Documentos de Académico
Documentos de Profesional
Documentos de Cultura
$XWKRUV'RXJODV3XOOH\EODQN
6RXUFH$QQXDO5HYLHZRI$QWKURSRORJ\9ROSS
3XEOLVKHGE\Annual Reviews
6WDEOH85/http://www.jstor.org/stable/2155891
$FFHVVHG
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=annrevs.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Annual Reviews is collaborating with JSTOR to digitize, preserve and extend access to Annual Review of
Anthropology.
http://www.jstor.org
Annu. Rev. Anthropol. 1989. 18:203-26
Copyright ? 1989 by Annual Reviews Inc. All rights reserved
NONLINEAR PHONOLOGY
Douglas Pulleyblank
of Linguistics,Universityof Ottawa,Ottawa,Ontario,CanadaKIN 6N5
Department
INTRODUCTION
The productionand perceptionof speech involve the interactionof a varietyof
semi-independent grammatical modules. The MORPHOLOGICAL component
governs the structure of a well-formed word, licensing certain types of
morpheme' combinations (for example, un+event+ful+ly, in-
ter+nation+al+is+ation) and disallowing others (for instance, *un+abil+-
ity, *inter+nation+ise2). The SYNTACTIC component governs the combina-
tion of words into well-formed sentences, allowing, for example, an English
sentence like "Which student did John say the professor failed?" while
disallowing a sentence like "*Which student did John see the professor that
failed?". Of particularinterestto the linguist are cases where ungrammatical-
ity does not follow from some logically necessaryproperty.In examples such
as those mentionedhere, a word like *un+abil+ity is unacceptablein spite of
there being a straightforwardinterpretation,the expected meaning being
comparable to that of in+abil+ity (note moreover the perfectly acceptable
un+able and abil+ity); similarly, the sentence "*Whichstudentdid John see
the professorthat failed?"is ungrammaticalin spite of there being a perfectly
straightforward(and unique) interpretation,namely "Johnsaw the professor
that failed which student?" Because the ungrammaticalityof such cases
cannot be attributedto some logically necessary property, they serve as
windows into the intrinsic structureof the language capacity (4, 5).
In addition to the morphology and syntax, one must distinguish "in-
terpretive"components of a grammar,components including the SEMANTICS
and the PHONOLOGY. These componentsassign meanings and pronunciations,
respectively, to the structuresgeneratedby the combinationof morphology
'Morphemesare linguistic units of which words are composed-typically, but not invariably,
the smallest formalunits thatcan be assigned independentmeanings (see 2). Boundariesbetween
morphemes are indicated here by the symbol "+".
2By standardlinguistic convention, ungrammaticalexamples are precededby an asterisk(*).
203
0084-6570/89/1015-0203$02.OO
204 PULLEYBLANK
'Here and throughoutthis paper, square brackets ("[1")are employed to indicate phonetic
transcriptions(of varying degrees of detail). Such transcriptionsrepresent in a standardized
alphabetthe actualsoundsof an utterance.For example, the word "phlegm"would be represented
as [flEm]:Orthographic"ph"representsa single sound [f]; the sound of orthographic"e" in this
word is representedphoneticallyas [E];orthographic"g" does not representa phonetic segment,
hence is excluded from the transcription.Aspects of transcriptionsthatare crucialare commented
on in the text.
4For simple verification, place a finger on the outside of the larynx and produce first a
prolonged [>e],then a prolonged [s], comparingthe presence vs absence of vibrationin the two
segments.
206 PULLEYBLANK
s Q m
Constituency
In a linear theory such as that illustratedin Example 1, the only constituents
into which phonological segments are groupedare defined morphologicallyor
syntactically.That is, phonological segments are groupedtogetherto make up
morphemes, to make up words, to make up phrases, and so on. But an
examination of data from a wide array of languages makes it clear that
segments can also be grouped into constituents on purely phonological
grounds.
For example, Clements & Keyser (9:59-60) discuss five independent
phonological rules of Turkish, all of which involve an environmentcrucially
referring to a consonant that is either word-final or followed by another
consonant. Such a systematic disjunction is problematicfor a linear theory
because the two environments share no formal properties.
