Part of the Phonological

Home , Akuapem dialect, Asante dialect, Twi

I DOCUMENT RESUME AL Cv:1 514 ED 022 189 48 By- Schachter, Paul; Fronikin, Victoria A PHONOLOGY OF AKAN:AKUAPEM, ASANTE, FANTE. California Univ., Los Angeles. Spons Agency-Office of Education(DHEW), Washington, DC. Bureauof Research. Report No- WPP -9 Bureau No-BR-5- 1094 Pub Date Sep 68 Contract - OEC -6-14 -028 Note- 272p. California, Los Angeles, California90024 Available from-Textbook Department,Student Store, University of WOO/ i EDRSPrice MF -$125 HC -S10.96 STRUCTURE. *DIALECT STUDIES.DISTINCTIVE FEATURES,MORPHOPHOWMICS, Descr" tors- *AKAN, DEEP GEWRATIVE GRAMMAR, VERBS OGY, SURFACE STRUCTURE,TOWLANGUAGES, sTRANSFORtIATION Halle Phonological System, Fante,Tshi,Twi Identifiers-Akuapem, Asante, *Chowtsky phonological In this _preliminary reportthe authors compare apart of the Asante, and Faniethe maiordialects of Akan. The comparition systems of Akuapem, which reveals the features common toall three dialects aswell as the features from one another. Thedescription of thephonological systems distinguish the dialects phonology as of these dialects ispresented within theframework of generative developed principally byChomsky. Hale, and Stanley.Although this study does not either a complete grammar or acomplete phonology of Akan, a attempt to present and "reasonably complete" treatmentof the finite verb is givenand deep structures relation to certainP-rules. Chapter 1 presents transformational rules are discussed in Segment 1 the inventory of systematicphonemes and Includes the setof Morpheme Structure Conditions which are tobe apphed to the dictionarymatrices ofmorphemes. of Morpheme SequenceStructure Cooditions whichapply ! Chapter 2 presents the set generally to all formatives.The first chapters thus presentall the general lexical redundancies in the language and cover a maiorpart of thephonological constraints surface Chapter 3 discusses all thegeneral Phonological Ruleswhich apply to the P-rules applying specifically tothe finite verb are discussed structures of sentences. surface structures in Chapter 4, whichalso discusses various otheraspects of Akan dominated by the Verb-Phrasenode. (JD) zii.- sr /0te 717-yr PHONOLOGY OF AKAN. KU E 3A E 11 FANTE"NI" 044- / y oar

AUL H E & VI IA F KIN

U.S. DEPARTMENT OFHEALTH, EDUCATION & WELFARE OFFICE OF EDUCATION

THIS DOCUMENT HAS BEEN REPRODUCED EXACTLYAS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATING IT.POINTS OF VIEW OR OPINIONS

STATED DO NOT NECESSARILY REPRESENT OFFICIALOFFICE OF EDUCATION

POSITION OR POLICY.

A L 001 5 1 4 KING PAPERS IN E I O. 9/UNIVERSITY OF CALIF lAs OS ELES/ AUGUST 1986 A PHONOLOGY OFAKAN:

AKUAPEM, ASANTE, FANTE

PAUL SCHACHTER

AND

VICTORIA FROMKIN

O.E. Contract 6-14-028

The research reported herein wasperformed pursuant to a contract withthe Office of Education, U.S. Department ofHealth, Education, and Welfare. Contractorsunder- taking such projects underGovernment sponsorship areencouraged to express freely their professional judgment inthe conduct of the project.Points of view or opinions stated do not, therefore,necessarily represent official Officeof Education position or policy.

Working Papers in Phonetics 9

September 1968

University of California. Los Angeles

Extra copies are available from:Textbook Department, Student Store, University of California, LosAngeles, California 90024

Price: $3.00

_ Preface

This is a preliminaryreport on researchthat we have conducted ()Ter the pastfew years, under U.S.Office of Education Cvntract0E-6-14-028, into the phonologyof the major Akan dialects ofGhana: Akuapem, Asante, andPante. No one realizes betterthan we just how"preliminary" a report this is, but wehope that, by issuingit in its present form, we can elicitcomments from interestedscholars that will assist us in preparing acontemplated revision of the report. We recognize, inparticular, that ourfirst-hand investigation of the Pantedialect has been farfrom thorough, and expect that moreerrors willbe found in our analysis of this dialect thanin our analyaes of theother dialects with wILLch we havebeen concerned.

It is a pleasure toacknowledge the assistancethat we have hadfrom native speakersof Akan, For Akuapem, our principal informants havebeen: Miss Doris Anakwa andMiss Mercy Martinson; for Asante,they have been: Miss Agnes Aidoo, Miss GraceAsafogyei, and Miss GraceMeenu, Mensah; Mr. and for Pantethey have been: Miss Christine Aidoo, and Mr, AlbertKodwo- 0, K. Brew, Miss JaneGarbrah-Aidoo, Mensah, We wish to offerMiss Agnes Aidoo ourparticular thanks for the manycontributions her intelligenceand in- sight into the workingsof her native languagehave made to our work. We have had the goodfortune to have theservices of Dr, John M. Stewartof the Institute ofAfrican Studies, University of Ghana, Legon as aconsultant throughoutthe course of ourresearch. The influence of Dr.Stewart's investigations of Akan isapparent throughout thisreport, although be is, of course,in no way responsiblefor our errors ofcommission or omission. Finally, we should liketo thank Mrs, ReneeWillen and Mrs. TheodoraGraham for theirassistance in typing, respectively, the preliminarymanuscript of the reportand the finished version.

P.S. V.F. Table of Contents

Introduction 0.10The Akan Language 0.20The Model 0.30The PhonologicalFeatures 0.31Segmental Features 0.32 Non-SegmentalFeatures 0.40The MorphemeStructure Conditions. 0.41If-Then Conditions 0.42 PositiveConditions 0.43 FurtherRemarks on Redundancy. The PhonologicalRules of MS Conditionsto P-Rule 0.51 &tension 18 Output 19 Syntactic and DiacriticFeatures 0.60 19 0.70The Treatment ofDialect Differences of the Text 20 0.80The Organization 21 0.90Symbols Defined.

Chapter1 Segment Structure 25 1.10Contrasting Segments 28 1.20Segment StructureConditions Segment StructureConditions for 1.21 29 Vowels . 1.22Segment StructureConditions for . 33 Glides . 1.23Segment StructureConditions for Consonants 35 1.24Segment StructureConditions for Non-Distinctive Features 38 Contrasting Segments . . 40 1.30Minimally Specified 40 1.40Boundary Units 1.50The Exclusion ofTraditional Phonemic Features 114

Chapter 2 Sequence Structure Segments 4 5 2.10Sequential Constraints on 46 2.20 The Structure ofthe Syllable. the Principal 2.21The Structure of 49 Syllable . 2.30The Structure of Root Morphemes 53 2.31Vowel Tenseness Harmony 55 2.32 Other Restrictions onVowels 58 2.33 Dialectal Restrictions 60 2.34Restrictions on Particular Lexical Categories 61 2.40The Structure of Derivational Morphemes. 61 2.50The Structure of Grammatical Morphemes 62 2.51 Verbal Grammatical bpiphenes 65 2,52 Nominal Grammatical Morphemes. 0 65

Chapter 3 General Phonological Rules

3.10Introduction . 69 3.20 Nasalization 71 3.21Regressive Non-Vowel Nasalization. . 72 3.22Syllable-Final Nasal Consonants. 73 3.23Nasal Prefixes . 75 3.24Homorganic Nasals. 76 3.25Progressive Non-Vowel Nasalization 77 3.26Vowel Nasalization . 78 3.30 Final Consongints . . 81 3.31 Final Lwi - 83 332 Glottal-Stop Deletion. . . 83 3.40 Medial Intervocalic Consonants . 84 350 Labialization. . . . 87 3.60Palatalization 89 3.61Palato-Labialized. Non-Vowels and Fronted Round. Vowels . 91 3.70Vowel Rules . 96 3.71Vowel Tensing * . 96 3.72 Vowel Backing. . 101 3.73 Pante Backness Harmony 102 3.74 Vowel Deletion and Addition. . 104 3.80 Systematic-Phonetic Segments 3.90 Tone Rules lgt 3.91 Downdrift . 106 3.92 The "Drop" or "Mid" Tone . 109 3.95 Tone Incorporation and Tone Simplification 394 Other Tone Rules 11;

Chapter4 The Finite Verb 4.10Surface Structure of the Finite Verb 117 4.20The Subject-Concord Prefix 118 . 122 4.30ASPect ...... 4.31STAtive and HABitual . 123 . 124 4.32 PROgressive. . . . . 4.33 PASt and PERfect . . . 126 4.34 FUTure . . . 132 4.35 OPTative and ltiPerative...... 134 137 4.36 CONsecutive. . . 4.40 NEGative , . 141 4.41Aspectual Changes in NegativeVerbs. 146 4.50INGressive ...... 150 4.60REDuplication. . . . 155 4.61General Rule of Monosyllabic RED 156 4.62 Subsidiary Rules AffectingMonosyllabic 0 0 162 RED. 4.63 General Rule of DisyllabicRED . . 169 4.64Subsidiary Rules Affecting Disyllabic . . 171 RED...... 177 4.70Verb Roots . . 4.71 Tone Raising in MonosyllabicVRs . . . 178 4.72 Replacement of Final NonArowelin Monosyllabic VRs by Glottal Stop . . . 181 4.73 Vowel Addition in Monosyllabic VRs . 184 4.74 Other Rules Affecting MonosyllabicVRs185 475 Structure of DisyllabicVRs...... 189 4.76 Tone Reversal in Disyllabic VRs. . . . 195 4.77 Other Rules Affecting DisyllabicVRs 197 4.78 Trisyllabic VRs...... 203 4.79 Final Glottal Stop in VRs. . . . 203 4.80SUBordinative. . . . 207 4.81 Rules Affecting VP-FinalSUB . . . . 209 4.90Tone-Changing P-Rules...... 215 4.91 Tone Changes in AffirmativeSimple Vs. 216 4.92 Tone Changes in NegativeSimple Vs . . 225 4.93 Tone Changes in IngressiveVs. . . . . 231 4.94 Tone Changes in ReduplicatedVs. . 233 4.95Tone Changes in Verb-Plus-Pronoun Constructions...... 236 4.96 Tone Changes in SubordinativeVs . . . 239

Appendix Phonological Rules . . . . 245

. 267 Bibliography . . . . 3

Introduction

0,10 The Akan Langumg* The most widely spoken languageof Ghana consists of a number of more orless mutually intelligibledialects* Until very recently, when Akan becamethe official name for the dialects spoken by allthe Akan peoples, there was no generallyaccepted name for the language as awhole. Ohristaller in 1875 used the nameTshi (Twi), a name whidh has been widely used for two ofthe dialects dealt with in this study, Akuapem Twi(hereafter, Akuapem) and Asante Twi (hereafter, Asante),The third dialect dealt with is Pante*Asante and Pante are the mostwidely spoken Akan dialecta* Akuapem is included in this studybecause of its historical role as thedialect originally selected i4 1838 as the written literary standard* Since 1673, when a book by W.47* Huller was published containing about 500 words of thePante dialectthese three dialects of Akan have been thesubject of serious linguistic investigation* The most complete grammarand dictionary of any Akan dialect are'4;hose published by Ohristaller in 1875 and 1881respectively*These are based primarily on the Akuapem dialect*Various other studies (cf. bibliography) have contributedgreatly to an radar- standing of the grammatical andphonological structure of one or another ofthe dialects, It is only very recently, however, that lauch comparativestudy of these dialects has been attempted. (As J, 14, Stewart pointed out in a paper presented to the Seventh West AfricanLanguages Congress, comparative study of the Akandialects heti been stimulated by "the current interest in theproblem of unifyiag Akan orthography.") The present work is an attemptto compare a Part of the phonological systems of Akuapem,Asante, and Pante* Sudh a comparison must reveal thefeatures common to all three dialects as well as thefeatures which distinguieh the dialects from one another*NO attempt has been made to reconstruct the commonhistorical antecedents whizlx underlie the different synchronicreflexes. It is possible however, that this study of thepresent-d*y phonological 4 0.10, 0.20 systems of the three dialects may shed some light on the phonological changes whidh have occurred through tine.

0.20 The Model.

The description of the phonological systems of Asante, Akuapen, and Pante will be presented within the framework of generative phonology as developed principally by 0homsky, Halle, and Stanley (cf. bibliography). This model presupposes a complete transformational-generative grammar, of which the phonological component is a dependent part.The entire grammar consistu of a syntactic component, a phonological component, and a semantic component. According to one formulation of the theory of transformational grammar, the syntactic component can be further divided into a set of"base rules", a lexicon or dictionary, and a set of "transformational rules". The output of the base rules specifies the "deep structures" of sentences: i.e., all the syntactic features, constituent relationsetc. which are relevant for the seaan- tic interpretation of the sentence. (This interpretation is provided by the semantic component.)The lexiconein- cludes all the morphemes in the language. Each morpheme lo represented by a phonological matrix consisting of columns of segments and rows of distinctive features (cf. Sections 1.30, ff.). (In a full grammar, each morpheme is also represented by matrices of syntactic and semantic features.) The transformational rules operate on deep structures to produce the "surface structures" of sentences. The surface structures of sentences may be represented as a sequence of morphemes with a labeled immediate-constituent structure, e.g.

)401Kumasen] (1) SEVPYS0E°)ASPESTAE°16(wo where the notationA.EX] indicates that the bracketed string X is a string of the category A. This can be represented equivalently as a tree diagram: 0.20 5

(2)

V ///1 SC

STA.

o 0 wo Kumase

((1) and (2) above representthe surface structureof the sentence 3wo Kumase'He's in Kumase'.)

The phonological componentof the grammar operates on the surfacestructures of sentencesto convert an abstract representatian of anutterance, such as(1), into a representation of the phoneticform of the utterance. Thus, tho surface structurerepresentations of sentences mustinclude all the syntacticfeatures that are relevantto the way sentences are pronounced.

In our description, weshall be concernedwith characterizing a numberof interrelated aspectsof Akan phonology, among them: (a) phonclogical redundancies;(b) the structure ofmorphemes and morphemeclasses; (c) the relation between therepresentations of morphemesin surface structures(the "systematic phonemic"representation) and in phonetictranscriptions (the"systematic phonetic' representation).The characterirstionof (a) and (b) will be in the form of aset of conditionswhiCh we, following Stanley, shall call MorphemeStructure Conditions(MS Conditions); the characterizationof (c) will be in the form of a set of rulescalled PhonologicalRules (P-rules). The model which we areusing will be presentedin greater detail below.While it is basedprimarily on the models of Chomsky, Halle,and Stanley, wehave altered these models where wehave found it convenientto do so. We make no claims about anytheoretical validityfor the changes we have introduced. 6 0.30, 0.31

0,30 The Phonolo ical Features.

The contrasting phonological segmentsof the lens, guage will be specifiedby a set of articulatorydistinc- tive features. We are assuming a universalset of phonetic features such as that proposed byJakobson and Halle(1956), In, making this- assumption, we arenot necessarily claim- ing adequacy for the particularset of universal phonetic features we use, which is essentiallythat of Chomsky and Halle (1968); that is, we are notclaiming that a different set of features may not be equallyadequate, or perhaps more adequate. (We havewithout special comment, changed the definitions of certain featuresthat we have taken from Chomsky and Halle, and have added otherfeatures, where we believe that these alterations makefor a better characterization of the phonological processeswith which we are concerned.) We shall follow the suggestionthat on the most abstract level--thesystematic-phonemic level--these features are classificatory in nature,and that each feature is defined by two values,'+' ana 1.! The following features will be used.Those marked by an asterisk are the distinctivefeatures which will be necessary to specifythe dictionary matrices ofmorphemes. Where the definition of a featurementions only the '+' value, the negative value of thefeature is defined as absence of the quality stated.

0,31Segmental Features,

Distinctive Segmental Features

*1, ±Vocalic: [+Vocalic] segments include vowels and liquids; 1-Vocaliej segmentsin- clude true consonants and glides. + *2, _Consonantal: [4-Consonantal] segments Xzclude true consonants and liquids; 1-Consonantal' segments include vowels nnd glides. + *3. _Coronal [+Coronal] segments are those produced with the blade of the tongue raised from its 'neutral' position. Dentals, alveolars1 and palatal consonants ars [+Coronal],as are front non-low 0.31

vowels, and the glide WI non- coronal segments include the labial and velar consonants, the glides/w/ and /h/ and the back and low vowels.

*4 tBack: (+Back] segments are produced with the back of the tongue retracted. In, Akan only the velar consonants, the g4des/4/ and /w/, and the back vowels are L+Backj; all other consonants -Rd glides and the front vowels areL-Back].

*5 tHigh: Segments in which the body of the `ongue is raised above the !neutrer poAtion are classified as(+Highj. Theirs include ths velar, palatal, and palatalized consonantsthe glidei a/ and /w/, and ths high vowels; L-High) segments include alveolar, dental, and labial consonants, the glide /h/, and the mid or low vowels.

6, tLows C+Ilow) segmentsrin Akan 4,nclude the glides /h/ and L?] and the low vowels. These are produced by lowering the body of tongue below the 'neutral° position.

*7. +Continuant: (+Continuant] segments are those in wIlich the air stream passes through the oral cavity without being com- pletay stopped. Continuants are the vowels, glides, liquids, and frica- tive consciants; noncontinuants include stops, nasal consonants, and affricates.

8. tVoiced: C+Voiced) segments are produced with vibrating vocal cords and include the vowels, the glides /w/ and /j/i the liquids, and the voiced consonants; I-Voiced] (voiceless) seggents include the glides /h/ andL?), and the voiceless consonants. 0,31 8

*9. +Tense: [...Tense] segments are produced with greater force andmuscular tension, and in Akan are thosevowels which are producedwith an advancedtongue position (Stewart,1967); L-Tensej vowels are those notproduced with this advanced tongueposition. In Akan this feature isrestricted to vowels.

*10. aasal: I+Nasal] segments are produced with a lowered velumand include nasal consona4ts and nasalizedvowels; I.-Nasal.' (oral) segments are produced with a raised velum,

*11. ±Tone: [+Tone] segments areethose.with contrastive high tone; L-Tonejsegments are those withcontrastive low tone. In the phonologicalmatricestone is distinctive only for L+Vocalicj segments (vowels). (Phonological Rules will spread tone to othersegments, and will further specifyscalar tonal values: e.g., the'mid' or 'drop' tone will be differentiated onthe systematic phoneticlevel.)

Redundant Segmental Features

12. +Sonorant: (+Sonorant] segments include vowels, the glides /w/ and/j/, the liquids /r/ and /1/, and the nasalconsonants; L-Sonorant] segments (obstraents) include non-nasal consonantsand the glide /h/.

13, tAnterior [+Anterior] segments include labial, dentalan4 alveolar consonantsand the glide Lw] (which areproduced in ftont of tip palato-alveolarregion); L-Anteriorj segments includevelar and palatal conponants,,t4e vowels,and the glides Lj] and Lh.j.

14. tRound: 1+Round] segments are produced with lip rounding and includelabialized consonants and roundedvowels. 0.31, 0.32 9

15. ±Palatal: [4-Palatal] segmentsare produced with the front of the tongue raisgd toward the hard palate; L-Back,-Lowj vowels, palatali;ed non-vowels, and the glide /j/ are Iqalatal]; all other segments are I-Palatal].

16. tStrident: [4-Strident] segmentsare[-Sonorant] continuants and affricates; [-Strident] segments are plosives and sonorants. (Stridancy is a feature restricted to non-vowels.)

17. tDelayed Release: This feature is restricted to sounds produced with complete closure, and distinguishes affricates from plosives, with the former being segments marked by a slow rather than an abrupt release, producing friction after the release of the stop closure.

18. tGlottal Constriction: In Akan only the phoneticrglottal stop is marked by the feature l+Glottal Constriction].

0.32 Non-Segmental Features

*19. ±Segment: This feature distinguishes systematic phonemic segments which are marked L+Segment],,from boundary units which are markedL-Segmenti.

[-Segments]are further divided into the following boundaries:

20. ±Syllable Boundary (SB): Symbolized as '='

21. tFormative Boundary (PB): Symbolized as '+' (formative and morpheme are near-equivalents)

22. Nord. Boundary (WB): Symbolized as 'O!'

23. ±Phrase Boundary (PB): Symbolized by a comma ','

±Sentence Boundary (SnB): Symbolized by a period 10 0.32, 0.40

The Boundaries ((-§egmeigts)) 20 through 24 are hierarchal. That is, a L+SpBJ implies the presence of all other boundary features; a L+PB] implies the presence of (+WB), [+1113], [+SB];aL+WB] implies [4493], [4.84; anC a 1+1111] implies [4.88].

0.40The Morpheme Structure Conditions. Like all languages, the Akan dialects reveal a high degree of phonological redundancy. In the model we are using, this redundancy is characterized by a set of Mor- pheme Structure (MS) Conditions.We shall be concerned with three types of redundancy: (a) constraints on the combinations of phonological features which occur in individual segments (which are stated by a subset of MS Conditions called Segment Structure Conditions); (b) constraints on the sequences of segments or features which can occur within one morpheme(which are stated by a subset of MS Conditions called Sequence StructureConditions); (c) constraints on combinations of (a) and (b) as they relate to different morpheme classes.The following are examples of these three types: (a) Voicing is a redundant feature for Akan fricatives. Thus sa and fa are possible, whereas *za and *va are not.Interirof the features listed 167ve we can state that all

[+Consonantal] -Vocalic segments are (-Voiced] +Continuant

(When features occur in vertical order enclosed in square brackets, this is to beinterpreted as a set of simultaneous features occurring in one segment.When features occur in horizontal order, this is to be interpreted as features belonging to sequentialsegments.)

