The Phonology of Nasal-Obstruent Sequences by Samuel Rosenthall a Thesis Submitted to the Faculty of Graduate Studies and Resear

Home , Delateralization, Hanis language, Obstruent, Prenasalized consonant, Sierra Popoluca

Iii ,

The Phonology of Nasal-Obstruent Sequences

Samuel Rosenthall

A Thesis submitted to the Faculty of Graduate Studies and Research in partial fulfùlment of the requirements for the degree of Master of Arts

Department of Linguistics © Samuel Rosenthall McGill University August 1989 Montreal, Quebec c ( ABSTRACf This thesis presents an analysis of the phonological processes that affect contiguous nasal and obstruent segments. These phonological p!ocesses include voice, manner and

place assimilation as well as deletion and coalescence. The goal of this thesis IS to accoun t for these seemingly disparate processes by introducing universal constraints on the representation of segments in non-linear phonology. Deriving these processes from the principles of a theory of representation is beneficial because such an analysis is not possible in a theory that appeals only to rules. The result is a theory of phonology with greater explanatory adequacy than a theory that relies on mIes. Chapter 1 con tains a review of the history of the representation of segments and the representation of assimilation as weIl as a discussion of the theoretical assump:ions used throughout the thesis. Chapter 2 contains a discussion of the phonological processes as they occur during the formation of prenasalized consonants. These processes are shown to ( be triggered by the representation of prenasaHzed consonants and a theory of underspecification. Chapter 3 proposes an analysis of the universal characteristics of nasal obstruent place assimilation which is then extended to explain sorne universal properties of consonantal assimilation in general.

c .. " RESUME • Le présent mémoire consiste en une analyse non-linéaire des processus phonologiques qui résultent du contact d'une nasale et d'une obstruente. Notre but est de démontrer que ces divers processus--voisement, assimilation de point et de mode d'articulation, élision et contraction--découlent tous de contraintes universelles sur la représentation des segments et non de l'application de règles spécifiques. Dans le chapitre l, nous présentons un bref historique de la représentations des segments et du traitement des processus d'assimilation ainsi que les présupposés théoriques que sous-tendent notre analyse. Après avoir décrit les processus phonologiques qu'implique le contact d'une nasale et d'une obstruente, nous démontrons, dans le chapitre 2, que ces processus sont déclenchés par la représentation des consonnes prénasalisées et la sous-spécification. Enfin, dans le chapitre 3, nous établissons les propriétés universelles de

l'assimilation de point d'articulation entre nasale et obstruente et les appliquons à l'assimilation des consonnes en général.

... ACKNO~EME~ITS 1 wish to thank rny advisor G. L. Piggott who se wisdom and red ink has !ouched every page of this thesis. Every appointment (good or bad) was a learning experience. 1 am very grateful. 1 also wish to thank aIl the linguists who have commented on different aspects of this thesis. In particular, 1 wish to thank Leslie Barratt who was a1ways helpful and aiw:\ys willing to answer that infamous question tt.at begins "Do you know a language ... " Louise Glackmeyer, Henrietta Hung, Alan Libert, Dave Lipscomb, Dean

Mellow, Alan Munn, Ben Shaer, and everyone at McGill must be thanked for making it worth my while to show up in the department day afler day. Henrietta Hung anJ Ben Shaer deserve special mention for not only the much appreciated proofreading and humour, but also for helping me survive the highs and lows (both thesis-related and unrelated) that 1 endured while writing this thesis. Shayna Shapiro and Harvey Yelen also helped me through the thesis with their friendship and sorne eleventh-hour proofreading. Only reill ( friends would help with such a chore on a subject that must seem so boring. 1 think they actually leamed some linguistics in the end. Lastly, l wish to thank rny parents Leonard and Devorah, my brother Gary, rny sister Wendy, and rny brother-in·law Gerry. Their encouragement, advise, and love made this thesis possible.

IV o TABLE OF CONTENTS

ChMHer 1: Introduction 1.0 IntrOOuction ...... , ...... , ...... 1 1.1 Historical Background...... 3 1.1.1 Nasal-Obstruent Assimilation ...... 3 1.1.2 Prenasalized Consonants ...... 6 1.2 Theoretical Preliminaries ...... 1 0 1.2.1 Feature Geometry ...... , ...... 10 1.2.2 'fheory ofUnderspecification ...... 13 1.2.3 Underlying Representation ...... 15

1.2.4 Syllable Structure.,. t •••• Il' Il .,' •••• ", l'' Il •• , •••••••• " •• , , •••• ,. Il ••••••• """ •• , 17 1.3 Principles and Parameters in Phonology ...... 18 1.3.1 A&P's Parameters of Assimilation ...... 22 ,.- 1.4 Conclusion ...... 24 1.5 A Note on Transcription ...... 24 Chapter 2: PrenasaliZed Stops 2.0 Introduction ...... 26 2.1 The Representation of PrenasalizedStops ...... 27 2.2 The Representation of Prenasalized Stops on the Root Tier ...... 29 2.2.1 Evidence for Two Root Nodes ...... 30 2.2.1.1 Autosegll1ental Spreading ...... 30 2.2.1.2 Prenasalized Fricatives ...... 1,31 2.2.1.3 Sinhalese Gemination ...... 32 2.3 Unification (Prer.asalized Stop Formation) ...... 35 2.4 More on the Representation of Prenasalized Stops ...... 36 ~- 2.4.1 The Contour Node Condition ...... 37 J ~ ,

\1 ( 2.5 Phonetic Adjustn1ent Processes ...... 42 2.5.1 Post-nasal Voicing ...... 43 2.5.2 Post-nasal Hardening ...... 44

2.5.3 Nasal Deletion ... il il ••••• il il il il il. il il' il. il il. il iI.t il il il" il il il il il il •• il ..... il il •• il il il il'. il il il il il il il il .. il •••••• 45

2.5.4 Coalescence ...... il Il •••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••• 49 2.6 More Phonetic Adjustment Processes ...... 52

2.6.1 The Hardening of Glides. il il il il il il. il il il il il il il il il" il il. il il il' il il il' il il il il il il il' il il il il il il il il il il il il. il il il il il t" 52 2.6.2 The Hardening of Liquids...... 53 2.6.3 De-Implosion ...... 55 2.7 Other Prenasalized Consonants...... 56 2.8 Conclusion...... 60 Chapter 3: Nasal -Obstruent Assimilation 3.0 Introduction ...... 62 ( 3.1 The Nature of Nasal-Obstruent Assimilation ...... 62 3.2 Inverse Contours ...... 64 3.3 A Theory of Nasal-Stop Assimilation ...... 70 3.3.1 A Case Study: Diola-Fogny ...... 74 3.4 Sorne Consequences of Inverse Contours ...... 78 3.5 Constraints on Progressive Assimilation ...... 83 3.6 Conclusion ...... 85

Chapter 4: Conc1udin~ Remarks 4.0 Residual Problems ...... 86 4.1 Concluding Remarks ...... 88 References ...... 90 c

VI o CHAPTER 1 Introduction 1.Q. Introduction This thesis is an investigation of various phenomena that occur in the context of adjacent nasal and obstruent segment:.;. These phenornena are deletion, coalescence, and a variety of assimilation processes. The purpose of the investigation is to account for sorne univers al aspects of the phenomena associated with adjacent nasal and obstruent segments

by appealing to the principles of a theory of representation. Adjacent ~ ",saIs and obstruents appear in two structures: 1. prenasalized consonants and 2. nasal-obstruent c1usters. AlI of the phenomena to be investigated occur in both structures, but the discussion here mostly considers these phenomena within the context of prenasalized consonants. Most of the discussion of prenasaiized consonants concerns the behaviour of adjacent nasals and stops because these sequences are more prevalent. The analyses of nasal-stop sequences are ~ Jo subequently extended to include nasal-fricative sequences. The nasal and the obstruent in a prenasalized consonant obligatorily share one place of aniculanon and often agree for voicing and continuancy. The shared place of aniculation is the result of the nasal assimllatmg to the place of articulation of the obstruent. In sorne cases, the obstruent in the context of a prenasalized consonant undergoes assimilation for [voice] and [contmuant) so that bath the nasal and the obstruent agree for these features. In other cases, the obstruent or the nasal deletes. As previously mentioned, aU the phenomena that occur in the context of prenasalized consonants also occur in nasal-obstruent c1usters, but with the exception of place assimilation, these phenomena are ooly discussed in the cootext of prenasalized

consonants. Place assimilation is discussed in the context of nasal~obstruent clusters because place aSSImilation has interesting characteristics that require separate investigation.

Place assimilation of nasals adjacent to obstruents has three universal characteristics: 1. the

nasal always assimilates to the obstruent, but the obstruent never a~similates to the nasal, 2. any language with nasal-to-fricative assimilation aIso has nasal-to-stop assimilation, but not the converse, and 3. the nasal is always to the left of the obstruent. The aim of the investigation in this thesis is to account for the phonology of nasal

obstruent phenomena by proposing constraints on the feature-geometry model phonological representation introduced by Clements (1985). The purpose of introducing new constraints

on representation is to account for significant cross-linguistic generalizations conceming the

behaviour of adjacent nasal and oh')truent segments that are overlooked in 'i theory that relies on only language-partÎ\.1ï1:.r rules. The advantage of adding constrainls on representation that are provided by universal grammar is the achievement of greater explanatory adequacy in phonology. Nasal-obstruent phenomena are suitable for achieving this goal because nasal-obstruent phenomena are so widespread. Furthermore, the constraints used to account for nasal-obstruent phenornena have applications beyond the

phonology of nasal-obstruent sequences. These constraints can also account for sorne ( universal characteristics of assimilation in general. The thesls is in four chapters. Chapter 1 briefly outlines the theories of non-linear representation, [eature speCIfication, and syllable structure assumed throughout the thesis. The first chapter also contains a discussion of the model of phonological principles and

parameters in WhlCh the analysis of nasal-obstruent phenomena is embedded. Chapter 2

contains the analysis of the phonological processes which occur durirg prenasalized stop formation. Chapter 3 contains the analysis of place assimilation in nasal-obstruent

sequences and a discussion of how the constraints apply to assimilation in general. Although place assimilation occurs in both contexts, place assimilation is discussed

separately because it requîres different constraints than the constraints required in the

analysis of prenasalized consonants and the different constraints introduced for place

assimilation have dlfferent applications in the theory of phonology. The fourth and îinal

chapter contains the conclusion and a discussion of sorne residual problems not directly

( accounted for elsewhere.

2 1.1. Historical Backuound • Since the aim of this thesis is to appeal to constraints on representation to account for nasal·obstruent phenomena, it is important to demonstrate the problems crcated uy

using rules in sorne previou~ analyses. The following discussion is divided into the different contexts used for discussing nasal·obstruent phe!lomena.

1.1.1. Nasal·Qbstru~nt Assimilatiml With the use of only the feature values '+' and '.', three individu al ruIes, (la), are needed to describe an assimilation process where a segment ber,omes labial before a labial,

alveolar before an aveolar, and velar before a velar. (1(1) obviously misses a generalization about the nature of the assImilation process. Chomsky and Halle (1968) (henceforth SPE)

introduce variables as feature coefficients, shown in (1 b), to capture the fact that the place

of articulatIon of the second consonant determines the place of articulation t)f the first consonant.

(1) R. Set of assi milation processes b. variable fea\ ure coefficients

[+cons] --> [X ant ) J _ ant ) [+cons] ..> [:::: J J_ [::::) r [a. Lf3cor ~ cor

[ +ant [ +ant ] -cor ] J_ -cor [-ant ] [-dnt ] +cor J_ +cor [-ant ] [-ant ] -cor J- o -cor Consider the following data frorn the English prefix in- and the corresponding rule.

(2) impossible intolerable i[1J]compIete

3 ( (3) Nasal- ;top assimilation C -cont a ant [+na~l] --> ~am. 1 l- ~cor ~cor

Li~ting three assimilation rules (one eaeh for labial, alv'eolar, and velar assimilation) misses

the obvious generalization conceming the behavlOur of nasal segments Yi hen followed by ~ stop, but variable feature coefficients adequately de scribe nasal-stop assimil&tion in one

rule. Furthe~mf)re, variables as fe~ture coefficients ensure that nasal-stop assimilation is highly valued by the slmphclty m{"tric (Cnomksy and Halle 1968), whereas listing three

indiviual mIes hke (la) IS le,s valùed Variable feature coefflciellts, although descriptively adequate, lack the explanatory power needed to capture sorne pr()perties of assImilatIon phenomena. For example a rule ( that asslmilates stops to foUowing nasal~ is just as highly \alued by the simplicity metric as (3) because both rules have tpe :iame number of :;ymbols in the minimal representation. However. (3) is extremely common but stop-nasal assimilation is unattested. The fIrst challenges to SPE approaches to assimilation address other problems

posed by the pror erues of assInlllation. The the0ry of autosegmental phonology (Goldsmith 1976) ,\ddresses the inability of the ,:,PE framework to accoUl.Jt for the phonology of supra.segmentals such as tone. In auto segmental representatlon, the tor.e

oecuples m own uer mdependent of the feature matrix and it is linked to the matrix by an

associatIon hne. Goldsmith proposes that tonal harmony can be more adeqJately charactenzed as the spreading of tone from one segment to the other segments. Since

harmony is represented as autosegmental spreadmg, the iterative aSSImilation rules used in SPE to characterize harmony are eliminated. Clements (1976} capitalize[ on Goldsmith's ( an llysi~ of suprasegmentals and accounts for a wide variety of vowel harmony rules by

4. .... _------_.

o claiming that the harmonizing feature, e.g. [round], [high], [ATR], occupies its own tier. Harmony proccsses, hke tonal processes, are represented by autosegmental spreading as shown in (4).

(4) matrices lX] lX] lX] lX] " ' .,."" ~ .,.-"'" ,. ..: -.,..::.:,... ~ --- ~ 0;- - feature tier lP]

THe amosegmental framework as proposed by Goldsmith captures the nature of single feature assimilation, !:lUt other assimilation rules such as gemination, which involves

the assimilation of aIl features, would he writtcn as SPE rule~. McCarthy (1979) proposes that autosegmental representation should include a melody tier attached to a core tier of timing units consisting of consonantal and vowel s101s (henceforth CV phonology). Geminates in CV phonology are represented as one segment, i.e. feature matrix, attached to two units on the core. Gemination, l11ustrated in (5), can now be represented as the

spreading of the segment to another unit of the core.

(5) v d --> d d C c i""J ." d

CV phonology can account for complete assir.1ilation, i.e. gemination, and single feature assimilation, but because of the assumptions about the structure of the matrix, CV phonology provldes no explanation for partial assirrulation which involves the simultaneous spreadmg of a number of autosegments. For example, nasal-stop assimilation which

involves the sprearling of the place features ([corl, [ant], [back], etc.). Steriade (1982)

daims partIal assimilation IS explainable by assuming the place features form their own

submatrix which can spread en masse to other segments. Although the probiem of partial assImilation can be resolved by the appropriate representation of segments, the problems ( previously mentioned with respect to SPE mIes remain. AIl assimilation processes in CV phonology are still described in terros of mIes but these rules are now spreading rules. Although the use of spreading to represent assimilation is valid, an unconstrained theoI)' of spreading creates the same problern for the simplicity metric. Given the fact that aIl assimilation mies are represented the same way in CV phonology, then there is no reason sorne processes, such as progressive nasal-stop assimilation, do not occur whereas regressive nasal-stop assimilation is widespread. This asymmetry, [Ifst raised as a problem

, for the simplicity metric, persists in autosegmental theories of assimilation. Phonological '1 'i theory should predict the occurrence of only regressive nasal-obstruent assimilation. However, this cannot be done in a theory that relies sole!y on language-particular rules. 1

propose here that the characteristics of nasal-obstruent assimilation follow from som~ aspect of universal grammar in an approach to phonology based on the principles and purametcrs discussed in section 1.3. ( 1.1.2. Prenasalized Consonants A prenasalized consonant consists of a consonant preceded by a brief nasal onset within the timing of a single segment. These consonants are problernatic for the SPE theory

of representation (If segments as feature matrices because they are best described as a [ nasal] stop with a [+nasal] onset, but the theory of representation in SPE does not allow multiple specification of features in the description of a single segment. Two attempts to account for prenasalized consonants, in particular prenasalized stops, within the SPE framework are worth discussing. Chomsky and Halle note the problem posed by prenasalized stops and they suggest that [dl!ayed release] might be the appropriate feature. Although Chomsky and Halle do not pursue their suggestion, it is not hard to show that [del.rel.j is inadequate. The release phase of an affricate and the release phase of a prenasalized stop are different gestures. Therefore, only by stipulation can [delayed release] describe the release ph(lse of both affricates and prenasalized stops. { Chomsky and Halle's suggestion that [deI. rel] might be the appropriate feature puts affricates and prenasalized stops ioto a natural class which is desirable because these are the only segments that have a phase change within the span of a single segment. A discussion of how this natural c1ass is derived is presented in rhapttf 2. Ladefoged (1911) suggests a new feature [prenasal] which Anderson (1976) interprets as the lowering of the velum "during the formation of the principal con striction." Anderson redefines [nasal] as the lowering of the velum "during the period of maximum articulatory closure, through the release of the closure." Therefore, while simple nasal segments have velum lowering throughout the time span, Le. from onset to release, prenasalized stops have velum lowering only at the onset, and simple stops have no velum lowering. The contrasts are shown in (6).

(6) N C Ne +prenasal pprenasal +prenasal +nasal pnasal -nasal The feature [prenasal] provides a means of representing prenasalized stops, but according to Anderson, [prenasal] violates the fundamental notion that the independent features should correspond to discrete articulatory mechanisms. Both [nasal] and [prenasal] refer to the same the same articulatory mechanism hence there is a redundancy. The only

difference between the two i~ the duration of velum lowering. Furthermore no modification of the feature system of SPE can account for the fact that rules can manipulate the timing of oral release of prenasalized stops. Anderson discusses languages in which prenasalized stops only occur when preceded by a nasal vowel and followed by an oral vowel. This phenomenon is more adequately accounted for by a single feature for nasality and the idea

that somehow the nasality of the vowel influences the nasality of the stop. This can be done by assuming (nasal] is suprasegmental and hence it can range over units larger than the segment. Anderson concludes that there is no need for [prenasal] and that prenasalization is more accurately explained by treating [nasal] suprasegmentally.

7 ( The problem of the representation of prenasalized stops is solved naturally in Goldsmith's (1976) Autosegmental Phonology framework, which accounts for many properties of the phonology and representation of suprasegmentals. Goldsmith proposes

(hat [nasal] 10 Guarani is suprasegmental hence [nasal1 can occupy its own tier independent of the feature matrix. Goldsmith proposes that prenasalized stops in Guarani are formed by a postoralization rule which associates a [-nasal] autosegment to a segment previously specified as [+nasal] on the [nasal] tier. The result is a stop attached to two autosegments on the [nasal] tier.