To illustrate,consonantsin Turkishmay be long (geminate)or short. When
a geminateconsonantoccurs before a vowel, it is retained;but when the same
consonant occurs either word-finally or before a consonant (the disjunction
under consideration), the geminate consonant is shortened:
4.
h a kk+ i h/k h/k + tLn
w w
an g
k an g a o
I nA o mA
A n gA roo~~~~~~~~~~~~~~~~ a
;
t o poei
~~5.
kAngaROO OnomAtoPOEIa
6See, for example, Kahn (21), Clements& Keyser (9), Steriade(37), Levin (24), McCarthy&
Prince (27), 1to (19, 20), and Hayes (17).
7See, for example, Liberman & Prince (25), Hayes (14, 15), Hammond (13), Halle &
Vergnaud (12); for a contraryview about the role of constituency, see Prince (30).
8Keepin mind thatphonologicalconstituentsare actuallyerectedover phonologicalsegments,
not orthographicalsymbols. For example, the sequence"oei"in onomatopoeiarepresentsa single
vowel, not three separatevowels. The structureshere are slightly simplified, and the reader is
referredto alreadycited work for details. Heads of foot ("4") and word ("W")level constituents
are indicated by a small circle. Main stress is indicated by capital letters, secondary stress by
small capitals.
NONLINEAR PHONOLOGY 209
9In Example6, "+c" = [+continuant], "?n" = [+nasal], "L" = low tone, "H" = high tone.
210 PULLEYBLANK
L stem: ba "grasshopperkiller"
bAaa "jealousy"
bala "build"
baaIa 1"cleave"
balaga "change a route"
7.
LH stem: saa "seed necklace"
samI "conversation"
saabI "roofing"
mW ragi "younger brother"
L stems: b a b a I a b a ga
L L
8.
H stems: b a b a\g/a b a Ig a
H H H
LH and HL stems are realizedby distributingthe two tones of the melody over
the available vowels of the stem. In the case of the sequence HL, the H and L
are assigned by a left-to-rightconvention, the final L subsequentlyassigned to
any leftover vowels to its right:11
HL stems: k a a k a r a k a r a g a
I I I I a9I
H L H L H L
LH stems are comparable except that in examples with three vowels, the
associations are modified in such a way that the initial L tone of the melody is
linked to the first two vowels:
LH stems: s a a s a m mw a r a g I
lI I I 10.
L H L H L H
Of particularinterest, however, are the cases where there are not enough
vowels for the tones of a particularmelody to be assigned to individual
vowels. It is precisely in such cases that contourtones appear-that is, tones
with a changing pitch curve. Correspondingto the HL cases of Example 9, a
falling tone is observedon a monosyllable;and correspondingto the LH cases
of Example 10, monosyllables exhibit a rising tone:
HL stems: ka LH stems: sa
HL
A A
LH 11.
LHL stems: bv I k a a y
/\
LHL
AA
L HL
12.
1-
bvI "falls" kaay "loses weight"
All such contours are the expected result of an account with the following
properties:(a) five tonal melodies are posited for Kukuya:L, H, LH, HL,
LHL; (b) these melodies are freely combined with stems of any number of
vowels; (c) combinationsof tones on a single vowel arepermitted,resultingin
surface contours. All three propertiesconstitute argumentsagainst a strictly
linear theory of phonology. Within a segmental theory of the type seen in
Example 1, it would not be expected thatsome single featureor set of features
would have a status completely independentof the others. That is, since all
featuresare simply componentsof unorderedfeaturematrixes, therewould be
no reason to expect the existence of melodic units composed of some subsetof
such unstructuredfeatures. In a nonlinearmodel, however, the existence of
melodic sequences (such as the tonal melodies of Kukuya) is completely
unsurprisingsince particularfeature classes can be assigned a status that is
independent of both other features and their assignment to particularseg-
ments.
In addition, since sequences of features cannot be assigned to a single
segment within a segmental model, the cases in Examples 11 and 12 would
require the postulation of additional, intrinsically contoured tone features.
The immediateproblemraised by such an approachconcerns the distribution
of such tones. A rising-fallingtone must be restrictedto monosyllabic stems;
rising and falling tones must be restrictedto monosyllabicor disyllabic stems,
with the additionalrestrictionon a disyllabic stem that such contours appear
only on the second vowel. Such clearly ad hoc restrictionsare unnecessaryif
contours are formally analyzed as consisting of a sequence of simple L and H
tones assigned to a single vowel-the nonlinear analysis.
This analysis can also explain contextual conditions on the creation of
contours even in languages that do not exhibit any evidence for tonal melo-
dies. For example in Yoruba (31), a rising contouris createdwhen a H tone
immediatelyfollows a L tone-a simple assimilationof the H toned vowel to
the preceding L toned vowel. Although the three lexical tones of Yorubaare
all level (high, mid, low), ri-singcontoursappearon the final syllable of words
like the following:
a. iwe "book"
b. ek6 "Lagos" 13.
c. akpa "scar"
For a linear theory, such cases presentthe same type of problem seen above.