(b) The feature [Vocalic] is redundant for primary syllable-initial segments, and for the second seg- pent in a.primary syllable.The initial segments are all l-vocalicj (glides or true consonants) and thesecond segments of these syllables are allL+Vocalic] (vowels). 0.40, 0.41 11

(c) Verb roots are either monosyllabic or disyllabic, i.e. are of one of thetwo following shapes: [=X sa], or[la X= X Aaare X contains no(+SB]. (For exceptions, cf. Section4.78.)

In dealing with redundanciesof these kinds by means of MSConditions, we follow Stanley, who states: "For every significantgeneralization that ern be made about the morphemes of thelanguage there is so. cnrres- ponding MS Condition." We alsofollow Stanley in making a formaldistinction between MS Conditions of twotypes: "If-Then Conditions" (I-TC) and"Positive Conditions" (PC). (WA have not found a need for Stanley'sthird type of MS Condition, "NegativeCondit:tons".)

0.41If-Then Conditions. An I-T Condition requires twoparts, which are soma- what similar to the two partsof a transformational rule in the syntactic component.One must state the structural description of the dictionary matrixto which the condition applies (the 'If') and thestructural change whidh takes place (the 'Then°. The fact that there are novoiced fricatives in Akan (a Segment Structure constraint) can bestated as an I-T Condition: r+Consonantali L+Continuant

T: [-Voiced

The above condipion is to beinterpreted as follows: If a segFent is bothL+Consonantal] and L+Continuantj it is also I-Voicedj.Thus, the feature specifyingvoicing can be leftblank in the dictionary maVrtxofrany morpheme which contains as one of itssegments lsJ or lf]. Since the If part of the conditionwill match the dictionary patrix, the condition applies andthe blank for the feature LVoiced] will be filled in automatically as a minus.

An I-T Condition can also bestated for sequential constraints. One generalization which will bediscussed in chapter 4 is that theinitial segment of the second 12 0.41, 0.42 syllable of a disyllabic VertRoot must be a trueconsonant, This can be stated by the (i.e. I-Vocalic9+Consonantaln. following I-T Condition:

I: VR(X= ]

T: (+Consonantal]

Since the MS Conditions areunordered, whenever one is applicable it will bemapped onto any lexicalmatrix which is a submatrix of thematrix swified in theCondi- tion. Thus a matrix which ismarkedL+Consonantal] by the above conditpan will stillhave unspecified the valuefor the feature LVocalic.By g Segment StructureCondition (of the 4-Then type ), allL+Consonantal) segments in Akan are alsoI-Vocalicj. Thus, in the lexical matrixof all disyllabic Verb Roots, theinitial segment of the second pyllable will have..blanks inthe cells for the features Nonsonantal] and LVocalic).These blanks will be filled by the Sequence Structureand Segment StructureConditions just discussed.

0.42 Positive Conditions.

Stanley states that "eachpositive condition con- sista simply of an incompletelyspecified matrix." For example, we would like toshow that all the vowelsin any one morpheme areeither tense or lax.This may be stated as iollows: [c+itrocalic]* + PC: + X Tense

: I does not include a +,

(As was noted in Section 0.32, aplus represents Morpheme Boundary.The asterisk is to beinterpreted as a symbol for iteration: i.e., in this case, thecondition applies to any number of segmeitsof the specified typewithin a single morpheme.The alpha is a variableranging over plus and minus: i.e., for every occurrenceof a in a rule, a = +9 or a = -; -athen assumes the oppositevalue of that assigned to ce. In other conditionsand rules, additional Greek letters--beta(0), gamma (Y), etc.--are similarly used.) 0.42, 0.43 13

Stating this positive condition permits usto mark pnly one vowel in a dictionarymatrix for the feature lTense]. It does not matter which vowel ismarked. Since tenseness is a property of theentire morpheme we do not want to state this fact in anI-T Condition which would have to stipulate the particularsegment, but rather show by a positive condition that thefeature applies to all vowels.We will mark the first vowel for thisfeature in the dictionary matrices, butit should be understood that this is an arbitrary decision, and that wecould just as easily have chosen to mark the lastvawel instead. It is the condition itself which makes thegeneralization and not the dictionary matrices, andtherefore we need not be concerned about which segment is marked.

0.43 Further Remarks on Redundancy.

The cells of the phonological matrices inthe lexicon can be specified aseither '+' or '-' or can be left blank. A blank is not to be consiAered athird value. Specifically, a blank in al4y cellin the dictionary matrix of a morpheme represents a redundant specification: i.e., a + or - which is predictable from the MSConditions. The dictionary matrices will therefore representnon-redundant specifications of the segments which comprisethem.According to tae Stanley model, once all the matriceshave been matched with the MS Conditions, all features will bespecified by one of thebinazy values. That is, no blanks will remain. We shall depart from the modelin this respect, sl.ace we believe that the features of certain segmentscannot be specified in any non-arbitrary fashion. For example, the negative morpheme in Akan is a low-tonenasal prefix -N-. Phonetically it is realized as a nasal consonanthomorganic with the consonant that follows it. Ong could arbitrarily assign any point of articulation, e.g.L+Coronali (/12/), and then by a later P-Rule adjustthe features to agree with the articulatory features of thefollowing consonant. We believe that this would fail to capture agenerality which is present: i.e., that the negative prefix isunmarked for point of articulation. If it were the case that in a phonetic meta-theory the alveolar nasal wasdesignated as the "unmarked" nasal in alllanguages, it might be reasonable to mark all such neutralizedfeatures with the unmarked values. Since to date the markedness of featureshas not been established in any generalphonetic theory, we shall 0.43, 0.50 14

the lexiconbut leave such featuresunmarked not only in structurerepresentations ofsentences. also in surface features acquire a (In other words,such neutralized specification only as aresult of theapplication of PuRRules.) that given asample of In summary,then, we can see enough to illustrateall the lexiconwhich is inclusive and the major segmentsand constraintsof the language, of MS Conditions,all thesignificant given the set will be generalizations aboutphonological redundancies the set of MSConditions(Segment revealed. That is9 Structure) should permitthe most Structure and Sequence i.e., representation ofthe dictionarymatrices: economical of blanks. matrices in which occurthe largest number Conditions for- Furthermore, theSequence Structure called thepolysystemic natureof lan- malize what Firth contrasts are foundin the fact thatonly certain guage: For example, the specific placeswithin the morpheme. Conditions on Akanroot morphemes pro- Sequence Structure never vide a formalaccount of thefact that vowels are non-vowels, of thefact that thepossible in contrast with are much morelimited contrasts ofsyllable-final consonants than those ofsyllable-initialconsonants, etc.

0.50 The Phonoloical Rules. (P) Rules will The input to theset of phonological structures of thesentences generatedby be the surface labeled and brack- component. These will be the syntactic with the eted strings ofmorphemes (cf.Section 0.20), by systematic-phonemicmatrices. morphemes represented onto a representa- The P-rules will mapeach such string sentence at thesystematic-phoneticlevel. tion of the to a '-', or The P-rules maychange featurevalues (a '+' add or deletewhole segments, maypermute vice versa), may by adding new rows, segments, mayintroduce new features integers(represent- etc. The P-rules mayalso substitute scalar values) for a'+' or For ex- ing relative (dictionary ma- ample, theunderlying forms ofmorphemes specified by markingtone bearj.ngunits as trices) can be L+Tone] or contrasting in abinary fashion,either high phonetic fact,however, low L-Tonej. It is an important low tone that a "downdrift"occurs, suchthat, whenever a 0.50 15 intervenes between twohigh tones, the secondhigh tone and whenever a high toneinter- is lower than thefirst, venes betweentwo low tones, the secondlow tone is lower than the first. This phenomenon will berevealed by the P-rules, so that thefinal phonetic output will bemarked for relative tone aswell as contrastive tone. Furthermore, there are certainphonological phenomena which do notapply to a single morpheme,but which do apply acrossmorpheme boundaries. All such phenomena willbe specified by P-rules. The P-rules will belinearly ordered.They will also be cyclical,A cycle consists of a sequenceof ordered rules which are appliedin the following fashion: first the rules are applied tothe innermost immediateconstitu- ents of the utterance. Once the rules haveapplied, the brackets are erased. Then the rules are applied tothe next innermost IO's etc.,until no more bracketsremain. For example, given theutterance cited above OwoKumase 'Re's in Kumase' with thestructure: 6[Kumase SEVP(VE SCE °3ASPE STIP 13VREwo]

the rules first apply tospi10:11 and then, on a second cycle, to sc[o],Asp(03, and vR(wo).When the brackets afe eritsed, th9third cycle of the rulesapplies to vi.0,0w3j and LKumase].Again the brackets areerased, --1412 and the fourth cycle appliestoup[o0wo Kumase).F4nal1y, a fifth cycleof the rules applibb tosbOwo Kumasej. The P-rules will have thefollowinm general form: A -"WY Y.The arrow (-0) means'rewrite as' or 'replace by'. TErslash (/) means'in the environment of'; Thusi the rule above is to beinterpreted as: "Replace A by B when A is preceded by Iand followed by Y."Furthermore, this rule is to be interpretedin such a way that all non- redundant features of Awhich are wt specifically mentioned in the rule remainunchanged. (For the effect of P-rules on redundant features,cf. Section 0.51.) For example, a P-rule such as: [gBack] [-Low] LaBackMr V (C) V 0.50 16

[+Vonsonanocalic present, and (where V = -Ctalb ( ) = optionally

C =L*Consonantal]) (-Low] interpreted as;ollowg: rewrite all is to be vowels before an (high or mid) vowels asL-Backj (ftiont) L-Back] vowel, andrewrite optipnal consonantfollowed by a L+Bgck] vcowels before anoptional allL-Low] vowels as Non-redundant fea- consonant followedby a L+Backjvowel. unchanged: e.g., highvowels tures of thevowels remain vowels remain remain high, midvowels remainmid nasalized nasalized, etc. noted that inP-rules of thefora given It should be segments, or above A and B areunits whichrepresent whole Since P-rules canadd seg- specific featuresof segments. segments, and addand deletefeatures, A ments or delete rule where Bis and B may benull.Thus we may have a null (B = 0):

V -1.0/ V Such a rule in Or we may have arule where A isnull. underlying monosyllabicverb stemswith the needed for the plionetic structure CVCwhich have assurface structures by a rule ofthe general feorm CVCV. This can be stated form: V/VR("c) state thefeatures of the (The specific rulewill of course vowel to beadded.) represented by Xend Y in The context orenvironment given above mayrepresent labeledcategory the formula This brackets, features, orsegments, or mayalso be null. will pertainonly to certain is necessarysince some rules other rules willrefer to phoneticcontexts, IC classes, restrictions. and other ruleswill have nocontextual from the abovethat there is noone-to- It is obvious matrices onecorrespondence betweenthesystematic-phonemic matrices. In some cases seg- and thesystematic-phonetic the systematic- ments which haveidenticalspecifications in systematic-phonetic phonemic matriceswill havedistinct 0.50 17 matrices. In other cases, the reverse will be true.We believe that such a position has been fully justified in previous publications.

Thc, necessity for linearly ordered rules can be illustrate4 in Akan by the following example.

A /u/ which occurs as the first vowel of a diphthong in a lexical matrix is always deleted. This deletion, however, must occur after the /u/ has labielizod the preceding nonvowel. For example, the undeaying representation of the word hwe [emi] 'look at' is /1=1/. By the labialization rule the A/ becomes labialized before /u/.The /u/ is then deleted, and the consonant becomes palatalized before /e/. The following derivation shows the ordering which takes place.

/hut/ Underlying fora [awl] Labialization rule [hwfi] [U]-Deletion rule (c46) Palatalization rule If the) [U]-deletion rule occurred prior to the labialization rule, the context for labialization would not be present and 1100 labialization could occur.

As will be shown in the sections below, the ordering of P-rules is nteessary in many other cases.

After all the P-rules have been applied any remaining boundary elements are removed. The final output will then be the sentence in phonetic transcription.

Our view of the phonetic transcription follows that of Ohomsky and Halle (1968):

(The phonetic transcription is not) a record of what is overtly present in the signal, but rather...a representation of what the speaker of a language knows about the phonetic properties of an utterance by virtue of his knowledge of the surface structure of the sentence and of the rules of the phonological component. 18 0.50, 0.51

It is suggested by Chomsky andHalle that in the final systematic-phonetic matrices which represent asentence all + and - feature values should begiven final scalar values. In our description such valueswill be used only for tone. For example, we will not specifyby an integral valu the duration of voicing or thedegree of nasalization. 0.51 Extension of MS Conditionsto P-Rule Outplq.

As was noted in Sections 0.41 and0.42, the MS Con- ditions are applied to(mapped onto) all matrices in the surface structurerepresentations nf sentencs prior to the application of the P-rules.We shall find, however, that some of the MS conditionsapply to strings whiCh ar the output of P-rules, as well asto those which are the input to this component of the grammar.To repeat these conditions as P-rules, or to specifythe redundant features in P-rules, would not only beuneconomical but would obscure the gnerality concerningthe redundancies whiCh persint.We therefore include the conventionthat when a it re- feature is specified as redundantin the MS rules, mains redundant with theapplication of the P-rules, if the non-redundant feature ischanged and the redundant feature is not mentpned..For example, at the systematic- phonemic level all L-Backj vowelsin Akan are redundantly L-Round], and are so specified in the MSSegment Structure Conditions. If a P-rIlle changesrthe featurespecifica- tion of a vowel fpom[+Back] to L-Back] and does not mention the featureLBoun0), this implies pat the segment which was redundantlyL+Round] becomes L-Round] simulta- neously with the change of thebackness feature.However, if a feature which is redundantin the underlying lexical matrix is specifically mentionedin a P-rule, the MS condition, will not reapply. Thus, g P-rule depigned tochange theL+114gh,+Baok,+Round] vowel Lit] to its L-Back] counterpart Dij (a segment type that doesnot occur as a systematic phoneme in Akan, but thatdoes occur at the systematic- phonetic level) must explicitlymention the retained specification of the feature LRoundj: i..,

+Back r-BackjiT +righRound L+Roun 19 0.609 0.70

0.60Syntactic and DiacriticFeatures. In addition to thephonological featuresdiscussed in Seations 0.30through 0.32, phonologicalmatrices may include features of twoother types: syntactic features and diacritic features. Syntactic and diacriticfeatures are propertiesof the entire formative. If a formative belongsto the syntactic category VerbRoot (VR), this thenbecomes a syntactic feature of eachsegaent of pe formative,and by convention eachsegment is markedL+VR] (cf. Chomsky and Halle, 1968, pp.174-176).We shall find thatthis is a necessary conventionin applying the MSConditions and the P-rules. On the other hand, weshall also find that there are soneformatives whidh areexceptions to the general conditionsand rules.These will constitute a very smallsubset of the entirelexicon, and we would fail to reveal themajor generalities aboutthe phonology of Akan if we did notin some way excludethese exceptions. For example, in disyllabicVerb Roots we shall seethat there are certainconstraints on the vowelswhich can ocaur. If the vowel of the secondsyllable is a high backvowel, /U/9 the vowel of thefirst syllable must alsobe /U/.Vs have found one exceptionto this condition: the Verb Root kamfo /kiLbe,k/ 'praise'. Each segment of thisVR will therefore be marked[-Sq SC n] (minus Sequence Structure Condition n), where n is thenumber of the Conditionwhich does not apply.This is a diacritic feature,and a property of the entiremorphene.Where formatives areregular, i.e. not exceptions, byconventionreach sygmen$will automatically be considered asmarked L+MSC DJ or L+Pmj, where n and m standfor the numbers of theMS aionditions and p-rules.

0.70 The Treatment ofDialect Differences The theory of generative grannarof which generative phonology is one part hasprimaray been conceivedof as a descriptionof the knowledge of anideal speaker-hearer of one dialect.Our attempt to use thismodel to describe and compare threedistinct dialects thereforecreates new problems. The solutions which havebeen adopted here are, for the most part, adhoc solutionsbut ones whidh we believe can adequately revealthe similarities andthe regu- larities which are found inthe three dialects,and which 0.70, 0.80 20 at the same timepermit the simplestmethod of differenti- ating between thedifferent dialects.The base forms (dictionary matrices) ofthe morphemes areconstructed with the aim offormulating a set of MSConditions and P-rules with the widestrange ofapplicability to all Some number of ruleswill apply to all three dialects. others the dialects, othersto two of thethree, and still will be restricted to onedialect. This is unavoidable innovations since a comparativestudy of Akan reveals some which are common to thedialects and otherinnovations dialect. (Fuxther comparative work unique to a single forma might make it possibleto postulate aset of base represent reconstructedproto-Akan which more directly which morphemes.) In the sections tofollow, the rules apply to all dialectswill be unmarked;those which have marked Ak forAkuapem, specific dialectapplication will be As for Asante, andFa for Pante. been to start An alternativesolution waald have dialects as basicand after derivingthe with one of the construct a surface and phoneticforms of this dialect, set of transferrules to derive theother dialects. lie rejected this approachsince we concludedthat such a description waald leadto fewer insightsregarding both the diachronic andsynchronic phonologyof Akan,

0.80The Organizationof the Text. This study does notattempt to presenteither a complete grammar or acomplete phonologyofAkan.Our of sentenceswill, in general, syntactic characterization which be limited tothose parts of Akansurface structures to the phonologyof the language. (In certain are relevant deep cases, however,it will be necessaryto refer to the structures of sentencesand the transformationalrules operation of theP-rules. Such syntactic which permit the necessary will base rules andtransformational rules as are the relevantsections.) Our phonology is be presented in of reasonably completeonly with respectto the phonology (which is, however,easily the mostcomplex the finite verb Other parts of the part of the Akanphonological system). phonological system aregiven less attention, or,in some cases(e.g., the phonologyof ideophones), none. Chapter 1 presentsthe inventory ofsystematic Segment phonemes. It also includesthe set of Morpheme 0.80, 0.90 21

Structure Conditions which are to be applied to(mapped onto) the dictionary matrices of morphemes.

Chapter 2 presents the set of Morpheme Sequence Structure Conditions which apply generally toall formatives. (Those applying only to the Verb Root are presented in Chapter 4.) Chapters 1 and 2 between them thus present all the general lexical redundanciesin the Akan language, and cover a major part of thephonological constraints of the language. Chapter 3 discusses all the generalPhonological Rules which apply to the surfacestructures of sentences. P-rules which apply specifically to the finiteverb are discussed in Chapter 4. This chapter also discusses various other aspects ofAkan surface structures dominated by the node VP-(Verb Phrase). , -- Finally, an Appendix presents the set of all of the ordered P-rules (those discussed in Chapters3 and 4).

0.90Symbols Defined.

V vowel, i.e. a segment which is An'true'f+Vocalic -Consonantal]

C Any non-vowel, i.e. a segment which is[-Vocalic] or[+Consonantal]--a true consonant, liquid, or glide

nasal consonant,i.e. a segment which is +Consonantal] or an archisegment which is +Nasal f+Segment] L+Nasal -VocalAc A glide, i.e. a segment which is[. Conson tea. +Vocalic l A liquid, i.e. a segment which is[+Consonant 22 0490

A salized vowel, i.e. a segmentwhich is +Vocalic -Consonantal rasal

Zero, or a null segmentalelement

The acute accent representshigh tone.

The grave accent representslow tone.

The vertical accent represents'drop' or 'midl tone.

AlE,I,O,U Cover symbolsrfor vowelarchi-segments, all of whieh areL±TenseltNasalotTone]

A Vowel segment whieh is[-Back,+Low] Vowel segment which is[-Backl-High,-Low]

Vowel segment which is[-Backl+High]

0 Vowel segment which is[+Back,-High,-Low] Vowel segment which is[+Back,+Righ] i.e. the / / Systematic-phonemic representation; dictionary matrix after theblanks have been filled in by the MSConditions Systematic-phonetic representations, or any representation between thesystematic- phonemic and systematic-phoneticrepresentations: i.e., any representationreflecting the application of anyP-rules (Note: In the text, where asystematic phonemic orsystematicophonepicetranscrip- tion iS used, e.g./hue/ or 1.910), this is to be understood as ashorthand version for the feature matrices.)

'Replace by' or 'Rewriteas' e.g. A B means 'Replace A by B' or'Rewrite A as B' 0.90 23

'In the environment of', e.g.(1)A -B/X I (2) A - B/ X, (3) A - B/X Read (1)ai-TReplace A by B-Men A follows X and precedes Y; or XAX XBT (2) means AX - BX, or 'Replace Aby B when it occurs before X (3) means n--.rB, or 'Replace Aby B when it occurs after X. 71 Feature specification for onesegment, e.g. 1+71] - [-P1)/ means 'Replace+71 by -71 P2

67-: [ -3 when 7 occurs in a segmentwhere F is n 1 2 specified as +, and 73 isspecified as -.'

Optional element. e.g. A-.8/ (C)V 'Replace Aby B when A occursbefore-eV, or when A occurs before V. The use of parentheses in a rule thusis an abbreviation for two rules. The rule given above collapsesthe two rules: (1) A - B/CV, and (2) A - B/ V When a rule includes anoptional element, first apply the rule withthe option chosen. Since these are ordered rules,if one is applied, the other cannotapply.