(7) -cont ] -cor [ +ant /~ [+nasal] [-nasal]

( Autosegmental representation also accounts for sorne interesting properties of prenasalized stops. For example, Sinhalese has a nasal hannony process which nasalizes aIl vowels on either si de of a nasal segment. Prenasalized stops, however, only propagate nasality leftward. This phenomenon has a natural explanation in the autosegmental framework. Since a [-nasal] autosegment is attached to the right of the prenasalized segment, it blocks the association of the [+nasal] autosegment to vowels on the righthand side. Steriade (1982) provides a generalized representation of contour srgments, i.e., segments that are specified as [oF] and [-oF] on the same tier.

(8) x y x,y arbitrary units in the core

""-/X Xis CorV

(8) applies to prenasalized stops, as weIl as affricates which are specified as [-cont] and

[+cont] on the [continuant] tier. (8) however does not prohibit impossible contours such as c [voice] contours or [coronal] contours. Except for tone, [nasal] and [continuant] appear to

8 o be the only features that forro contours. This fact is complicated by the existence of prenasalized voiceless stops, e.g. [mp], which include a [voiceJ contour. However, there is a dependency between [nasal] contours and [voice] contours because the latter do not exist independently of the former. Recent views of the organization of the auto segmental tiers (Clements 1985, Sagey 1986), (11S,.1JSsed in 1.2.1, fail to properly constrain contour

segments. In chapter 2, 1 introduce a condition on the representation on contour segments that delimits the set of possible contour segments. Tones are excluded from the discussion. Prenasalized stops are often the result of morphophonemic processes which involve other phonological processes, su ch as the voicing of an obstruent or the despirantization of a fricative. Herbert (1986) notes that the cause of obstruent vOlcing and despirantization follows from a tendency for languages to have only prenasalized voiced stops. The same tendency also seems to cause the deletion of the nasal which occurs wh en the nasal is followed by a voiceless stop or a fricative. Herbert, as weIl as others, account for voicing by means of linear rules such as (9). (9) Post-nasal voicing C --> [~voice] / N_ Linear rules such as (9) simply fail to explain the relation between the tendency for

languages to have prenasalized voiced stops and the occurrence of roles such as (9). 1 propose that these facts follow from the representation of prenasalized stops and Archangeli' s (1984) theory of underspecification. Autosegmental repesentation can account for sorne phenomena associated with assimilation and contour segments, but there are many facts which autosegmental representation cannot account for without stipulation or relying on language-particular rules. Although the analysis of nasal-obstruent phenomena proposed here requires roles, the rules are accompanied by a theory of representation from which the characteristics of nasal-obstruent sequences follow. The analysis of nasal-obstruent phenomena proposed here is based on the feature-geometry model of autosegmental. phonology and the

Q · :

( underspecification theory of segments. These two theories of representation as weIl as the other theories used in this analysis are introduced in the following section. 1.2. Theoretical Preliminaries 1.2.1. Ecature Geometr,y

The faet that partial assimilation treats certain groups of features as a submatrix is captured by the feature-geometry framework (Clements 1985) in which the features are grouped into larger tiers (called c1ass tiers, shown in small capitals in (10». These class tiers are in tum organized under the root node which represents the melody tier of older models. The core tier is represented by CN slots.

(10) feature-geometry (Clemen13 1985) C/V 1 ROOT

LARYNGEAL [spre~ ( [COl\3~1e'd] '\. [voice] SUPRALARYNGEAL

MANNER [nasal~ [c'orit] '" [strident] PLACE

[cor] [ant] [dist [round] [high] [back] [low]

The class nodes are arranged hierarchically to capture the independence and the interdependence of certain features. Features doroinated by the same c1ass node behave as a group in assimilation processes. For example, the place features assimilate together because ( place assimilation is represented by the spreading of the place node which dominates the

10 o place features. Features dominated by different class nodes do not participate together in assimilation. For example, [voice] and [coronal] assimilation is prohibited because these

features are dominated by different class nodes. If there is a rule of [voice] and [coronaI]

assimilation, then feature-geometry predicts that such a mIe is indistinguishable from complete assimilation because the only c1ass node that dominates both [voice] and [coronaI]

is the root node. The modifications to this feature-geometry model made by Steriade (1985) and

Sagey (1986) radically alter the feature arrangement below the place node as well as

elsewhere.

(11) Peature-geometry (Sagey 1986) mot

laryngee.l [VOic~ [cons1r) supra.laIyngeal sOft-Pala~ 1 place [nasal)

dorsal [~V] [beck] corenal [ant~ [dist]

The terminal place features are attached to articulator nodes which represent the independently functioning articulators of the vocal tract. The coronal node represents the

use of the ton gue tip and blade, the dorsal node represents the use of the tongue body, and

the labial node represents the use of the lips. The articulator nodes represent monovalent feature specification so the presence of an articulator node in ttle representation implies

l' c activity in that part of the vocal tract. For example, a velar segment has a dorsal node, but no other articulator nodes. One advantage of the monovalent aniculator nodes is that it eliminates impossible processes such as [-coronal] assimilation since only the coronal node, representing the presence of the active articulator, can spread. Sagey uses the aniculator nodes to constrain the range of segments with multiple articulation, called

complex segments. Labio-velar segments such as [kp] are represented by the dorsal and labial node (but no coron al node since there is no tongue-blade activity), attached to a single place node. The aniculator node theory c1aims that aIl complex segments must contain two (or more) independent articulators.

Another interesting consequence of the aniculator node theory is that it provides

,1 new conditions under which segments may be adjacent, thus supponing the hypothesis that spreading domains are restricted to adjacent nodes (Archangeli & Pulleyblank 1986, Steriade 1985). For example, Sanskrit n- rettoflexion changes ln! to [1').] when preceded by Ir 1 or Isl, but the process is blocked by intervening coron al segments. Steriade (1985)

accounts for the blocking effect of intervening coronal segments by proposiIlg that retroflexation occurs only when the coron al nodes of the target and the trigger are adjacent. The intervening coron al segments block spreading because their coron al nodes violate the adjacency of th\': trigger and target. Noncoronal consonants do not have any tongue-blade

activity. Therefore, they do not block retroflexation be~ause they do not have a coron al node. The feature-geometry assumed here is based on the modifications of the arrangement of features above the place node made by Archangeli and Pulleyblank (1986) (henceforth A&P), Piggott (1987) and Schein and Steriade (1986). Clements (1985) and A&P show the supralaryngeal node dominating the features [cont], [son], [nasal] and [strident]. Sagey a!1d Schein and Steriade show the root node dominating [cont], [strident] and [son] and the supralaryngeal node (which dominates the soft-palate node in Sagey's c mode!) dominates [nasal]. Piggott (1987) argues that [nasal] must link directly to the root

12 o node so that nasal stability phenomena can be explained. The feature-geometry proposed here has the root node dominating [nasal] and [son] and the supralaryngeal node dominating [cont] and [strident]. The arrangement of features below the place node is the same as in (11).

(12) x 1 rt. rnuall [sonoXU\t) lar. [consonan1al] [~ice] s-la.r. ~n1lAu.&.r\t] 1 lstrid~nt] place

1.2.2. Theory of Underspecification Archangeli (1984) proposes a theory offeature specification in which certain values for features are not present in underlying representation. This theory, stated in (13), is called underspecification. (13) there is no feature such that sorne value is specified for that feature in every phoneme (Archangeli 1984: 39).

The statement of underspecification given in (13) states that one or both feature values are absent in underlying representation. Funhermore, only one value of a given feature can be

present underlingly. Consider the feature specification of features [L], [il], and [A] for segments A, R, and C.

(14) a. fully specified b. underspecified A B C A B C [0] + [0] + [2.] + + + [2.] [6] + [6] + The fully specified mau-ices in (14a) violate (13) because ail three features are specified in

every phoneme. By removing the feature values [-il], [-A], and [+L], the feature matrices

11. ( can conform to (13). The choice of underlying specification is restricted to non-redundant feature values that distinguish the sounds of the language in tlnderlying representation. Segments A and C are distinguished by the specifications of [Cl] and [dl. B is distinguished from A and C by not being specified for [Cl] and [dl. The absent feature values are inserted by universal co.1text-free default rules. For example, the default rules for the system in (14b) are shown in (15). The process by which

the default rules are chosen is discussed pre~ntly. (15) [] --> [-0] [] --> [-~] [] --> [+1:] The insertion of the default values is regulated by the Redundancy Rule Ordering Constraint (A&P 1986) which roughly states that default values are inserted as late as possible in the phonology or at the first instance the feature value is referred to in the phonology. For example, if the default value [+high] is not referred to by a rule, the default

( value will be inserted as late as possible. Therefore. it will he absent from the phonology. If a rule makes specifie reference to [+high], then the default value is inserted as early as possible with respect to the rule. In other words, the feature is introduced at the same time the rule applies. Archangeh and A&P allow for language-specifie changes to the set of redundant values. For ex ample, A&P consider [+high] to he the default value for [high]. However, it is possible that a language might have [+high] underlyingly. A Complement Rule is established to change the default value in this case from [+high] to [-high]. [-high], as the default value in this case, cannot appear underlyingly. The selection of default feature values is a function of the phonology of the language. Archangeli (1984) also proposes universal co-occurrence restrictions to further streamline specification in underlying representation. For example. a vowel specified as c [+low] cannot receive the l+high] default value. In order to avoid this contradiction,

14 .~. < .-

o Archangeli proposes that there are universal co-occurrence restrictions that automatically in sen the appropriate default value. 1.2.3. Underlyini: Representation A&P consider underlying representation (henceforth UR) to consist of a set of macro nodes and unorganized feature matrices. The features are mapped onto the geometry by universal grammar and mac.ro nodes and segments are linked by one-to-one, left-to-right association.

(16) {o 0 o}

+F +F +F -0 l -0 -H j

The problem with A&P's theory of UR is that it cannot include underJying geminates unless p:-especification is permitted. The word scene /si:n/, for example, must have two macro nodes associated to the vowe1, but this does not seem p')ssible if association lines

are always drawn one-to-one, left-to-right. The rightmost macro ncde in Isi:nl would not he linked to a segment. The Iii in Isi:nl must be underlyingly linked to two macro nodes to ensure the Iii surfaces as a geminate. Underlying geminates therefore force associations lines to exist underlyingly, but this does not mean that all segments are underlyingly associated to macro nodes.

Piggott (1985) proposes a theory of UR in which the association of skeletal points

to segments must satisfy certain constraints. Segments need not be associated to skeletal points but skeletal points in roots must he associated to segments. Affixes, on the other hand, are allowed to have unassociated skeletal points.

11: ( 17) Representation of roots (Piggon 1985) ( xxx] [x x ] '" [ x xx x x ] [ 1 1 1 1 1 1 1 1 A BC ABC AB C

1 propose that underlying representations consist of a set of skeletal points and root nodes which dominate unordered matrices. Since association !ines are unnecessary in UR

(except for geminates), Piggott's .!onstraints must he slightly changed. Root morphemes cannot allow the number of unassociated skeletal points to ex cede the number of unassociated root nodes in underlying representation. Root nodes in affixes, on the other hand, may be linked to skeletal points. This theory of underlying representation maintains Piggott's constraints and A&P's c1aim that feature-geometry does not exist underlyingly.

(18) x x x x n. n. rt. Tt. [~] [+~] [-6] [:] ( A&P do not inc1ude the root node in UR because they c1aim its appearance is predictable from lin king the features to the macro nodes. Since there are association lines from the

skeletal tier in UR, then root nodes must also he present to anchor the association lines. Since the feature-geometry is absent from underlying representation, then there must be a way to map the features onto the geometry. The mechanism by which features are mapped onto the geometry can be presented in two ways. One, the features present in UR are mapped onto a complete geometry with aIl c1ass nodes present. For example,

assume Isl is underlyingly [+cont, +ant]. This Isl would be represented with a complete set of cIass nodes a!though the laryngeal node wouid not dominate any features. A&P argue that generatmg the complete geometry is undesirable because it poses problems for The

Locality Condition whlch requires aIl phonological processes to be adjacent. For example,

vowel hamlOny processes require intervening segments he skipped. This can he done by ( removing cenain nodes from the representation. One way to remove unwanted nodes is to

16 ~------~I·~------

o say that they are not there in the frrst place. This leads to the other way to map features onto the geometry. A&P propose a Node Generation convention so that undesirable intervening nodes are absent frorn the representation. (19) Node Generation (A&P) A rule or convention assigning sorne feature or node A to sorne node B creates a path from A to B. In the case of /s/ mentioned above, the [+cont] specification will generate the supraiaryngeai node and the [+cont] specification will generate the coron al node. The place

node will be generated to iink the coronai ~,ode to the supralaryngeal node. Nod\ Generation ensures that clas5 nodes are present only when needed to dominate sorne feature. 1.2.4. Syllable Structure The theory of syllable structure and syllabification assumed here is from Piggott o and Singh (1985) (henceforth P&S). Every syllable (a) consists of two obligatory constituents, the onset (0) and the rime (R). The rime consists of one obligatory constituent, the nucleus (N), and one option al constituent, the coda (C). The onset and nucleus dominate the timing ticr which consists of skeletal points (Kaye and Lowenstamm 1985, Levin 1985) rather than eN slots. The minimal (unmarJœd) syllable must contain an onset and a rime which dominates a nucleus, as shown in (20a). A syllable with the coda is shown in (20b).

(20) a. a b.

oA R 1 N 1 x x x x x Syllables are constructed in accordance with the princip les and well·formedness

conditions outlined in P&S.

17 (21) Initial Syllabification Principle (ISP) ( Every skeletal slot is assigned to a position in a syllable. maximizing onsets.

(22) Syllable Well-fonnedness Conditions (wfc) a. Every node of syllable structure must dominate at least one skeletal slot. b. A segment which is specified [+vocal~c] must be asso:iated with at least one nuclear slol. c. A slot which is linked to a flegment unspecified for the feature [vœalic] must be associated \Vith a nucleus unless it precedes a slot linked to a [+vocalic] segment, in which case the fonner is assigned to the onset. d. A [-vocalic] segment is not associated with the nucleus. Vowels are syllabifieJ as nLclei, then the ISP syllabifies the consonants. A string such as ICI VC2C3 VC41 is initially syllabified as I$Cl V$C2C3 VC41 by the ISP. If the c1uster Cl C2 is a pennissible onset, the syllabification remains. If the cluster is not a permissible

onset, C2 is resyllabified 'lS a coda.

Wfc's (22c&d) ensure long vowels and diphthongs are initially syllabified in the

nucleus. The glide component of a diphthong may remain in the nucleus or may be ( resyllabified in the coda depending upon parametric restrictions.

1.3. Principles and Parameters in Phonolo~ Although autosegmental representation provides a more appropriate explanari:m of

the nature of assimilation by using spreading, assimilation is still governed by rules. The

fact that sorne assimilation processes are impossible ÎliC s that a generalization is being

overlooked. It seems logical that the range of possible a:3~i~nilation processes should follow

from universal grammar thus explaining their wide distribution and simplifying learnability.

Proposing that sorne characterisucs 'Jf phonology are part of universal graI11J1lar leads to the conclusion that phonological properties can be accounted for in terms of principles and

parameters in the same way as Chomsky (1981) has done for syntax. Chomsky proposes

principles and parameters are part of universal grammar, and the setting of the parameters

produces a set of core grammars which define the set of possible human grammars. Since ( the principles are part of universal grammar, they need not be learned. Parameters are

18 supplied by universal grammar but their setting is triggered by the language-particular factors. Lastly, there is idiosyncratic information that must he learned. This infonnation is

peripheral to the core because it does not follow from any aspect of univers al grammar. The

language learning task is greatly simplified since the child only learns the peripheral aspects

of the grammar. The concept of a core and a periphery component in phonology is very recent (Piggott 1988). Hence the properties of the components are not completely defined.

Sorne properties of the core and periphery cliscussed by Piggott (1988) are mentioned

below. The use of princip les in phonology was introduced to autosegmental phonology when Goldsmith (1976) proposed constraints in the forrn of well-fonnedness conditions on autosegmental representation. One well-fonnedness condition is the constraint prohibiting

association Hnes from crossing which ensures that the features on tier T that are joined to ., features on rier tare simultaneous . (23) a. ABC b.* A C B ...... T - 1 1 1 1>( abc abc...... t

This well-formedness condition accounts for the opacity of segments with respect to harmony processes. It is weIl known that a hannonizing feature [aF] cannot hannonize

over [-aF). This fact has a natural explanation in autosegmental phonology. If [aF] were

to harmonize over [-aF], the association Hnes would cross. Therefore [-aF] blocks

harmony. Steriade (1982) notes that the Crossing Constraint also constrains epenthesis

processes. For example, assume C in (23b) is epenthetic. This epenthesis process is ill

fonned because C cannot precede B on tier T and follow b on tier t.

The most widely discussed phonologie al principle is the Obligatory Contour

Principle (henceforth OCP) (Goldsmith 1976 and recently modified by McCarthy 1986)

which prohibits idenrical adjacent autosegments on any rier. For ex ample , the OCP dictates

that long vowels must De represented as shown in (:l4b).

19 ,., , " , \

(24) a.* x x b. x x ( 1 1 ,,/ Vi V'1 V The OCP, like the Crossing Constraint, not only constrains autosegmental representation. but also constrains the application of phonological processes. McCarthy (1986) examines cases where phonological processes are blocked from applying in environments where an OCP violation would arise.

Another constraint on represen~ation is Steriade's (1982) Shared Feature Convention which forces identical features in a linked structure to he shared by the linked structure. This aecounts for the fact that some assimilated structures are subject to the same constraints. (25) Shared Feature Convention (SFC) (Steriade 1982) [aF] [aF ] ~ ~G ~G] [~G J --> ~ ( [rH -rH [rH] [-yH] 1 1 1 1 x x x x The Shared Feature Convention eliminates some arbitrariness from representations by

forcing the identical features in a linked structure to be reY'\resented uniquely. In the feature geometry framework, the SFC applies to class nodes. This means that two segments that are linked by a class node share aIl other identical class nodes. Therefore, it is henceforth called the Shared Node Convention. The Crossing Constraint, the OCP, and the SFC account for a number of phonologie al phenomena by constraining representation. Sorne universal properties of assimilation can be explained by the representation of segments. For example. there is [+coronal] and [+labial] assimilation, but there is neither [-labial] nor [-coron al] assimilation. Sagey accounts for the asymmetry by representing coronality and labiality as c monovalent articulator nodes.