NONLINEAR PHONOLOGY 213
Pre-pausalforms:
H stem: ba "oil palms"
baai "cheeks"
baiga "show knives" 15.
baadma "liana"
balaga "fence"
As seen in Example 15, all H tones of a stem are affected by this rule. In an
autosegmentalapproach, this is straightforwardlyrepresentedby the follow-
ing rule (18):
H -- M/_11- // 16.
J VV \/ t V 17.
H ->M H M H M
214 PULLEYBLANK
H -- M /- ll18.
A single rule of skeletal deletion accountsfor both the examples with short
vowels (Example 21) and the examples with long vowels (Example 22):12
a. n a g a m o b. b i m o s e
II II I I I I I I I A
C V C V C( C VCV C V( Deletion 23.
0 0
n a g a m b i m o s e
I
C
I I lI
V C V C
I I
C V
I
C
KI
V C V
I
Processes of this type that affect solely the skeletal level of representationare
entirely consistent with a sequential analysis of long vowels.
This rule, which affects not the skeletal level of representationbut the actual
melodic composition of a segment, applies in a manner so as to affect an
entire geminate:
12I am simplifying the process somewhat here. For details, see Piggott (29) and Halle &
Vergnaud (12).
NONLINEAR PHONOLOGY 217
a. OLDn t u b. o I u (
I AVI III
K nt u o I u ) Palatalization 26.
c vvc v v cv c cv
V CC Skeletal Level
a. [] b. [g +g 2
Under the sequential analysis of Example 32a, a linear theory would in-
correctly predict the application of spirantization,producing forms such as
NONLINEAR PHONOLOGY 219
This rule is to be read, "a nasal has the same values for [coronal], [anterior],
and [labial] as a following consonant," the Greek variables a, ,B and 8
indicating dependencies between feature values.
As pointedout by K. P. Mohanan,and developed in numerouspaperssince
then (e.g. 1, 8, 33), there are a variety of problems with such a linear
approach. Consider two examples.
The first problem is that of feature dependency. In a linear theory, the
formal notation employed to express dependenciesbetween featurevalues is
that of variables with interdependentvalues, as just seen. This notation,
however, allows the expressionof a wide numberof unattesteddependencies.
For example, althoughit is common to have the value of a feature[F] on one
segment be dependenton a value of the same feature[F] on anothersegment,
it is undesirableto allow a general mechanismwherebyboth + and - values
of [F] depend on the values of some second feature [G]. For example, a rule
identical to Example 36 but with revised featuredependenciesas in Example
37 is impossible:
Although formally just as simple as Example 36, with the Yoruba data just
cited this rule would have bizarreeffects like (a) deriving a bilabial nasal [m]
before an alveolar [d] (thatis, a [+anterior, -coronal, +labial] nasal before a
[+anterior, +coronal, -labial] consonant), and (b) deriving a palatal nasal
before a doubly articulatedlabial-velaras well as otherwise unattestedcon-
NONLINEAR PHONOLOGY 221
sonants before labials and palatals. Even the rule's only plausible result (the
derivation of a velar nasal before a velar consonant) must presumably be
considered accidental.
Although this problem might be solved by some form of condition on
interfeaturedependencies, an additional shortcoming of linear theories of
assimilation is the problem of natural classes. Rules such as Example 36
clearly demonstratethat a single rule can affect more than one feature. But if
this is the case, then what are the-featurecombinationscapableof functioning
as sets in phonological rules? If one adopts for the sake of concretenessa set
of 20 features [note Chomsky & Halle (6)], then there would in principle be
190 distinctcombinationsof two features, 1140 distinctcombinationsof three
features, and so on. But allowing such free combinationof features vastly
overgeneratesthe types of sets actually attested in phonological processes.