Select one of the items withinthe braces.This is also a schematic notationwhich collapses rules. e.g. A-,B/...{g} a))

represents the followingtwo rules: (1) A-0B/ C, (2) A - B/ D. If the output after applying (1) fitsthe conditions for application of (2), apply(2).

a, as ys Variables stand for either + or e.g. 7 - 01F / a F

'If F is + then Pis + and if Fis - 2 1 9 2 9 then F is 1 24 0 .90

111 --alP1 oil2 'If F2 is +, then Fi is -, andif 72 is -, then F v l <> Angle& brackets used in P-rulesspecify that if one item is chosen, the othermust also be chosen, e.g. +F F 1 1 A -7 2 2 4+71 3 This means that if 74 in the preceding segment is not +then 7will remain un- 9 3 changed. When a rule includes angledbrackets, first apply the rule with the items inthe brackets chosen. If a rule includes both paran- theses and angled brackets, expandthe parentheses before the angled brackets.

Word boundary

Morpheme boundary

Syllable boundary

Pause, phrase boundary 9 Sentence boundary 25

Chapter 1

Segment Structure

1.10 Contrasting Segments.

Twenty-seven contrasting segments (systematic phonemes) are needed to differentiate all possible morphemes in Akan: eight true consonants (C), /p, b, t, d, k, gs s, fl; three glides (G), /ilmw,mh/; and sixteen vowels (V), /i, 1, /Y, e, s, a, el a, sl u, ff, 0, 0, 0/. Ten segmentat binary feAturesare ntedpd to phese segments: Vocalic], LConsonantalj, LContinuant 9 LBack], [Coronal], [Voiced], [High], [Low], [Nasal], Tense]. Besides these ten distinctive features, the following four redundant features will be specified br Morpheme Structure Conditions, and will be utilized in the Phonological Rules to define classes pf sounds anO. to produce,the systematic- phonetic output: LSonorant], LAnterior], LRoundj, and lPalatal]. (The other tOree redundant segmental features list'd in Section 042--LStrident],[Glottal Constriction], and LDelayed Releasej--are introduced by P-rules.) Underlying morphemes in Akan are also differentiated by the presence or absence of high tone, which is marked in the lexicon only for vowel segments.Thus each pf the aboye vowel segments will be marked by the featureL+Tone], prL-Tone], with [+Tone] signifying high tone or pitch, and l-Tone] signifying low tone or pitch.

An inventory of the contrasting segmental units, with all features exclusive of tone marked, is given in Table 1.

It should be kept in mind that the segments listed in Table 1 are those needed for the underlying systematic- phonemic matrices of Akan formatives, and that segments of other kinds may occur in the final phonetic matrices which result from the application of the P-rules. Thus, for example, while there are no underlying palatalized or labialized non-vowel segments, such segments do result from applying certain P-rules to surface structures.Similarly, while there are no underlying nasal non-vowels, voiced non-vowels 2.1. vocalicconsonantal +_jhwpbfkgtds _ _ + + + + + + +3 + eagle 5 + + + + + + + + + + + + + + + + +ITialootiffutt + + + 4.6.5.3. lowhighbackcoronal +_ __+ + + _ _ -_ + + + + + + + + + + + + + + + + + 10. 8.9.7. nasaltense*voicedcontinuant _+ _+ +_ _+ +_ _+ _-+ +_ _+ +_- -+ +- + +_ +_ + ++ +_ +_ ++ ++ +- ++ +_ +_ + + + + ++ + 13.12.11.14. palatalroundanteriorsonorant + +_- _ ++ + + + - _ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + FULLY SPECIFIED SYSTEMATIC PHONEMES OF AKAN (Tone not included) Table 1 (*We believe, for reasonsforwithtionsriTense) all which are statedfeatures.) are includedwe actuallyarein Sectionworking, in theirrelevant 1.23, table,which that requireshowever,to thenon-vowels specifications thatin conformity allin Akan.segments with be the specified model These specifica- 1.10 2? preceding nasalized vowels willbe realized t11,.tattil1y as nasals as aresult of the applicationof a P-rule.For example, the underlying matrixof the Verb Root ma 'give' may be represented(after certain MorphemeStrucEre Condi- tions have applied) as follows:

Segment + + Vocalic

Consonantal =IP Coronal Back High 1M Low Continuant Voiced + + Tense* 1M Nasal Tone* 1M

(*The features [Tense] and[Tone] are applicable only to vywels.) The first c4umn contains aminus in the row representing the featurealasal], while the second column of thematrix contains a plus in this row,specifying the nasality of the vowel segment. A P-rule will change the -to a + in the first column, by virtue of thenasality of the following vowel. In other words, nasalityof non-vowels is predictable by rule, and thus need notbe specified at thesystematic- phonemic level. By limiting the underlyingsegments to only the distinctive features, wereveal phonological generalities about Akan throughexplicit statements asprovided by the P-rules. (In the case of the nasalnon-vowels, an alternative solution would havebeen to treat nasality as a redundant feature of non-vowelspreceding nasalized vowels, and to use the MS Conditionsrather than the P-rules, to reveal the generality. However, this would have necessitated a larger inventoryof systematic phonemes than webelieve is necessary.) 1.102 1.20 28

It should also be kept inmind that a number ofthe fratures which are needed tocontrast certain underlying segments are redundant inthe case of certainother segments: i.e., they can be predictedfrom the presence of other features in thesegment. Since such features are predictable, blanks may beleft in the cells oflexical matrices which correspond tothese features, providing we formally state the conditionsby which the cells maybe filled in. Such conditions are theSegment Structure Con- ditions.There are also certainfeatures that are never distinctive: i.e., that are neverneeded to contrast underlying segments. Certain of these featureswill also be specified by theSegment StructureConditions. (The remaining non-distinctivefeatures are inserted byP-rules.) By stating 21 SegmentStructure Conditions we are able to leave blanksin 263 of the 378cells in the matrix of Table 1. Every blank which appearsin the lexicon, and which can be filled inby a matching processwhich matches the lexical matrices tothese stated conditions,reveals a generalizationabout the phonologyof Akan.

1.20 Segment StructureConditions. All of the SegmentStructure Conditions(SgSCs) are If-Then Conditions (cf.Section 0.41). In our presentation of these conditions, weshall first coverconditions that specify some of theten distinctivefeatures. Condi- tions specifying suchfeatures for vowels willbe presented in Section 1.21, thosefor glides in Section1.22, and those for true consonantsin Section 1.23. Finally, in Section 1.24, we shallpresent the Sg$Cs whichspecify four non-distinctive features.

In our presentation, weshall not always attemptto state SgSCs in theirsimplest or most elegantform. In- stead, our aim willbe to state theconditions in such a way as toreveal with maximum claritythe redundancies relevant to each classof sounds. Thus one rule pertaining to vowelsstates that all vowels areredundantly continuants, and anotherrule pertaining toglides states that all glides arealso redundantlycontinuants. These two rules couldhave been combined asfollows: 1.20, 1.21 29

I: [-Consonantal]

T: C+Continuant]

But we have preferred instead to treat theredundancy for vowels and that for glides separately.

1.21Segment Structure Conditions for Vowels.

There are five vowel archisegments representedin underlying (dictionary) matrices, as follows:

-Back +Back

+High I U -Low 1

-High E 0 -Low

-High A +Low

Table 2

Each of these archisegments may be dividedinto tense and lax vowels, and the non-mid vowels maybe further divided into oral and nasal vowels. (The mid vowels are all oral vowels.) The complete inventory of phonological vowels (with the exception of tone) is shownin Table 3. 30 1.21

i -Back +Back

0 +Tense -Tense +Tense -Tense

+High i T a 0 +Nasal -Low i i u u -Nasal -High e e o o -Nasal -Low

-High g a +Nasal +Low a a -Nasal

Table 3

Some of the features shown in Table 3 areredundant. These redundancies, as well as certain other redundanciesin- volving features not shown in Table 3, arestated in Segment Structure Conditions 1 through 5 below.

SgSC 1: I: [+Vocalic]

ronsonantall +Voiced +Continuant

This condition states that all[+Vocalic] segments are true vowels: i.e., there are no liquids in the underlying forms of Akan morphemes. It states further that there are no voiceless vowels, andthat all vowels are continuants. (It is true that the surface forms ofAkan utterances Oave segments which are phoneticallyrealized as the liquid lrJ: e.g., kyere(t9irij 'show', Ak-As horo Calrt's1/1Pa Dim?) 'wash'. Such occurrences ofCr] result from the operation of two P-rules: P 15, which replaces a /d/ between oral vowels by[r], and P 03, which replaces a final /t/ by [r]. P 15 is applied in the derivationof 31 1.21

in the [tgirE] from underlyin$/kidii/a and I' 03 is applicd derivation of Ak-As[hurid/Fa Lhur?] from underlying /hist/o)

SgSC 2: I: (+Vocalic] L+Back

[7Coronal -Low T:

This conditionstates that all Akanback vowels are non-low vowels, andthat baft vowels arenon-coronalo thatL+Coronal] sounds are "produced (Given the definition posi- with the bladeof the tongue raisedfrom its neutral tion" (Chomsky andHalle, 1968, po304), back vowels, which by raising thebody of the tongue,rather are produced L-Coronal).) than the blade,must necessarilybe classified as

SgSC 3: I: (+Vocalic] L+Low

T: -Corona]] -High

vowels which This condition states,first, that those distinctively low areall classified asfront vowels. are classify the low (For a justificatipn oftht decision to vowels of Akan asL-Back), cf, Section 1.24,Segment Struc- 200 Second, it statesthat all such vowels ture Condition redundancy of the under- are non-coronal, (The non-coronal lying low tensevowels will in some casesbe changed by P-rulep, which willdifferentiate between a;.ow tense coronal vowel Ls) and a lowtense non-coronalvowel Le].) Finally, this condition alsostates that lowvowels cannot be (This pall of the rule, aswell as its counter- high vowels. -[-Low]), will have to be part in the nextrule (L+High] stated in any phonologyunless a set ofuniversal marking conditions are utilizedwhich wouldinclude the factthat: 32 1.21

r+Vocalici L+High and

r 3 L-Low] [-High]

At the present time such aset of universal marking conditions has merely beensuggested, and is vyry incom- plete. We are therefore requj.red tostet, the incompati- bility of the specificationsL+High] and L+Low] for vowels.

SgSC 4: I: +High -Back

1119 [-Low I T: +Coronal

This condit.on states that ahigh vowel cannot be low (see above) and aisothat a front high vowel is redundantly coronal. (In the discussion on the features given by Chomsky and Halle(1968 )9 they suggest that all vowels are redundantlyL-Coronal . However, their suggested correlate of thefeatureCoronal] (see above) would appear to contradictthis, since certainly in the articulation of high or midfront vowels the blade of the tongue is raised abovethe neutral position.. Furthermore, we have foundthat by utilizing theintrinsic phonetic content of the feature, we areable to state a number of the P-rules in a waywhich seems to provide alinguisti- cally significant generalizationwhich otherwise could not be stated.)

SgSC 5: I: rrocal -High -Low

T: [-Nasal]

This condition statesthat the mid vowe1.1/s, e, 3, 0/ are intrinsically non-nasal. Phonetically, when they 1.21, 1.22 33 precede a nasal segment,the velum may lower duringthe articulation of the vowel;phonologically, there are no morphemes which aredistinguished by contrastsbetween nasal and oral vowels ofthis set.Furthermore, there are no morphemeswhich contain a nasalconsonant followed by a vowelfrom this class. Since we are predicting all prevocalic nasal consonantsfrom a following nasal vowel, this is further justificationboth for the lack of nasal consonants in the underlyingmatrices and for this Segment Structure Condition. (A few words, such as Ak-As anova [lut6A] 'morning', would seem to constituteexceptions. The root of anopa, i.e., -am,is a compound derived ultimately from da[di] 'day' and -21 [IA] 'good, genuine'. The plural of da is Ak-As nna[Antd/Fa nda LAdAil and it seems likely that thisplural form may be involved in the derivationof anopa. In any case, it may be noted that the unusual[313]sequeLce isdialectally restricted. Thus Pante has AnIpaUAW, while the Akyem dialect of Akan--which is not ealt wita in detail in this study--has adopa[Ad5p6.).)

1.22 Segment StructureConditions for Glides. There are three glides indictionary matrices: /j, w, h/. (The glide P] is introduced bycertain P- rules.) These three glides are allI-Vocalics-Consonantal, +Continuantl-Nasal]. Using all the otherdistinctive features which pertain toL-Vocalic] segments, the glides can be fullyrepresented as in Table 4.

+High -High -Low +Low +Voiced -Voiced

+Coronal -Back i

t-Coronal w h -Back

Table 4 1.22

Segment Structure Conditions 6 through10 state the segmental redundancies which applyspecifically to these glides

SgSC 6: I: [-Voca].ic -Consonantal:I 4 [+Continuant] T: -Nasal

This condition states that allglides are oral continuants.

SgSC 7: I: [Vocalic ] -Consonantal

[7aLow I T: aVoiced

This condition states that theglides /j w/ which are[+High] are redundaptly [-Low] and[+Voicep, whereas the glide /h/ which isL-High] is redundantly +Low] awl [-Voiced], (We are able to use alpha notation for the features LHigh) and[Low] in the case of glides where we were unable to usethe notation for vowels sincethere is no glide which isboth [-High] and[-Low].)

SgSC 8: I: -Vocalic -Consonantal aCoronal [+High 4, T: [-mBack]

This cond#ion states that[Back] is a redundant feature for theL+High] glides /j/ and /w./. Specifically, the1+Coronal,+High] glide /j/ is[-Back], while the [-Coronall+High] glide /w/ is[+Back]. (In classifying /j/ asL+Coronal], we once,more differ from Chomskyand Halle, who classify it asL-Coronal]. Our decision has 1.22, 1.23 35 again been based on the phonetic correlates of the feature and on the ueeds of the phonology.)

SgSC 9: I: [-Vocalic -Consonantal -Voiced

T: r-Coronal] L+Back

This condition states that the voiceless glide /b/ is redundantly non-coronal and back.

(For the specification of the feature [Tense] in relation to glides, cf. Section 1.23, SgSC 16.)

1.23Segment Structure Conditions for Consonants. In dictionary matrices of Akan formatpes, eight 'true' consonants (i.e. segments which are L+Consonantal, -Vocalic]) are needed to distinguish morphemes. All of these segments are redItudantlyL-Vocalic,-Low,-Nasal]. The consonant segme.ts are presented in Table 5.

+Voiced -Voiced

+Continuant -Continuant

+Coronal -Back d s t -High

-Coronal -Back b f P -High

-Coronal +Back g k +High Table 5 1.23 36

Segment Structure Conditions10 through 16 state the segmental redundancies ofthese segments.

SOC 10: I: [+Consonantal]

T: r-Vocalic L-Nasal J

The specification of thefeature [Vocalic] is redundant for(+Consonantal] segments since, as was notedin Section 1.21, there are nosystematic-phonemic liquids in Akan. The condition alsostates that all underlying consonants inrAkan are[-Nasal]. (As has preyiously been mentioned,L+Nasal) consonants, as well asL+Nasal],glides, do occur at thesystematic-phonetic level. The L+Nasalj specification of these segmentsis always the resultof P-rules--cf. Section 3.20. One suc# rule, P 06,nasalizes all voiced non-vowelsthat precedeL+Nasal] vowels. We are aware thatthis might appear to be astrange analysis, since it is highlylikely that historicallythe reverse process occurred: i.e., the nasalizationof vowels resulted from theinfluence of nasal non-vowels. However, in a description ofmodern Akan, an attemptto predict the nasalization of vowels fromthat of adjacent non-vowels would result in positingunderlying forms which are sofar from the surface forms asto require a set ofcomplicated rules which would obscurerather than revealthe generalities of the presentlanguage. On the other hand, the nasalization of non-vowelsunder the influence ofadjacent vowels may be statedquite straightforwardly. If it is true that, historically,the vowels werenasalized under the influence ofnasal non-vowels,this process must have occurred farther backthan Proto-Akan.)

SgSC 11: I: r+Consonantall LeBack 4 T: [eHigh]

This condition statesthat all back consonants are high, and all non-backconsonants are non-high.High sounds are defined asthose "produced byraising the body 1.23 37 of the tongue above the level that it occupies in the neutral position" (Chomsky and Halle, 1968p. 304). This feature is used to distinguish the p Imary articulation of velars (and palatals), which are+High], from other primary articulations. (The C+Highcspecification might also be used to represent the secondary articulation of palatalization, but we have chosen inste0 to represent this secondary articulation by the feature L+Palatal]. Our reason for doing this is that we have wished to show the essentially assimilatory character of palatalization. In Akan, non-vowels are palatalized when they are immediately followed by high or mid front vowels--cf. Section 3.60. Since these vowels and the palatalized non-vowels may both be characterized as [+PalatAl], the use of this feature specification, rather than L+High], to mark the palatalized non-vowels, permits a simple and natural statement of the palatalization rule, P 11.)

SgSC 12: I: r+Consonantall L+Coronal

T: r-Back 1 L-High J

This condition states that the Alveolor ([+goronal]) consonants /t, d, s/ are,redunOantlr L-Backj and L-High]. Since the features LBackj and LHighj refer to articulations using the body of the tongue, and since these consonants are articulated with the blade of the tongue, phonetically as well as phonologically these are redundant features.

SgSC 13: I: r+Consonantal] L+Back

T: f-Coronal 1 L-ContinuantJ

This condition states that in the lexical representation of morphemes there are no velarrfricatives, and no velarized consonants, i.e. the only L+Backj consonants are the velar stops /k, 38 1.23, 1.24

SgSC 14: I: 140Insonaztall LoContinuant J 4 T: -Voiced -Back -High

This condition states that undeaying fricatives are voiceless, and that there are no palatal or velarfrica- tives. As a result of this condition, the segments/s/ and /f/ need not be marked for voicing,backness, or height.

SgSC 15: I: [4-Consonantal]

T: [-Low]

This condition merely stat's that the featureL±Low] is distinctive ia Akan only forI-Consonantal] segments: i.e., vowels and glides.

SgSC 16: I: r-Vocalicl LaVoiced J

T: [-aTense]

This condition (which pertains to glides as well as to true consonants) states that tenseness is neverdistinc- tive for Akan non-vowels, since all voi9eless non-vowels are[4-Tense] and all voiced non-vowels 1-Tense]. While we are stating here that tensenessis redundant for non-vowels, a more satisfyingsolution would be to leave the specification for this feature blank, since the "tense" and"lax" non-vowels do not form natural classes with the tenseand lax vowels respectively.The only reason that we are including thj.s condition, then, is to illustrate the wayin which the ITense] feature is mariced if we follow the Stanley model's requirement that all cellsof matrices be specified.

1.24Segment Structure Conditions for Non-Distinctive Features As has been stated previously, there are anumber of redundant features which we will wish to use todesignate 1.24 39 classes of segments in theP-rules, and which are required for phonetic specification. The values of four of these features are given in SegmentStructure Conditions 17 through 21.

SgaC 17: I: [-Consonantal] +Voiced

T: [+Sonorant]

The vowels and the glides/j, WI which are [-Con- ponantal] and L+Voi9ed], are sonorants. /h/, which is l-Consonantal] and 1-Voiced], and all true consonants, are obstruents.

SgSC 18: I: [+Vocalic] g T: [-Anterior]

SgSC 19: I: -Vocalic aBack

-Coronal [ OHigh t L-aAnterior T: 0AnterioA SgSC 18 says that all vowels are[-Anterior). SgSC 19 says that pe glide14] and the labial and alveolar consonants arc i+Anterior , w4le theglides [j] and[h] and the velar consonants are1-Anterior].

SgSC 20: I: -Consonantal;

aBack 1 [0Corpnal J Al, 11; raltound ] Lealatal

Back vowels and glidell areredundantly rouad, and front vowels and glides areredumiantly non-round.' (Akan low vowels are all classified as1-Back). This classifica- tion permits a mezt...mally stapiestatement of the redundancy 40 1.24, 1.30, 1.40 of the roundness feature.) For reapons stated in Section 1.23, we are utilizing the feature L±Palatal]. Vowels and glides produced wiDh the tip or blade of the tongue raised Do the palate are L+Palatalj; other vowels and glides are L-Palatal].

SgSC 21: I: [I-Consonantal]

T: PRound 1 L-PalatalJ

This condition thus makes all true consonants redundantly non-labialized, and non-palatalized. The labialized and/or palatalized consonants that occur at the systematic-phonetic level result from the application of P-rules.

1.30 Minimally Specified Contrasting Segments.

Given the set of Segment Structure Conditions specified in Sections 1.21 through 1.24, the inventory of fully-specified contrasting segments presented in Table I can be replaced by the inventory of minimally-specified contrasting segments shown in Table 6. This inventory of segments is presented merely for descriptive purposes and has no explicit theoretical status in this model of generative phonology.Given the Segment Structure Conditions and the lexicon, all the necessary generalities are made about the segmental redundancies in the language. The Sequence Structure Condi- tions (cf. Chapter 2) will enable us to leave many additional blanks in dictionary matrices.