20 o Piggott (1988) proposes that configuration al constraints, i.e. constraints on structure, are part of the core. Furthermore, he proposes that there are configurational parameters that are also part of the core. For example, Piggott and Singh's theory of syllabification states that onsets are obligatory, but the complexity of the onset, Le., whether or not it can dominate more than one skeletal point, is parametrically determined. Similarly the complexity of the rime and the presence of the coda are aIso the result of parame ter settings, Hayes' (1981) metrical theory provides other examples of parameters that are part of the core. As a last example, Piggott claims that the occurrence of a geminate in a panicular language is related to a parameter (The Gemination Parameter) which allows one segment to be linked to two adjacent skeletal points. Substantive constraints, as opposed to configurational constl'aints, are part of the periphery. For example, Fula does not permit geminate continuants (Paradis 1987), but this restriction on the geminate

inventory is language-specifie. The geminate inventory of Fula can he captured by the setting of the gemination parameter, but there is an added restriction to prohibit geminate continuants. Plggott recognizes that sorne substantive constraints, such as the universal

redundancy constraints, might he included in the core. It is important to note that peripheral restrictions cannot override the constraints derived by the core, otherwise the notion of universal grammar is trivialized. The periphery consists of the aspects of a cenain language that are not detennined by the core. The core and the periphery together characterize the phonology of a language. The fact that configuration al constraints account for a wide range of phenomena with an appropriate level of explanatory adequacy gives credence to the role of principles and parameters in phonology. However, the status parameters that concern aspects of phonology such as the range of possible phonological processes remain to he investigated. The most exhaustive attempt to fonnulate parameters of assimilation is provided by A&P (1986).

"1 1.l.1. A&P's Parameters of Assimilation Assimilation processes are represented in the A&P framework as the setting of certain parameters plus learned infonnation concerning the trigger. The flfst p3rameter called the insertldelete parameter regulates the insertion or deletion of phonological infonnation. The particular piece of infonnation to he inserted or deleted must he specified because anything, e.g. features, nodes, association Hnes, can be subject to insertion or

deletion. One aspect of the information that can he factored out is whether structure (association lines) or content (features) is affected. Rence a second parameter called the content/structure parameter determines whether content or structure is affected. For a parameter to say insen or deI ete a feature is insufficient because there is no specific feature for the parameter to affect. Similarly, to affect an association line, the parameter must 'know' from which tier the line must start. A&P use the tenn 'argument' to define the trigger of the insert/ delete parameter. The argument combined with the content/structure parameter determines the phonological information that is affected by the insert/delete parameter. The hypothesis that aIl phonological processes involve adjacent segments leads A&P to define adjacency as a relative notion. For ex ample, two vowels can he considered adjacent it' adjacency is defined on the rime tier (A&P 1986) or two vowels can be considered adjacent with respect to a feature such as [round] on the [round] tier. The targets of a process must be defined in terms of the tier on which they are adjacent because it is possible for nonadjacent vowels on one tier to behave as adjacent vowels on another (A&P 1986, 1986a). For example, two vowels may behave as adjacent segments on the [round]

tier, but not on the rime tier. If a process treats these vowels as adjacent segments, then adjacency must be defined with respect to a particular tier. A&P introduce the maximal/minimal parame ter which determines the tier on which adjacency must hold. The maximal setting means adjacency is defined at the macro tier and the minimal setting defines c adjacency at the }owest relevant tier.

22 , ,. ~ .. 1 , :î l " o Another parameter determines the directionality of the assimilation. If the direction of the assimilation is in the same direction as the initial association, the parameter is set for 'same direction'. If the assimilation is in the opposite direction of the initial association, the directionality parameter is set for 'opposite direction'. Bach pàrameter has a default setting which, when set, decreases the amount idiosyncratic information learned for each process. AlI that must be known about a cenain process is the trigger and the marked parameter settings. The fonn of phonological processes in A&P's framework and the default settings of the parameters are given below. (26) 1 a. inseI1cletidelete b. maximaldef/minimal c. contenldef/structure d. same directiolldef/Opposite direction II Arguments (definition of trigger) Consider the following example of nasal-stop assimilation.

(27) n p --> mp x x 1 1 rt. rt. [+nas1/1 3-1. 5-1."" [-nasJ ~-"'-J pl. pl. 1 1 cor lab

This process is characterized by the following parameter settings.

(28) 1 a. inseI'tclef b. maxima~ef c. structure d. opposite direction II place node Since two of the parameter settings are the default setting, the idiosyncratic (leamed) infonnation of this assimilation process is reduced to 'structure' as the focus, assimilation is in the 'opposite direction of initial association', and the trigger is the place node.

23 c There are numerous problems with A&P's parameters. However most are not discussed here. With respect to nasal-stop assimilation, their framework does not provide any explanation for the fact that sorne assl:nilation processes, such as stop-nasal r assimilation, are unattested. Sorne assimilation processes might he more marked than others because there are fewer default settings, but their framework does not prohibit any unattested processes. According to A&P, stop-nasal assimilation is actually less marked because the directionality parameter would have the default setting. The A&P parameters must he abandoned because these parameters give no insight into the characteristics of nasal-obstruent assimilation nor do they prohibit impossible assimilation processes. The approach to nasal-obstruent assimilation proposed in chapter 3 is based on the propenies of the representation of segments and assimilation. The limits of the range of possible consonantal assimilation processes are shown to follow from constraints on the representation that are expressed in tenns of configuration al constraints c that are part of the core. 1.4. Conclusion In summary, this thesis assumes the geometrical representation of features and the theory of underspecification. These two aspects of phonology are crucial throughout the analyses presented in chapters 2 and 3. Although the nature of the core in phonology is unclear, the number of cross-linguistic generalizations support the need for sorne aspects of phonology to follow from universal grammar. The purpose of the approach to the study of nasal-obstruent phenomena here is to develop constraints on the representation of contour segments and consonant clusters from which the phonological phenomena within nasal obstruent sequences naturally follow. These constraints suppon Piggott's c1aim that configuration al constraints are part of universal grammar. 1.5. A Note on Transcription Prenasalized consonants are transcribed as a superscript nasal followed by the oral c counterpart, e.g. mb, nd. Only the partkular prenasalized consonant being disc:ussed is

24 jj\ " ,i ~

o represented in the aforementioned manner. In many of the languages discussed, all nasal· consonant elusters are in faet prenasalized consonants, but will not he shown as such when not pertinent to the discussion. Vowels are not always transcribed in the detail given in the original sources. Mid front vowels are transcribed as lei, low back vowels as lai, and high front vowels as IiI. Vowels are given in the proper transcription only when relevant to the discussion of prenasalization. The palatal glide is tra!1scrihed as Iyl regardless of the transcription found in the original source and the palatal voiced stop is transcribed as Ij/. ( CHAPTER 2 Prenasalized Consonants 2.Q. Introduction Steriade's (1982) representation of contour segments shown in 1.1.2 eliminates sorne problems posed by prenasalized consonants. However, as mentioned in 1.1.2, Steriade's proposal cannot limit contours to only those involving [nasal] and [cont]. It was also noted in 1.1.2 that [voice] can fonn a contour only when there is a [nasal] contour as well. The fact that there is only a limited number of contour segments is shown here to follow from a condition on the representation of contour segments. The proposed condition

is developed to account for (he representation of prenasalized stops in the feature-geometry framework. This condition is also used in the account of the phonological processes that occur during the fonnation of prenasalized stops. Herbert (1986) proposes a three step model for the derivation of synchronie ( prenasalized consonants derived from underlying nasal-consonant sequences. The three steps are: 1. underlying independence of the nasal and the consonant, 2. phonetie adjustment (voice, manner, and place assimilation), and 3. unification (the merger of the underlying sequence into a single segment). No arguments are made in this chapter to support or to undermine Herbert' s derivational stages; the model is used here to separate the different aspects of prenasalized consonant formation. Although prenasalized consonants can aiso be derived in other ways (Piggotl 1989), the discussion in this chapter

only concems prenasalized consonants derived from nasal-consonant sequences. The chapter begins with a discussion of the representation of prenasalized stops in the feature-geometry framework. This is followed by an account of Herbert's unification step. The remainder of the chapter is a discussion of the proposed condition on representation and how this condition accounts for Herbert's phonetic adjustment stage c with particular attention to Bantu languages. ,.;"

o 2.1. The Re.presentation of PrenasaUzed Stops The first question relevant to the representation of prenasalized stops concems the representation of the time duration. Feinstein (1979) proposes that prenasalized stops in Sinhalese are simply tautosyllabic nasal-stop sequences. Feinstein's syllabification analysis is ambiguous because it only states that the nasal and the stop feature matrices are tautosyllabic without any reference to the duration of the segments. Feinstein's proposaI

can be represented formally as (la) 01' (lb). (la), however, does not capture the traditional notion of prenasalized stops as a brief nasal onset before the oral segment with a shorter duration than a nasal-stop sequence. (lb), on the other hand, seems to capture the traditional characterization of prenasalized stops.

(1) a. 0 b. 0 A 1 x x x 1 1 A nasal stop nasal stop o Both (la) and (lb) capture the observation that prenasalized stops behave as nasal segments with respect to processes sensitive to the left edge of the prenasalized stop and behave as oral segments with respect to processes sensitive to the right edge. For example, bidirectional nasal spreading in Sinhalese is triggered by a nasal consonant (Feinstein 1979), but the nasal component of a prenasalized stop can trigger only leftward spreading. The oral component blocks spreading but cannot trigger spreading. Sagey (1986) provides both acoustic and phonological evidence that something similar to (lb) is correct. Sagey shows that prenasalized stops are marginally longer in duration than single segments, but prenasalized stops do not have the duration of nasal-stop sequences. The phonological evidence is drawn from syllable structure, compensatory lengthening, and reduplication. The syllable structure of many languages, e.g. Sinhalese, prohibits consonant clusters in the onset, but prenasalized stops occur. If (lb) is the correct t •

( representation, the prohibition against consonant clusters can he stated as a restriction on the nurnber of skeletal points dominated by the onset. Sagey (1986) provides a representation ofprenasalized stops below the skeletal tier based on Clements' (1985) feature-geometry framework. Sagey proposes that prenasalized stops have one Toot node and the features [+nasal] and [-nasal] are dominated by a single soft-palate node.

(2) PreIIMa]jzed S1O~S (Sagey 1986) x 1 root lar(f1-contl ,-laI. A sOft-pel. place A [+nasal] [-I\83ell ( Sagey main tains Steriade' s (1982) position that contour segments are represented as

branching terminal features. According to Sagey, distinctive features ar~ phonologically ordered on any given tier. Hence [+nasal] is phonetically realized before the [-nasal]. An alternative is to suppose that adjacent specifications of the same feature are each associated with its own class node. Sagey rules out the latter option by postulating that branching c1ass nodes are prohibited.

(3) Comour segments may branch for terminal features only. No branching class nodes are allowed. (Sagey 1986;50)

Sagey provides three arguments for (3), only two of which are discu3Sed now. The first argument is from the behaviour of tones which also form contours. Sagey considers

tone to he dor.unated by the laryngeal node, yet tone can participate in harmony processes that are not blocked by intervening laryngeal nodes. However, there is an alternative explanation. The fact that intervening laryngeal nodes do not block tonal spreading can be c accounted for if tone is considered to occupy its own tier (A&P 1986) or dominated

28 o directly by the skeletal point. Furthennore, if Sagey is correct and tone is dominated by the laryngeal node, it is not possible to account for tone stability which is the result of deleting the tone bearing unit. Tone phenomena are more adequately accounted for by placing tone at a level superior to the root node. The second argument Sagey uses against branching c1ass nodes is motivated by prenasalization in Guarani. Prenasalized stops in Guarani are considered to be derived by interaction with nasal spreading (Goldsmith 1976, van der Hulst and Smith 1982).

(4) Guarani (van der Hul3t and Smith 1982)

a. [ +l\M] [-nu] (+na3] [-na" ] L_<Ç:O ",,-'fi. l , , l , ne-tupI. --> ne-tup a. [ndetupa] 'thy bed 1

b. [+1\.83] [+na:s] [-c'nas] [+I183]

l", ---~,""',.,...,. l , ne-tupI. --> ne -tupI. [nêtüpi] 'thy gcd 1

The representation of prenasalized consonants in (2) accounts for prenasalization in Guarani because [-nasal] attaches to a segment already specified as [+nasal]. However, Uuarani can also be used as evidence against Sagey's representation of prenasalized

consonants. Piggott (1988a) argues that prenasalized stop formation in Guarani is not

typical because most cases of nasal spreading do not ereate a prenasalized stop. Piggott claims that prenasalized stops that are formed as a result of nasal spreading are more arJequately explained by abandoning Sagey's prohibition against branching class nodes.

The next section discusses an alternative to Sagey's prohibition against branching class nodes. 2.2. The Representation of Prenasalized Stops on the Root Tier

In the feature-geometry assumed here, Sagey's representation of the nasal contour

in a prenasalized stop would he (5). (5) x ( 1 root A [+nasal] [-nasal] Piggott (1988a) proposes an alternative to Sagey's representation of contour segments which is stated in (6). (6) Anode may immediately dominate no more than one value for a given feature. ln the case of the representation of prenasalized stops, this means there are two root nodes

are linked to one skeletal point as shown in (7).

(7) x A root root 1 1 [+nasal] [-nasal] Piggott argues that the interaction of prenasalized stops and nasal spreading is evidence for the representation in (7). Further evidence for (7) is drawn from gemination in Sinhalese and the representation of prenasalized fricatives. The evidence form each phenomena is discussed separately. 2.2.1. Evidence for Iwo Root Nodes

2.2.1.1. Autose~mental Spreadin~ As previously mentioned, Piggott (1988a) rejects Sagey's representation of prenasalized consonants because there is no evidence that prenasalized consonants are derived from nasal spreading. Piggott shows that nasal spreading in Malay, Warao, Sundanese, and Capanahua does not nasalize the opaque consonant. The absence of derived nasal contours is not surprising if the opaque segment blocks spreading and cannot be a target of spreading. Piggott captures this in a thellry of spreading that prohibits the spreading of a feature to a position already specified for that feature. c '. .

(8) Spreading Theory (in part) (Piggott 1988) a. anode [or feature] (X) may spread only to a position o not specified for X. b. the spreading of a node [or feature] (X) may be mested only by a position specified for (X).

Guarani now appears to be a counterexample to (8) because [nas~l] spreads to a segment specified for [nasal]. Piggott reanalyzes GUf.rani so that the Spreading Theory is preserved. Piggott proposes that the In/ of the morpheme ne is represente"d as one skeletal point dominating two root nodes, one (the leftmost) is specified as [+nasal] and the other (the

rightmost) is not specified for [nasal]. The leftward spreading of nasality in (4b) spreads to the unspecified root node creating a configuration where one skeletal point dominates two

[+nasaI] root nodes. The two identically specified root n~jes are then reduced to one root node by the OCP. Prenasalization occurs in (4a) as a consequence of the absence of nasal spreading. Piggott rejects the [-nasal] spread mIe assumed by van der Hulst and Smith. The oral stem cannot spread [-nasal] in Piggott's analysis: the [-nasal] specification is inserted as a default value. Therefore, the unspecified root node of /n/ of ne in (4a) is specified as [-nasal] by redundant feature value insertion thus creating the configuration of a prenasalized stop. 2.2.1.2. Prenasalized Fricatives

According to Sagey's representation of a contour segment, prenasalized fricatives

and prenasalized affricates would he indistinguish~ble. In the feature-geometry used here, both would have the structure as shown in (9b&d).

(9) a. affIicate b. prenasall2ed affrica'œ x x 1 1 rt. rt ~ ~ :.-1. [-nas] (+xw] ,-1. (-n&3] [-cont(f'[ +cont] [-cont(f'[ +cont] place place ( c. fricative d. p:r:enualized fricative x x 1 1 n. ft. ~ ,-1.'" [-nas] [+nu] ,-1. [-lW] ll+cont1 [-cont(l'[ +cont] place place

There is no way to associate the [cont] contour with the [-nasal] feature to get a

prenasalized affricate and there is no way to associate only the [+cont] to the [-nasal]

feature to get a prenasalized fricative. This problem is relevant in Rundi where Herbert

(1986) reports there are both prenasalized fricatives and prenasalized affricates.

No such problem arises with Pigg(\tt's.. proposal that one node cannot dominate

more than one value for a given feature. The prenasalized affricate and the prenasalized

fricative each have a unique representation.

( (10) prenasalized fricative pren.asalized af!rica1e x x A A l't. ft. ft. rt (+DM~ ~~&I (+nu&o/j)nu&1 ,-Jar. ,-Jar. ,-Jar. ,-Jar. ~ ~ (-cvnt] place (+cont] (-cent] place (+cent]

2.2.1.3. Sinhalese Gemination

Geminates in the feature-ge,metry are represented by two skeletal points

dCtminating one root node, as shown in (11).

(11) x x

""-/rt.

(

32 o Given Sagey's representation of prenasalized consonants, a geminate prenasalized consonant should surface as NeNe. However, this is not correct. Consider the behaviour of prenasalized stops in Sinhalese where prenasalized stops and single segments altemate with the corresponding nasal-stop cluster and geminate, respectively, as a result of adding the singular morpheme.

(12) Sinhalese (Feinstein 1979) singular plural root gloss a. kanda kafldu kandw hill b. hombe hotnbu hombw chin c. potte potu potw core d. redda redu redw cloth

Since prenasalized stops and the simple segments bath have one root node linked to one skeletal point, there is no reason why the prenasalized stops do not surface as NeNc in

the singular. Prenasalized consonants. however, never geminate as NeNc (Herbert 1986).1 The NcNe geminate is prohibited in an analysis in which prenasalized consonants have two root nodes because the NeNc geminate would violate the Crossing Constraint.

(13) ... x x C>

The NC-NC altemation in Sinhalese follows from the representation of the NC sequence and the representation ofthe singular morpheme -a (which becomes [a]) given

1The [NNCJ clusters in Fula are derived from a nasal followed by a stop as shown (i) (Paradis 1987). Although a geminate configuration is created, Le. the nasal is attached to two skeletal points, it is not the result of a gemination mie such as C --> CC as in Sinhalese. i.

nasal 5top , Il ,- 1

in Rosenthall (1988). The singular morpheme introduces a skeletal point that must link to the root node tier in accordance with one-to-one, left-to-right association. Furthennore,

Rosenthall proposes that the post-nasal obstruents in the roots of (12) have r~t nodes which are not associated to the skeleton. These root nodes will associate with the skeletal point introduced by the singular morpheme. The obstruent It! of Ipotw/, for example, links to one skeletal point by one-to-one, left-to-right association, but it also links to the introduced point, hence fonning a geminate.

