected (underapplication). B approaches these
phenomena from the point of view of a fully parallel,
nonserial variety of Optimality Theory.
Although not mentioned in the abstract, a major new
proposal is that of recursive evaluation of paradigms:
according to this model, the harmonic evaluation of a base
word takes precedence over the harmonic evaluation of a
derived word, but the constraint ranking within each
evaluation remains the same. So, for example, the stress
patterns of "origin, original, originality" are subject to
the same constraint ranking, but the entire tableau for
"origin" outranks that of "original" which in turn outranks
that of "originality".
The dissertation consists of six chapters. Chapter 1,
"Phonological relations between words", is a general
overview in which the theoretical tools used in the
dissertation are introduced. B assumes the Item-and-
Arrangement model of morphology, in which morphologically
complex words are derived from the Underlying
Representations of their bases plus affixation. She does
not defend this model or compare it with other
morphological models, but seems to take it for granted.
The relationship between morphology and phonology is
governed by the Correspondence Theory manifestation of
Optimality Theory, which B discusses in detail in this
chapter. Crucial for her is the difference between Input-
Output faithfulness, which compares a surface form with its
Underlying Representation, and Out-put-Output faithfulness,
which compares two related surface forms. Throughout the
dissertation, B shows that when Out-put-Output faithfulness
outranks Input-Output faithfulness in the context of
recursive evaluation of paradigms, phonological processes
apply as expected in base words but either overapply or
underapply in derived forms in order to achieve identity
between the derived forms and their bases.
Chapter 2 explores more deeply the novel proposal of the
dissertation: transderivational correspondence. Central
to Transderivational Correspondence Theory (TCT) is the
idea that surface forms of related words are compared
against Output-Output faithfulness constraints such as OO-
MAX (no deletion), OO-DEP (no insertion), and OO-IDENT[F]
(no changing the value of the feature F). Because of
recursive evaluation, according to which harmony in a base
form takes precedence over harmony in a derived form,
Output-Output faithfulness can be achieved either by
violating in the derived form a phonotactic constraint
which is not violated in the base form (underapplication of
a phonological process) or by using in the derived form a
marked segment in a context where it would be unmotivated
in the base form (overapplication of a phonological
process). This chapter also provides a foretaste of the
case studies explored in detail in chapters 3 through 5.
Chapter 3, "Sundanese", is an in-depth analysis of an
instance of phonological overapplication resulting in
transderivational faithfulness, and Chapter 4, "Tiberian
Hebrew", is a case study of a language with
underapplication of phonological processes for the sake of
transderivational identity. Unfortunately I do not have
room to summarize these chapters in this review and still
give sufficient attention to the chapter on English, which
in my opinion in the most interesting of the three.
Chapter 5, "English" is a discussion of a wide variety of
effects in English (some dialect-specific and some cross-
dialectal) demonstrating the different behavior of class 1
and class 2 affixes. The first effect discussed is stress:
class 1 suffixes cause stress shift ('origin, o'riginal),
class 2 suffixes do not ('obvious, 'obviousness). B's
analysis is that there are two types of OO-faithfulness
constraints in English: OO1 (governing faithfulness
between words with class 1 affixation and their bases) and
OO2 (governing faithfulness between words with class 2
affixation and their bases). OO2-faithfulness then
outranks the constraints governing stress placement, but
OO1-faithfulness constraints are dominated by the stress
constraints. Also, while underived words of five or more
syllables show the "initial dactyl" effect in secondary
stress placement (e.g. ,Tatama'gouchee), words of this
length with class 1 suffixation will displace secondary
stress to the right in order to match the primary stress of
their bases: o,rigi'nality (cf. o'riginal), a,risto'cratic
(cf. a'ristocrat), etc. B's analysis is that OO1
faithfulness constraints outrank the ALIGN-L constraint
that forces word-initial secondary stress in longer words.
The second issue addressed in this chapter is "closure
effects", i.e. cases where words with class 2 suffixation
are so faithful to their bases it led derivationalists to
assume class 2 suffixes were added after the base word had
been completely derived. The effects in this section are
dialect-specific; B gives examples from New York
City/Philadelphia English, London Vernacular English,
Northern Irish English, Scottish English, and Adelaide
(Australian) English. In each of these dialects there is
vowel allophony between open and closed syllables or
between word-final and non-word-final position, but the
class 2 suffixes (inflectional -s, -ed, etc.) are invisible
for these purposes, with the result that pairs like
"pause/paws", "staid/stayed", "brood/ brewed", "bowler
(hat)/bowler (one who bowls)" are not homophonous in the
relevant dialects. Class 1 suffixation, however, will
trigger alternation, so that "classic" has a different
vowel from "class" in New York/Philadelphia, and "polar"
has a different vowel from "pole" in London. The two
closure effects B examines more closely are dentalization
in Northern Irish English and cluster simplification in all
dialects of English. In Northern Irish English, alveolar
[t d n l] have dental allophones (I'll use the ASCII-
friendly transcriptions [t% d% n% l%]) before [r @r], e.g.