1.40 Boundary Units

In Section 0.3we stated that the boundary units, which are marked as L-Segmentsj, are hierarchically related. This hierarchical relationship can be formally stated in MS Conditions, which we shall call Boundary Conditions (BCs). 1. vocalic jhwpbfkgtdsceaffeSiTiTooutfuty- - 0 0 0 0 0 0 0 0 + + + + + + + + + + + + + + + + 4.3.2. backconsonantalcoronal +000-- -0 +- 0- 0--0+ 0 0 0 +000000000-0 0 - + + + + + + + + 0 0 + + + + + 0 0 0 0 - - 0 0 0 0 + + +0 +0 -0 -0 +0 +0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 - - - + + + + + + 0 0 0 0 0 0 0 + + 6.8.7.5. highvoicedcontinuantlow 0 00 00 00 -0 +-0 0+0 -00 +00 -0 +-0 0+- 0- 0+ + + 0 0 0 0 0 0 0 0 0 0 0 0 + 0 0 0 0 0 0 0 0 - + + - + - 0 0 0 0 +0 + 10.12.11. 9. nasaltenseanteriorsonorant 0 00 00 00 00 00 00 00 00 00 00 0- 0+ 0- -- ++ -+ +- - 0 0 0 0 0 0 0 0 0 0 0 0 0 + - + 0 0 - + - + 0 0 0 14.13. palatalround 0000000000000000(.0000000000000000000000000000 Table 6 00000000 (Segmental Redundancies SYSTEMATIC PHONEMES OF AKAN (Tone not included) left blank) 42 1.40

BC 1: [-Segment]

T: 1-F1...14, where EF114] u :3.1 segmental features ...

BC 2: it [+SnB] I T: ...74411

+Fil t+SB4

BC 3: I: [+PB]

T: +WB +FB [+SB]

BC 4: (+WB)

41, T: [::111

BC 5: C+FB)

44 T: [+SB]

Some of these boundary units may be includedip th, lexism: e.g., compounds mayinclude a unit marked 1+FB.1 (Formative Boundary) as in Ghanani((hana + ni) 'Ghanaian', where ni (from o + ni'persorir) ii-a derivational suffix addwd to the name ora place to designate a personwho comes from that place. (If rules of derivational morphology are included in the grammar, many compounds can bederived by rule, and thus need not be listed as separatedictionary items.) (For a discussion of the status of certain syllable boundaries in the lexicon, cf. Section2.20.) 1.40 43

By convention formative boundaries areautomatically added at the beginning and endof each dictionary (lexical and grammatical) entry. This is necessary since certain MS Conditions and P-rules applyonly to strings of segments preceded gnd followed by a1.+FB]. Thus while Ghanani has only one 1.+FB] unit specified in thedictionar57177on- vention th, form as it would appear inthe surface structure would bea+gana+ni+]. (In certain cases we will require special elles for the deletion of[+FB], One such rule, P 14, eliminates formative boundariesbetween subject-concord prefixes and aspectual morphemes. For further discussion of this rule, cf. Section4.32.) Other conditions and P-rules apply only to strings preceded and followed byl+WII] (Word Boundary), We shall follow the convention suggested by Chomaky and Halle(1968. p. 366) that"the boundary # is automatically inserted at the beginning and end of every stringdominated by a major category, i.e. by one of the lexical categories'noun', 'verb',ladjective',or by a category such as'sentence', 'noun phrase', 'verb phrase' which dominates alexical category.° The derived surface structure of thesentence 3w3 Kumase 'He's in Kumase' would be, roughly:

+Kumase+

Using the above convention, this tree maybe represented by the labeled bracketing:

SE#VPI#V4SCE+°+3ASPESTAL+93+16[#+113+#]#]Npl#N[#+Kunlase+#44] 44 1.40, 1.50

(Verb Root (VR) is a lexical category in Akan.) A word is defined as a string of units which is bounded by at least two occurrences of # on each side, where no two co- terminous units within this string are themselves AIL Omitting the labeled brackets from the above string, we can illustrate this definition. In the string ###+o++0+#+wo+N##+Kumasei.###, 0+3++0+#+w3+## is aword, and ##+Kumase+0is a word. But +o+, the Subject Concord formative, is not a word, nor is the Stative morpheme, nor is the Verb Root +wo+. The Sentence Boundary (DB]) and the Phrase Boundary ((PB)) representa phonetic pause.The Sentence Boundary is automatically inserted after a word which is followed by no further segments.The Phrase Boundary is inserted by P-rules (which are not formalized in this phonology) in certain syntactic constructions.

1.50The Exclusion of Traditional Phonemic Features. The segments discussed in Sections 1.10 through 1.30 are what have traditionally beancalled morphophonemes. We believe that Chomsky (1964) has adequately demonstrated that a "taxonomic-phonemic* level between the morpho- phonemic (or systematic-phonemic) and systematic-phoretic levels has no theoretical justification. It is for this reason that the set of contrastingconsonantal segments in Tables 1 and 6 is much smaller than would be found in a taxonomic phonology. While it is true, for example, that "phonemically" one finds contrasts betw,en,palotalized and non-palatalized segmentse.g., kielue Ltgit9ej 'divide (reduplicated)' keka [kikid 'remain (reduplicated)', kese [kisf] 'big'--palatalization can be predicted if boundaries and syntactic information are available for the statement of P-rules, and if the rules are ordered.

In the next chapter, which deals with sequential redundancies, and particularly in Chapter 3 and pertinent sections of Chapter 4, which diseuss P-rules, we shall attempt to justify our analysis, showing that phonetic phenomena such as palatalization, labialization, nasalization of consonants, etc. are predictable, and hence need not be specified in dictionary matrices. 4 5

Chapter 2 f Sequence Structure

2.10 SequentialConstraints on Segments. The blanks which arepresent in the dictionary matrices representingAkan formatives mayreflect either segmental or sequentialredundancies. The segmental redundanciesi.e., those presentwithin a single segment regardless of the contextin which the segment occurs-- are specifiedby the set of SegmentStructure Conditions (SgS0s) presented and discussedin Chapter 1. The sequential redundanciesi.e.,those that depend upon the phonological or syntacticcontexts in which segments occur-- are specifiedby a set of SequenceStructure Conditions (SqSCs), to be presented anddiscussed in the present chapter. (Certain SqSCs that applyspecifically to verb roots are, however,given separate treatmentin Section 4.75.) Some sequential constraints aregeneral: i.e., they are reflected in allformatives in the language;others are reflectedonly in formatives belongingto particular lexical or grammaticalcategories.For example, regular verb roots (VRs) in Akan areeither monosyllabic or disyllabic, while regular nounroots do not show this restriction. Su.h a condition on thestructure of VRs can be formalized as aPositive Condition(PC):

IX=(X)] PC : VR (Where X does not include asyllable boundary(=) and where the parenthesesindicate an optional element)

If a dictionary matrixwhich belongs to thecategory vR does not meet thiscondition, it must bemarked with a diacritic feature (cf.Section 0.60) indicating thatit is exceptional in thisrespect. (There are a few trisyllabic verb roots,probably derived ascompounds historically, which aremarked with such a diacriticfea- ture.) 46 2.10, 2.20

By a convention establishedin Section 0.60, the syntactic category of a formativeautomatically becomes a distinctivefeature of all its segments.Such syntactic features enable us to state thecategorial limitation of a conditionwithin the condition itself. For example, there i3 a sequential constraint ondisyllabic VRs to the following gffect: if the vowel of the second syllableis /U/ (i.e. L+Back,+High]), then the vowelof the first syllable must also be /U/.This may be stated formally(as an I-TC) in twodifferent ways.The first statement--(a), below--utilizes the spreading of syntacticfeatures to the segments; thesecond--(b)--uses labeled brackets instead.

+Back x[+Vocalic]X = X (a) I: +High +VR

T: r+Back 1 L+High J +Vocalic DE (+Vocalic] X = X +Back X] (b): I: VR [+High

T: r+Backl L+HighJ

The (a) and (b) forms of thecondition are equivalent. In the writing of MS Conditionsand P-rules, we shall sometimes use one notation andsometimes the other. Most SqSCs apply to all formativeswithout reference to a particular lexical category.In such cases, no syntactic feature (or labeledbracket) is included in the condition.For example, syllable-finalnon-vowels are, in all formatives, limited tothe set of stop consonants. Hence no syntactic feature isrequired in the formalization of a condition to thiseffect.

2.20The Structure of the Syllable. Specifying the structure of theAkan syllable makes it possible to state manygeneralities about sequential constraints in a simple fashion. While no theory at 2.20 47 present provides a viable definition for either a phonological or a phonetic syllable, there seems to be general agreement that the syllable is an important phonological unit in many languages, possibly in all languages. For example, native speakers intuitively 'know' how many syllables occur in an utterance. (pailey a4d Milner (1967) have suggeste0 that the feature LVocalicj be repladed by a feature ISyllabic]. 9n the phottetic level all segments which are marked L+Syllabjxj belong to separate pyllables.. Vowels are inherently L+Syllabic], but become l-Syllabicj if they are replaced by glides; non-vowel ponorants occurring between two obstruents become l+Syllabic], etc. While this might provide a solution on the phonetic level, it would not help to define the phonological syllable in Akan, since on the systematic-phonemic level, Akan syllables may have two vowels, one of which is later deleted by P-rules.) Although we cannot define the concept "phonological syllables precisely, utilizing this concept enables us, for example, to avoid stating separately those constraints on morpheme-medial consonants which may be more generally stated as constraints on syllable-final and syllable-initial consonants.Furthermorethere are some P-rules which pertain only to monosyllabic Verb Roots, others only to disyllabic roots. These can be stated in a clumsy fashion without reference to the syllable, but we believe that the fact that they can be stated very much more simply if the syllable is referred to in itself indicates that the syllable has a phonological status which should be revealed.

Every Akan phonological syllable has at least one vowel segment which bears tone, and every syllable has one and only one distinctive tone. Tone is non-distinctive for consonants in the dictionary representation of morphemes. These statements pertain to both lexical and grammatical formatives.There are, however, sequential constraints which pertain to syllables in lexical morphemes which do not pertain to those in grammatical morphemes, and vice versa. Christaller differentiates between a principal syllable, and a secondary syllable. We shall adopt this terminology, although our description of the principal (lexical) syllable and secondary (grammatical) syllable will not be equivalent to Christaller's, since his description concerned the phonetic structure of syllables, while ours concerns the systematic-phonemic, or phonological, structure. 48 2.20

In Akan the occurrence ofsyllable boundaries is predictable from the sequences ofsegmental units. This fact requires some furtherdiscussion of the model we are using. Thus far, we have been employing thenotion of redundancy to mean that a feature valuein a dictionary matrix"can be left blank and laterpredicted' (Stanley, po 433). However, as Stanley has pointed out: If, in some envXronment E, thevalpe implies the valugL+g] and the value L+gim- plies the valueL+f] then it would be arbitrary which value we actually chooseto indicate in the dictionary (To) say that a certain fully specified matrix is highy redundantin some language is actually to say that manyof its feature values are interrelatedin ways deter- mined by the constraints of thelanguage, and it is simply the statementof these constraints (in the MS Conditions) which constitutesthe most natural characterizationof the redundanoy of language. Once these constraints have been stated, it is true that they maybe utilized... in giving dictionaryrepresentations in their most economical form; but thisis a secondary fact, and these redundancy-freerepresentations play no real role in a theoryof redundancy. (Stanley, 435) In other words, it is theMS Conditions themselves which define the redundancies,and not the blanks which may be leftin the dictionary matrices. We have chosen, neverthelesb, to represent thedictionary matrices in their most economical form: i.e., using blanks.This presents nc problems until wewish to describe the inter- relationship of certain featuresin a non-arbitrary way and are unable to make anon-arbitrary choice.

Such a situation exists indescribing the Akan principal syllable.The occurrente of syllableboundaries (l+SB]s) can be generalized in anIf-Then Condition as follows:

SqSC 1: I: [ ] [-Vocalic][+Vocalic]

T: r-Segmentl 1.+SB 2.20, 2.21 49

This condition states that a syllableboundary occurs before every non-vowel which is followedby a vowel. In a formative with the form CVCV, a syllableboundary occurs before both the first and the secondnon-vowels since they are both followed byvowels, but in a formative with the form CVC, a syllable boundary occurs onlyat the beginning of the formative, sin9e the final consonantis not followed by a vowel, but by aL-Segment]. Given the hierarchy and dependencyof boundary units (cf. Section 1.40), it is unnecessary to useSqSC 1 to insert a syllable boundary at thebeginningrof q formative, since the presence of a formative boundaryL+FBj in this position automatically implies--cf. BC5--that a syllable boundary is present as well.SqSC 1 is, however, required to correctly predict the placement ofmedial syllable boundaries in formatives of the shapes CVC=CV,CVV=CV, etc. But while it is true that medialsyllable bwndaries may be correctly predicted if (and onlyif) the LtVocalic] specification pf subsewent segments is known,it is also true that the LiTocalicj specificationof these segments may be correctly predictedif (and only if) it is known where syllable boundaries occur. For eximple, it may be predicted that the first two segments after amedial syllable boundary are a true consonant and avowel.Yet at the same time the occurrence of thesyllable boundary may be predicted from the occurrence of twosuch segments after it. Since we wish to utilize blanks in thedictionary matrices for descriptive purposesi.e.,since the blanks show at a glance which features areredandant--we shall represent the dictionary matrices withmedial syllable boundaries given, enabling us to leave blankthose segmental features which depend on their presence.

2.21The Structure of the PrincipalSyllable.

The Akan principal syllable.p may besymbolized as:

= C(V) V (C) =

and specified by the Positive MorphemeStructure Condition SqSC 2. 50 2.21

SqS0 2: +High +Vocalic <: -Back> +Back aTone fr+ConsonantallN PC: = [-Vocalic] <(aTone ) `L-Continuant Ji ONasal ONasal 1 yTense 1YTense J Me3gmen1) (where = is a feature complex specified as

(Note: Where [-Segment] is not specified for a unit, this unit is understood to be L+Segmentj.)

SqS0 2 combines into one MS Conditi^n a number of separate sequential constraints:

(1) It states that there are four possible thonological structures for the Akan principal syllable: a) =CVVC=, (.') =CVC=, (c) =CVV= (d) =CV=. Examples are: a) =CVVOIll, An from /kusg/ 'wait', (b) =CVOs', bam from /bab/, 'embrad61-, (c) =CVV= hwe from /hue/ 'look-if', (d) IIICVIII, wo from /wo/ 'be in'. -Riamples can also be c-ited for syllables in disyllabic or polysyllabic morphemes: e.g., kita /kite/ 'hold', a CV=CV formative, nanse /dIdsx/ 'spide17-a OVC=CV formative, tweri /tuedi/ Tealragainst', a CVV=CV formative, baakam /bagkab/ 'amber bead', a CVC=CVC formative, -berante /bedadtx/ 'young man', a CV=CVC=CV formative, etc.

(2) It states that the initial segment of a principal syllable must be a true consonant or a glide.

(3) It states that the second segment of the syllable must be a vowel.

(4) It states that in a syllable with two vowels the following conditions obtain:

(a) The first vowel is /U/. (This vowel will be deleted by a P-rule after the preceding non-vowel has been labialized because of its presence.)

(b) Tht second vowel is /I/, /E/, cr i.e., one of the L-Backj vowels. (c) The Imo vflu4s agree in tOeir specifications for the ;futures iTonej, aassa], and LTense): i.e., if one is L+Tone], so is the other; it one is L-Nazal], 221 51 so is tho other, etc. (SqSC 6 will stipulate that all vowels within an entire morpheme agree in tenseness.There- fore tile specification arfiliiiiiess agreement in SqSC2 is unnecessary, and is included only forexpository purposes.)

(Given this condition, together with certain Segment Struc- ture Conditions, it 1.8 unnecessary to specify my ofthe features of the-ftEst vowel of a vowel sequence-indictionary matrices. In addition t9 the features predicted by this con4tion, the features L-Consonantalo.Voiced,+Continuant) and L-Coronall-Lowj are predicted for this vowelby SqSC 1 and Sq8C 2 respectively.)

(5) It state; that a final segm;nt, if [-Vocalicj, must also be L+Consonantalj and L-Continuant): i.e., a final non-vowel is a true consonant and not aglide, and furthermore, it is a stop consonant rather than oneof the continuants /f/ and /14/.

(6) It states that, given a principal syllable with four segments, the third segment must be avowel and the fourth a true consonant.

If there were a set of universal markingconventions which stipulated an unmarked structure for syllables,ii is likely that this structure would be CVwith 9 standing for a true consonbalt: i.. a segment which isL-Vocalic, +Cons9nantal]. This wculd leave as unmarked for the feature LConsonantal] the first segment of themajority of Akan syllables, since in most of the formativesthe initial segment is a true consonant.However, since we are working with a model which does not include such a setof conven- tions, we are required to mark the initial segment forthis feature. SqSC 2 permits us to leave many cells blank in dictionary matrices: e.g., the seven blanksin the following partial-matrix for the VR Ak-la saigiAsdwa 'skin': +gusk+

Segment + + + + Vocalic 0 0 0 0 Consonantal + 0 0 0 52 2.21

(In a full matrix for this formative, a number of other cells are left blank by virtue of pe condition under disciassion: e.g., the[High] and,LBack] cells for /u/, the LBack] cell for /e/, and the LCon4nuantj cell for /k/.The Akuapem-Fante phonetic form Lgwen results from the application of P-rules whichplabialize the /g/, delete the /u/, and replace the /k/ by L?]. Additional P-rules ore inyolved in the derivation of the Asantephonetic form Ldale2.1.) It may be noted that, while SqS0 2 has constrained final non-vocalic segments of primary syllables to the set of stop consonants, none of these consonants is left t ct by the P-rules in this position.Specifically, /p/ - w, -[r], /k/ - [7], /b/ -) [m], /d/ - En], and /g/ - 13 In some cases, subsequent P-rules may apply to alter these final non-vowels still further. A syllable-final consonant that is not also morpheme- final shows further cert in constraints, which are captured in Sq80 3: roronall SqSC 3: I: X [+Consonantal] =OBack yHigh I T: aCoronal OBack ylligh [+Voiced

This condition states that a syllable-final, morpheme- medial consonant (which has already been specified as a stop by SqS0 2) is redundantly voiced, and is homorganic with the initial segment (necessarily anon-vowel) of the following syllable. Phonetically, all such syllable-final, morpheme-medial consonants are realized as nasals, as the result of the operation of a P-rule which nasalizes voiced syllable-final consonants.

(An alternative way of accounting for the (systematic- phonetic) limitation of syllable-final, morpheme-medial consonants to nasals homorganic with the following non- vowel would hgve been to leave consonants in this position specified as L-Vocalic] archisegments until thJ P-rules, and to have let the latter account for the homorganicness, as well as the nasality, of these consonants. We have, 53 2.21, 2.30

however, chosen to expressthe generalizationabout the homorganic character ofthe consonants in aSequence Structure Condition,in conformity withthe requirement of the model with which we areworking that allfeatures of segments b specified, wherever possible,in inputs to the P-rules.) Below is a partialredundancy-free dictionarymatrix for the medial consonantsof nanse 'spider',and the same matrix mapped ontocertain Sarrind ontoSqSC 3:

Dictionary Matrix BoUgSCs By Scift 3 +dad = sx+ +did = sI+ +did = st+

Coronal 0 + 0 + + + - - Back 0 0 0 - - - High 0 0 0 - - Voiced 0 0 0 - +

(The [-Back,-tligh) specificationof /s/ is providedby SgSC 12, the L-yoicejspecification by SgSC 14. SgSC 14 depends on theL+Consonantal,+ContinuantJspecification of /s/, which has beenomitted from the abovepartial matrices.) SqSO 3 concludes the listof SqS0s that applyto principal syllables. There is one SqSC(SOO 8) that may be said to apply tosecondary syllablesi.e.,the syllables condition, since it also of grammaticalmorphemes--but this constrains the number ofsyllables that may occurin grammaticalmorTEiiii, is best considered inconnection with a general accountof the phonologicalproperties of such morphemes, andis reserved forpresent;41on in Section 2.50.

2.30The Structure of RootMorphemes The root or stem morphemes arethe formatives which Christaller called "simplestems", the "realprimitives to of the language"(1875, p. 17). He limited such stems sequences of oneto three syllables,and differentiated 54 2.30

them from derived or compound words.With regard to the latter, however, he addeda note to the effect that "Derivatives in form...may be consideredas primitives in

meaning, when ncrilipler word still existing in tgrlin- guage can be shown, from which they are evidently derived" (p. 18).

In present day Akan (with some exceptions), this limitation on the size of formative strings still exists. (Furthermore most morphemes consist of eitherone or* two syllables, with three-syllable morphemes being considerably less common, and molt of them probably derived stems or compounds.) Noticerhle exceptions to this constraint are loan words and ileophones.The dictionary patrices gf thpse two setq Jf formatives can be marked l+Foreignj or L+Ideorhom.si. A more complete phonology of Akan would present the special conditions and Phonological Rules which are applicable to formatives marked with these features.

We can therefore state a condition defining the set of all root morphemes in either of the two following, equivalent, ways. SOC 4: PC: +X (=X) (=X)+ (where X does not include a syllable boundary.) SW 41: PC: +1Sylji+

(The subscript and superscript notation in Sq80 41 means that any formative must have a minimum of one (the subscript) and a maximum of three (the superscript) syllables.)

As was indicated above, a formative such ap &mires/ 'England', or Amerika 'America' would be marked L+Poreignj and thereby nanifrained by Sq8C 4.A compound like Ayaasekese 'a large couxt in a king's house' is derived from -ase (itself a compound, literally 'fire-bottom' or'(a p-ace ) under the fire') plus kese 'large'. The dictionary entry for such a compound would be something like: NININ[gzaiNta+seniCkesej]. (The a of a+se is a nominal prefix.Ordinarily it would not be listed in the dictionary entry but be inserned by a Transformational

It 2.30, 2.31 55

Rule. In many compouuds, however, medialnominal prefixes are omitted from thecompounded word, and we list the pre- Zix here since it is retained. An alternative listing would mark the noun se in this compound asminus the rule for deletion of nominal prefixesin compounds, thereby re- taining the prefix.) The phonological string of this compound would be/+++ga++a+sx+++kesx++/, and the constraint on the number ofsyllables per formative would apply.

2.31Vowel Tensenessliarmony.