(14) singular fonnation2 (Rosenthall1988) singular morpheme: x x 1 a

a. /kandw/--> [kanda] xxx xx xxxx x 1 1 1 1 1 1 kan d w + a --> kan d w a

b. Ipotw/-- > [potta] ( x x x x x x x x x x 1 IlL, ~ 1 pot w+ a --> pot w a The ev id en ce for Piggott's alternative to Sagey's representation of prenasalized consonants is convincing. A representation of prenasalized consonants that has two root nodes enables Piggott to provide a more explanatory theory of spreading. Further evidence

is provided by languages in which prenasalized fricatives and prenasalized affricates are contrastive. Such a contrast is impossible with Sagey's representation. LastJy, the behaviour of prenasalized consonants in gemination environments is properly explained by a representation of prenasalized consonants where there are two root nodes. Thus far, the only aspect of the representation of prenasalized consonants that has been argued for is mat prenasalizcd consonants contain one skeletal point and two root

2The Id! and the Iwl in (14a) and the Iwl in (14b) link to the introduced skeletal point. The It/ in (l4b) must also link to satisfy syllable structure constraints. The !wl in both cases c delink due a restriction prohibiting branching onsets. See Rosenthall (1988) fordetails.

34 o nodes specified as [+nasal] and [-nasal] respectively. This aspect of the repres~ntation is sufficient for the discussion of the unification stage of prenasalized stop fonnation. 2.3. Unification (Prenasalized StoP Fonnationl Unification, i.e., Herbert's process that merges a nasal-obstruent sequence into a prenasalized consonant, is explained by the principles of syllabification. Consider the data from Ndali. (15) Ndali (Vail1972) a. /iN+ puno/ [imbuno] 'nose' b. /iN+ nyenyel [inyenye] 'bee' c. /iN+ oshi/ [inoshi] 'sheep'

The underlying representation of the prefix iN- contains two skeletal points. The Initial Syllabification Principle forces the skeletal point of the nasal of the prefix and the skeletal point of the word-initial obstruent under the same onset. However, Ndali does not allow

onsets to branch. Therefore, one of the skeletal points deI etes and the nasal and the word initial obstruent are attached to one skeletal point thus merging the nasal and the obstruent into one segment. Phonetic adjustment causes voice and place assimilation.

(16) [imbuno] (relevant structure only) x x x x x x --> (J A 1 N+ P u n 0 o N 1 1 x x a A Il N P u no Cross-linguistically, it is common for the nasal segment of the prefix to delete when the fOot is nasal-initial. (17) Swahili (Welmers 1973) a. /N+ mende/ [mende] 'cockroach' Umbundu (Schadeberg 1982) b. /N+ mola! [mola] '1 see' Shona (Fivaz 1970) c. /N+ nomwe/ [nomwe] 'seven' ( Nasal deletion before nasal-inital roots is easily explained. The two adjacent nasal segments violate the OCP (cf. Yip 1988). Therefore the prefix deletes. The Initial Syllabification Principle also accounts for the presence of the nasal

segment before vowel-initial roots, as in (ISe). Since the ISP dictates that vowel-initial stems have an onset position to ml, the skeletal point of the nasal of the prefix is syllabified as the onset. The surface form of the nasal before vowel-initial stems varies across languages. A palatal or velar nasal appears in sorne languages whereas other languages have a default prenasalized stop, usually velar or palatal. 1 assume the prenasalized stop that surfaces before vowels receives itf place features via redundancy roles. (18) Ndali (Vai11972) a. IiN+ oshi/ [inoshi] 'sheep' Venda (Ziervogel et al 1972) b. /N+adza/ [itJgadza] 'spreading out' Swahili (Welmers 1973) c. /N+ukil [nYuki] 'bee' ( The analysis of unification proposed here follows from the principles of syllabification and association and the appropriate underlying representation of the prefix. This analysis of unification maintains Herbert's (1986) clairn that prenasalized stops may he derived from nasal-stop clusters, but the analysis here goes further because without appealing to rules it accounts for the deletion of the prefix before a nasal-initial root. The above discussion, however, does not attempt to explain the phonetic adjustment processes that occur during unification, e.g. the place and voice assimilation. These assimilation processes are examined in the remainder of this chapter. 2.4. More on The Representation of Prenasalized Stops Thus far, the discussion of prenasalized stops has concemed only the repesentation at the skeletal and rOOl node tier. These tiers have been sufficient for the discussion of unification and the deletion of the nasal prefix before nasal-initial roots, but more must be known about the structure of prenasalized stops in order to account for the phonetic

( adjustment processes which take place during unification. 2.4.1. The Contour Node Condition Having established that prenasalized stops must be represented as one skeletal point dominating two root nodes and using the feature geometry proposed in section 1.2. 1, prenasalized voiced stops have the following representation.

(19) x A ft. rt. (+IIU~IIUa\1 (+voice] s-Jar. ~ place (-cont]

Sagey's (1986) objection to branching class nodes is partially motivated by the lack of a constraint to limit possible contour segments. For example, if the root node can fonn a contour, then any two segments can be dominated by one skeletal point. It is possible however to propose a restriction concerning possible contour segments using feature geometry proposed in 1.2.1. First, the notion of a contour segment must he defined. Given Piggott's claim that a class node may immediately dominate no more than one value for a given feature, then contours, which dominate [aF] and [-aF], must be defined as one class node immediately dominating two class nodes which dominate [aF] and (-aF]. (19) satisfies the definition of a contour segment because the skeletal point dominates two root nodes which dominate [+nasal] and [-nasal], respectively. Furthermore, (19) reflects the fact that the nasal and oral components of a prenasalized voiced stop (the least marked prenasalized consonant) agree for place of articulation, voice, and continuancy. The following condition on representation captures these phonetic agreements and also restrlcts the set of possible contour segments. (20) Contour Node Condition (CNC) (first approximation) Given the configuration: A c A B C B,C on tier T must dominate the same nonempty set of subordinate class nodes. The statement of the CNC stipulates that the set of c1ass nodes must he nonempty. This means that class nodes Band C must dominate at least one class node. This stipulation suffices for the present discussion but it is derived in the final form of the CNC. The CNe ensures prenasalized (voiced) stops agree for voicing, continuancy, and place of articulation and that affricates have a single place of aniculation. The CNC aIso predicts (21a,b) are possible contours, but (21e,d) are not possible contours.

(21) 8.. x b. X c. • x A 1 1 ft. It l't. 11.

l'''~'''PI ,-Jar.A s-la.r. ~./1"1· c ,-Jar. 1oc.Pl.L.·I-e<.Pl 1 loc.Moc.Pl 1 place place place d.· x 1 11

'1s-Jar. A place place 1 1 [~P) h)I.P)

The articulator nodes, which are class nodes, are excluded from forming contours because aIl articulator nodes are tenninal c1ass nodes. Hence there is no set of class nodes they can dominate. (21a) is the representation of prenasalized stops as previously mentioned and (21 b) c is the representation of affricates. (21c and d) are ill-formed contours because the laryngeal

3R '0 node and the place node are tenninal class nodes. Hence they do not dominate any other class nodes. The CNe as stated in (20) is too strong because it excludes prenasalized voiceless stops and prenasalized fricatives from the set of contour segments. The strong version CNC, as stated in (20) (and revised in,{24», is assumed for the present discussion and these exceptions to the CNC are discussed in detail in section 2.7. Prenasalized affricates, as shown in (22), are also counterexamples to the CNC.

(22) (relevant structure only) x A rt. rt.

1+nas~nase.ll ,-laI. ,-laI. ~ [-cent] place [+cont]

Although the [-nasal] root node dominates the sarne supralaryngeal node as the [+nasal] root node, the latter root node does not dominate the [+cont] supralaryngeal node. The CNC, on the other hand, correctly predicts the following is not a possible contour segment.

(23) (relevant structure only) ,.. x A rt. rt.

1+nas~nase.ll ,-bar. ,-laI. ~ [-contI place (+cont)

The [-nasal] root node in (23) does not dominate the [-cont] supralaryngeal node, creating

an ill-formed contour that is excluded by the CNC. However, since (22) must he permitted, ( there is an asymmetry for which the CNC cannot account. This asymmetry can be explained by assuming that there is a relation within the the representation of contour segments that allows (22) but not (23). Shaw (1987) recognizes that contour structures should exhibit sorne properties derivable from branching. Shaw c1aims that these properties can be related to headedness which defines one branch as more salient than the other. The saliency asymmetry can be defined in tenns of the relation between the head and the non-head. Shaw proposes the right branch ,of a contour segment functions as the head, thus accounting for the lack of structure preservation in reduplication in Nisgha. Shaw accounts for the fact that only the fricative component of an affricate reduplicates by proposing only the head reduplicates. 1 propose here another characteristic of the head is that it must dominate the same class nodes

as the non-head. The CNC Can now be restated in terms of headedness. (24) Contour Node Condition (revised fonn) Given the configuration A ( A B C B, Con tier T where C is the head, every class node dominated by the non-head must be dominated by the head. The é.symmetry shown in (22) and (23) is now accounted for by the CNC. Prenasalized affricates are licit segments because the [-nasal] root node is the head, therefore, it must dominate the same class nodes as the non-head. (23) is ill-fonned because the [+nasal] root node is not the head and it dominates a class node that is not dominated by the head. Right headedness, as used here, accounts for the particular properties associated with the right branch of contour segments. For instance, the obstruent (the right branch) of a prenasalized stop always de termines the place of articulation of the prenasalized stop. ---- Hualde (1987) supplies similar evidence for the right headedness of affricates by showing that the place of articulation of the three affricates in Basque is determined by the articulation of the fricative component. Therefore, the notion of right-headedness can be ( used to determine the articulation of contour segments.

40 o The head of a contour requires two properties. First the head must be the right bran ch of the contour and second the right branch must [-nasal -son] for root node contours and [+cont] for supralaryngeal node contours. The fact that [-nasal -sonorant] and [+cont] are the features associated with the right branch of the contour is supplied by universal grammar. There is an apparent problem with the choice of heads because [-nasal] and [+cont] are considered the default values for the respective features. Therefore, it is conceivable that there is a point in the derivation where the head does not have its feature value. This problem can he avoided by assuming that the right branch is the necessary condition for headedness and the feacure valnt! is the sufficient condition. Phonological

processes that do not refer to the feature values of the head can he sensitive to headedness, i.e. sensitive to branching, when the feature value is not present. Postnasalized stops, which are found in a number of languages, imply headedness might be parameterized because postnasalized stops would have [+nasal] on the right branch. There are reasons, however, for excluding postnasalized stops from the set of contour segments. First, postnasalization is never the result of unification, but only appears as a result of interaction with nasal vowels. Secondly, postnasalized stops do not QCcur underlyingly. Based on these observations, Piggott (1989) proposes that postnasalized stops have a representation different from the prenasalized stops that are derived from unification. It is now possible to delimit the set of possible contour segments. As a consequence

of the CNe, contours can only he formed at the root and supralaryngeal nodes. Therefore, aIl features dominated by the other class nodes cannot form contours. However, since the root node domina tes [consonantal] and [sonorant] and the supralaryngeal node dominates [strident], the set of possible contour segments is not exclusively limited to [nasal] and [cont] contours. In order to properly limit the set of contour segments, it must be stipulated that contour segments are restricted to [+cons] root nodes and contours formed by [strident] must be accompanied by a contour for [cont].

41 il

( Thus far, the CNC is able to capture the shared phonetic properties of contour segments and the CNC also delimits the set of possible contour segments. The CNC is also crucial in the account of the assimilation processes of the phonetic adjustment step of Herbert' s model. 2.5. Phonetic Acijustmeot Processes The fact that both components of a prenasalized stop often agree for voicing, continuancy, and place of articulation is unproblematic when the obstruent is a voiced noncontinuant. However, many languages have no phonemic voiced stops yet have only prenasalized voiced stops which are created by the obstruent asslmilating to the voicing of the nasal during unification. Furthermore, sorne fricatives assimilate to the continuancy of the nasal segment and beco;ne stops during unification. Along with these phonetic adjustment phenomena, a fricative or a voiceless stop oflen triggers the deletion of the nasal, or the obstruent becomes its nasal counterpart and the nasal deletes, i.e. coalescence. ( The linear rules for these phone tic adjustment processes, deletion, and coalescence are given below. (25) a. Voice assimilation C --> [+voice] /N_ b. Continuancy assimilation C --> [-conl] /N_ c. Nasal deletion N --> f/J/_C

d. Coalescence +cons ] +cons +nasal +cons ] --> f/J ( [ +nasal ] [aplace aplace

1 2 1 2 On the surface, these processes have nothing in common. However, all four processes are shown below to follow from the CNC, the theory of underspecification (Archangeli 1984, A&P 1986), and Piggott's (1988) Spreading Theory. (

42 2.5.1. Post-nasal Yoici0K Many languages have a process called 'post-nasal voicing' which voices an obstruent wh en the obstruent is preceded by a nasal. Consider the following data fro~ Kikuyu.

(26) Kikuyu (Annstrong 1967) a. /N+ tem-al [lldemeete] 'eut Ip. perf. ind.' b. /N+ tom-al ["domeete] 'send' C. /N+ kom-al [lJgomeete] 'sleep' d. /N+ ker-al [lJgereete] 'cross'

Since stops are voiced only when preceded by a nasal and voiceless elsewhere, the Kikuyu stops need not he specified for [voice] because the the value for [voice1 is predictable. Post-nasal voicing is shown here to follow from the theory of underspecification in Archangeli (1984) and A&P (1986). Consider the phonemic inventory of obstruents in

Kikuyu (Armstrong 1967) and the proposed underlying specification for [voice] and [nasal].

(27) Kikuyu P t c k ~ 3 'Y N(= m,n,Jl,g) [voice] + + + [nasal] + The specification for [voice] given here is consistent with Kikuyu phonology since stops are voiced only post-nasally and voiceless otherwise. Therefore, stops are not specified for [voice], but fricatives are specified as [+voice]. However, a Complement Rule must be established so that [-voice] is the default value. Post-nasal voicing is based on the premise that the stops which are not specified for [voice] receive their specification for [voice] from the nasal. Therefore, the default rule [+nasal] --> [+voice] applies as soon as the nasal is required to spread its laryngeal node. As a result of node generation discussed in chapter 1, no laryngeal node is generated for stops in Kikuyu because there is no feature, i.e.[voice], to license il. Post- ( nasal voicing, therefore, is simply the spreading of the laryngeal node from the nasal to the obstruent. (28) Post-nasal Voicing (relevant structure only)3 x.. ~~ " ~, " rt. rt. /l /'/~-l\Ual] [+nu~] l/ Jar. 1 [+voice)

2.5.2. Post-nasal Hardening Many languages also have a process called 'post-nasal hardening' (Herbert 1986) that apparently despirantizes fricatives during unification. Consider the data from Kikuyu in (29). (29) Kikuyu (Armstrong 1967) c a. /N+ ~or-al [mbureete] 'lop off 1p. perf. ind.' b. /N+ "(Or-al [I,)goreete] 'buy'

The labial and velar fricative must surface as a fricative in all environments except post nasally. This can be related to the fact that the fricatives are unspecified for [cont]. Tt.e

proposed specification for Kikuyu obstruents is given in (30).

(30) Kikuyu P t c k ~ 3 'Y N(= m,n,Jl,I,)) [voice] + + + [cont] [nasal] + Complement Rule: [ ] --> [-voice] Default Rules: [ ] --> [-nasal] [ ] --> [+cont]

Further support for the feature specification in (30) is drav n from the universal propenies of phonemic inventories. Kikuyu has a series of voiced fricatives, but no voiced stops

3Hereafter, the right root node is shown specified as [-nasal] for exposition only. The right c roo.t node is not specified for [nasal] during unification.

44 ·0 which is unusual. The voiced segments, as proposed in (30), are nOL underlyingly fricatives because they are not specified for [cont]. Therefo:e, Kikuyu has only stops at the phonemiclevel. Post-nasal hardening, like post-nasal voicing, is a case of spreading anode to a segment that is not specified for that node. The default rule [+nasal] --> [-cont] must apply at the same time as spreading so th,at the nasal can spread the supralarygneal node to the voiced fricatives.

(31) Post-nasal Hardening (relevant structure only) X.. , " ,, " " l't. rt. /1 //'l-nasal] [+I\83~1 l/ ,-Jar. 1 [-co nt] 1\ -J 2.5.3. Nasal Deletion Nasal segments often de1ete during unification when followed by a nasal or by certain obstruents. Consider the fcllowing data.

(32) Ndali (VaiI 1972) a. /iN+ puno! [imbu no] 'nose 9/10 class' b. /iN+ tunye! [indl~nye] 'banana' c. /iN+ kunda! [i%,lgunda] 'dove' d. !iN+ ~ale! [imbale] 'plate' e. !iN+ fuwa! [ifuwa] 'hippo' f. /iN+ satu! [isatlil] 'python' g. !iN+ nyenye! [iny<:nye] 'bee' Swahili (Welmers 1l973) h. /N+ boga! [mlxllga] 'vegetable' 1. /N+ devu! [rlde'lfu] 'beard' J. /N+fimbo/ [fimbo] 'stick' k. /N+ simba! [sim!!>a] 'linn' Kikuyu (Annstrong 1967) 1. /men-a! [meiieete] 'know 1p. perf. ind.' The two deletion processes are stated in (33). a

(33) Nasal Deletion ( a. IN+ m,n, fJ/--> [m,ii,fJ] ex. (32 g,l) b. IN+ f,s,sl --> [f,s,s] ex. (32 e,fj,k)

(33a) is explained by the OCP as discussed in section 2.3. The deletion process identified in (33tb) must now be explained. Although deletion triggered by fricatives seems natural, there is no explanation for the fricatives /f,s,sl to trigger deletion and the fricatives IP, yI to

trigger hardening. Post-nasal voicing and hardening (32a-d) in Ndali, like in Kikuyu, are accounted for by proposing the stops are not specified for [voice] and the fricatives are not specified for [co'nt]. The proposed underspecification of Ndali obstruents is given in (34). (34) Ndali (obstruents only) p t c k f s S ~ 'Y N(m,n,jl,fJ) [voice] + + [cont] -- - [strident] + + + [nasal] + Complement Rule: [] --> [-voice] Default Rules: [ ] --> [+cont) ( [ ] --> [-strident] [ ] --> [-nasal] The only difference between the phonemic inventories of Ndali and Kikuyu is that the latter inc1udes voiceless fricatives. These voiceless fricatives are also the only strident phonemes hence they are specified as [+strident]. Nasal deletion is not actually the deletion or the delinking of the nasal segment, but rather the prohibition of the [+nasal] root from unifying with the obstruent. Now, this

prohibition against association in Ndali must be related specifically to If, s, SI. Since t~ese

fricatives are [+strident], they generate a supralaryngeal node. Therefore,the supralaryngeal node of the nasal cannot spread. Since the nasal cannot spread its supralaryngeal node to the obstruent, the linking of the nasal root node to the skeletal point dominating the strident

fricative: would creaW a configuration that violates the CNe because the two root nodes would flot dominate the same supralaryngeal node. The nasal does not link to avoid ( violating the CNC. Thus it does not surface.