"train" [t%reyn], "ladder" [laed%@r], "pillar" [pIl%@r],
etc. Class 1 suffixes cause alternation, as in "element"
[-nt] vs. "elementary" [-n%t%ri], but Class 2 suffixes do
not (underapplication of dentalization), as in "late"
[leyt], "later" [leyt@r] (*[leyt%@r]). Under B's analysis,
the OO2-faithfulness constraint outranks the dentalization
constraint, so that "later" has the same alveolar consonant
as "late" has, but the dentalization constraint outranks
the OO1-faithfulness constraint, so that "elementary" has
dental consonants at the cost of unfaithfulness to
"element". The second closure effect B examines is cluster
simplification, which happens word-finally ("condemn" with
[-m]) and before class 2 suffixes ("condemning" with
[-m-]), but not before class 1 suffixes ("condemnation"
with [-mn-]). These data receive a similar analysis to the
other closure effects: OO2-faithfulness outranks the
markedness constraint (in this case the constraint banning
syllable-final [mn]) which in turn outranks OO1-
faithfulness. The next issue examined in this chapter is
what B calls "aggressive closure", namely why class 1
suffixes may attach to bound roots (electr-ic), but class 2
suffixes may not (*electr-ful). That bound roots cannot
appear unaffixed is attributed by B to a constraint BOUND
ROOT that says so. This outranks OO2-DEP (requiring every
segment in a class 2 affixed word to have a correspondent
in its base), which outranks IO-MAX (requiring every
segment in the input to have a correspondent in the
output), which outranks OO1-DEP. High-ranking BOUND ROOT
prevents "electr" from ever standing on its own as a word.
The ranking OO2-DEP >> IO-MAX rules out "electriful" since
it is better to have no surface segments matching the input
/electr+ful/ than to have all those surface segments in a
class 2 affixed word without any surface base word. But
"electric" is allowed because of the ranking IO-MAX >> OO1-
DEP; it's better to have a surface form matching the input
/electr+ic/ than to delete all those input sounds when the
affixation is only class 1. B's conclusion from the
closure effects she examines sets her analysis distinctly
apart from serial derivation analyses: Class 2 affixation
is not added after the base word has already been formed;
rather, the constraint ranking simply requires class 2
affixed words to be highly faithful to their bases.
The third issue discussed in this chapter concerns the
alleged productivity and transparency of class 2 affixes
with respect to class 1 affixes, and the fact that class 2
affixes are ordered outside class 1 affixes. In
derivational theory, both of these effects can be viewed as
a direct result of level ordering, but since B's analysis
is nonderivational, she is forced to conclude that these
effects are not attributable to phonology. B concludes
this chapter as she ended the two previous chapters, with a
comparison between TCT and serial OT on the one hand and
derivational theory on the other, and shows the superiority
of the TCT approach.
In Chapter 6, "Outstanding Issues and Concluding Remarks" B
draws her conclusions about TCT, stating, "The strong claim
of this theory is that all morpheme-specific behavior
follows from the rank of the relevant set of faithfulness
constraints (OO-Identity, IO-Faith or BR-Identity)" (p.
234). She then moves on to discuss some consequences of
TCT and to refute potential counterexamples. In conclusion,
she examines the morphology-phonology interface, summing
up, "The strong claim of TCT is that in the domain of
paradigms, the selection of an OO-correspondence relation,
played out in the rank of the faithfulness constraints
proper to that relation, is sufficient to model phonology's
sensitivity to morphological information.
In general, I find Transderivational Correspondence Theory
to be a very promising field for future research. My only
criticisms with the theory concern matters discussed in
chapter 5. First of all, when discussing why class 1 but
not class 2 affixes may attach to bound roots, B assumes
"that bound roots are lexically marked as such and
prevented from surfacing on their own by an inviolable
morpho-phonological constraint BOUND ROOT (roughly,
'unaffixed bound roots cannot be words')" (p. 203). This
"solution" is nothing more than restating the problem: In
effect, B's answer to the question "Why can't 'electr'
stand by itself without an affix?" is, "Because it's a
bound root," and her answer to the question "What is the
definition of a bound root?" is, "A root that cannot stand
by itself without an affix"--a circular definition. Her
commitment to the morpheme-based Item-and-Arrangement model
of morphology apparently prevented her from considering the
possibility that /electric/ might be an input listed as
such, whole and without internal boundaries, in the
lexicon. Such an approach would also account for the
presence of exceptions to the generalization that class 2
affixes can't attach to bound roots, which B mentions in a
footnote (she lists "hapless, feckless, gruesome, fulsome"
in fn. 108 on p. 231) but for which she provides no
explanation at all.