ScIPC 5: PC: +XVocalic]Tense *X+ (where X does not include[+FB]: i.e., +)

This condition states that allthe vowels in apy one morpheme are either tense orlax. (The asterisk is a symbol for iterapion, meaninghere 'any number of segments specified asL+VocalicP.) In dictionary matrices we have ;Lrbitraray decided to mark the firstvowel for the feature lTense] (cf. Section 0.40) but any vowel would serve.

We could state this condition o4vowels in an alternative but equivalent condition,SqS0 5i. SOG 51: iar+Vocalicl [11-Vocaliclx+ ipTense aTense (where X and Y include MD +9 and I includes noL+Vocalic] segment)

Formulated in this manner, the conditionspecifies that any two vowels within a morpheme agreein tenseness.In a three syllable formative, the conditionwould apply to two of the vowels, and then apply again. This formulation avoids the useof the iterative symbol, which has certainadvantages. Since the number of vocalic segments within aformative is limited by SqS0 4, a condition onformatives which refers toindefinitely many such segments may seemcounter-intuitive. Tenseness harmony in Akan 16 notconfined to single formatives. It generally characterizes sequencesof root morphemes and adjacent grammaticalmorphemes within word 56 2.31 boundaries, and applies, in a morelimited way, across word boundaries. Examples oftenseness hamony within nouns are As afesLafxej 'year' and As afiepone] 'houses'. ([a] is the phonetic form taken by/a/ or /e/ before a tense vowel in the nextsyllablecf. Section 3#71.) B9t4 of these words consist of an9minal prgfix (La.' andLaj re- spec4vgly), g noun root(Ifx] and Lfi]) and a nominal suffix (Lej and Lei).All the vowels of afes arelax, all those of afie tense.Eomples of tenseness harmonywithin verbs are-Wks obsyleLobsjer] 'he came and did (it)' and Ak-As obeyii[obejii 'he came and removed(it)'. Both of these words consist of asubject-concord prefix((p] gnd lo] respectj.vely), an ingressiveprefix (Lbe] find Lbej),..a verb root (LjeJ andI0J), and a past suffix(Lx] and LiJ). All of the vowels ofobeyee are lax, allthose of obeyii tense. While it might have beenpossible to extend SqS0 5 to apply to suchtenseness-harmony constraintswithin the word, we have preferred tocharacterize the latter by means of aP-rule (P 69), for a numberof reasons. One of these is the way inwhich tenseness harmony acrossword boundaries (whidh mustobviously be characterized by a P-rule) operates.Across word oundar es a1.-Tensej vowel at the end of oue wordis made +Tense whenthe following word begins with a rootwith a t+Tensevowel, but not when the followingWM begins with a grammatical (i.e., a prefix) withaj+Tense] vowel.Thus,the I- ense final vowel of Ak AmmaLammaj 'Amma' is madcL+Tense when it immediately precedesthe root yi 'remove' asin Ak Lmma yiiCamma jii] 'Amma removed it'. But this vowel is not made[+Tense] when it immediately precedes the[+Tense] prefix [be], as in Ak Almabeyii lemmabejii] 'Amma came and removed it'.These facts.would indicatethat the vowel of the prefix realized aslbej in beyii, is inherentlylax (as are, in our opinion,all vowels in grammaticalprefixes), and is still lax atthe point in theoperation of the P- rules at which therule of vowel tensing acrossword boundaries applies. Further, there seems tobe a close relationbetween vowel harmony acrossword boundaries andvowel harmony within word boundaries(but across morphemeboundaries) in the case of compounds. In both casvs, the vowelharmony is strictly regressive,changing aL-Tense] vowel before 2.31 57 a[i-Tense] vowel (but not a [-Tense] vowel after a(+Tensb] vowel) to [+Tense], is limited in its effect to onesyllable, and never changes L+Tense] vowels to[-Tensa]. An example of the operation of vowel hammy within acompound is provided by Ak Kumaseni[kamasini] 'person from Klause', whidh consists of the compound stem Kumase 'Sumas&(in turn analyzable into the root kum(0731-47nd the noun ase last]) and the derivational suffix (originally aroot) ni [a] 'person'. We observe in this form: (a) that the vowel tensing is regressive, since the tense vowelof Ent] changes the preceding[x] to [i], but the tense vowel of (0110 does not change the following [a] to [m] (or[4); (b) that the tensing is limited ia its affect to onesyllable, since although the LT] ofCast] is made [+Tense] by the following l+Tenrae] vowel, the[a] of last] is unaffected; (c) that lax vowels have no effect on preceding tensevowels, since the [0] of (kOm] is xvit changed to[0] before the lax [a] of last]. Similar examples may be cited for the operation of tenseness harmony across word boundaries: e.g., As Kum ase [k0m asi ji] 'Kill this father-in-law's which consists a the words kum[km] 'kill', ase last] 'father- in-law', and [ji] 'this (one)'.

. Such evidence points to the desirabilityof a unified treatment of tenseness harmony acrossword boundaries and tenseness harmony within the word.Since the former car, only be dealt with by meansof a P-rule, we shall deal with the latter in this way as well, and limitSqS0 5 to the characterization of tenseness harmonywithin single morphemes (which, in any case, is very much moregeneral than is tenseness harmony across morphemeboundaries, whether or not these boundaries are also wordboundaries). Vowel harmony has been discussed in manyarticles dealing with the phonology of Akan. Stewart (1967) discusses the phonetic realizationof what we are calling "tenseness", and presents apparently conclusiveevidence for his position, which he summarizes asfollows: The vowels of Akan fall into two sets on the basis of vowel harmony: I, s, a, o, uand el 3 (our a or m) o, u. The relation of the first set to the second is one of unmarked to marked, so that one would expect to find a constant articulatory feature extending throughout 58 2.31, 2.32

harmony spans with i, e, 3, 0, u. Tongue raising is unacceptable because there is noconstant tongue height, and tenseness isunacceptable as in the unmarkedvowels laxness is conspicu- ously absent.AdvanciAs of the root of the tongno. on the other hand, isacceptable, and appeaxs to account, moreover,for all the phonetic phenomena whidh have been foundto be associated with this particular type ofvowel harmony. (p. 202) It sOuld be noteO that we use theterms 'tense' and 'lax', or L+Tense] and1-Tense] only as classificatory (phonological) features.For final phonetic specification, all C+Tense) vowels can beinterpreted as vowels in whidh .phe tongue root is advanced, in contrastwith their l-Tensej counterparts where thetongue root is not advanced.

2.32 Other Restrictions on Vowels

In morpheme7initial position theconsonant /g/ is followed only by 1-Nasal] vowels.Fuxther, if there is only one vowel in a morphem-initilsyllable introduced by /g/this vowel must be L+TenseJ.These restrictions are captured inScISC 6 and ScISO 7 which might havebeen combiued, with some notational complexity,into a single condition):

ScISC 6: I: + ack (l+Vocalic]) [+Vocalic] +Voiced

T: [-Nasal]

ScISC 7: ronsonantal I: + +Back [+Vocalic]a+Consonantalp= +Voiced

T: [+Tense]

SW 6 accounts for the fact that,while /cux/ (dwe [datix] 'kernel') and /gi/ (as in /gidg/ (gyina(d;,'inf] 2.32 59

'stand') are possible morpheme-initialsyllables, */g0T/ and */e/ are not. (For morphemes with initial/gVV/, such as /gui/, onlythe second vowel isspecified as[-Nasal] by Sq0C 6. tiowever, this entailsthat the first vowel is also L-Nasalj, sinos SqSC2cf. Section 2.21requires that two vowels in the samesyllable agree with respect to nasality.) SqSC 7 accounts for thefact that vibile /gi/, Cd4s) 'fire') are /02/ CIE [gu]'pour% and /go/ (al possible morpheme-initialsyllables (as is, e.g., /gep/ (gya(w) [d4e(w)?) 'leave%*/gI/, */gx;/, and */ga/ are impossible (as is, e.g.,*/gap/).

Neither of these restrictionsapplies to vowels following a syllable-initial/g/ that is not morpheme-initial. Thus, in our analysis, theunderlying forms of the verb roots pan(e), 'melt' and boa'help' are /die/ and /buga/ respectively (cf. Section4.77).

There is no evidence at presentwhich provides any explanation for these anomolousrestrictions. It seems likely that in proto-Akan, anunderlying /g/ existed and whether or not these samerestrictions existed in the very earliest forms of the language,examples given by Koelle indicate that the restriction wasprobably present in the Asante spoken in the earlynineteenth century.

(It may Iv noted that Christaller'sdictionary lists one item, maLgaga] 'the noise made by the chattering of the teeth -ot-by the cutting ofbones' that is an apparent exception to SqSC 7. This word is an"ideophone": i.e., it belongs to a class of words,found in many African languages, which C. M. Doke,who first used the term,characterized as"vividrepresentation(s) of an idea in sound" (1935, p. 118). Such words are oftendistinguished from other formatives in a language bythe fact that they deviate from the general phonologica;constraints.A dictionary entry marked by the featureL+Ideophonej would therefore not be constrained by the moregenerol morpheme structure conditions.The modern word Ghana(Lgana]) is also exceptional.) There is one further restriction onvowels in root morphemes, having to do with thedistribution of the low tense vowels /e/ and /67. This restriction is to the following effect: the low tense vowels are neverthe olz vowels in root morphemes, exceptthat /8/ may occur 60 2.32, 2.33 as the only vowel if it isimmediately preceded by /e. What this restriction indicates is that theC+Low,-Tense) vovels (a] and[1]are "normal", or "unmarked", and that the occurrence ofL+Lowl+Tense] vowels depends either upon harmony with a non-low tense vowel in the same morpheme (i.e., upon the limitation imposed by SqS( 5) or upon the fact that the vowel is immediately preceded by a consonant that cannot be followed by a lax vowel (i.e., upon the limitation imposed by SqSC 7).This restriction upon the distribution of low tense vowels is captured in SqSC 8:

SqSO 8: r+Vocalici Y+ L+Low

LL T: [-Tense]

(where either; (a) no L+Vocalic,-Low) segment occurs in X or Y; or, (b) X does not end in a L+Consonantals +Back,+Voiced] segmenti.e., /g/)

What thjs condition says is that anyC+Low] vowel must also be L-T:nse] unless Qne of the following conditions is met: (a) thereris a L-Lowj vowel somewhere else in the morpheme; (b) the L+Low] vowel is immediately preceded by

The condition predicts that while there are morphemes of such shapes as /ba/ (ba 'child'), /ka/(ka 'say°, /kasa/ (kasa 'speak% etc., */be/, */k6/,lild */kese/ are impossible. It does not, however, exclude such morphemes as /suek/(sue (swe?] 'be smell% bisa (bisa 'ask% or /k5di/ (kaammtkinI 'species of yam% which contain[-Low] vowels as well as+Low] vowels, or /go/ (as [dae] 'fire% in which the(-Low vowel is immediately preceded by /g/.

2.33Dialectal Restrictions Sections 2.30, 2.31, and 2.32 preseat themajor general sequential constraints in Akan. Were we writing a phonology of any one ofthe three dialects we are con- sidering, there would be other constraintswhich pertain to the individual dialect. 2.33, 2.34, 2.40 61

For example, the mid tense vowels[e] and Co] never occur in an Asantemonosyllabic roo6. Where Akuapem has le] in such a root,. Asante generally has[x] (as in al 'receive', Ak [dtfej/As[MID, and where Akuapem has Lo], Asante generally hasLu] (as in so(w) 'hatch', Ak [sow?]/ As [su?]). We can deal with this constraint by aP-rule, unique to Asante, which change; underlying/e/ and /0/ in monosyllabic roots to[I] and lu] respectively. (This P-rule is not formulated in the presenttext.)

There are other dialect-specificconstraints which can be similarlydealt with by means of dialect-specific P-rules.

2.34Restrictions on Particular LexicalCategories

The Sequence Structure Conditions SqSC1 through SqSC 8 are applicable to all the lexicalcategories which are generated by thesyntactic component of the Akan grammar. We deal in detail with SequenceStructure Condi- tions which are specific to the VerbRoot in Chapter 4. It is only the phonology ofthe Verb which is more or less complete in this description. There mcq be specific cw- straints relating to Nouns, Adjectives,and Adverbs whiwi are not dealt within this treatment. We are also not presenting thespecial phonological properties of ideophones or loan words. Loan words would require additional systematic phonemes.For example, b391 [bool] 'ball' has a final lateralliquid which is not present in any native morphemes

2.40The Structure of DerivationalMorphemes Derivational morphemes are not added bytransformational rules in our description, but areincluded in dictionary matrices in the same way ascompounds. For the most part they may be treated aslexical morphemes, since they obey the sequential constraintapplicable to the latter. The P-rules which apply across wordboundaAas also apply, in general, within wordscomposed of a root morpheme plus a derivational morpheme. 62 2.40, 2.50

Christaller lists the suffix -fo in hisdictionary as"a noun that is DA4 almostexoluiffely used as a suffix in names of persons, especiallyin the pl." (130).When added to a noun root, -fo mayadd the meaning "agent" as in Ak okuafo,'farmerc-from kua 'farm' plus fo 'person'. The surra-7ni has a similarmeaning, and comes from the noun o-ni 'person'.Thus, osikani 'rich person', is dertierrram sika 'money' airinperson'.

As was noted in Section 2.31, voweltensing across word boundaries is regressive, andappliee only once to a final lax vowelof a word preceding anotter wordwith a tense vowel.It is never progressive: i.e., a tense vowel does not tense a following vowelwhich is lax.This provides justification for dealingwith these derivational morphemes not only as lexical morplomtsbut as words.The suffix -fo is phpneticelly always Lfuj,with a lax vowel. on the otser handalways The suffii -ni (LnIj, from/d1/)9 has a tense 'Owl, and its progengdoes tense a preceding vowel.Thus in Ak Ghanani Lganinij 'a Ghanaian'the lying lax /a/ beforririi tensed, butin Ak Twifo 'Twi people', both the underlying tense/i/s Urfa .14 the underlying lex /u/ of -fo remain unchanged.

There are a number of other derivationalaffixes which adhere to these same restrictions.The diainative euffis we, derived originallyfrom ba, a stem meaning 'child', is iiso always lax even when farowing a root with tensevowels.

No additional Sequence StructureConditions are needed, therefore, for derivationalmorphemes.

2.50The structure of GrammaticalMorphemes Grammatical morphemes, or formatives, arecomposed of secondary syllables, whichdhow different constraints from those of the lexical or priaarysyllables. Grammatical morphemes may be added or altered bytransformational rules. Consequently most grammatical morphemes arelisted in a special subsection of theLexicon, (the so-called "second lexicon", in which morphemes that areinserted post-transformationally are listed). Like root morphemes, most grammatical morphe:aes are representedin the Lexicon by fetture matrices, with the columns representingsegments and the rows representingphonological, syntactic, anddiacritic features. However, in some cages, thedictionary matrices 2.50 63

of grammatical morphemes maycontain archisegments which are notfully specified by MorphemeStructure Conditions: i.e.,they remain partiallyspecified in surface structures. And in other cases there may be nodictionary matrix corresponding to the grammaticalmorpheme. In such cases the syntactic brackets of thegrammatical morpheme are carried f down in the derivation untilP-rules either add features to it or delete it. For example, whether thereduplicating affix (RED) is introducedinto deep structures throughthe exercise of an optional choicein the base rules, or whether it is present in somepartieular lexical entry, itsphonetic form is always specified entirely byP-rules (cf. Sections 4.60 - 4.64).On the other hand, some grammaticalmorphemes that are not listed in thelexicon (the "zero morphemes") are never given aphonetic form.Thus the Stative (STA) morpheme and the Imperative(IMP) morpheme have no phonetic segmental form, but areactualized by their tonal effects on the segmptsof other formatives. These will be symbol- i0] and IM[0] in Chapter 4.Thera is no ized asSTA P historical evidence that thesemorphemes ever had segmental forms. (The Habitual Morpheme (HAB), however,while it has no surfacesegmental form, must bespecified as having low tone in order to accountfor the operation of certainP- zules (cf. Section4.31). This suggests that at somepoint the habitual morpheme was alow-tone vowel. The segmental features, other than tone, have beenlost. We have no non- arbitrary means of deciding whatthe other features for this vowel are, and willtherefore list the HAB m4rpheme in the grammatical sectA.onof the Lexicon as the arehi- segpoot l+Voealic,-Tonej(and refer to itrin Chaptor 4 as LOD. A P-rule will changethis to a L-Segmentj after HAB the low tone of the morphemehas had its effects.The "zero Nominal Prefix (NomP)--cf.Section 2.51--is treated in a similar fashion, except thatin this case it is possible to suggest, o4 the basisof the non-occurrence of overt l+High,-Nasalj prefixes, thatthe archisegment underlying the zero NomP hos these-features aswel) as the features l+Vocalic,-ToneJ.) The grammatical morphemesthat aro represented by feature matrices in the lexicon areall Aonosyllabic, with the syllable being either asingle lax vowel, or a non- vowel followed by a single laxvowel.Thus the segmental structure of possible grammaticalmorphemes can be symbolized as +(C)V+ (and that ofsecondary syllables asgi(C)Vss). 64 2.50

The segments of all grammatical morphemes that are listed in the Lexicon will be marked, aptomatically by convention, wtth the syntactic featureL+G] (standing for l+Grammaticall).Similarly,,the segments of all lexical morphemee will be marked as (All of the segments mentionpd in SqSCs 1 to 7, above, are to be understzod as marked L-G], although this specification has notbeen inde explicit in the formalization of theconditions')

The shape of grammatical morphemes can bespecified by a Sequence Structure Condition, SqSC 9: SW 9: fr+G ) +Vocalic + PG: '14ocalic -Tense

There are three grammatical morphemes which are realized phonetically as prefixed nasal consonantshomor- genic with the following non-vowel. These are the negative morpheme (NEG), the Optative morphuae(OPT), and one of the NomPs. Since all these morphemes have to be specified for the feature tone, and since in ouranalysis, tone on the phonological level is restrictedto vowele, we have chosen to represent thesemorphpmes in the dict4o- nary by the following featurematrix: L+Vocalics+Nasa1J. The NEG morpheme and the NomP will also bespecined as L-Tone], i.e., low tone, end the OPT morpheme asL+Tone], high tone. (The NomP is further specified as j+High]--cf.,Section 2.51.) A P-rule (P 02) will change the fpature I+Vocalicj toI-Vocalic] and also edd the feature L+Consonantalj.The P-rule for homorganic nasals will then apply to add the other needed features. (Alternatively, we could have specified the nasal _prefixes as arlhisegments with the features fjasal,+Tone].It is likely, however, that at auearlier WI:age these tone-bearing nasals werenasalized vawels, and, at a still earlier stage, syllablesconsisting of nasal consonants and oral vowels.Since we are positing no underlying nasalconsonants we feel there is justification for specifying these morphemeswith underlyiag vowel segments.) This treatment of the nasal prefixes permits us to specify all grammatical morphemes by SqpC9. FurtUrmore, it also peamits us to represent all nominalprefixns by an underlying single vowel segmeat(cf. Section 2.52, Sq5C 11). 2.51, 2.52 65

2.51Verbal Grammaticallaukemea. The grammatical morphemes which coastitute part of Wile verbal word Include the Subject Coulowd (SC) prefixes, the STAtive, HABitual, PROgressive, PASt, PERfect, FUrure, OPTative, IXPerative, and CONsecutive aspactual affixes and the NEGetve, INGressive, REDuplicativa, and SUBordina- tive morphemes. These are discussed in detail in Chapter u

As was stated in Section 2.50, the STA und 111110 morphemes have no matrix representations in the Cctionary. All the other verbal morphemes are either of thtN shape/V/ or of the &ape /CV/. Here, it may be noted thac the tyo INGrestilive prefixes ul-Ach have the phonologi9a1 gorm /bel s:nd /ko/, are derived from the verb roots /be/ 'come' and /k3t/ 'go' respectively.As a verb root., /kat/ conforms to the constraints on primary syllables. As a gramwatical morpheme, which is introduced into the surface ntructure by a transformation, it does not have the final consonant ard conforms to the constraints on secondary, or gramwaticall syllables. The viatematio phonemic forms of the SC prefixes are given ia Section 4.20. All of the SCs as listed in the dIctionary have redundant low tone, which may be specified by SqSC 10.