46 ----.---.. ----r~.----- .. ----.-~-- .. ----

o (35)

s-laI. ,-Jar. / [-cant] I!+s1lidentl place

The use of [strident] as the trigger for nasal deleêon is not arbitrary because it appears that strident fricatives never undergo post-nasal hardening. Nasal deletion can also be triggered by voiceless stops.4 (36) Venda (Ziervogel et al. 1972) a. /N+ pala/ [pala] 'scratching' b. /N+ tsumbulol [tsumbulo] 'rolling in dust' c. /N+ kakol [kako] 'fault' Swahili (Welmers 1973) d. /N+pembel [pembe] 'hom' e. /N+ tembo/ [tembo] 'elephant' f. /N+ kukul [kuku] 'chicken' The deletion facts in Venda and Swahili follow from the CNC provided that stops are specified as [-voice] :md that [+voice] is the default value, unlike Kikuyu and Ndali. Since Venda and Swahili stops are specified as [-voice], the [+voice] laryngeal node of the nasal cannot spread to the stop. The two root nodes in this case do not dom\nate the same laryngeal node. Hence the nasal does not link thus avoiding a violation of the CNe. An alternative to prohibiting the linking of the nasal segment that precedes a strident fricative is to form a prenasalized affricate, as seen in Venda and Kihungan. (37) Venda (Z;ervogel et al 1972) a. /N+ vule

41n Swahili and many other languages the voiceless stops are aspirated as a reult of nasal deletion. 1 do not propose an account of aspiration.

47 b. /N+ zwirnaJ [Ildzwima] 'hunting' Kihungan (Clements 1987) c c. /luN+ vaatisl [lumbvaatis] 'dress' The fonnation of a prenasalized affricate is one type of intrusive stop formation discussed in Clements (1987). Stop intrusion, therefore, is a strategy used by sorne languages to avoid violating the CNC. The prenasalized affricate is formed by spreading the supralaryngeal node of the nasal to the root node of the fricative. The spreading of the suprarlaryngeal node in this case must he considered a marked spreading process. This spreading process can conform to Piggott's Spreading Theory because the supralaryngeal node of the obstruent dorninates [strident] and the supralaryngeal r.ode of the nasal dominates [cont]. Hence the spreading of the supralaryngeal node need not be blocked because the two supralaryngeal nodes are identically specified. Prenasalized affricate

formation in (38) is similar to the representation of intrusive stop formation in Clements (1987).

(38) c x , .. .. , , .. , ,,

Languages with intr'.iSlVe stops also have voiceless stops becorning affricates as a result of prefixation.

(39) Venda a. /N+ fula! [pfula] 'pasture' b. /N+ sengaJ [tsenga] 'hear a court case' Kihungan c. /liN+ sey! [lutsey] 'mock' d. /luN+ fut! [lupfut] 'pay' c

4R o Clements (1987) proposes the node dominating [continuant] (the oral cavity in Clements (1987» spreads to the stop as in intrusive stop fonnation. A role then causes the nasal

segment to delete, but the obstruent surfaces as an affricate because it has the nasal' s specification for [cont]. The affrication of voiceless stops is accounted for by the CNe. Although intrusive stops prevent CNC violations at the supralaryngeal tier, a voiceless segment will create a CNC violation at the laryngeal tier. As in Clements' analysis, the nasal root node delinks, but the nasal' s supralaryngeal node remains attached to the obstruent creating a voiceless affricate.

(40)

ft. ft. x 1 [+l\83al~ 1/j2-nasal1 --> 11. ,-- 1 laI. MU. 1 1 [ +~lCe 1 / [-vaice ) ,-Jar. ,-Jar. IÎ\[-VOlee] ,-laI. ,-Jar. [-con~rident] [-cont~+'lIidentl place Languages with intrusive stops differ from languages with deletion insofar as the fonner have a language-specifie process to avoid CNC violations at the supralaryngeal node. The process is to spread the supralaryngeal node from the nasal, as shown in (38). CNe violations at the laryngeal node, however, cannot be avoided. Therefore, the nasal

cannot link aft~r supralaryngeal node spreading thus crearing a voiceless affricate. 2.5.4. Coalescence Schadeberg (1982) claims that voiced stops in Umbundu only cccur when a continuant is preceded by a nasal, i.e., the voiced stop of a prenasalized stop is the result of post-nasal hardening. (41) Umbundu (Schadeberg 1982) a. /N+ vanya/ [mbanya] '1 look' b. /N+ fela! [fela] '1 dig' c. /N+ seva/ [seva] '1 cook' ( d. /N+ popya/ [mopya] '1 speak' e. /N+tuma/ [nurna] '1 send' f. /N+ kwata/ [!.)wata] '1 take'

Post-nasal hardening is seen in (41a). (41b&c), on the other hand, are cases of nasal deletion which have been observed in other languages. (41d,e,&f) are interesting. Voiceless stops should either undergo post-nasal voicing when not speclfied for [voice], as seen in Kikuyu and Ndali, or trigger nasal deletion when specified, as seen in Venda and Swahili. Instead, voiceless stops-in Umbundu trigger coalescence. Voiceless stops in Kihehe and Si-Luyana also trigger coalescence. (42) Si-Luyana (Givon 1970) a. /N+ pokol [moko] 'vulture' b. /N+ tabi/ [nabi] 'prince' c. /N+ kukul [!.)uku] 'chicken' d. IN+ supal [supa] 'soup' Kihehe (Odden and Odden 1985) e. IN + teeful [neefu] 'reed mats' f. IN+ kaanzi/ [!.)aanzi] 'walls' ( "the nasal deletes before voiceless fricatives" (p. 498n.) (43) shows the specification of features in Umbundu from which hardening, deletion, and coalescence can be derived. (43) Umbundu pte k v f s N(= m,n,jl,!.) [voice] ---- [cont] --- [strid] + + [nasal] + Default Rules: [] --> [+voice] [] --> [+cont] [] --> [-strident] [] --> [-nasal] The approximant Iv 1is not specified for [cont] hence it can undergo post-nasal hardening. The fricatives Ifl and Isl are specified as [+strident] therefore a CNe violation occurs during

unification and the nasal cannot link, as in Ndali. Coalescence, like nasal deletion, is the result of prohibiting a root node from linking c to the skeleton. The [-voice] specification of the Umbundu stops creates a CNe violation o because the [-nasal] and [+nasal] root nodes do not dominate the same laryngeal node. Since prenasalized stops consist of two root nodes linked to one skeletal point, there are two association lines that may be prohibited from linking to avoid a CNC violation.

Coalescence is the result of two processes: 1. nasal-stop assimilation and 2. prohibiting the [-nasal] root node from linking to the skeleton. As a result, the [+nasal] root node remains with the place features of the obstruent. (44) Coalescence (relevant structure only) x "A'LJ , , 1''\,

The fact that there are two ways to resolve CNC violations implies there might be a parameter that can detennine whether a language resolves CNC violations by prohibiting the linking the right or 1eft root node. Obviously this is not al ways the case since Umbundu

exhibits both phenomena. Furthermore, if there is a deletion/ coalescence parameter, there

should be other phenomena for which it accounts. Since 1 do not know any other effects of

nasal deletion/coalescence, 1 do not c1aim there is a parameter that dictates linking.

However, there might be a markedness convention for resolving CNC violations. The

difference between Swahili and Umbundu, for example, is that Swahili has only the unmarked form of resolving CNC violations which is to link the head. Umbundu has the unmarked form of resolving CNC violations created at the supralaryngeal node, but has a language-specifie constraint that links the lefttmost root node wh en a CNe violation is created exclusively at the laryngeal node. ; ,

( Coalescence applies to both voiced and voiceless stops in Uma Juman (a Borean language). (45) Uma Juman (Blust 1977) a. /n+ pugut/ [mugut] 'to rob' b. /n+ bagi?/ [magi?] 'to sbare' c. /n+ tadav/ [nadav] 'to dive'

The CNC cannot predict the coalescence of the nasal and the voiced stop because the CNe is satisfied. However, Uma Juman does not have prenasalized stops which means that coalesence in Uma Juman is a simplification process which prohibits the linking of the obstruent to avoid a configuration in which there is a branching skeletal point. The two phonetic adjustment processes (post-nasal voicing, post-nasal hardening) and the two deletion processes (nasal deletion and coalescence), which superficially have nothing in common, follow from the CNC and underspecification. Post-nasal voicing and hardening occur wh en the obstruent is unspecified for [voice] and [strident], respectively. --( As a result, the nasal can spread its feature to the obstruent. Deletion and coalescence occur when the obstruent is specified for [voice] or [strident] and a CNC violation occurs. 2.6. More phonetic Adjustment Processes Thus far, only the behaviour of obstruents during unification has been examined, but other consonants behave in interesting ways during unification, For instance, glides and liquids become their corresponding obstruent stops, e.g. w ,.. b, y ... j, r - d. The hardening of these sonorants is interesting because there is a change in sonorancy as weIl as a change in continuancy. The hardening of glides and liquids is discussed separately because each has its own set of interesting behaviours during unification.

2.6.1. The Hamenjn ~ of Glides Consider the following altemation between glides and prenasalized stops.

(46) Swahili a. !N+watil [mbati] 'hut poles' Ndali b. !N+w/ [rnb] (no data given) C c. !N+ yukil [Jljuki] 'bee'

'\? o The fact that the labial and the palatal glides alternate with the labial and palatal stops is not too surprising because the glides and the stops have the same place of articulation. The hardening of glides to stops can he accounted for by post-nasal hardening as proposed in 2.5.2. Glides are not specified for [cont] because aU vocoids are redundantly [+cont]. Therefore, the nasal can spread its supralaryngeal node since the glide will lack a supralaryngeal node. A more interesting phenomenon that occurs during unification with glides is desonorantization which cannot be the result of spreading because the nasal itself is [+sonorant]. Desonorantization can he accounted for by following A&P's proposaI that [+sonorant] is the default value for [sonorant]. Post-nasal hardening creates a configuration where the nasal root node and the glide are linked to a [-cont] supralaryngeal node. The

default value [+son] for glides cannot he inserted because the only noncontinuant sonorants

are nasals. Therefore, thr only value for sonorancy that can he inserted is [-sonorant]. Hence the glide surfaces as an obstruent. Glides in sorne languages do not trigger any phonetic adjustment, but rather the default prenasalized stop surfaces. The absence of place assimilation indicates that the glides in these languages are semi-vowels which are in the nucleus as pan of a light diphthong.

2,6.2. The Hardenin~ of UQuids Liquids also harden during unification, as shown in (47). (47) Umbundu a. /N+larnda/ [ndanda] '1 buy' Kihehe b. /N+limi/ [ndimi] 'tongues' Kikuyu c. /N+ reek-ia/ [fldeeketie] 'finish 1p. perf. ind.' Shona d. /N+reful [fldefu] 'tall' The hardening of laterals poses two problems. The first is desonorantization and the second is delateralization. Before discussing the beha"iour of laterals during unification, the ( representatit:)n of laterals in the feature-geometry needs to mentioned. Levin (1987) proposes that [lateral] is dominated by the coron al node. Placing [lateraI] under the corona! node has significant consequences for the analysis of unification. If [lateral] is dominated by the supralaryngeal node as in Clements (1985), then a lateraI segment eould block post nasal hardening. But if [Iateral] is dominated by an articulator node, then post-nasal hardening is not impeded because the nasal and the lateral can he linked at the

supralaryng,~al node. However, as a result of unification with laterals, the [+nasal] root node dominates a [+lateral] coronal node. Since [+nasal] and [+lateral] are incompatible, the [+lateral] feature is deleted. It must be assumed that preserving contours by deleting features is only possible below the place node, otherwise the analysis of unification phenomena is undermined. Delateralization, therefore, is an artieulator node adjustment process. After [lateral] has been deleted, [-sonorant] must be assigned to the right root node, which is specified as [-nasal], because of the [-cont] specification of the shared ( supralaryngeal node. Another aspect of the hardening of liquids is the debate conceming the specification of [cont] for laterals: they are [-cont] in (Halle and Clements 1983), but are [+cont] in SPE. The specifkation of laterals for [cont] can he parameterized so in sorne languages they behave as stops and in other languages they hehave as fricatives. In languages such as Umbundu and Kihehe, liquids are not specified for [cont] and hardening occurs as predicted. Liquids in Mwera (a Yao language) trigger coalescence (Harris 1950) indicating a CNC violation. Therefore, liquids are assumed to he specified as [+cont]. (48) Mwera (Harris 1950) a. /N+ limi/ [nimi] 'tongues' b. /N+ limbal [nimba] 'hand pianos' Since [+cont] is the default value for [cont], nonlateralliquids are not specified for [cont]. The lack of specification of [cont] for /r/ seems universal beeause there is no language in which /r/ triggers nasal deletion or coalescence. The /r/ becomes [-sonorant] by c the mechanisms used for glide and lateral desonorantization.

54 o 2.6.3. De-implosion The following data from Shona show that implosive stops become plosives during unification.

(49) Shona (Fivaz 1970) v a. /N+ Sata/ [Illbata] 'pincers' b. /N+ Sudo/ [mbudo] 'mouse hole' c. /N+ âove/ [Ddove] 'wetplace'

Implo&ive stops must he underlyingly distinguished from voiced stops beca\&se there is a surface contras t, e.g. [bara]/ [Bara]. In order to account for de-implosion, a Complement

Rule must be established so that the plosives in Shona are underlyingly specified as [-constricted glottis] and the implosives are not specified for laryngeal features. Voiceless stops must be specified as [-voice] since these segments trigger coalescence. Voiced plosives trigger prenasalization as expected.

(50) a. /N+ pasa/ [masa] 'mats' b. /N+ tete/ [nete] 'thin (adj.)' c. /N+ buwo/ [mbuwo] 'horseflies' d. /N+diki/ [ndiki] 'small (adj.)' Shona stops have the following underlying feature specification.

(51) Specification of Shona stops. [voice] [constr. gl.] [nasal] p,t,k b,d,g S,ô m,n,!) + Default Rules: [ ] --> [+voice] [ ] --> [-nasal] Complement Rule: [ ] --> [+constr. gl.] The implosives will not generate a laryngeal node hence the nasal can spread its laryngeal node. The voiced stops, on the other hand, do generate a laryngeal node which should block laryngeal node spreading from the nasal. However, the laryngeal node of the plosive is not identically specified as the laryngeal node of the nasal at the time of spreading.

Therefore, the nasal can spread to the plosive. The final product is a prenasalized plosive. .,' 4.,

( The account of post-nasal hardening proposed for fricatives in section 2.5.2 can be used to account for the hardening of glides and liquids. After the nasal has spread its supralaryngeal node to the glide or liquid, "nly [-sonorant] can he assigned to the glide or liquid thus the sonorant surfaces as its nonsonorant counterpart. The hardening of implosives in Shona can he managed by assuming the specification of segments for [voice] and [constricted glottis] given in (51).

The CNC, along with the theory of underspecification and Spreading Theory, provides an analysis that unifies the phonetic adjustment processes and the deletion

processes that occur during prenasalized stop fonnation. The version of the CNC used thus far is admittedly problematic because il exclu des prenasalized consonants other than

prenasalized voiced stops. A revised CNC that accounts for the full range of prenasalized consonants is proposed in the following section. 2.7. Dther Prenasalized Consonants

( Although the CNe as stated in 2.4.1 is sufficient for the discussion of phone tic adjustment, il is too strong because it dictates that the subordinate set of class nodes

dominated by the non-head must be dominated by the head. This fonn of the CNC excludes the prenasalized consonants shown in (52). (52) a. prewalized voiceles, s'tOp b. prenasalized fI1ca1ive x x A A rt. l't. l't. rt [ +nasal],. [+IIIS~l1UoIl ,-Jar. ,-Jar. [ +voice)

,-Jar. [-cont]A/\. place [+cont] 1 place c c. pxenasalized Y'OiceJess frica.Üve o x A 11. rt. [+nase.1.] -nual] Jar. 1 lf· (+voJCe) [-Y'Oice] ,-laI. ,-Jar. ~ [-cont] place [+cont]

In each representation in (52), there is a class node that is dominated by the nonhead but

not dominated by the head. Note that the nodes that violate the CNe in (52) are the laryngeal node and the supralaryngeal node, but the place node is always dominated by the

head and the nonhead. The CNC must he revised so that the laryngeal and supralaryngeal nodes are option al members of the set of class nodes dominated by the two root nodes. (53) CNC (final form) given the configuration A o A B e B,e on rier T, C is the head, the nodes in the set S={ (lar., s-lar.) place }that are dominated by the nonhead must he dominated by the head.

The previous version of the CNC stipulated that the set of class nodes dominated by the

branching configuration had to he nonempty. This stipulation is unnecessary in the version of the CNC in (52). If Band C are the place or laryngeal nodes, there is no class node in S

that can he dominated by the branching configuration. Hence contours formed at the place and laryngeal tiers are ill-formed. Since the laryngeal and supralaryngeal nodes are oprional members of S, there are four possible subsets of S. (54) SI {lar., s-lar., place} S2 {s-lar., place} S3 {lar., place} S4 {place} These sets make interesting predictions concerning possible prenasalized consonant inventories. A language which has SI can only have prenasalized voiced stops because aIl

subordinate class nodes must he dominated by the two root nodes. A language which has J 1 ~

S2 can have prenasalized voiceless stops, but must allow prenasalized voiced stops because there is no way to eliminate the possibility that both laryngeal nodes are dominating

[+voice].5 A language which has 83 can have prenasalized voiced fricatives, ~ut must

~now for prenasalized voiced stops because there is no way to eliminate the possibility that both supralaryngeal nodes are dominating [-cont]. Funhermore, sinc(; the laryngeal node is a member of 53, the prenasalized fricatives and stops must be voiced. A language that has prenasalized voiceless fricatives has 54 and must allow for the full range of prenasalized consonants because there is nI') way to eliminate the two laryngeal nodes or the two supralaryngeal nodes from being identically specified as [+voice] and [-cont], respectively. Since there are only four subsets of S, there are only four possible prenasalized consonant inventories.

(55) prenasalized prenasalized prenasalized prenasalized voiced stops voiceless stops voiced fricatives voiceless fricatives SI yes no no no S2 yes yes no no c S3 yes no yes no S4 yes yes yes yes Examples: 51: Ndali, Kikuyu S2: Luganda, Kinywaranda 53: Swahili, Terano S4: Rundi The subsets of S appear to conform to a markedness convention. SI is the unmarked subset and S4 is the most highly marked. S2 appears to he less marked than S3 and both are less marked than S4. The markedness hierarchy of the subsets can he related to the propensity for contour segments to maximize the number of similar features. Therefore, SI is the least marked because all the features dominated by the two root nodes

are the same, but 84 is the Most marked because it allows dissimilar features for [voice] and [cont]. c SThe identicallaryngeal nodes collapse to one node by the Shared Node Convention.