Problems such as these receive straightforwardsolutions within an appro-
priate nonlinear framework. Let us begin by assuming that features do not
constitute unorderedsets, but that they are intrinsically structuredso as to
reflect certain articulatoryproperties. Following work by Clements (7) and
Sagey (33), let us assume that the overall set of features(which is labelled the
ROOT, following a proposalby K. P. Mohanan)is divided into LARYNGEAL and
SUPRALARYNGEAL sets. If we think of the vocal tractas essentially a system of
tubes throughwhich air passes to produce speech, then one can think of the
laryngealfeaturesas determiningwhether, among otherthings, the column of
air passing throughthe system is vibratingor not; the supralaryngealfeatures,
on the otherhand, determinethe shape and volume of the vocal tract, thereby
determiningthe quality of the resulting sound. Within the set of supralaryn-
geal features, a further subset is the class of PLACE features, with further
subsets under place consisting of the features related to individual articula-
tors-for example, LABIAL for the movements of the lower lip, CORONAL for
the tip or blade of the tongue, and DORSAL for the body of the tongue. The set
structureso far described can be representedby a tree as follows:
A ,. rootnode
...... laryngeal node
...... supralaryngealnode
... place node 38.
o~~-~ ~. --. ....labial node
- - - - -o -\ jJ
....- coronal node
...... skeleton
L 1- ? 1_ | ..... root node
laryngeal .. \ \ ....\\\
node ......
...... supralaryngealnode 39.
9,_
)< j place node
.............
/____ \ coronal node
........
labial node ......_L_ _ _ _ _
As can be seen from this table, the sequences [a..e] and [:..o] are possible
only with an intervening consonant other than [h]; the sequences [a..a] and
[.. ], on the other hand, are possible with adjacent vowels, or when an
intervening consonant is [h].
Puttingaside the problemof the cases involving [h], the patternin Example
40 is straightforwardlyderived by positing a rule that assimilates a vowel to
an immediately preceding vowel. Assuming that the sequences [a..a] and
.. ] do not appearunderlyinglyin Tiv [as can indeed be argued(32)], vaa
andyox are thereforederived by the applicationof assimilationto priorforms
/vae/ and IyjoI. The problem is how to account for the applicabilityof this
rule of vocalic assimilation to sequences of vowels with an intervening[h].
To achieve this end, let us assume that the requiredrule is one that spreads
the supralaryngealset of specifications, that is, a rule along the following
lines:
v v
O supralaryngealnode
1"Thevowels [a, e (both produced without lip-rounding)cannot co-occur with [L, ol (both
producedwith lip-rounding)because of a rule causing agreementin roundingin such forms (32).
The surface tonal patterns of verbal forms depend on how a verb is inflected. Since this is
irrelevanthere, tone is not indicated below (but see 31).
224 PULLEYBLANK
1o1 o o J A supralaryngealnode
Assimilation is applicablein the first two cases but blocked in the third case
by the supralaryngealspecification of the intervening consonant:
a. a a b. a h a c. a s e
VV V CV V C
O Q o o o supralaryngealnode
CONCLUSION
The approachto phonology presented here is one that crucially involves a
numberof semi-independentsubtheories. Segments are externallyorganized
PHONOLOGY 225
NONLINEAR
I thank Alan Bell and Larry Hyman for discussion of certain parts of this
paper.
Literature Cited
1. Archangeli, D., Pulleyblank, D. 1989. Lengtheningand ConsonantGemination
The Contentand Structureof Phonologi- in LuGanda.In Studies in Compensatory
cal Representations. Cambridge, MA: Lengthening, ed. L. Wetzels, E. Sezer,
MIT Press. To appear pp. 37-77. Dordrecht:Foris
2. Aronoff, M. 1976. Word Formation in 8. Clements, G. N. 1985. The geometry of
Generative Grammar. Cambridge,MA: phonological features. Phonol. Yearb.
MIT Press 2:225-52
3. Bagemihl, B. 1988. Alternatephonolo- 9. Clements, G. N., Keyser, S. J. 1983.
gies and morphologies. PhD thesis. CV Phonology: A Generative Theory of
Univ. British Columbia the Syllable. Cambridge, MA: MIT
4. Chomsky, N. 1980. Rules and Repre- Press
sentations. New York: Columbia Univ. 10. Durand, J. 1986. Dependency and Non-
Press Linear Phonology. London: Croom
5. Chomsky, N. 1981. Lectures on Gov- Helm
ernment and Binding: The Pisa Lec- 11. Goldsmith, J. 1979. Autosegmental
tures. Dordrecht:Foris Phonology. New York: GarlandPress
6. Chomsky, N., Halle, M. 1968. The 12. Halle, M., Vergnaud, J.-R. 1987. An
Sound Pattern of English. New York: Essay on Stress. Cambridge, MA: MIT
Harper & Row Press
7. Clements, G. N. 1985. Compensatory 13. Hammond,M. 1984. ConstrainingMet-
226 PULLEYBLANK