SciSC 10: rVocaliolx SOL L-Tone

2.52 Nominal IT,arixetioallallahemes The nominal word ocw.rs in surface structures under the node N(oun), as shown ia the following tree diagram:

NomP NR NS where NomP at-anis for Nominal Prefiz, NR for Noun Root, and NS for Nominal Suffix.

every NR in the lexicon ls marked either 4s[±Plurai] Of it is the :cool; of a count noun) or [-Plurall (If it is the root of a non-count noun). (By a convention of obligatory specification; whenever the 'belt of a count noun ie 66 2.52

selected, either the[4-Plural]or the[-Plural] specifica- tian of that root must be selected.) Every NR is also marked for the particular NomP or NomPs itseleV.e.Thus the count-noun NR ba/bit/ 'child' is marked as 1+3/5]. This means that, if /ba/ is specified as[-Plural] it takes the prefix which we arg numbertng"3" (i.e., /5-/ while if it is specified as L+Pluralj it takes theprefix which we are numbering "5!'ci.e., /N/).Similarly, the ponzcount-noun NR sika /sib)/ 'gold, money' ismarked as 1.4.4.1, which means that It takes the prefix which we are

numbering " 4": i.e.,/0/ There are six NomPs, all of which conform to the following Sequence Structure Condition:

SOC 11: I: [+NomP]

T: r+Vocalic 1 124ane J

This condition says thati in their underlyingforms, all tomir.al prefixes are low-tone vowels. (They are also lax, iu conformity with SqSC 8.) The distinctive specifications of the six NomPs appear in the lexicon, asfollows:

r+Low 1 i.e., /1/, as in ,afe(e)[&ff(6)] 'year' Nose' 1: -Nasalj

[-HighI NomP 2: -Low i.e., /1/, as in Ak ana[Ini] 'mother' -Back

fa-High 1 NomP 3:1-Low i.e., /V, as in oba[IA] 'child' L+Back

f+High 1 NomP 4: i.e.,011s:.4s in sika 'some

(+High 1 NomP 5: i.e.,(11], as,inmma[Mmti] 'children', L+Nasali nsa[itei] 'liquor', nka [13ki] 'scent' r+Low NomP 6: i.e.,/1/, (which if; changed tpaiNt by +Nas al.1 a P-rule), as in ampaublafpad 'punting pole', ankaa[kkaa] 'lemon' 2.52 67

(NomP 6 is rather rare.As Christaller (1875) notes: "The prefix am occurs seldom, and appears in some cases as a variation of the prefix 'a'" (p. 22)0, These prefixes may be altered by various P-rules. Thus /a/, /e/s and /0/ may be changed to [6] (ultimately [i]), [6], andLOJ rebpectively by the tenseness-harmony rule, P 69; in Fante /s/ may be changed to /1/ (by a P-rule not included in the present phonology), etc.

In some cases, a Noun Root selects different NomPs in different dialects. For example, in the singular the noun root dan(e) 'room' takes5- in-Akuapem but 6- in Asante. (The plural is formed with in both dialects.) Such dif:erences pust be 4ndicaped in the lexicon.Thus dan(e) is marked 1+3/1 Akj and 1+2/1 As). Similarly, the noun root ponko 'horse', which forms its plural with6.- in Akuapem but in Asante (and which forms the singular with 5- in both dialects), is marked 1+3/1 Akj and 1+3/5 As].

We have not distinguished the NomPs,as inherently singular or plural, although some, e.g., o-,,are, in fact, always singular. Others, however, such asa-, may occur in either singular or plural forms: thus, Effie 'palm nut' (plural mme), abosam 'guardian spirits' (singular obosom). Historically, the singular A- and the plural 6.- are very likely derived from distinct prefixes. We have, however, decided to classify the prefixes according to their phonological shape rather than according to their syntactic properties or presumed etymologies.

The Nominal Suffix (NS) is homophonous with the morpheme which, in Chapter 4, we refer to as the VP-final SUBordinative morpheme: i.e., it is a high-tone mid-vowel archiphoneme, which, as a result of the operation of P- ruleshas a segmental realization only in Asante, and then onlyin certain cases. (It is quite likely that the NS and the VP-final SUB are, in fact, the same morpheme, and that the VP-final SUB is simply an occurrence of NS that is transformationally inserted in certain syntactic contexts.) The dialectal differences in the treatment of the NS by,the P-rules are illustrated by such forms as Ak-Fa afe [6..fx]/ As afee [WU 'year' and Ak-la ohoho [h6126]/As ohohoo [Mann] 'stranger'. Since the phonology of the VP-final SUB is discussed in detail in Section 4.81, and since all of this discussion also applies to the NS, the latter will not be discussed any further in the present section. 68 2.52

Tbe personal pronouns of Akan may be said to belong to the lexical eatecory Noll". Root, since, like other mem- bers of thie categcry. they select particular Nominal Prefixes. (In surfac toms the NomPs occur only when the pronouns are used disjunctively: e.g., Ono? Ninhuu no. 'Him? I haven't seen 'Aim,' where the NomP o- of ono 'him' is present in the disjunctive pronoun but not inISobject pronoun.) Their phonology is not accorded any further discussion in this study.

No further discussion is actorded, either, to the phonology of till various other grammatical morphemes-- e.g., the interrogative particle/a/ or the conjunction /4/ (na)--whichmay occur in surface structures. 69

Chapter 3

General Phonological Rules

3.10Introduction. This chapter is concerned withcertain"general" Phonological Rules (P-rules): i.e., P-rules whoseappli- cation is independent of thesyntactic status of the forms serving as inputs to them.Those P-rules, on ths other hand, that apply specifically tothe finite verb, are discussed in Chapter 4, and those thatapply to other specific syntactic categories are, in general,mot included in ths present study at all. Even within the realm ofgeneral phonological phenomena expressibleby P-rules, this chapter does not aim at completeness.For example, we have not attempted to deal withthe assimilation of vowelsto adjacent vowels, or with theepenthotic vowels that may be inserted between adjacent consonants. Nor have we attempted to deal with intonationbeyond phrase boundaries. As was explained in Section0.50, the function of P-rules is to convert representationsof surface structures into systematic-phoneticrepresentations. The representations of surface structuresthat serve as inputs to the P-rules are in the formof labeled-bracketed strings of morphemes and boundary units,eadh of which is represented by a matrix in which the columnsstand for units(either segmental units or boundaryunits) and the rows for features. In our description, most ofthe cells in these matriceswill be filled in by a + or a -,representing the feature value assigned to that particular unit.These binary values will either have been present in thedictionary representations of the morphemes, or willhave been added by the processof mapping the dictionary matricesonto the Morpheme Structure Conditions (which also are statedin the form of feature matrices).There will be some units insurface structures, however, which are not fullyspecified as to featurevtlues-- cf. Section 2.50. The P-rules will enter valueswhere needed, change feature values, add and deleteboth segments and boundary units, etc., and by such operationsproduce a final string of segments which represent,for the most part, the 3.10 70 equivalent of a traditionalphonetic transcriptionof the sentence generated. Just as the MorphemeStructure Conditionspresent a formal set ofgeneralizations about thephonological redundancies of morphemes, sothe P-rules expressfurther generalizations about thephonological system.A case in point is the distributionof nasal and oralvoiced non- vowels in syllable-initialposition.As we have seen(cf. Section 1.30), no nasalnon-vowels are postulatedat the systematic-phonemic level in Akan. It is, however, true that at the systematic-phoneticlevel nasal voiced non- vowels do occur, andthat in syllable-initialposition they are generally incomplementary distributionwith their oral counterparts,the nasal. non-vowelsoccurring only before nasalizedvowels, the oral non-vowelsonly before oral vowels.The latter generaltzationis captured by a P-rule whichchangesethe L-Nasaljspecification of all voiced non-vowelsto L+Nasalj in appropriat,contexts. (EXceptions to the generalization,such as dam Ldam71 'extinguish' or Ak-As mmaImma?) 'doesn't come', are accounted for by theopention of other P-rules.) A formal statement ofall the P-rules withwhidh we deal, both those discussedin this chapter andthose discussed in Chapter 4, isgiven in theAppendix. In this appendix all of the ruleswill be linearlyordered: i.e., numbered according to theorder in which they areto be applied. In the present chapter, nosuch ordering is attempted, since, in thefirst place, the rulesdiscussed bele are, from thepoint of view of linearordering, interw. spersed among the rules tobe disoussed inChapter 4, and since, in the second place,it has seemed to usappropriate, for expository purposes todiscuss all rulesrelating to such phenomena asnasalization palatalizationetc., in blocks, regardless oftheir linear ordering inrelation to other rules.For example, somenasalization rules must precede palatalizationrules while others mustfollow them. Furthermorethe ordering of some ofthese rules is different in differentdialects. (It is possible that such ordering differences reflectthe way in whieh changes occurred in time after thespeakers of the severaldialects had become spatiallyseparated.) While we do mention sone of these orderingphenomena in the presentchapter, we have not attempted toreflect them in the orderin which the rules are discussed. 3.10, 3.20 71

For the purposes of this chapter, wehave adopted an interim numbering of the rules, usinglower-case Roman numerals: i.e., P i P ii, etc.Atter each such number at the head of a rule will appear apa:enthetic notation re- ferring to the number of the rule inthe linearly-ordered set to be found in the Appendix. (The numbers of tha rules in the Appendix are all two-placeArabic numerals: i.e., P 01, P 02, etc.) In the text of this chapter,however, only the interim (Roman) numbers willbe referred to.

3.20 Nasalization.

All published descriptions ofAkan phonology before the present one have included nasalconsonants as phonological segments. This is true of the description of"morpho- phonemic" segments as well asof"phonamic" segments. Our decision to eliminate nasal consonants asunderlying segments is, however, supported byStewart, who writes (personal communication): "To me it is thoroughly counter- intuitive to recognize nasal consonantsat the morphopho- nemic level in any of the languagesof the so-called 'Akan' group My only reservation is ia the caseof the homorganic nasal prefix, which might perhapsbe better treated as a vowel."(Interestingly, we had decided inde- pendently to represent the homorganic nasalprefixes as nasal vowels--cf. Section 2.50.) It is worth noting, in this connection, that inBaule, a language closely related to the three dialects we aredescribing, the nasal consonants occur only as homorganic nasals, orbefore nasalized vocalic sounds (Stewart, 1956).

In the dialects we are describing,nasal non-vowels occur (at thesystematic-phonetic level) in the following environments:

(1) in syllable-initial positionbefore nasal vowels: e.g., ma[mi] 'give', na [a] 'and'; (2) in syllable-final position, bothmorpheme- medially, before voicpless-homorganic consonants: eog., adanseLadansx] 'testimonY', kom o lkomOrWrrgoiter!, and in morpheme- al position: e.g., namlam] 'meat'; (3) as syllabics,especially before voipeless consonants or npals: e.g., ntadeLntadx] 'clothes', ago Lim] 'palmair: 72 3.20, 3.21

As we shall show in the ediately following sections, all these occurrences of"sal; non-vowels arepredictable if we start with aninventory of systematicphonemes which includes only oral non-vowelsbut both oral andnasalized vowels. In most cases, nasalizedvtwels occurring at the realizations of L+Vocalic systematic-phonetic level are 1) +Nasal] systematic phonemes.Th re area however, some cases in ttldhvowels which are-Mosel] at the systematic- phonemic level acquire al+Nasalcspecification as the result of the applicationof P-rules. Such cases are die- aussed in Section 3.26.

3.21 RegressiveNon-Vowel Nasalization. SAlable-initial nasal non-vowels arein most cases phonetic actual.zations ofeither voiced oral true consonants or oral glideswhich precede a nasalized,vowel.This can be stated asthe following P-rale:

P i (P 06):

[+Consonantal] [+Vocalic] +Voiced [-Vocalic) -4, [+Nasal)/ +Nasal

[-Consonantal]

(When two slashes appear on theright of the arrow in aP- rule, features between thetwo slashes are featuresof the segments affected by therule, while featuresafter the second slash are featuresof segments adjacentto those offected by the rule. Thus,the abQve rule saysthat a L-Vocalicj segment becomesL+Nasalj if the segmentis marked either as[+Consonantal,+Voiced] or as[-Consonantal], and if, further, itimmediately precedes a(+Vocalicl+Nasal] segment.) Some applications of therule are illustratedbelow: Underlying form After P i /1)1/ 'give' [221] /dR/ 'and' mnR] /ja/ 'receive' [pi] (or QC) /WW2/ 'scrape' Nwinf] (or [Rini]) /MI/ 'fear' [Rfl] 3.21, 3.22 73

It is this rule thataccounts for the factthat (with certain exceptionsattributable to the operationof subsequent P-rules) there are nosurface syllables in which a voicedoral true consonant or anoral glide is followed by a nasalizedvowel. Voiceless oral true consonants, on the other hand, are notaffected by the rule, andthus we find such contrasts as: nsa[nal) 'liquor' vs. nsa [nsa] 'hand', 22m[kW] 'confederate' vs. ma[pamn 'sew', kyi(t9In 'squeeze' vs. kzi[tqi?] 'hate', etc.

3.22 Syllable-Sinal NasalConsonants.

SqS0 2 (Section 2.20)specifies all syllable-final non-vowels as stopconsonants. If these segments are vYiced, they are realizedphonetically as nasal consonants. SqJC 3 (same section)specifles all morpheme-medialsyllable- final consonants asL+Voicedj, with the consequencethat all such segments arerealized phonetically as nasal consonants. The P-rule for syllable-finalnasal consonants can be stated as follows:

P ii (P 03):

[4-Voiced] -6[4.Nasal]/ [+Consonantalla

The following examplesillustrate thisrule: Underlying Form After P ii

dub /dub/ 'extinguish' [dum] t on( o) /ted/ 'forge' [ten] sen /Beg/ 'hang' iss13) tumpan /tub=p1g/ 'bottle' [tum=p113] Asante /a+sadmtr/ 'Asante' [a+san=tx] Donk°, /pog=ko,/ 'horse' [powsko]

(The forms presented in squarebrackets are notnecessarily final phonetic formsand are used merelyto illustrate this step in thederivation.) 74 3.22

There are a number of dialect differences in the ultimate phonetic realization of the final nasal consonants which result from the operation of P ii.For example, in Pante a word-final velar nasal (or a velar nasal followed only uy a word-final glottalstop) is always changed to an alveolar nasal.The replacement of velar by alveolar nasals in Pante, however, is not limited to word-final (or pre-glottal-stop) position, but also occurs intervocalieally.Furthermore, the intervocalic replacement of velar by alveolar nasals (but not the word- final replacement) occurs in Asante as well. (For a discussion of circumstances under which the intervocalic velar nasals that undergo the change in question aru derived, cf. Sections 4.73 and 4.740 Since it seems advisable to express the replacement of velar by alveolar nasals in a single rule, we shall state the change in question in the following form:

P iii (P

[4] Cn1/7 {No3ra

(where the morpheme boundary + is not also a word boundary) (In dialect-specific rules, the relevant dialects are referred to, as above, to the right of the rule. For examples of the operation of this rule, cf. Section4.74.)

As is mentioned in Section 4.74, Asante has arule for vocalizing a word-final velar nasal or a velarnasal followed only by a word-final glottal stop. (In the latter case, the rule also deletes theglottal stop.) The rule may be stated as follows: P iv (P71): oc +Vocalic (13] <[?]>- aBack <0> /aBack .1# :As aTense OTense

(Since [33] is [4-High,+Nasal], these features carry over to the L+Vocalic] segment which.replacesit.) This rule, which replaces word-final L4J in Asante by anasalized high vowel agreeing in backness and tenseness with the preceding vowel,is responsible for such dialectdifferences 3.22, 3.23 75 as those in the final segments of Ak[m123]/As [mif] 'community', Ak [di13]/As (dII] 'name', Ak [tov]/As(t50] 'sell', etc. (For an account of the nasalization of the first vowel of As [di] and [t30], cf. Section 3.260 (In sopa Asante subdialects, the replacement pf word- final Lg] by a nasalized vowel is limited tpLls which follow [-Back] vowels.Such dialects have Loin and [dri], but have (to13] instead of [t50].)

There are other dialect-specific rules affecting final nasal consonants, such as the rule Oigausse0jr Secticr) 4.72 wOigh changes certain final Lmjs or Lnuls in Verb Roots to Ms. Such rules, however, are not formulated in the present chapter.

3.23 Nasal Prefixes.

As was noted in Section 2.50, there are three grammatical morphemes which are realized as homorganic-nasalconsonant prefixes.These are the negative morpheme (NEG), the optative morpheme (OPT), and one of the nominal prefixes (NomPs). In dictionary matrices, these morphemes are specified as nasalized-vowel archisegments with low tone in the case of NEG and the NomP, high tone in the case of OPT. (In the case of the NomP thepdictipary matrix also specifies the archisegment as l+Highj.)

There isalso one grammatical morpheme, another of the NomPs (cf.Section 2.52), which is realized as a pro- fix consistingof a low vowel plus a homorganic nasal consonant. Infhe dictionary matrix, this morpheme is specified as a+Low,+Nasalj mment (whiq, like all NomPs is redundantly+Vocalic,-Tonel): i.e., /a/.

A P-rule is needed to change the[+Vocalic] specifi- ca4on of the three morphtmes realized asehomorganic nasals to L-Vocalicl+Consonantalj, and to add * 14rocalic,+Oonso- nantall+Nasal] segment after the NomP /a/. This rule which must precede the rule that makes nasal consonants homorganic with thb non-vowel that follows (cf. Section 3.24), may be stated as follows:

P v (P 02): r+Cocalic r +Vocalil onsonantal / + +Nasal + L<+Low> -40 j +G 76 3.23, 3.24

(The presence of the two formativeboundaries in ths environment prevents the rulefrom applying to thenasal- ized vowel of such agrammatical morpheme as/WI the lst-person-singalar Subject Concordprefix.These boundaries ensure that the rulewill apply only tosingle-segment grammatical morphemes.) Given the surface structureof the centence3mfa Kofi 'He should take Kofi', whichis, roughly: ]iqp[Kuim s[vp[v(sc[5]oprz[i]va[fitt]

+VocaliC1

+Nasal I 1 (whereorT(N)stands for the featurematrix +Tone +G +OPT this rule operates tospecify the OPT segment as ahigh-tone nasal consonant. (At this point in thederivation, the other features are notyet specified.) Similarly the rule specifies the NEG morpheme orthe hamorganic-nasalNomP as alow-tone nasal consonant. In the case of theNomP the rule adds a nasalconsonant after the vowel. /5/9

3.24 HomorganicNasals. There are many cases inAkan where, withinword boundaries, we find nasalconsonants that arehamorganic with a followingnonvowel. Within morpheme boundaries, the obligatoryhomorganieness of nasalconsonants with following non-vowels isaccounted for by SqS0 3(cf. Sec- tion 2.21), and by Pii (cf. Section3.22).For cases in volving the homorganicnessof nasals withfollowing non- vowels across morphemeboundaries (out within wordboundaries), another rule isneeded. Such .mses include, among others, words with oneof the nasal prefixesdiscussed in Section 3.23, and, in mostdialects, words involvingthe reduplication of monosyllabicVerb Roots with finalnasal consonantsin which the finalnasal consonant of the reduplicating syllable ismade hamorganic witilthe initial nowavowel of the VR(cf. Section 4.62, subsection(b)). (In the case of reduplicatedforms of disyllabicVRs that end in a nasal consonant,the situation issomewhat more complex, although heretoo the final nasalof the reduplicating prefix is oftenmade homorganic withthe initial non-evowel of the VR--cf.Section 4,640 3.24, 3.25 7?

The rule for homorganicnasals across morpheme boundaries may be stated asfollows:

P vi (P 25 ): -Vocalic Coronal r+Consonantali [[YHigheCoronal fiBack OBack Laasal yHigh

(This statement of the rule doesnot provide for dialect differences in the treatment ofreduplicating prefixes.)

Some examples of the operationof P vi are:

Before P vi After P vi [pi+pad 'punting pole' [Am+pad LL-4-ban 'doesn't come' (i+ba?) [i+dia] 'should sleep' -tht] [1+6] 'palm oil' [134-gu] Ekgm+kam) 'kill and kill' [kav+kam] (It should be noted that Pvi must follow the palatalization rule P xviii--ct.Section 3.60. For example, given tho under.LyIng form/N+kge 'side' and the surface form Ak,Ot964], it mustbe the case that Pxviii haschanged /k/ to ltg] before P v3. applies,since the homorgapic nasal is the palatal,lp], rather than the velar,l4].)

3.25Progressive Non-VowelNasalization. After the homorganic-nasalrule, P vi, has applied, another rule, which appliesin Akuapem, Asante, and some Fante subdialects, nasalizesall voiced non-vowela(within word boundaries) thatfollow nasal consonants,This rule may be stated:

P vii (P 72)

[-Vocalic] -Vocalic -[4.Nasal]/Xf ] : -pan L+Voiced L(+)+Nasal (where X does not include #ff, andwhere -Fan indicates non-applicabilityin some Fante subdialects) 3.25, 3.26 78

items illustrate somephonetic con- The following in trasts between, onthe one hand,Akuapem and. Asante, of P vii isobligatory, and onthe other, which applicatian fails to apply: those Fantesubdialects(Fan) in which P vii Pan Ak-As LibP] [ilm6.7] ' doe sn ' t come (Adij [Anti] 'days' [(340] 'should pour' [PO?) ' dos sn t do ' [11j6?] 'doesn't die' [13140] L;333,10] and As a result ofthe operationof P vii, Akuapem Asante (and someFante subdialects)lack a distinction nasal voicednon-vowels afternasal con- between oral and. between the sonants within aword. Thus the difference of affirmativefprme of the VerbRoots initial consonants not ba (/ba/pm]) 'come' and maOA/ Lap 'give' is Mud, in these dialects,Enegative forms: e.g., Ak-As [iota] 'doesn't come', Ak-As mmaiimil?) 'doesn't give'. mma contrast in suchforms: There is, however,still a phonetic contrast between theoral and nasalvowels.For, i.e., the nasalized to some while it is truethat all vowels are vicinity of nasalconsonants, thereis none- degree in the between vowels whose theless a clearphonetic difference representations arenasal (or whidh are ex- underlying discussed in Section plicitly nasalizedby P-rules, as which are merelygtven a slightnasal 3.26) and vowels segments. coloration becauseof the nasalityof adjacent

3.26 VowelNasalization. occurrences As was notedin Section3.20, uhile most vowels at thesystematic-phonetic level cor- of nasalized at phe systematic- respond to occurrencesof such vowels there are casesin which L-Nasaljvowels phonemic level, of P-rules. We are madeL+Nasalj throughthe operation in the presentpection In. shall discusstwo such cases by bothrof these cases,the replacementof the L-Nasalj L+NasalLspecifteation represents anassimilation to the the L+Nasalj segment. In one of the cases, an adjacent the other, it is a adjacent segmentis a vowel; in consonant. 3.26 79

SqS0 1 (Section 2.21)the MS Condition specifying the shape of the Akan principle syllable, requires that two successive vowels in a syllable agree in nasality, tenseness, and tone.At the phonetic level, however, the requirement of nasality agreement between successive vowels extends beyond the syllable.In fact, whenever it happens thatat some point in a derivation, two successive vowels within a wordrshow different specifi9stions for the feature [Nasal], the 1-Nasal] vowel is made l+Nasal] by application of a P-rulp. Thiq rule applies rpgardlees of the order in which the L-Nasalj vowel and the L+Nasalj yowel occur. Thus the rule applies when the inherently1-Nasal] past- aspect suffix (cf.. Secticin 4.33) occurs after a Verb Root that ends with,a L+Nasalj vowel, and it also ppplies when an inherently1.4asal] vowel occurs before a IA-Nasal] vowel that has beam derived through application of the 4sante rule (P iv, Section 3.22) which replaces a finalIA] by a nasalized. high vowel. An example of the first type is AY maeLmATJ 'gave', derived from deep structure /ba/ 'give'

+ /I/ (past-aspectsuffix). (For details of the derivation9 cf. Section 4.91, pubstetion (d).) An example of the second type is As 1bITJ 'w41', derived fr9m deep-ptructure [bag]as follows: /bag/ -'lbaKi] by P ii; lban] Cbaf] by P iv; [be] - [bitY] by the P-rule under discussion.