58 o The CNC main tains that affricates must have one place node because the place node must he a member of S. Hualde (1987) suggests that there are nonhomorganic affricates,

thus contradicting the claim made by the CNC. However, not enough is known about nonhomorganic affricates to consider them here. It was noted at the beginl1ing of the chapter that [voice] contours only occuJ:' in the context of [nasal] contours. This observation follows from the CNC. Since the repI'f.:sentation of prenasalized consonants includes two root nodes, each root node dominates a separate laryngeal node. Therefore, there is no CNC violation because the laryngeal nodes arc not linked to one root node. Since only prenasalized consonants have two root nodes, the [voice] contour can only occur in the context of a prenasalized consonant. The final form of the CNC fOl'ces a reexamination of the motivation tor headedness within contour segments. One of the motivations for headedness given in 2.4.1 is to account for the asymmetry between prenasalized affricates (56a) and the ill-formed contour (56b).

(56) a. b. x III X /" A n. It. n. n. [+W~lIBSell [+W~lIBSall s-lar. ,-laI. s-laI. ,-laI. !'v~ A/\. [-cont] pliee [+contl [-cont] place [+cont]

This motivation for headedness is lost in a language that has prenasalized affricates (56a) and prenasalized fricatives (52b) because in these languages the supralaryngeal node dominated by the nonhead is not dominated by the head. However, there are independent reasons for the ill-fonnedness of (56b). Consider (57) in which a [+nasal] root node branches. ( (57) ... x 1 n.

,-lar.~+nasalJ ,-laI.

(-cont1AA place [+cont]

(57), which is the representation of a "nasalized affricate", is an impossible segment althou,gh it is well-formed with respect to the CNC. The fact that (56b) and (57) are impossible segments cannot he not the result of the CNC, but rather it is a result of the branching [+nasal] root node. The way to eliminate (56b) and (57) but preserve headedness is to restric.t branching root nodes to only [-nasal] root nodes. There has been no mention yet of the mechanism which allows contour segments to occur. For example, English and many other languages do not have prenasalized ( consonants and many languages do not have affricates. Since contour segments involve branching, their occurrence in a language is regulated by the configurational properties that the lahguage allows. Hence, the occurrence of contour segments can he related to a parame:ter which allows skeletal points or [-nasal] root nodes to branch. (58) Contour Segment Parame ter a. [-nasal] root nodes can branch (yes/no) b. skeletal points can branch(yes/no) English, for example, allows root nodes to branch, but does not allow skeletal points to branch. Therefore, English has affricates, bUl does not have prenasalized stops. These parame ter settings by themselves are insufficient because not all homorganic stop-fricative pairs are affricates in English. The fact that English has only alveolar and palatal-alveolar affricates is determined in the peripheI). 2.8. Conclusion The Contour Node Condition is introduced as a constraint on the representation of ( contour segments, but the Contour Node Condition, alûng with underspecification and

60 ,- .. '

o Spreading Theory, provides a unified account of the phone tic adjustment processes, deletion, and coalesence which occur during unification. Furthermore, the Contour Node Condition is the well-formedness condition needed in phonology to delimit number of possible contour segments. Representing prenasalized consonants as two root nodes dominated by one skeletal point is necessary to account for the behaviour of prenasalized consonants with respect to gemination and to distinguish prenasalized affricates and prenasalized fricutives. Furthermore, the interaction of nasal spreading and prenasalization is more adequately accounted for in an analysis in which prenasalized consonants have two root nodes. i, IL ; Il

c CRAPTER 3 Nasal-Obstruent Assimilation 3.Q. Introduction The assimilation of the nasal to the place of anicualtion of the obstruent was not

discu~~~d in the precetiing discussion of the phonological processes that occur during prenasalized consonant fonnation. The fact that the obstruent detennines the place of

articulation of pr~nasalized consonants is considered a result of headedness. However, the role of headedness in detennining the place of articulation has not been fully explored. The aim of this chapter is to provide an account of nasal-obstruent assimilation. There are three interesting characteristics of place assimilation within adjacent nasals and obstruents that a theory of assimilation based solelyon rules cannot explain: 1. the nasal is always the target, 2. the assimilation is always regressive, and 3. every language with nasal-fricative assimilation also has nasal-stop assimilation, but not vice versa. This c chapter accounts for the se charactenstics of nasal-obstruent assimilation by introducing a condition on the representation of hornorganic clusters. The discussion in this chapter concems place assimilation in ambisyllabic nasal-obstruent clusters rather than prenasalized consonants because the required condition on representation has consequences that affect the nature of linking in consonant clusters rather than the linking in contours. The chapter begins with a discussion of the nature of nasal-obstruent assimilation and sorne properties of homorganic nasal-obstruent clusters. The chapter continues with the introduction of the proposed condition on representation and the account of nasal-obstruent assimilation. The chapter ends with a discussion of the consequences of the proposais. 3.1. The Nature of Nasal-Obstruent Assimilation Targets of assimilation processes are given their features in one of two ways. One, the trigger spreads to the target, thus filling in a feature that is not there underlyingly. This feature-filling assimilation is used in the preceding chapter to account for voice and manner c assimilation in prenasalized stops. The second way a target can get its features is from a o feature-changing process whereby the target loses its underlying features in favour of the features resulting from assimilation.

(1) a. feature-fillinc 1>. featœe-chencinc • • l,/'# [+P] [+G]

Feature-filling assimilation (la) can easily be included in a principle-based approach to phonology because there is no need to appeal to rules other than rules of the fonn 'Spread [X]' meaning a trigger specified for a feature automatically spreads to a segment unspecified for that feature. Therefore, feature-filling assimilation is relatively cost-free to the grammar. Feature-changing assimilation (lb) requires the added process of delinking the feature underlyingly linked to the target, which should he more costly to the grammar because the spreading is not automatic. Many cases of nasal-obstruent assimilation, in particular nasal-stop assimilation,

might be considered to be instances of feature-filling in autosegmental phonology. For example, Chomsky and Halle (1968) claim that intramorphemic nasal-stop clusters in English are derived from.l stop preceded by an archiphoneme nasal. Words such as hand, lamp, and link have the underlying representations /haNdI, /laNp/, /liNk/, respectively, and, assuming that the segments are not linked underlyingly, the homorganic nasal-stop cluster is the result of the stop spreading ils place features to the nasal archiphoneme. A feature-filling analysis of nasal-obstruent assimilation can also be used to explain why in some languages only certain nasals undergo place assimilation. For example, Kiparsky (1985) notes that only the corona} nasal in Catalan and Axininca Campa undergoes place assimilation. This can he explained by assuming that the coron al nasal is unspecified for place features and the following obstruent spreads its place features to the nasal. The other ( nasals in these languages block place assimilation bccause these nasals are specified for place features. The feature-fùling analysis of nasal-obstruent assimilation seems correct in .cases in which only one nasal undergoes place assimilation and other nasals block assimilation. However, there are cases "n which aIl nasal segments undergo place assimilation. These

cases must be feature· ""nanging assimilations. For example, consider the data from Malayalam compounds, Diola-Fogny reduplication, and Spanish. (2) Malayalam (Mohanan 1986) a. Isam+ gittam/ [s8%Jgittam] 'music' b. lmiin +cantal [IllÏiJ1.canta] 'fish market' c. /baalan+pooyi/ [baa1ampooyi) 'the boy went' Diola-Fogny (Sapir 1969) e. ni+ gam+ gam [niga!Jgam] '1 judge' f. ku+ ban+ ban [kubamban] 'they seN' Spanish (Hanis 1984) intervocalic nasal preceding a stop g. [presumo] [presunto] ( h. [tine] [tinte] The focus of this chapter is on an account of feature-changing nasal-obstruent place assimilation. The proposed account of nasal-obstruent assimilation depends crucially on the representation of homorganic nasal-obstruent clusters, which is discussed in the following

section. The discussion henceforth concems nasal-stop assimilation in particular because it

is more common. Any proposai to account for nasal-stop assimilation will be extended to aIl nasal-obstruent assimilation. 3.2. Inverse Contours Feinstein (1979) states that the difference between homorganic nasal-stop clusters and prenasalized stops is the location of the syllable boundary. Thus Feinstein is able to capture many of the phone tic similarities of prenasalized consonants and homorganic nasal stop clusters. The representation of homorganic nasal-stop clusters should reflect the fact that the two segments agree for place of aniculation and continuancy. The only difference c

64 o between homorganic nasal-stop clusters and prenasalized stops is the difference in timing which is reflected in the number of skeletal points in the representation.

(3) 1. nasal-S1Dp clus1er b.prenasalized S1Dp x x x 1 1 A rt n. rt. It [+W~'tJ1IJleJ.l [+~WeJ.l s-Jar. s-Jar. 1 1 place place

Prenasalized stops, as discussed in chapter 2, are represented as one skeletal point dominating two root nodes, but homorganic nasal-stop c1usters, as shown in (3a), have two skeletal points. The representation of a homorganic nasal-stop cluster is a150 unique because [nasal] is the only feature that distinguishes the segments on the root tier. No other configuration with two skeletal points can have two root nodes dominating the same set of class nodes. The only other possibility of a configuration where there are two root nodes is when there is a distinction for [sonorant] or [consonantal] dominating the same set of cIass

nodes. Like contours, the root nodes in (3a) are assumed to be restricted to [-consonantal]. Configurations in which the two root nodes differ for [sonorant] are discussed in 3.3.1 where it is shown that these configurations contain two supralaryngeal nodes. Henee (3a)

is unique to homorganic nasal-stop c1usters. The relation between the class nodes in a contour is one class node dominating two cIass nodes which are on the same tier. The relation between the root nodes and the

supralaryngeal node in the homorganic nasal-stop cIuster is henceforth ca1led an 'inverse contour'l because there are two class nodes on the same tier are linked to one immediately

11 thank G. L. Piggott (p.c.) for this tenn. ( subordinate c1ass node. In other words, contours involve a one-to-many relation of class nodes and an inverse contour involves a many-to-one relation of c1ass nodes. Geminates, as shown in (4), are also inverse contours because there are two skeletal points linlced to one root node which is also a many-to-one relation.

(4) x x

"'/rt

lar.A s-lar. 1 place The laryngeal and the place nodes cannot he the superordinate class node in an

inverse contour because these nodes do not dominate any class nodes that can he the subordinate c1ass node. The supralaryngeal node can he the superordinate class node of an inverse contours in a limited numher of cases which are discussed in 3.3.1. The possibility ( of the laryngeal node as the subordinate class node of an inverse contour is also discussed later. For the present, the only inverse contours are homorganic nasal-stop clusters and geminates. The natural class formed by the similar representations of geminates and homorganic nasal-stop c1usters explains why geminates and homorganic nasal-stop clusters have many common properties. Both geminates and homorganic nasal-stop clusters display Inalterability (Hayes 1986), which is the inability of the components to act independently as the target of rules. For example, velar stops in Tigr..nya become spirants when adjacent to and preceded by a vowel. However, spirantization fails to apply to geminate velars regardless of the fact that the first half of the geminate satisfies the structural description. Homorganic nasal-stop clusters exhibit the same behaviour. Sanskrit n-retroflexion, which changes the first Inl to the right of Irl or IsI to (1)], does not affect nasals in homorganic

nasal-stop clusters. c (5) Geminates Homorganic nasal-stop clusters Tigrinya (Hayes 1986) Sanskrit (Schein and Steriade 1986) o /kAlbi/ [?axalib] PrJ)a /fAkk/'JN [ftJck/'JA] lq'andati

Schein and Steriade (1986) account for the inalterability of geminates and homorganic nasal-stop clusters by proposing the Unifonn Applicability Condition which states that the structural description of a rule must be satistied by all nodes mat are defined as the target. Geminates in Tigrinya are not affected by spirantization because only one of the two skeletal points satisfies the structural description which requires that the skeletal point he adjacent to the vowel. Homorganic nasal-stop clusters block Nati because the target of Nati must be a nasal and the obstruent linked to the nasal in a homorganic cluster does not satisfy the structural description. Geminates and homorganic nasal-stop clusters also display Integrity (Hayes 1986) which is the ability to resist epenthesis. (6) Geminates Homorganic nasal-stop clusters Palestinian (Hayes 1986) Kolami (Steriade 1982) nakJI [?akil] /rœlp/ [melaptan] nimml [?imm] /mind/ [mintan] In languages such as Sinhalese and Pali, among others, geminates and homorganic nasal-stop clusters are the only possible medial c1usters. Steriade (1982) proposes that medial clusters in these types of languages are pennissible when the segment in the coda is

segmentally linked to the segment in the onset thus enabling the coda segment to avoid stray erasure. The fact mat geminates and hornorganic nasal-stop clusters involve segmental linking accounts for their similar behaviour with respect to Inalterablity, Integrity, and stray syllabification. Interestingly, only complete and partial assimilation produce the properties associated with segmental linking. There is no reported case of the output of spreading a single feature or an articulator node that has these properties although the segments are

linked. Using the tenns developed here, the properties of segmentallinking are restrlcted to inverse contours. - ( A condition on the representaûon of segmentallinking is necessary to ensure that only the class of inverse contours display the propcrties attributcd to linkcd segments. The

Inverse Contour Condition, (7), is introduccd as a condition on the rrorescntaûon of linkcd segments. The Inverse Contour Condition states that scgmentallinking is restrictcd to c1ass nodes and ensures that only inverse contours occur in certain environments.

(7) Inverse Contour Condition (ICC) Two adjacent class nodes A, B on tier T, in domain D, must dominate the same class node C, where C;t: lar. (yes/no) The ICC cannat be allowed to pennit inverse contours to be fonned by two root nodes linked to the same laryngeal node which would force two segments ta have the same laryngeal fealures. Such an inverse contour displays Inalterability to roles which affect laryngeal features but does not display Integrity. Therefore, two root nodes linked to one laryngeal node must he excluded from the class of inverse contours. Once the ICC is set in ( a given language, then that language restricts c1usters in certain environments to inverse contours. If A and B are the skeletal points and C is the root node, that language a1lows

only geminates in domain D. If A and B are root nodes and C is the supralaryngeal node, that language has only homorganic nasal-stop clusters in domain D. A language that has both geminates and homorganic nasal-stop clusters has the ICC set so that A and B are skeletal points and C is the supralaryngeal node. The geminates in this case are derived, as shown in 3.3.1. If the ICC is not set in a given language, that language can still have inverse contours but that language may have other medial consonant clusters as weIl. It is

shown in 3.4 that a11 consonant clusters that are linked by a class node are inverse contours irrespective of the ICC. The ICC is only a constraint that forces two superordinate nodes to dominate one subordinate node, but the ICC is not the only means by which inverse contours are formed. Inverse contours, i.e., clusters that display the properties of

segmentallinking, that are not created by the ICC may he the result of assimilation or the ( OCP.

68 -;---..- -- o Since inverse contours involve branching, it is reasonable to assume that headedness is relevant. Like contours, 1 propose that inverse contours are right-headed. Therefore, the inverse contour in (3a), which is formed by two root nodes and ·he

supralarygneal node, must he right-headed. It is now possible to postulate that all branching configurations are righ'.-headed. As with contours, [-nasal] is the head of an

inverse contour which means that only a [-nasal] segment can be the right branch of an inverse contour in which the root node is the superordinate node. As discussed in chapter 2, the right branch is the necessary condition for headedness because [-nasal] is a redundant value. Hence [-nasal] might not be present at the time when headedness is relevant. If a mIe is sensitive to [-nasal], the Redundancy Rule Ordering Constraint inserts [-nasal] at the appropriate time. The fact that the right branch is the necessary condition for headedness is also important because it enables geminates to be rightheaded even though the skeletal point does not dominate any features. This means that the he ad of a geminate lacks the sufficient condition for headedness. As previously mentioned, the only medial cIuster in Sinhalese, for example, is a geminate or homorganic nasal-stop cIuster. The ICC in Sinhalese is set to apply to the

skeletal point as nodes A and Band node C is the supralaryngeal node with no domain limitations. This allows for both geminates and nasal-stop clusters in Sinhalese. Now consider the following data from Hindi. (8) Hindi (Ohala 1983) a. phenki 'handfuI' b. kendha 'shoulder' c. tamba 'copper' d. cimta 'tongs' e. dhemkana 'to threaten' f. nemda 'a mg' g. temga 'medal' Ohala (1983;108) states, "in Hindi all nasals are homorganic before stops", but this is not obvious from the data. However, the nonhomorganic clusters in (8d&e) are derived by vowel deletion and the non>omorganic clusters in (8f&g) are in loanwords. These J.a 2 .. ~ f' J, '~ .. ,,,, g .. ;

( counterexamples to the homorganicity of nasal-stop c1usters in Hindi are naturally separable, but Ohala does not have the mechanism to formally state the generalization.

Ohala's insight is properly captured by the ICC. By assuming that vowel deletion l~ves an empty skeletal position, then the nasal and the stop (8d&e) are not adjacent ifnodes A and B of the ICC are skeletal points. Therefore, these cases do not satisfy the adjacency requirement of the ICC. (8f&g) are loanwords which can he assumed to have an empty position separating the nasal and the stop.2 As a last example, Kolami (a Dravidian language), discussed in Steriade (1982), allows homorganic nasal-stop clusters with only underlying voiced stops. A nasal preceding a voiceless stop is nonhomorganic. The ICC in Kolami applies to two root nodes and the supralarygneal node in the domain of root nodes linked at the laryngeal node. The ex amples of the application of the ICC above show that the ICC can account for some medial cluster restrictions that involve linking. Sorne changes to the ICC are made ( in 3.3.1 to account for a wider range of clusters that display the properties of segmental linking, but for the present discussion the ICC captures the natural class formed by geminates and homorganic nasal-stop clusters. Most importantly, the ICC can he used to account for cases where adjacent nasals and obstruents must he homorganic. The next section demonstrates how the ICC can also account for feature-changing place assimilation. 3.3. A Thern:y of Nasal-StOl' Assimilation Feature-changing nasal-stop assimilation follows from the ICC, the CNe and headedness. The assimilation occurs as a result of the following sequence. First, there is the underlying nasal and stop. For example, consider Iml followed by Id!. Second, the ICC forces the two root nodes to dominate the same supralaryngeal node. As a result of the ICC, the supralaryngeal node dominates two place nodes. This configuration, however,

2 Claiming an empty position separates the nasal and the stop in loanwords is just as valid as claiming that ICC fails to apply to nonnative words since it is c inconceivable how features such as [native] are leamable.

70 o violates the CNe. Therefore, the left place node delinks because it is the nonhead. The rmal result is a homorganic nasal-stop cluster in which the place features are those of the

obstruent. Thé steps of the assimilation process are shown in (9).