A further criticism is the fact that TCT does not predict
any affix-ordering generalizations. B views this as an
asset to the theory, since there are exceptions to the rule
that class 2 affixes in English must always stand outside
class 1 affixes: She mentions the sequences -ability,
-ization, -mental, and -istic as cases where class 2
suffixes precede class 1 suffixes. But it isn't clear that
-able, -ize, -ment, and -ist always behave like class 2
suffixes anyway: In "'comparable" and "ad'vertisement",
-able and -ment affect stress placement like class 1
suffixes; in "publicize" and "publicist", -ize and -ist
cause Velar Softening, which is commonly held to apply only
before class 1 and not class 2 suffixes (hence no softening
in "picnicking", for example). And all four can attach to
bound roots, as in "abominable", "baptize", "altruist",
"detriment". And even if -ability and the others *are*
exceptions to the affix-ordering generalization (AOG), the
fact remains that except for these few fixed cases, affix
ordering usually does hold, and TCT does not exclude the
possibility of monstrosities like *innonlegible and
*tendernessous. In acknowledgment of this, B states, "the
AOG is not a phonological fact. It is, if anything, a
morphological phenomenon. In the general case, violating
the AOG has no impact on the phonology, since paradigmatic
relations are evaluated locally, in pairs of words." But
once one has conceded that, what is to stop us from
attributing *all* output-output mismatches under class 1
affixation to the morphology (see Bybee 1985, Bochner 1993,
Ford et al. 1997, Green 2001)?
When B discusses closure effects, I wish she had chosen the
Adelaide English example rather than the Northern Irish
English example to discuss in detail, because the Adelaide
example proves more clearly her point that class 2 affixed
words are not always phonologically identical to their
base, i.e. OO2-faithfulness can sometimes be violated. In
Adelaide English, there is vowel allophony of the /o:/ and
/u:/ phonemes before [l], depending on whether the [l] is
tautosyllabic or not: one allophone appears in "holy" and
"Julie"; the other in "goal" and "fool". Furthermore, [l]
is velarized in coda position but not in onset position.
Under class 2 affixation, the vowel allophony is faithful
to the base, so the vowels of "goalie" and "fooling" match
those of "goal" and "fool"; however, the quality of the [l]
does not match, being non-velarized in the former pair and
velarized in the latter pair. Thus the Adelaide facts
really show the superiority of TCT to a derivational
account, so in my opinion this would have been a more
interesting case study than Northern Irish dentalization.
In one or two places I also find myself disagreeing with
B's judgments on English pronunciation: on p. 215, for
example, it is crucial to her analysis that "condemnable"
be pronounced without the "n"; but in my own pronunciation
and in the recommended pronunciation of all the
dictionaries in my office, the "n" *is* pronounced in
"condemnable" (another way in which -able behaves as a
class 1 rather than a class 2 suffix). And on p. 218 she
discusses the stress identity effect allegedly seen in
"con'demn/con,dem'nation", but again both I and my
dictionaries agree that the stress of the latter word is
",condem'nation".
I found some minor mistakes that do not affect B's
argumentation in any way. For example, on p. 170 she gives
"postal" as an example of -al suffixing to a Germanic root;
in fact, the "post" from which "postal" is derived is
Latinate. Nevertheless, there are cases where -al is
attached to Germanic roots: the adjectival -al in
"bridal", and the nominalizing -al in "burial" and
"withdrawal". Another instance: in (159) on p. 185 she
implies that "pass/passive/passing" is an example of a base
in its unaffixed form, with a class 1 suffix, and a class 2
suffix, respectively. In fact, "passive" is not related
(semantically or even etymologically) with "pass", but the
other examples "class/classic/classy" and "mass/massive/
massable" prove the point just as well.
Finally, to turn from substance to style, it seems very
likely that this book was not professionally edited but was
supplied in camera-ready form by the author to the
publisher. Among the recurring stylistic irritations are
the use of hyphens after adverbs ending in -ly (as in
"morphologically-related words"), misuse of the term
"begging the question" when what is meant is "raising the
question", and the location of footnotes at the end of the
chapter instead of at the bottom of the page. (This was
presumably Garland's decision rather than Benua's, but it's
annoying to the reader and completely unnecessary in the
age of computerized word processing.)
All in all, I find TCT to be a very interesting and
exciting proposal, and one that deserves more research.
Above all, TCT provides important insights into allophonic
alternations that behave differently depending on the kind
of suffix added (e.g. the Adelaide English case mentioned
above). I am less certain of its usefulness in the
analysis of alternations of different phonemes within a
paradigm, and B did not actually ever discuss this issue.
(I'm thinking of things like Trisyllabic Shortening and
Velar Softening in English.) But that need not detract
from the important work that TCT can do, especially with
the tool of recursive constraint ranking.
BIBLIOGRAPHY
Bochner, H. (1993). Simplicity in Generative Morphology.
Berlin: Mouton de Gruyter.
Bybee, J. L. (1985). Morphology: A Study of the Relation
between Meaning and Form. Amsterdam: Benjamins.
Ford, A., R. Singh & G. Martohardjono (1997). Pace Panini:
Towards a Word-Based Theory of Morphology. New York:
Peter Lang.
Green, A. D. (2001). A nonderivational word-based
morphology and its phonological consequences. Talk given
at the Conference on the Lexicon in Linguistic Theory,
University of Duesseldorf, 22 August 2001.
Antony Dubach Green is a research associate at the
University of Potsdam (Germany). A member of the research
project "Optimality-Theoretic Constraints and the Lexicon"
within the Research Group "Conflicting Rules", he is
researching OT approaches to lexical organization and the
phonology/morphology interface.
|