The rule for nasalizing a vowel adjacent to a nasalized vowel may be stated as follows:

P viii (P 73):

r+Vocalic (+) [+Vocalic] LOW:M:VWi. L+Nasal

(This rule incorporates a notational device suggested by Bach (1968), in which the environment is to beinterpreted as either preceding or followingthe segment affected by the rule. Thus P viii is an abbreviated form of the follawing:

(+444-1:::lic](+) [4-Vocalic]-[4-Nasal)/ (+)1+Vocalicio.) 1 L+Nasal

To turn now to tile nasalization of vowels under the influence of adjacent 1+Nasal] consonants, we find that this 80 3.26 is a considerably more restricted phenomenonthan is vowel nasalization under the influence of adjacentL+Nasal] vowels. In the first place, it is only[-Nasal] vowels that wispede nasal consonants, neverthose that follow such consonants, that are changed toL+Nasal]. ample, while the vowel of dumLdamfl (underlying form: /dub/) 'extingrish' is naiirIzed under theinfluence of the following m), the vowel of Ak-As mmu(mmu?] (underlying form: /N+bu/ 'doesn't bewl.' is not nasaliz9d under the influence of the precedingLm]. Second, the L+Nasalj consonant responsible for the chang, mustbe in the same PYllable as the vowel: compare FaLten?) and Ak-As ltu=n0] (underlying form:/tud/) 'forge'. (For details of the derivation of such forms, cf.Sections 4.73 and 4.740 Finally, it is only high vowels, never mid orlow vowels, that are affected. Thus while dum and Ak den [dfv) 'difficult' have nasalized vowels, Ak den[ail 'what', Ak dan [doe] 'clock', and dam(dam] 'madness' do not. The rule for nasalizing high vowelsbefore nasal consonants in the same syllable may bestated as follows:

P ix (P 74):

[...Consonantal] [Vocalic]-(+Nasal)/ = +High +Nasal

It may be noted that, as a result of acombination of the operation of this rule andthe progressive non- vowel nasalization rule, P vii,discussed in Section 3.25, certain systematic-phonemic contrasts maybe eliminated at the systematic-phonetic level.For examplein Akuapem and Asante (and some Fantesubdialects), while affirmative forms involving the Verb Roots dum(/dub/) 'extinguish' and num (/MW) 'suck' are(except in the optative) dis- tincte.g., odumCodumfl 'he extinguishes' va. onum LonBe] 'he sucks'negative forms,involvtngthig7Rs are homophonese.g.,Ak-As onnum Lonnamn 'he doesn't extinguish, he doesn'tsucke77Eishomophony results from the fact that, in negative formsin these dialectsthe initial /d/ of dum is nasalizedafter a nasal prefix by P vii, while the vowel is nasalizedby P ix. (In the case of num, the initial /d/ ofunderlying /Mb/ is nasalized br P i--cf. Section 3.21and thevowel is inherentlyC+Nasalj.) Such an elimination of contrasts doesnot occur in negative forms involving such verb roots as am(/gab/) 'bemoan' and am (/j1b/) 'wave', because, eventhough the initial 3.26, 3.30 81 non-vowels of the roots arehomophonous in negative forms in these dialects(both being realized as palatal nasals, as the resultof the application of differentP-rules), the vowel of Dm, not being high,is unaffected by P ix, and hence contihues toecontrast withthe vowel of 014: compare Ak-As onn amLoppam?) 'he doesn't bemoan' ana onnyam iopple] 'he oesn't wave'.

3.30 Final Consonants.

At the systematic-phonemiclevel, morpheme-final non-vowels are restricted, bySqSC 2 (Section 2.20), to the six stop consonants/p, t, k, b, d, g/, (and syllable- final but morpheme-medial non-vowels arerestricted, by SOC 3 (same section), to the threevoiced stop consonants /b, d, ig/). We have already, in Section3.22, discussed P-rules which apply to the voicedfinal consonan s: i.ec, P ii, which changes syllable-final/b, d, g/ t9m, n, gj respectively; P iii, which changesword-finalrigto Lnj in Fante; and P iv, which changesword-final IA1 to a nasalized high vowel in Asante.

The three voiceless stow", /p, tak/, are changed, in morpheme-final position to Lw, r,?J respectively. Since these latter segments do not constitute a"natural class" within the feature system with which we areworking, it would perhaps be simplest not toattempt to formulate Os rule for the change of morpheme-final/p, t, k/ to 4?) entirely in feature terms but, ins 3adto lw r, 9 9 state, it in some such form as thefollowing:

[w] /[-Corona]] [+Consonanta1] -Back -Voiced [r] /{+Coronall

/[+Back 1 (The segment [r] is a liquidi.e., a[+Vocalic,+Consonantal] segment; this is a segment type notfpund at the systematic- phonemieflevel in Akan.The segment 0] is a glide specified as l+GlottalConstriction]; this feature specification 82 3.30 too is not found at thesystematic-phonemic levelin Akan.) shedding some light In the interest,however, of possibly non-vowels in on thehistorical developmentof the final Akan, wn may propose thefollowing rathercomplex rule, which is stated entirelyin feature terms,and whidh covers the changes not only ofthe voicelessfinal non-vowels but nonmwvowels as well(and which, sine, of the voiced final iii): it replaces rule Pii, is given theinterim number P P 1(P03): L+Nasal] 4+Voiced] (a)

(+Voiced] +Anterior] (b) -Nasal

[+Consonantal] -0 [-Consonantal] -Coronal (c) <+Glottal> <-Anterior -Nasal

[+Vocalic] /1±Coronall (d) -Nasal

(The four subparts ofthis rule are"conjunctively ordered": rule--(a), (b), (c), (d) above-- i.e. each subpart of the applies in turn if theconditions for itsapplication are That is, seg- met.However, no recyclingis permitted. ments that haveundergone the changespecified in subpart (b) of the rule may,where appropriate,undergo subpart resubmitted for the (c) or subpart(d), but may not be application of subpart(a). Similarly, segmentsthat have undergone the changespecified in (c) may notbe resub- those that have mitted for applicationof (a) or (b), and undergone the changespecified in (d) maynot be resubmitted for applicationof (a), (b), or(c).) Line (as of the rulechapges the voicedsyllable- final consonants/b, d, g/ to Lm, n,Ki] respectively. Line (b) voices theanterior vpicelessroralconsonants /p/ and /t/, changingthem toLb] and Ld] respectprely. Line (c) changes thenon-coronal oralconsonants Lbj and /k/ to glides, and, inthe case of thenon-anteriQr adds the fyatlireof glottal constriction;thus Lb.' 4- 1w] and 4/ -0]. Line (d) changes thecoronal oral consonant Ld] to theliquid Lr]. The operation,ofthis rule may beillustrated as follows:

.11 3.30, 3.31, 3.32 83

'extinguish"forge"hang"dance"reach"stand' Underlying/dub/ /tad/ /seg//sap/ /dut/ /sik/ Form

[Beg)] After (a) [dum] [Wu] OM1111110

After (b) [Bab] [dud] [sin After (c) [saw]

After (d) [dux.]

Outpvt of [sin P ii [dulm] [tOn] [se4] [saw] [dur]

3.31Final [w] [fl. In Akuapem andlin sopFante subdialects(which we shall refer to asFe), a Lw] which isOrived from an /p/ (cf. P Section 3.30) underlying morpheme-final In is normally retained,and occurs in surfaceforms. Asante, however, and in someother Fpnte subdialects (which we shall refer to asFa4), a Lw] in morpheme-final position is replaced by aglottal stop, by means ofthe following P-rule:

P x (P 29):

1w] - [fl/ + :As,Fa2

This rule is responsiblefor such dialectdifferences as Ak-Falasaw?]/As -Fa2 isa?] 'dance', Ak-Fa1 [piw?]/As-Fa2 [pi?] 'be thick',Ak-Fal [dow?]/As-Fa2 (do?) 'cultivate', etc., (For anexplan*tion of the glottalstop that follows the NJ in the Ak-Falforms cited, cf. Section4.74, subsection (a).)

3.32Glottal-Stop Deletion. There is a general,non-dialect-dependent, P-rule that deletes allglottal stops that arenot immediately followed by a pause. The glottal stopsdeleted by this rule may have had oneof a number ofdifferent origins.

Al 84 3.32, 3.40

For example, they may reprepent reflexes of underlying morpheme-final /k/ (cf. Piil, Section 3.30); they may represent reflexes of underlying morpheme-final /p/, which has first been changed to Liqj by Pii1 and then changed to [?] by Px (cf. Section3.31); or they may have arisen in various other ways discussed in Chapter 4.Regardless of the derivation of the glottal stops, the following rule is obligatory:

P xi (P 04):

[+Glottal Constriction] 0/ X

(where X is not a [+PB]) This rule is responsible for the absence of the glottgl step after Losiin Losi ho] 'he stands there' (cf. Losi?j 'he stands' )2 after[odo] in As-Fa2 Lodo hoj 'he cultivates that place' (cf.As-Fa2 (odon 'he cultivates'), etc. (Although P xi occurs early in the linearly- ordered set of P-rules, there is a condition on it to the effect that it reapplies immediately whenever the conditions for its application are met.Hente it applies, for example, to outputs of P x, even though P x occurs at a laterpoint in the linearly-ordered set.)

3.40 Medial Intervocalic Consonants. Some rules that affect medial intervocalic consonants have already been mentioned:viz., P (Section 3.21), which nasalizes a voiced syllable-initial consonant that is followed by a nasalized vowel, whether the consonantin question is initial or medial within a morpheme or word; and P iii (Section 3.22), which, in Asanteand Pante, changes an intervocalicLn] to Ln]. In addition to these rules, which affect certain intervocalic consonantsthat there are several P-rules are followed by nasalizedvowels9 that affect intervocalic consonants(within a disyllabic or longermorpheme) that are followed by oral vowels. When the systematic phoneme /d/ occurs inintervocalic position within a morpheme, it is as we have seen, changed to [nj if the following voyel isyNasal]. If, on the 9ther hand, this vowel is L-Nasal 1 the /d/ is changed to lr]. Thus systematic-phonemic /flu/ 'call' is realized phonetically as (f(x)re], systematic-phopemic/a+bvdo+be/ 'pineapple' is realized phonetically as Lab(y)robeit,etc. (For an account of the optional deletion ofLI] and Luj in the examples cited, cf. Section 3.750This change may be stated in a P-rule as follows: 85 3 AO

P xii (P15):

f+Vocalici [+Vocalic]/V +Corona' L-Nasal +Voiced [ (Alternatively, since this rule follows P i, we maystate it in the following form:

Pxiil: +Consonantal +Coronal -1+Vocalic)/V V +Voiced -Nasal E.

This formulation presupposes that anunderlying /d/rfollowgd by a nasalized vowel has alreadybeen specified as L+Nasalj. Thus the only cases in which thecondition for P xiii are met are those in which the/d/ is followed by an oral vowel.) 1 P xii (or P xii), by changing Vile underlying[-Vocal- icj specification of /d/ to L+Vocali1J, replaces/d/ by the liquid Lr] in the specified context.This rule, in combination with P i, accounts for the factthat [d] never occurs morpheme-modially in Akan: i.e., it is always replaced either byrimi or by 41, acepprding towhether the following vowel is L+Nasal] or L-Rasalj. (In some subdialects of Asante, there is a set of exceptions to thisgeneralize- pion.These are words in which a morpheme with amedial IC followed by an oral vowelis followed by a segmental form of the Nominal Suffix(cf. Section 2.52) or of the homophonous Subordinative suffix(cf. Section 4.81). In such worgl.s in those Asante subdialectsthere is alternation between Lr] and Ld] at the phonet4cleyel.Thus /bidi+NS/ 'witch' may be realized as Lbiriej orL'Oidie], and, /jadi+SUB/ 'be ill' may be realized asLjarxej or Wafts].) It should be noted that P xii is arule restricted to the morpheme, and does notapplyacross morpheme boundaries,. Tkius one finds intervocalic Ldjs inwords such as oda Loded 'he sleeps', whichconsists of the Subject Con- cord prefix o- plus the verb rootda, or Ak-As akade [akade], in which kade is a compound noun rootformed from ka 'smell' andade'rEing'. In the case of progressive verb forms, however, the initial/d/ of the Progressive 86 3.40 morpheme /di/, is realized phonetically as[r] after a Subject Concord prefix, in spite of the factthat, at the surface-structure level, it is preceded by amorpheme boundary.Thus /o+0,1+bai 'he is cming' ts realized phonetIcally as AkLoriba], not as Lodxbaj.This suggests ttle need for a P-rule whicheliminates morpheme boundaries (l+FB]s) between Subject Concord prefixes and Aspectaal prefixes (of whidh the Progressive isone). This rule may be stated:

P xiii (P 14):

(+FB: -100/[+SC] LEASP]

We must assume that P xiii preqedesP xii in prder to account for the occurrence of Lrj, ratherthan Ld], as the initial segment of the Progressive morpheme.

When the systematic phoneme /g/ occurs inintpryo- calic position within a morpheme, iVis changed to 1.4.1 (by P i) if t4e foll9wing vowel isL+Nasal]. If the following vowel is L-Nasalj, and if themedial syllable beginning with /g/ is of the shape CV or CVO(i.e., if it is not of the shape CVV or CVVC), the/g/ is deleted by means of the following rule:

P xiv (P 08):

[g] 01/V V(C)=

This rule operates, for example, tochange the underlying forms of therVerlo Root; boa/bugs./ 'help' and /wigek/ 'finish' to Lbuaj andLwleI respectively. (Actually, in the case of /wigek/, PiilSection 3.30), since iteprecedes P xiv, would already have changed thefinol /k/ to L7j, so P xiv would operate to changelwige] tQ lwie?). The final phonetic form of this morpheme isL4ie?j, with the change in the initial segment the resultof the palatalization rule, P xviii (Section 3.60)0 Note that, if the /g/ is followed by a sequenceof two vowels, it is not deleted. Thus the /g/ of hodwo(w) /hoguep/ 'become slack' is not affected by P xiv,and is pealized,,after the application of variousP-rules, as the 1.09) of Lhodlmo(w)]. Similarly, after the /I/-insertion rule, P xix (Section 3.61)1 hasinserted an /i/ after the /g/ of kagya(w) /kagap/ 'wither', with theresult that the 3.40, 3.50 87

/g/ is followed by the vowel sequence lie], P xiv fails to apply, and the /g/ is realizedtfter the application of wkrious P-rules, as the ida, 041 kadje(w)?].

(We have only investigated the application of P xiv in detail in the case of Verb Roots.Noun Ro9ts segm to present a number of exceptions: e.g., 11.gye Lbudjsj 'chin', which is, presumably, derived from under ying /bugx/.)

Intervocalic consonants other than /d/ and /g/ do not normally undergo P-rules that depend upon the intervocalic position of the consonants (although they may undergo various P-rules, such as nasalization, labialization, and palatalization, which affect certain syllable- initial consonants followed by certain vowels, whetheror not the consonants are intervocalic). There is, however, a rule whidh affects an intervocalic /k/ in a disyllabic Verb Root (but not, for example, in a disy4abic Noun Root), and which operates to replace the /k/ byaLg] (which is ultimately either nasalized by P i or deleted by P xiv). Since this rule is specific to Verb Roots, it is discussed in detail in Chapter 4--cf. Section 4.77, subsection (a).

3.50Labialization

Within a morpheme, a syllable-initial non-vowel immediately followed by a [+Round] vowel is always labialized in Akan: i.e., the non-vowel itself always acquires aC+Round] specification. This labialization process may be stated in a P-rule as follows:

P xv (P 09):

i-Vocalic] [4-Round]/- L+Round]

The [-I-Round] vowel that causes the labialization may be the only vowel of the syllable, or it may be the first vowel of a two-vowel sequence. (As is specified by SqS0 2, Section 2.21, the first vowel of a two-vowel sequence withjm a p4mary syllable is always /11/, and heace obligatorily I+Roundj.) In the former casei.e., when the labialized non-vowel is the initial segment of a CU(C) or CO(C) syllable--the rounding of the non-vowel is not always easily discernible. (It is, perhaps, more discernible in Fante than it is in Akuapem or Asante.) However, just as instrumental evidence has shown that there is 88 3.50

labialization of the initialsegments of English soonand coat (though lessthan in swoon andquote) so,irriel aiff ident, would it show thatrthereis abialization,of the initial segments of Akan somLswem?] 'serve' and kojkwo?.1 'go'. (Except where we areparticularlyconcernirwith labialization, we do not, inthis text, normallyindicate or ko. it in phonetiotranscriptions of items such as som That isingtead of[sw0m/ and [kwo?), we wouldusuaillf transcribe Lailmn andLk*? respectively.)

When the labializednon-vowel is the initialsegment of a CUV(C) syllable,the /U/ which immediatelyfollows the non-vowel isdeletedso that thelabialization of the non-vowel is, gt thephonetic level, the onlysign of the 1 presence of aL+Roundj vowel in theunderlying structure. The /U/-deletion rut.e maybe stated as fellows: P xvi (P10):

[U] -40 0/ (+Vocalic]

(That is, any[U) that immediately precedes avowel is deleted.) Some examples of theoperation of P xv andP xvi (which must apply in thatorder) are: Before P xv After P xvAfter P xvi

sua 'besmall' [suan [swus?] [ewe] twe 'antelope' [kW (kweT] [kwT] hwe 'look at' [hue] Chwue] [live]

(In the case of tweand hwe, palatalizgtionlaterrapplies, resulting iatheMnal ABietic formsLW] and iv/cis-- cf. Section3.600 It may perhaps seemsomewhat curious toportulate underlying OVV(C) syllablesin which the first 11Nel is obligatorily /U/, and tofurther postulate thatthis /U/ permits us is obligatorilydeleted. Doing so, however, to eliminate labializednon-vowels at thesystematic- phonemic level. That is, we can saythat labialization of non-vowels occurson1 .zwhen a1+Round] vowel follows the non-vowel in theunderlying form. 3.60 89

3.60Palatalization. In al; dialects, the[+Back] non-vowels 1k, g, w, h, ow (or[2]).1 are palatalized, insylla4e-initial position, when they occurimmediately before theL+Palat 1] vowels /I/ and AI, and arerealized phonetically as to, d4., 4, 9, pm(or LC)] Thus /ke/ -* LW c( tri 'aivide°, /ge/ -.die) (Eye 'receive), /III/ -* Nil(we 'nibble% /11i/ -91] (Me 'border°, and /wid/(after becomingLowYn] through the application ofother P-rules) -. [p47n) (nwen(e) 'weave°. (The labialized(+Back) non- vowels Lkw, gw,hid1 which may, as a result of the[0- deletj.on rule, P xvi(Section 3.50), occur immediatelybefore L+Palatal) vowels, arealso palatalized inall dialects, being realized as[tog, 44, CO respectivelyucf.Section 3.61 for details.) There is one major exceptionto this generalization: palatalization does not occurif the following syllable(within the same morpheme) beginswith /t/ 91' /8/. Thup /ketc/ 'mat' isrealized phonetically as limbs], not Ltgetc], while/kcsx/ 'big' is realized phonetically asLkcsx], not ltocsx].