(9) x x x x 1 1 --> 1 1 rt n. It. It. /1 ~ ~ [+1\83] ,-l. ,-1. [-nu] [+1\131 ,-1. [-nu] ....-1 ~ ~ [-cont] pl. pl. [-cont] pl. pl. 1 1 1 1 ]ab cor Jab cor x x --> 1 1 It. n. ~ [+nI31 ,-1. [-1\13] ~ [-cont] pl. 1 cor The fact that the nasal is always the target and assimilation is always regressive (two of the three universal characteristics of place assimilation within adjacent nasals and obstruents) can be explained by headedness. The nasal is the target to the left of the obstruent because the nasal is never the head of an inverse contour. Assimilation is always

regressive because the illicit contour formed by the branching supralaryngeal node is corrected by delinking the left place node, which is the nonhead. Progressive feature changing and feature-filling stop-nasal assimilation are prohibited because su ch an assimilation creates an inverse contour that is headed by a nasal. This analysis of feature-changing assimilation explains Webb's (1982) insight concerning the prominence of regressive assimilation rules. Webb observes that progressive assimilation mIes do not alter the primary place features. Webb accounts for

this by proposing that the coda position is weaker than the onset position and the segment

in the coda is "weak by position". Therefore. place assimilation is a1ways from the stronger

'il ( position to the weaker one. Webb' s attempt to account for the difectionality of assimilation is essentially correct, but it is insufficient because Webb has no way to include the notion of the relative strength of syllabic positions in a theory of representation. Headedness, however, does explain the dominance of the onset over the coda. Since the head and the onset are always on the right in a linking configuration, the nonhead, which is always in the coda, will always he the target of assimilation. The 1ast characteristic of nasal-obstruent assimilation to be accounted for is the fact that languages with nasal-fricative assimilation also have nasal-stop assimilation, whereas the opposite is not true. Nasal-fricative assimilation is accounted for by setting the ICC to apply to the mot nodes and the place node. Since the supralaryngeal node is not mentioned in this case, assimilation occurs regardless of the continuancy of the obstruent. If the obstruent is a stop, the two supralaryngeal nodes will conflate to one by the Shared Node Convention. ( The ICC also makes interesting predictions conceming gemination rules. According to Webb (1982), gemination is more commonly regressive rather th an progressive. Regressive feature-changing assimilation can be accounted for in the same way as nasal- stop assimilation. In the case of gemination, the ICC forces two skeletal points to dooûnate the same root node. As a result, the root node dominates two branches. The right branch does not delink because it is the head. Hence the gemination is regressive.

(10) x x x x x x 1 1 --> 'V --> 'V rt. rt. rt. n. 1 1 'A 1 ,-1. ,-1. ,-1. ,-1. ,-1. 1 1 1 1 1 pl. pl. pl. pl. pl.

Progressive gemination, which does occur, is not problematic if it is the result of feature filling assimilation. Feature-changing progressive gemination (which is prohibited by c headedness) can occur when structure preservation or syllabificati0l1 blocks regressive

'7') If- -.. -,-- '..

o gemination, although this claim requires verification. One example of progressive gemination as a result of structure preservation occurs in Sinhalese. (11) Sinhalese (Feinstein 1979) a. /potw +a/ [potte] 'core' b. /redw +a/ [redda] 'cloth'

The ICC dictates that there must he a geminate. Since there are no geminate glides in Sinhalese, the stop must fonn the geminate. The discussion of assimilation thus far has used both the CNC and the ICC, but there is an apparent redundancy in a theory that has both. The CNC forces the existence of inverse contours because the CNe dictates that there is only one class node dominated by the branching configuration. (12) A A contour B,,/ C o D inverse contour Therefore it appears that the CNe can he subsumed under the ICC. However, there are reasons for keeping both conditions. Consider an inverse contour, which automatically crea tes a contour because the one class node dominated by the two superordinate class nodes will branch to the two class nodes it dominates.

(13) A B inverse contour ""'/C A D E contour Such a configuration, as discussed, is ill-fonned but there is no way to eliminate (13) without positing a constraint on contours. Therefore, the CNC and the ICC are both necessary. The ICC, which is panicular to certain domains, creates a situation in which the CNC can apply thus accounting for feature-changing assimilation. More precisely, the ICC ( creates an environment that triggers the universal CNC from which the properties of nasal obstruent assimilation naturally follow.

3.3.1. A Case Study: Dio1a-Fo~ny Medial clusters in Diola-Fogny are limited to geminate nasals, homorganic nasal .. stop clusters and homorganic liquid··obstruent c1usters. The absence of nonnasal geminates can he attributed to a constraint in the periphery that disallows two skeletal points to

dominate a single [-nasal] root node. As shown in (14), the nasal preceding a stop after reduplication becomes homorganic to that stop. Also, a nasal preceding another nasal as a result of reduplication becomes homorganic to the righthand nasal. If two obstruents become adjacent after reduplication, the left obstruent deletes.

(14) a. Ini-garn-garnl [niga%Jgam] '1 judge' b. I-kob-kob-en/ [kokoben] 'yeam' c. Ini-nan-naol [ninannan] '1 cried' The analysis here is essentially the same given by !to (1986). Ito c1aims that the ( obstruent in (14c) deletes because it cannot satisfy the Coda Condition which requires that one association line be exhaustively linked to the coda segment. Deletion is avoided in (14a&c) because the the coda is linked to an onset hence the segments in the coda avoid the Coda Condition. Ito assumes that nasal-stop assimilation is feature-changing in derived environments. Furthermore, ho assumes that nasal-stop assimilation in Diola-Fogny follows a sonority restriction which disallows nasals to assimilate to consonants that are more sonorous. sucb as glides or liquids. Another way to derive the same facts concerning deletion is to say that Medial clusters in Diola-Fogny are limited to a well-defined set of inverse contours. The advantage

of using inverse contoUT..) ~s tbat there is no need to refer to the number of associaton lines. Since medial c1usters in Diola-Fogny are necessarily homorganic and nasal-stop clusters trigger feature-cha"dng assimilation, the ICC is set to apply so that two adjacent skeletal points must be linked to the same supralaryngeal node. The sonority restriction used by Ito ( o is subsumed by using inverse contours because the segment to the right of the nasal, i.e. the head, must he [-nasal -son] so the trigger of assimilation is always less sonorous than the nasal. The homorganic liquid-stop sequences are derivecl independently by the OCP. This will he discussed presendy. Since the ICC in Diola-Fogny is set to apply to two

skeletal points, it would apply to two obstruents. The resulting configuration would he a geminate with two skeletal points linked to a [-nasal] root node. This would arise via the steps shown in (15). The two skeletal points dominate the same supralaryngeal node as a

result of the ICC. As in nasal-stop assimilation, this triggers the delinking of the 1~ ft place

nod~. Now the two nondistinct root nodes are collapsed by the Shared Node Convention henœ producing the configuration of a geminate. However, this geminate is ill-fonned because of a :-estriction that disallows two skeletal points from dominating a [-nasal] root node. Therefore, the left skeletal point, the nonhead of the ill-formed inverse contour, deh!tes and what remains is the rightmost obstruent. o (15) 8.. x x 'b. x x c. x x d. x 1 1 --> 'v' --> >x/ --> i rt. 11. 11. 11. rt. 1 1 ~ 1 1 ,-1. ,-1. ,-1. ,-1. ,-1. ,-1. 1 1 1 1 1 1 pl. pl. pl. pl. pl. pl.

The analysis of Diola-Fogny based on the ICC is essentially the same as !to's analysis. The main difference is the use of the ICC to account for feature-changing nasal stop assimilation rather than Ito's sonority restriction. Now consider the homorganic liquid-obstruent clusters. Ito uses the OCP to account for the homorganicity of liquid obstruent c1usters, but the OCP is insuffjcient because the OCP cannot prohibit obstruent

liquid clusters. Note that the homorganic liquid-obstruent clusters obey headedness if headedness is extended to inc1ude root nodes that are respectively specified as [+sonorant] and [-sonorant]. The head of an inverse contour is [-sonorant], as in aIl cases, but now the nonhead can be either [+nasal] or [-nasal +sonorant]. By extending headednes'i to include a ( contrast for [sonorant], aIl three possible medial clusters of Diola-Pogny are inverse contours because aIl three are linked and rightheaded. However, only geminates and homorganic c1usters are produced by the Iee. Including segmentally linked homorganic liquid-obstruent clusters in the set of inverse contours raises two problems. First, homorganic lateral-obstruent cluslers create a

problem for Levin's (1987) c1aim that [lateraI] is dominated by the coronal node. H [lateral] is do ninated by the coronal node, homorganic lateral-obstruent cannot be linked at the place node because the obstruent is not lateraI. This is not a problem if [lateraI] is dominated by the supralaryngeal node. However, if [lateral] is dom;nated by the supralaryngeal node, then [lateral] should block the hardening of laterals post-na sally. 1

assume that the place of [lat.!ral] in the feature-geometry is parameterized so that il can he dominliced by either the coron al or supralaryngeal node. Such parameterization implies that there is a complementary distribution of languages with homorganic lateral-stop cIusters

( that display the propertie~ of segmentallinking and languages with post-nasal hardening of laterals. This claim requires verification. Secondly, homorganic liquid-obstruent cIusters raise the issue of inverse contours formed by two supralaryngeal nodes dominating one place node. Segmentally linked nasal

fricative cIusters and liquid-obstruent c1usters have similar configurations of c1ass nodes since both have two supralaryngeal nodes linked to one place node.

(16)

homorga.nic wal-fncative closter homorga.nic lit! uid -0 bstruent clUS1er x x x x 1 1 1 1 n. rt. ft rt /1 ~ [+nas]/1 3-1 s-l'" [-n.es] [+son] s-l. s-l. [-son] ~ [-cont]~ pl [+cont] ([-cont]/[ +la.t]) pl. [+cont]

Previously it was mentioned that inverse contours formed at the supralarygneal tier are ( restncted. This restriction seems to be in the fonn of a dependency that aHows the

76 o supralaryngeal as the superordinate node only if the supralarygenal nodes are linked to separate root nodes. Consider the opposite. Given two homorganic obstruents that are [ son -nasal] but differ for [cont], the Shared Nooe Convention dictates that there is only one root node. The resulting configuration is shown in (17).

(17) x x

rt. "'/A sol. s-1.

""'-/pl. If the rightmost supralaryngeal node is [-cont], then (17) contains an ill-fonned contour

because the head is [-cont]. If the righthand supralaryngeal node is [+contl, then (17) is a geminate affricate. Since homorganic obstruents do not create geminate affricates, homorganic obstruents that differ for continuancy cannot fonn inverse contours. This conclusion is confirmed by the fact that no language has homorganic obstruents which differ for [cont] that display Inalterability or Integrity. Homorganic liquid-obstruent clusters in Diola-Fogny create another problem because [-cont] is associated with the right branch of an inverse contour. This problem can be avoided by assuming that the features that characterize the right branch must be assoclated with the hlghest node on the right branch. Therefore, the only feature necessary for detennining the headedness of an inverse contour is [nasal] or [sonorant]. To summarize thus far, the universal characteristics of nasal-obstruent assimilation follow from the ICC, CNC and headedness. The ICC creates a configuration in which the CNC is violated. This is resolved by delinking the nonhead of the ill-fonned inverse contour. Since inverse contours must be rightheaded, the nasal is al ways to the left of the obstruent. Assimilation is al ways regressive because the CNe violation is never resolved by delinking the head.

77 ( The appeal to inverse contours in Diola-Fogny produces the same lesults as Ito's analysis. Claiming that deletion is avoided when the obsnuents form an inverse contour is similar to Ito's use of segmentallinking. The difference is that the ICC also accounts for feature-changing assimilation and the theory of inverse contours accounts for the suprising fact that aIl the segmentally linked clusters (except geminates) have the obstruent on the righthand side. The following section provides more examples of inverse contours and discusses the relation between the OCP and inverse contours.

3.4. Sorne ConseQuences of Inyerse COMouIS The medial clusters of DioIa-Fogny display an asymmetry that cannot he explained

in a theory of phonology without the notion of headedness. This asymmetry cannot he accounted for by appealing to the OCP because the OCP applied at the place tier is insensitive to the specification of the root nodes. More examples of this asymmetry are provided here to support the need for the inverse contours. The OCP, however, is still a necessary constraint that is not incompatible with the inverse contours. It is shown here that the application of the OCP can trigger nasalization rules so that the resulting configuration is a proper inverse contour. The application of the OCP to class l'iodes below the root node is the source of much debate. McCarthy (1988) claims the OCP must apply to the place and the laryngeal tier, but not to the supralaryngeal tier. The failure of the OCP to apply to the supralaryngeal

node can be accounted for b} the proposaIs made here. The OCP cannot apply to the supralaryngeal because the resulting configuration would have one supralaryngeal node dominating two place nodes thus violating the CNC. McCarthy's evidence for the OCP at the place tier is mostly drawn from dissimilation phenomena. For example, triliteral roots in Arabie cannot contain more than one labial consonant. This fact is captured by the OCP because the two adjacent labial consonants in the root would violate the OCP. McCarthy

uses the OCP to account for languages that hmIt their ambisyllabic clusters to geIJJinates

78 o and homorganic nasal-stop clusters. This is the point where the OCP and the ICC overlap, but the OCP makes incorrect predictions. The OCP account of homorganic nasal-stop clusters fails to account for the absence

of homorganic stop-nasal clusters in languages that ? Uow only homorganic nasal-stop clusters. For example, stop-nasal clusters in Marshallese do not block epenthesis but nasal stop clusters do. (18) "AlI contiguous consonants except noninitial identical consonants and homorganic full consonants are separated by an excrescent vowel although Jlj, nt do not manifest excrescence j,/l, en do." (Bendor 1968) The clusters in Marshallese are more adequately accounted for by the theory of inverse contours rather than the OCP because it is impossible for the OCP to apply to nasal-stop clusters in Marshallese but not to stop-nasal clusters.

Barra Gaelic appears to have OCP effects at the place tier because epenthesis in Barra Gaelic is blocked by homorganic nasal-stop clusters and homorganic sonorant-stop c1usters. This imphes these homorganic c1usters are linked. (19) Barra Gaelie (Clements 1986) a. /baLg/ [baLak] 'bellows' b. Id'reLt/ [d'reLt] 'due'

Clements (1986) proposes the Twin Place Node Constraint to ensure homorganic clusters in Barra Gaelic share one place node. The Twin Place Node Constraint, which states "representations do not contain certain identical adjacent place nodes within [a] domain", can be subsumed under the I)CP. The Integnty of clusters in Barra Gaelic can be more adequately explained by using inverse contours although the ICC is nol operative in Barra

Gaelic because there are nonhomorganic c1usters as well. The OCP is more appropriate

because linking occurs only when there are two identical place nodes. Interestingly, the

OCP In Barra Gaehe must apply to sonorant-obstruent and nasal-obstruent cll1sters. The

resulting configurations ot~y the extended notion of headedness used in the discussion of

Diola-Fogny.

70 c The need to preserve headedness in linked segments has interesting consequences in Sierra Popoluca. Sierra Popoluca has a rule that nasalizes a voiceless stop when

followed by a homorganic nasal. Clements (1985) claims the OCP applies to the place tier

thus creating the environ ment for nasalization.

(20) Sierra Popoluca (Clements 1985) a. /capmeyni/ [c8%pmeyni] 'the ocean' b. /petne?/ [p~lDe?] oit is swept'

Note that the underlying stop-nasal clusters do not surface. The nasalization rule converts

these clusters into geminates with twodifferent specifications for [voice].

(21) x 1 rt. [+IWal] Jar. 1 (+voice] ( ,-Jar. 1 place

Interestingly, Sierra Popoluca also has oral geminates where there is a difference in the

specification for [voice], e.g. kg. The status of these segments as geminates might seem

dubious. However, Selkirk (1988) discusses homorganic stop clusters in Klamath that

differ only with respect to [voice], but must he considered geminates. The nasalization mIe

in Sierra Popoluca, therefore, is actually a repair strategy to correct ill-formed inverse

contours. The nonhead becomes [+nasal] eliminating the contrast for nasality at the root tier

thns also eliminating the role of [nasal] in determining headedness.3 After nasalization, the

righthand skeletal point is the head.

3 Nasalization in Sierra Popoluca is more complicated than shown here. H the voiceless stop is preceded by a glottal stop, nasalization does not occur. But if the voiceless stop and the nasal are linked at the place node by the oep, then there is a violation of headedness 1 propose that in these cases the segments are merged by ( the OCP, but then separate to avoid violating headedness. ~ ~ ~~~~------o Medial clusters in Ponapean are either geminates or homorganic nasal-stop clusters and other clusters are separated by epenthesis. However, geminates are excluded in reduplicated forms (lto 1986). As a result of reduplication, the geminate undergoes Nasal Substitution which nasalizes the first half of the geminate. The result is a homorganic nasal stop cluster. (22) Ponapean (lto 1986) a./pap-papl [pampap] 'swimming' b./kik-kik/ (kiljkik] 'kicking' c./tat-tat/ [tantat] 'writhe' Ito correctly proposes that the OCP applies first then Nasal Substitution applies. Using the terms developed here, nasal substitution applies to the left root node to create well-fonned inverse contours in Ponapean. The OCP and inverse contours are compatible, but the OCP, as used to trigger the merger of identical place nodes, makes incorrect predictions. The fact that linked segments confrom to a certain type is properly captured by using the notion of inverse contours. The OCP is required in situations in which the ICC does not apply, but the outcome of the OCP is subsequently checked by conformity to inverse contours. Languages with homorganic nasal-stop clusters and geminates as the only consonant clusters are more adequately accounted for by the lec because the OCP cannot exc1ude stop-nasal clusters. As previously mentioned, the ICC is a constraint that forces two superordinate nodes to dominate one subordinate node, but inverse contours can arise by other means, e.g. assimilation. This implies al1 linked structures are inverse contours irrespective of the presence of the ICe. The only impossible inverse contour discussed thus far is homorganic obstruent clusters that differ for [cont]. ln presenting the analysis of nasal-stop assimilation, it was shown that it is also impossible to have a configuration with two root nodes linked to one supralaryngeal node which in tum dominates two place nodes. The c1aim that this configuration 1S iil-formed 1S supported by the faet that nonhomorganic clusters that agree

81 ( for continuancy exhibit neither Inalterability nor Integrity. Hence these clusters cannot he inverse contours. There are a handful of cases of spirantization and hardening which imply that

nonhomorganic segments are linked at the supralaryngeal node. The cases of post~nasal hardening discussed in chapter 2 are not problematic because the segments are homorganic.

There~()re, the configuration conforms to the CNC. Catalan exhibits continuant harmony which is apparently problematic for the c1aims made thus far. Consider the

environments of the Catalan lb, d, g] - [~, a, y] altemation.

(23) Catalan (Wheeler 1979) [~, 3, "i] lb, d, g] a.N_V c. J_C [e')\)~a] 'august' [ubzekta] 'object' b./[+cont] _ d. /[-cont]_ [bizJ3e] 'bishop' [unsigdifisil] °a bit difficult'

The environments (23b,d) imply that there is a continuant hannony process involved. This ( harmony process can be explained by assuming that the segments that undergo the alternation are not specified for [co nt] and that [cont] spreads from the adjacent segment.

Such a spreading rule must he progressive because these segments a1ways appear as stops when 'to the left of an obstruent, as shown in (23c). To ensure that only stops appear to the

left of consonant c1usters, Mascaro (1984) posits a rule that makes the segment into a stop when it is dominated by a rime. If it is assumed that [-cont] is the redundant value for [cont] in Catalan, th en there is no spreading in the context (23d). The two segments are [-cont] by coincidence. This leaves only the case of [+cont] spreading in (23b). Note that the sc!gment that spreads is the voiced strident. If the supralaryngeal node is spreading, th en at least 10/ in (23b) should become [z]. Since this does not happen, then only the

feature [cont] spreads and there is no violation of the CNC because the segments are linked by a feature, not by a c1ass node. Further evidence for considering continuant harmony in ( adjacent nonhomorganic segments as feature spreading rather than node spreading is that o the resulting configurations do not display the properties associated with segmentallinking. Hence there is no evidence that these sequences are inverse contours. 3.5. Constrajnts on Proœssiye Assimilation Deriving feature-changing nasal-stop assimilation from the proposed theory of representation is significant because impossible assimilation processes such as stop-nasal assimilation and progressive, feature-changing, nasal-stop assimilation are prohibited. The absence of progressive, feature-changing, nasal-stop assimilation, however, is an instance of a more general constraint against progressive consonantal place assimilation. In this section, headedness and inverse contours are used to account for the more general constraint against progressive consonamal place assimilation. PartIal assimilation that affects sequences other than nasal-stop sequences is difficult to find because such an assimilation often creates a geminate. The assimilation in

question is something like ps --> pf, but this does not seem to occur. In fact, the regressive counterpart ps --> ts does not occur either. The absence of partial asismilation within two

obstruents can be accounted for by inverse contours and headedness. If either typl!S of the above place aSSImilation were to occur, the result is a [-cont] and a [+cont] segment linked at the place node. The different specifications for [continuant] mean there must be two supralaryngeal nodes and since these segments are not affricates, then there must be two root nodes. However, the root nodes do not contrast for either [sonorant] or [nasal]. Therefore, the inverse contour formed by the two supralaryngeal nodes and the place node is ill-formed because the dependency between the root tier and the supralaryngeal tier is violated. The ill-formedness of the inverse contour can be resolved by changing the continuancy of one of the segments, hence forming a geminate. Therefore, thf; absence of partial assimilation other than nasal-obstruent assimilation is predicted by invf;rse contours and headedness. Webb's (1982) hypothesis that progressive consonantal place assimilation rules do not occur speclfically states that progressive aSSImilation rules do not alter the primary place c features. This hypothesis has an explan~tion in the theory of representation developed here, but the meaning of 'primary place features' in the feature-geometry framework is unclear. If il is assumed that the primary place features are the articulator nodes, then Webb's observation is incorrect. Sagey (1986) discusses a number of examples of pseudo metathesis, shown in (24a&c), which are better analyzed as the spreading of an articulator node. (24) labial (Margi) a. ltubi/ [tubWi] 'to repent' coronal (Sanskrit) b. Iprna/ [prI)a] dorsal (Parne) c. lci-kaol [cikYaol 'ear 3per. sing. poss.' Since the articulator nodes can spread progressively, then another meaning must be attributed to 'primary place features'. By saying that the primary place features are the place node, Webb's hypothesis can be maintained. ( There are two apparent counterexamples to regressive nasal-stop assimilation. The stop in nasal-stop assimilation in Kannada and Sanskrit assimilates the retroflex nasal. (25) a.Kannada /kaélI)+ dl [k8I)çll 'to appear' b.Sanskrit Iph8I)+ ta! [phélI)ta] 'spring pass. part.'

These apparent counterexamples can be accounted for in two ways. First, the retroflex nasals in (25) might be spreading only the coronal node to the stop. If this is the case, then (25) can be accounted in the same way as (24). Second, the analysis of nasal-obstruent

assimilation proposed 10 3.3 does not exc1ude the possibility of progressive feature-fi!ling assimilation. Therefore, the nasal can spread its place node to a stop that is unspecified for

the place node. Progressive feature-filling assimilation from the nasal to the stop cannot he excluded because the result is a homorganic nasal-stop cluter that is right-headed.

(

84 -_._~------

o 3.6. Conclusion The universal characteristics of nasal-obstruent place assimilation follow from the ICC and the CNC. Nasal-obstruent assimilation is always regressive because the CNC violation created by the ICC is resovled by deHnking the nonhead of the contour. Languages with nasal-fricative assimilation must also have nasal-stop assimilation becuase the ICC is set to apply to the place node. Hence, the specification of the supralaryngeal node is irrelevant. Impossible assimilation processs such as progressive place assimilation and rare assimilation processes such as continuant assimilation in nonhomorganic clusters follow from the theory of representation developed here. Progressive place assimilation does not occur because the head place node does not delete. Putative cases of continuant spreading must be feature spreading rather than node spreading. This c1aim is supported by the fact that nonhomorganic clusters that agree for continuancy do not exhibit the properties of segmentallinking. The notion of inverse contour is introduced to account for the range of permissible segmentallinkings which exhibit Inalterability, Integrity, and stray syIlabification. The lCC, which produces inverse contours by forcing two superordinate to dominate one subordinate node, is relevant only in languages with stray syllabification or feature changing assimilation. However, segments linked by c1ass nodes in languages without the ICC surprisingly obey headedness. This means that aIl segmental linking at the class nodes, regardless of their derivation, are inverse contours.

Dt: c CHAPTER 4 Conclu ding Remarks 4.Q. Residual Problems One residual problem worth mentic:ning is the representation of certain types of geminates which lie outside the domain of the CNC. Geminate glottalized stops universally surface as CC' not C'C' and similarly geminate aspirated stops universally surface as CCh

not Chch. These geminates can he represented as shown in (1).

(1) x x

""-/rt. A lar. s-lar. 1 place

(1), however, is ambiguous because it can also be interpreted as C'C' sin ce the single c laryngeal node is a property of both skeletal points. The signifie an ce of the ambiguous interpretation of (1) is attenuated by the fact that C'C' and CC' are never contrastive.

Hence, it is reasonable to say that (1) is the correct repesentation of glottally articulated geminates and the laryngeal node is somehow associated with the release phase of the stop. This can be achieved by using Kingston's (1985) theory of Articulatory Binding which states that glottal articulation which accompanies the articulation of a stop occurs close to the release of the stop. Glottal artIculation in continuant clusters is not confined to any part

of the cluster. Therefore (1) is the appropriate representation of gloually articulated geminates and Articulatory Binding ensures the proper phonetic interpretation. Sierra Popluca (discussed in 3.5) and Klamath (Selkirk 1988) have geminates

within which there is a change in voicing, e.g. kg. It is unclear how these geminates can be

accounted for by Articulatory Binding. If these geminates are like (1), then [+voice] is articulatorily bound to the release and there would be no specifIcation for [-voice). c However, [-voice] is present in these geminates because rules need to refer to [-voiee J.

86 Since both specifications appear to he present, (2a) seems more appropriate although (2a) violates the CNC. Selkirk (1988) proposes that these geminates, in fact aIl geminates, must have two root nodes. Selkirk's proposaI is adopted here for only the cases where geminates have a [voice] contour 'fhese geminates, shown in (2), are now consistent with the CNe because each laryngeaI node is domianted by an individu al root node.

(2) x x 1 1 l't. l't. lar. 1 [ +voice) ,-Jar. 1 place

A similar problem occurs with the representation of geminate affricates which should be (3).

(3) x x V n. A s-lar. s-lar. [-con~+cont] place

The facts concerning the precise pronunciation of geminate affricates are scarce, but certainly affricates never geminate as [tsts]. This problem was raised when discussing the behaviour of prenasalized stops in a gemination environment. The conclusion for

prenasalized ~tops is that there are two root nodes which wou Id prohibit [NCNC]. The representation of affricates proposed in 2.4.1, however. has one root node hence there must another explanation for the behaviour of geminate affricates. One way to account for the behaviour of geminate affricates is to have two root nodes in the representatlon, as shown in (4). ( (4) x x It.V1 rt. 1 1 ,-Jar. ,-Jar.

[.con~ +cont] place

(4) shows a geminate stop followed by a fricative thus producing [tts], but (4) can be altered to create [tss]. Selkirk's representation of geminates would also dictate that genrinate affricates are represented as shown in (4) because all geminates have two root nodes. As stated above, not enough is known about the gemination of affricates to make any generaJizations. The proposaIs made here '\bout representation and assimilation only apply to contiguous consonants and no attempt is made to extend the proposaIs to vowel sequences or to long-distance aassimilation. Research on vowels would entail examining coalescence, ( deletion, and diphthongization to observe any generalizations for which the CNC, the ICC, or headedness might be relevant. A study of long distancr. assimilation would entail investigating the role of headedness and the CNC in linked clusters that are not contiguous and the CNC in consonantal harrnony. The theory ofrepresentation proposed here makes a number of interesting predictions that are worth further investigation. For example, headedness might explain mirror-image gemination roles. The regressive gernination would

be the result of headedness and the progressive gemination would he the result of structure- preservation.

4.1. Concludin~ Remarks The CNC and the ICC are configuration al constraints that are properties of core phonology. The CNe is operative in every language since every language has sorne restrictIon on the range of contour segments. The differing inventories of prenasalized consonants are the result of selecting which nodes are subject to the CNe. The ICC is also (

88 o part of the core, but unlike the CNC, the ICC is a parameter. A laJ'lguage without the ICC does not have feature-changing assimilation or stray syllabification. A language with the ICC permits only a certain set of linked segments called inverse contours. Inverse contours, however, are not exclusively formed by the ICC. AlI segmentallinhng that involves class nodes (except the laryngeal node), whether fonned by the ICC, assimilation, or the OCP, produces linked segments that are right-headed. In Sierra Popluca and Ponapean where segmentallinking produces ill-formed inverse contours, a nasalization mIe creates proper inverse contours. The configurational constraints developed here account for the universal characteristics of nasal-obstruent assimilation and also lirnit the number of possible assimilation processes involving two contiguous consonants. The superficially dIsparate processes of assimilation. deletion, and coalesence that occur during prenasalized consonant formation follow from the CNC, Spreading Theory, and underspecification. The CNC also delimits the range of possible contour segments J although sorne stipulation is required to limit contour segments to [nasal] and [com]. Furthermore, the univesal characteristics of nasal-obstruent assimilation also follow from the CNC and right-headedness. The inclusion of the proposed configuration al constraints in phonology accounts for many aspects of assimilation that are missed in a thtory that appeals only to mIes. The fact that the phonology nasal-obstruent sequences follows from these configurational constraints supports the role of the core in phonoJogy. The constraints introduced here provide eVldence that princip les and parameters, in the form of constraints on representation, can define the set of phonological systems.

-' ..

QO , o References Anderson, S.R. (1976) "Nasal Consonants and the Internai Structure of Segments," Language 52,326-344. Archangeli, D. (1984) Underspecification in Yawelmani Phonology and Morphology, doctoral dissertation, MIT. Archangeli, D. and D. Pulleyblank (1986) "The Content and Structure ofPhonological Representations," ms., U. Arizona & USC. Archangeli, D. and D. Pulleyblank (1986a) "Maximal and Minimal Rules: Effects of Tier Scansion," Proceedings of the Northeastern Linguisitc Society 17. Armstrong, L.E. (1967) The Phonetic and Tonal Structure of Kikuyu, C.U.P., Cambridge. Bendor, B. W. (1968) "Marshallese Phonology," Oceanic Linguistics 7, 16-35. Blust, R.A. (1977) "Sketches of the Morphology and Phonology of Bomean Languages,"

in Court, Blust, and Watuseke, eds., Pa pers in Borneo and Western Austronesian Languages, Pacifie Linguistics, Canberra. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. and M. Halle (1968) The Sound Patterns of English, Harper and Row, New York. Clements, a.N. (1976) "The Autosegmental Treatment ofVowel Hannony," in Dressler, W. and O. Pfeiffer, eds., Phonologica 1976, C.U.P., Cambridge. Clements, G.N. (1985) "The Geometry ofPhonological Features," Phonology Yearbook 2,225-252. Clements, G.N. (1986) "SyIIabification and Epenthesis in the Barra Dialect of Gaelie," in R. Koen and H. van der Hulst ,eds., The Phonological Representation of Suprasegmentals, Foris, Dordrecht. Clements, a.N. (1987) "Phonological Feature Representation and the Description of Intrusive Stops," Proceedings of the Chicago Linguistic Society, 23.

Of) ( Feinstein, M. (1979) "Prenasalization and Syllable Structure," Linguistic lnquiry 10, 245-278.

Fivaz (1970) Shona Morphophonemics and Morphosyntax, U. Witwatersrand Press, Johannesburg. Goldsmith, J. (1976) Autosegmental Phonology, doctoral dissertation, MIT. Halle, M. and G.N. Clements (1983) Problem Book in Phonology, Bradford Books, MIT Press, Cambridge.

Harris, J. (1984) "Autosegmental Phonology, Lexical Phonology, and Spanish Nasals," in M. Aronoff and R. Oehrle, eds., lAnguage Sound Structure, MIT Press, Cambridge. Harris (1950) The Mwera Language, C.U.P., London. Hayes, B. (1981) A Metrical Theory of Stress Rules, doctoral dissertation, MIT. Hayes, B. (1986) "Inalterability in CV Phonology," Language 62,321- 351. ( Herbert, R.K. (1986) Language Universals, Marlœdness Theory, and Natural Phonetic Processes, Mouton de Gruyter, Berlin. Hualde, J. (1987) "On Basque Affricates," Proceedings 01 the West Coast Conference on Formai Linguistics 6. Ito, 1. (1986) Syllable Theory in Prosodie Phonology, doctoral dissertation, UMass. Kaye, J.D. and J. Lowenstarnm (1984) "De la Syllabicité," in F. Dell, D.Hirst, and 1.R. Vergnaud, eds., Forme Sonore du Langage, Hennan, Paris.

Kingston, J. (1985) The Phonetics and Phon%gy of the Timing o/Oral and Glottal Events, doctoral dissertation, UC Berkeley. Kiparsky, P. (1985) "Sorne Consequences of Lexical Phonology," Phon%gy Yearbook 2,83- 136.

Ladefoged, P. (1971) Preliminaries to Linguistic Phonetics, Univeristy of Chicago Press, Chicago. ( Levin, J. (1985) A Metrical Theory of Syllabicity, doctoral dissertation, MIT.

91 , 1

o Levin, 1. (1987) "A Place for Lateral in the Feature-Geometry ," paper presented at meeting of the L.S.A., San Fransisco. Mascaro, J. (1984) "Continuant Spreading in Basque, Catalan, and Spanish," in M. Aronoff and R. Oehrle, eds., Language Sound Structure, MIT Press, Cambridge. McCarthy, J.J. (1979) Formai Problems in Semitic Phonology and Morphology, doctoral dissertation, MIT. McCarthy, J.J. (1986) "OCP Effects: Gemination and Antigemination," Linguistic Inquiry 17,207- 263. McCarthy,1. J. (1988) "Feature-Geometry and Dependancy: A Review," ms., UMass. Mohanan, KP. (1986) The Theory of Lexical Phonology, Reidel, Dordrecht. Odden, D. and M. Odden (1985) "Ordered Reduplication in Kihehe," Linguistic Inquiry 16,497-503. Ohala, M. (1983) Aspects of Hindi Phon%gy, Motilai Banarsidass, Delhi. Paradis. C. (1987) " On Phonological Constraints," McGill University Working Papers in LinglJistics 4.2. Piggott, G.L. (1985) "Deletion in Multi-tiered Phonology," paper presented at the Canadian Linguistic Association Meeting. Piggott, G.L. (1987) "On the Autonomy of the Feature Nasal," Proceedings o/the Chicago Linguistic Society 23.

Piggott, G. L. (1988) " Parameters of Nasalization," ms., McGill. Piggott, G. L. (1988a) "Prenasalization and Feature Geometry," paper presented at NELS 19, Cornell.

Piggott, G. L. (1989) "Variability in Feature Dependency: The Case of Nasality," ms., McGill. Piggott, G. L. and R. Singh (1985)" The Phonology of Epenthetic Segments," Canadian Journal of Linguistics 30, 415-451.

Q? ( Rosenthall, S. (1988) "Syllabification and Sinhalese Phonology,t' McGill Working Papers in Linguistics 5.1. Sagey, E. C. (1986) The Representation ofFeatures and Relations in Non-linear Phonology, doctoral dissertation, MIT.

Sapir, D. J. (1969) A Grammar of Diola~Fogny, C.U.P., Cambridge. Schadeberg,T. (1982) "Nasalization in Umbundu/' Journal of African LangUllges

and Linguistics 4, 109-132.

Schein, B. and D. Steriade (1986) "On Geminates," Linguistic 1nquiry 17, 691-744.

Selkirk, E.O. (1988) "A Two-Root Theory of Length," paper presented at NELS 19, Cornell. Shaw, P. A. (1987) "Non-conservation of Melodic Structure in Reduplkation," Proceedings of the Chicago Linguistic Society 23. Steriade, D. (1982) Greek Prosodies and the Nature of Syllabification, doctoral ( dissertation, MIT. Steriade, D. (1985) liA Note on r:oronal," ms., MIT.

Vail, L. (1972) "The Noun Classes of Ndali," Journal of African Languages Il :3, 21-47. van der Hulst, H. and N. Smith (1982) "Prosodie Domains and Opaque Segments in Autosegmental Theory," in H. van der Hulst and N. Smith, eds., The Structure of Phonological Representations Il, Foris, Dordrecht.

Webb, C. (1982) UA Constraint on Progressive Consonantal Assimilation," Linguistics 20,309- 321. Welmers, W.E. (1973) African Language Structure, University of California Press, Berkeley. Wheeler, M. (1979) Phonology of Catalan, Basil Blackwell, Oxford. Vip, M. (1988) "The Obligatory Contour Principle and Phonologieal Rules: A Loss of < Identity,"Linguistic Inquiry 19,65-100.

93 o Zeirvogel, D., P.J. Wetzel, and T.N. Makuya (1972) A Handbook of the Venda Language, University of South Africa Press, Pretoria.

0.1