The non-dialect-dependentpalatalization rule may be stated as follows:

P xvii (P 11):

[-Nasal] li+Vocal y [-Vocalic.] L+Palatr +Back +Palatal [Wesel +Round]

(where X (V)(C) = t/s

This rulp specifies thatall [-Nasal] back non - vowels - -i.e., Lk, g, w,h]- -aresubject to palatAlization in the appropriatecontext, but that onlyround L+Nasal] back non-vowels aresubject to palptalization.Thus [ow] is subject topalatalization, butLfij (which arises as a result of the applicationof P i to an underlying/0/ sequence - -cf.Section 3.21) is not.This accounts for the fact that, e.g., underlying/hi/ 'where' is realized phonetically as[al, not as [gly. (As we shall see in Section 3.61,[Sid, which, like ow] is 1+Nasal,+Round], is subject topalatalization.) Fante offers a few apparentexceptions with regard to the palatalization ofnon-nasalized Lb] before 3.60 90

[hen] [+Palatal] vowels: e.g.,(e)hen [(x)hen] 'we', hen Comparison of theseforms with forms that occur 'ship'. the underlying in other dialects,however, reveals that syllable-initial non-vowelsin these items areprobably Um)] and /j/ rather than/h/o Thus for 'we' Akuapemhas is Asante has(3IT]. In the case of'ship' the situation somewhat more obscure,since both Akuapemand Asante have [h]. (The initial[9]: i.e., the palatalizedform of Akuapem form isLge10, the Asante formNIT].) However Christaller's Dictionarylists a dialectal form za of the (presumably(jev]). Whilethen, the development Fante forms in questionis not altogetherclear, it seems likely that theydo not constitutegenuine exceptions to the palatalizationrule. In Pante, thepalataliption process hasbeen ex- tended so that,noto4ly theL+Back) non-vowels, but, in addition, the L-Backjtrue consonants arepalatalized be- nj are fore[+Palatal] vowels. Thus Ep, b, f, m, t, d, s, all palatalizedin this context inPante, being realized Compare as[pj, bj, fj, mj2 ts, dz, sj,p] respectively. Ak-As Lpej/Fa Lpjej(21 Iseek9, Ak-As[abe]/Fa [abje] (abe 'palm% etc. (Except in the case ofits), [dz], and GT the effects of palatalization arenot easily dis9ern- Ltsx?) ible when ttlefollowing vowel is/I/o Thus while Fa 'feel', FaLftx] 'take', and Fa[pI] 'this is' are easay distinguished fromtheir Ak-As counterparts[tx?), [dx], [bji] 'some' andmni] respectively, Fa[pjii] 'many' and are not soeasily distinguishedfrom their Ak-As counter- In this text, except parts[pii] and (bi] respectively. particularly concernedwith palatalization, where we are instead we do notn9rmally inOicate itin such items, but Lpii] and Lbi] for the Pante, aswell as for use, e.g., For Pante, then, the the Akuapem-Asantepronunciations.) following additionalpalatalization rule maybe proposed: P xviii (P75) +Vocalic] [-Vocalic I [+Palatal]/ . Lra +Consonantal +Palatal

This rule may beassumed to follow Pxvii, and to account not palatal- for the palatalizationof all true consonants ized by thatrule. (Note that, since thenon-vowels 3.60, 3.61 91

Affected by P xviii are restricted to trueconsonantft-i.e., l-Vocalic,+Oonsonantal] segments--neither the glideLh] nor the liquid[r] is palatalized by it.) We have thus far not accountedpr the occurrence of DA, the palatalized counterpart ofLg], in items such as Fya(w) [dbe(w)7] 'leave', ma[dje] 'fire', etc. Our postulated systematic-phonemic formsfor these items have initial /ge/: i.e., /gap/ 'leave', /ga/'fire'. Since the low tense vowel /a/ is[-Palatal], it cannot be the occurrence of this vowelafter the /g/ which causes the palatalization. While we might, in seeking to provide anaccount of such phonetic forms as1d4a(w)?] and WO], have postulated /giap/ and /gia/, insteadof /gap/ and /ga/, as the underlying forms, we have not done sofor several reasons. In the first place, such underlyingforms would represent unique instances of two-vowel sequences(within the syllable) in which the first vowelis not /U/.That is, there is never any need forpositing such underlying syllables as /kia/, /hia/, /gie/, etc. Second, and per4- haps more important, .phere is go9d evidencefor positing a rule whichinserts Li] before La] under several sets of circumstances. A detailed presentation of thisevidence is deferred u4i1 Section 3.61, sincethe evidence bears not only uponLdj.a), et9.: but als9 upon the palato-labialized non-vowelsLtl, ld411], and l94.1, and upon the fronted round vowel For the present we shall simply assume that there is a rulewhich inserts[i] between !Ed and [a], and that this ruleprecedes P xvii, so that /ga/ Igia] (by the [Il-insertion rule) Id4ie] (by P xvii). This last form is subject to an obligatory/I/- deletion rule (which will also be presentedin Section 3.61), such that [djia] [dje].

3.61 Palato-Labialized Non-Vowels and Fronted Round Vowels.

It may happen, as a result o; theapplication of the labialization rule, P xv, and theLU]-deletion rule, P xvi (cf. Section 3.50); that a sequence is derivedin which one of the labializedL+Back] non-vowels [kw, gw, hw] is imme4igtely followed by one of the(+Palatal] vowels [I] and LEJ. Under such circumstances, the palatalizationrule, P xvii (cf. Section 3.60), applies asit does in other cases in which backnon-vowels are followed by palatal vowels. Palatalization operates in such cases tochange 92 3.61

[kw] to lt94], Ligw] to ki.40, and paw] to [c4]. The following are some representative derivations: 'leopard"kernel"look at' Underlying Form /kui/ /gux/ /hue/

After P xv Ikwui] [gwux] Lhwue]

After P xvi lkwi] kswx] Lhwe]

After P xvii [tv.p.] [d4V] [94e]

The above derivations of palato-labialized non-vawels all involve underlying forms in whichL+Back] non-vowels are followed by /UI/ or/UE/ sequences. We also, however, find palato-labialized non-vowels occurring in certain cases where, in the underlying form, al+Backi non-vowel is followed by one of the sequences /us/ or/ 5/. Such cases show significant dialect differences.Specifically, in Asante, the phonetic realizations of all of the following underlying sequences--/kue/, /k115/, /gue/, /ha5/-- involve palato-labialized non-vowels, while in Akuapem and Fante, only the phonetic realizations of the underlying sequences /kue/ and /ka5/ involve non-vowels ofthis type. (The phonetic realizations of underlying /gue/ and /ha5/ sequences in Akuapem and Fante involve labialized,pather than palato-labialized, pon-vowels.)Thus we find Ltcmen 'cut' (from /kuet/) and 1t915] 'scar' (from /k115/) in all three dialects, but where Asante hasCdage] 'market' (from /010/) and 1.945 J 'scent' (from /ha/), Akuapem and Fante have [pre] and11145] respectively. To account for such cases, we posit a rule, wit4 appropriat, dialectaLrestrictions, which inserts anLI] (i.e., an Li] or an Li]) between a(U] and a low tense vowel in pertatn cases. This rule will, e.g., change /lag/ to Lkardj in all three dialects, and will change /gue/ and Aire/ to Lguie] and[hare] respectively in Asante only. (For the subsequent development of forms such as[kui5], lguie], and batin, see below.) (Interestingly enough, Christaller (481 ) cites kwia as the Pante form corresponding to Ak twa ltctie?]emirit the time he was writing. Furtheras Stewart(1966a) has pointed out, the Asante forms cited in Koelle's Polyglotta Africana(1854) suggest that pa the mid-Oanteenth century, Asante had someOing like igwie] and LEwilj where it now has[dal's] and 1945J.) 3061 93

In Section 3.60, we suggested that forms suph as [d4ej 'fire' resultedfrom the application of an 1I3- insertion rule changing underlying /ge/ to [gie], which, in turn, thep became subject to the palatalization rule. Clearly the 113-insertion rule that is needed to generate (d&a]can beecombldied with the [IJ-insertion rule needed to generate Lt945.1, etc. This rule may be stated as follows: P xix (P07):

[g] +Low [i]/1{[+Back]} :As [+Vocrlic] +Tense [k] :Ak-Fa

The rule says that [I] is inserted between [g] anda low tense vottl in all dialects. 4 says, further, that in Asante LI.' is inserted between lkiebbl and a low tense vowel 011 [gu], and [hi] are the only sequences of (+Back] and,(+Vocalic] segments thatoccur immediately Detore low tense vowels), and that in Akuapem and Pante LI.' is inserted between Lku] and a low tense vowel.

4fter P xix has applied, the labialization rule (P xv), the LIN-deletion rule (P xvi), and the palatalization rule (P xvii), apply in that order. Tben another rule is necessary in order to delete the LI]s that were introduced by P xix. This rule may be stated: P xx (P13):

[1]-1° OV LI-Vocalic]

(That is, any [I] that immediately precedes a vowel is deleted.) The derivations of twa [410] 'scar', Ak-Fa ma [gwe)/As dwa [de] 'market', and Ak-Fa hua (Tati)/As hwa [9115] 'scent'may now be illustrated: 3.61

'scar' 'market' 'scent' Underlying Form /k05/ /gue/ AU/ Ak-Fa As Ak-Fa As

After P xix (ktig] (lgue]) [guiej Wks]) Cida]

After P x-v (kwaig] Egwuo] [gwuio] Dwas]Dwais]

After P xvi (k0.5) [gwe] [gwie] [65] [Ma]

Atter P xvii [441'1] [dltlia] [9415]

After P xx [40] (Mlle] [945]

Related to the phenomena wehave been discussing is the phenomenon of thereplacement of an underlying/u/ (or /01) by the fronted roundvowel (ii] (or[u]) before an /a/ (or /5/) at the beginning ofthe next syllable. (The sequence(ume] arises normally from the deletion of a medial /gi--cf. P xiv, Section3.40.) This phenomenon shows a certain amount ofdialectal and subdialectal vslae- tion. For example, in Akuapem t4echange ofra41to(unal is most common after aL+Coronalj consonant (i.e., It, d, n, 4), while in Asante it is equally commonafter other segment types. Thuo,in both Akuaoem aqd Asante we find f94ms such as tua[tile] 'stick's dua [dig] 'plant', nua(05] 'sibling', and sua (sia] 'study', butforms such as[kig] for kua 'doll', [ba] for bua 'close',etc. are more common inAsante than in Akuapem(whidh would usually have [k46],[big], etc.) (We hove not been able to investigate the distributionof [ii=t1 sequence in Pantein anydetail.) It seems reasonable toderive the[i=e] sequence (and the[a=s] sequence) in some such way as thefollowing: [iale]. That is, we wouldpropose [ine] [unie] -0 [U=sie] 001 r . that ijLs Iinsertedbefore an Li epreceded byral.), just as it is insertedbefore an[e] preceded by a sequence consisting of a[+Back] non-vowel plusbir.within a single syllable. Thereafter the gllable-finalul is fronted under the influente of theLI] (a process an4ogous to palatalizatios of(+Back] non-vowels before[4), and, finally, theCrj is deleted, just as it is betweenthe palato-labialized non-vowels ande51 3.61 95

In order tooctend [I]-insertion so as to cover both the cases we havejuat been discussing and the cases discussed earlier inthis section, we may rewrite P xix as follows:

P xix1(P07):

+Back = ocalic] 0 - igh r+Low +Tense 1(+Back] : As [+Vocalic] [k] J : Ak-Fa

(This statemept of the Rule is very general as far as the insertion of LI) after riej= is concerned, since it does not allow for dialect differences dependent upon the non- vowel that precedes the Vd].We have also not attempted to account for the sporadic occurrence of,ponted round vowels in other contextsue.g., Ak noa [nna] [ait]'cook', Ak-As 2111 [pad - bite] 'appear'--which also presuppose application of an [I]-insertion rule.)

As for the rule that changesNItoa)before this may be stated:

P xxi (P 12):

1+Segment] [tekci]' ,[a]

The rule need not refer any more specifically tortil), since Vii] is the only segment thatoccurs in the context in question.)

There is no need to alter the [I]-deletion rule (P xigg, above), since it is already stated so as to delete anyLq jemmediately followed bya vowelowond so will apply to anIJ thathas been inserted after Lujam just as it applies to an [I] that has 1?een inserted in the other contexts specified in P xixi.

We may illustrate the derivation of the fronted round vowels as follows: 96 3.61, 3.70,3.71

'stick' 'plant' Istuayl Underlying form /ta=16/ /da=6/ /sitagis/

Before Pxixl [tam;) (daau csa.i] 1 After P xix LdUie [sibai] After P xxi [ta=1.4) [463 After P xx [ta.6] [da.;] [sa.i]

3.70Vowel Rules. There are a number ofP-rules, pome general, some dialect-specific, which operate uponL+Vocalic,-Consonantal] segments in certain contextsto change their phonetic quality. We have already discussedthree such rules: i.e., P viii and P ix(Section 3.26), which nasalize vowelsin the vicinity of L+NasalJvowelp and cpuonants, and P xxi Oection 3.61), which changesvall to NJ in the context l=111. There are other rules whichadd or delete vowels in certain contexts. The rule& of this type which wehave already discussed are: Vile LUJ-deletion rile, P xvi (Section 3.50), and the LIJ-insertionand LI]-deletion rules, P xixi and P xxrespectively (Section3.61). In addition to the rules wehave already discussed, there are a number of othersthat pertain to vowels.These rules are discussed in theimmediately following sections.

3.71 Vowel Tensing.

SqSC 5, in Section 2.30,specifies that the vowels occurring within a singlemorpheme are either all tense or all lax.As was pointed out inthat section, however, vowel tenseness harmonyin Akan is not limited to the morpheme, but manifests itselfto some extent both across morpheme boundaries and acrossword boundaries. Vowels occurring in grammaticalmorphemes, whether the morpheme is a prefix or asuffix, are all idherently lax, as is specified bySqSC 9 (Section 2.50). But when the first (or only) vowelof the first root syllable that follows a grammatical prefixis a non-low tense vowel, the vowel of the prefix isnormally realized as phonetically tense. Similarly, when the last(or only) vowel of a root syllable that immediatelyprecedes a grammatical suffix is 97 3.71 a non-lowtense vowel, the vowelof the suffix is normally realized as phoneticallytense.For example, given a verb formed with the VR/fiti/ 'enter, prick' (of whichthe first and last syllablesboth have non-low tensevowels), any affixesincluded in the verb normallyhave tense vowels. which Thus the verb Ak-Asokofitii 'he went and entered'9 , in the surface structure,'onsists of the subject-concord prefix /3/, the ingressiveprefix /ko/, the VR/fj.ti/ and the past suffix/x/, is realized phonetically asLokofitii]. It should be noted that itis only the non-low tense vowels in root syllablesthat tense aprecediEFOTTollow- ing vowel in a grammaticalmorpheme.Thus prefix vowels of verbs formed with aVR such as /keri/ 'weigh',in which the first vowel is tensebu$ low, doot become phonetically tense: e.g., Ak orebckariLorxbekmri 'he is coming to weigh'. (For the cnange of /e/ to La19 seebelow.) Simi- larly, a suffix vowel of averb formed with a VR such as /bise/ 'ask', in which the lastvowel is tense but low, do not becomephonetically tense: e.g., Akobisae lobisex] 'he asked'. Apparent exceptions to theclaim that it is only non- low tense vowels thattense preceding vowels ingrammatical porphemes are provid9d bywor4s such as Ak-Fa obegua Lobegwen/As obedwa Lobed6401'he comes to skin', okogyam (okod6em?] 'he goes to condole', etc. However, if we investigate the derivationof the root morphemes insuch words, we fipd that, at somepoint in the derivation,there is always aLiq or a non-lowtense vowel--before the low tensee] that occurs as the first vowelin the final phonetic form.Thus the underlying formof the root Ak-Fa Lgwe?)/AsWile?) is /guek/, and the[u] of the root j.s present in thederivation until it is deletedby the 0)-dele4on rule, P xvi(Section 3.50). In the case of the rootLdjom?], while the underlying formis /geb/, the M-insertion rule, P xixl pection3.61), obligatorily inserts an[1.) between the lg] and theCa], and this (J.] remains present in thederivation until it isdeleted by P xx (Section3.61). In order to accountfor these apparentexceptions, then, it would seem tobe necessary only to seeto it that Or tensing theprefix vowels precedes the ;ule responsible the $he LUJ-deletion andLI)-deletion rules (and follows 11)s-insertion rule). 98 3.71

Such a solution, however, is not in fact possible, given the constraints of the model with which we are work- Jag. This is because the model requires that the P-rules operate cyclically (cf. Section 0.50). That is, the entire cycle of applicable P-rules must apply first to the "innermost" constituents of an utterancei.e., the morphemes--before any rules are applied to larger constituents, such as words. Since the vowel-tensing rule.applies across porpheme boundaries, it is clear that the LUJ-deletion and LIJ-deletion rules, which operate within the morpheme, must be applied before it. Thus, even if the vowel-tensing rule were placed prlier in the cycle of P-rules than the NJ-deletion and 1I)-deletion rules, it would simply be inapplicable at the morpheme level, and, before it became applicable at the word level, the morpheme-level vowel- deletion rules would inevitably have applied.

We shall not attempt to solve this problem here, but shall, instead, merely note it, and acknowledge that our rules as presently formulated do not give a fully adequate account of the phenomena in question. The rule for vowel tensing within the word may be stated as follows: P xxii (P69): ocalic [4-Vocalic] [4-TenseP X-14w Y. +G -- +Tense -G

(where: (a) X and Y do not include ##; (b) if the affect'd segment precedes X, X does not include a l+Vocalic,-G] segment; (c) if the gfected Bement follows Y, Y does not include a l+Vocalic,-GJ segment) (In the above statement of P xxii, the representation of the context in which the rule applies: i.e.,

+Tense --- -Low -rouilicG

is an abbr,viation for the two contexts: X1+Vocalic, etc.] and L+Vocalic, etc.] Y .) 3.71 99

Unlike vowel tensing within the uord,vowel tensing across wordboundaries (including the internal word boundaries of compound words--cf.Section 2.30) is strictly regressive: that is, a lax vowel in the lastsyllable of a word is madeOnse when the first vowel of thefollowing word is markedL-Lowl+Tensel-G], but a lax vowel in the first syllable of a word is not madetense when the last vowel of the preceding word isesomaEked. Thus iq the sentence Ak Kwame fiti ne nsaLkwaami fiti nsaJ 'Kwame pricks his hand' pie u4derlying final /f/ of Kwame ismade tense before fiti LfitiJ, but theunderlying /777 ne 'his' is not made tense after fiti. Vowel tensing across word boundaries alsodiffers from vowel tensing within the word in thatonly 9ne syllable (i.e., the last syllable of the word beforetherL+Tense] root) is affected. Thus, while in Ak wakofitiLwamkofiti] 'he has gone and pricked', all of theprefix vowels have been pense4 under the influence o; thefirst yowel of the root LfitiJ, in Ak Kwame LkwaamI fitiJ, only the last vowel of Kwame has bean tensed. (According to our observations, it seems to be the casethat a word-final non-vowel prevents the tensing of apreceding vowel. Thus Ak 3 on fiti 'Opong pricks is realized phonetically as Lopozj fiti] rather than as[om fiti]. In our formulation of the rule for voweltensing across word boundaries, we shall assumethat this "protection" of a laxvowel by a following non-vowel is general.)

The rule for vowel tensing acrossword boundaries may be stated asfollows: P xxiii (P83): -Low [+Vocalic] [+Tense]/.....#00 -Tense +Tense -G

As the rule specifiesthe vowel responsible for tensing across word boundariesmust, like ttle vowel responsible for tepOrng within wordboundaries, be L-Lovq. Thus the final LTJ of Kwame isnot tensed before th9+Lowe+Tense] vowel of the firari011ableof, e.g., Ak kariLkeri] 'weigh' in Ak Kwame kari Lkwaamfknri] 'Kwame weigi-(it)'. Further- more(agirririn the case of vowel tensing withinword boundaries), the vowel responsible fortensing must be the 100 3.71 first vowel e_tf a root morplleme: i.e., a morpteme marked with the feziture L-G] OZ. 6action2.50). Thus the final vowel of Kwame t3 ;lot tezseei in Ak:Kwamerefiti Lkwaamf rifiti] 'Kwame is rrinting it',siza--747717-113 a gram- maticR1 morpheme (with an inherently laxvowel that has been tensed by P xxiv).

As in the case of P there are apparent exceptions to the claim that it ic Eully non-llwtenae vowels that are responal'hle for tenciag pzecedingvowels! e.g., Ak-Fa Kwamema Lkwusmlgwe?]/As !imam iwa [kwaami dlgej 'Kwame skins it', Kwame am[kwaamI d4emfl 'Kwame condoles', etc. In this case7.71O1 tLe protlem posed bythese apparent exceptions is not rest,1-,Id Borely by pointingout (as is true) that at sore poirn ithz Orivation..og roots 1j.ke Ak-Fa Lgwe?]/As LAW] or lik6idlaml a Luj and/or Li] precedes the(e), since, given 4141s cyclic ordering of the P-rules, the iul and/or Li] will heve been deleted at the morpheme cycle before P xxiiiapplies at the phrase cycle.Once more, we shall simply leave thsproblem unresolved. We have mot yet dealt with the questiona the phonetic realization of, 6.e.. the firstvowel of Ak-Aa Lkarij 'weigh'(from /keriii. It is in :act the case that, in Akuapem and Asgnte, whenever aL+Tense,+Low) vowel is followed by aL+Tense] vowel in the next syllable thej+Tensea+Low] vowel is normally rgalized 4c.7-hor,&itiliy as(s) (or Li0), rather than !.el Or 1..T. Th .l. is so whether the t+liowj vowel is inherently +Tense] oi- !)as acquired its[+Tense] specification as a result of the gpplication of P xxii or P xxiii. Thus in Ak-ts afi Lafifl 'has left', theperfect-aspect prefix a-/117: which has been mad, tense throughthe operation ofPXXiva iB realised as Ls].Similarly in Ak-As oba fi[oho fll 'a child leaves', the vowel of ba/ba/ 'child', which has bee; made tense through theoperation of P mor, is realized asUm]. We have also not yet dealtwith the fact t at, in thosg 2ontexts in littich Akuapemgnd Agante Oave il, Fallte has Cal: e:g., N.Lkeri]/Ak-As Lkarijs Fa Lefi? /Ak-As (afifl, Pa Lobe fiall/Ak-As [ob.fil. To deal with these phenomena, the followinglow-level P-rule may be proposed: 3.71, 3.72 101

P xxvi (P84):

+Vocalic 1 [+Coronal] : Ak-As - /...(0,4+Vocalic] +Low +Tense +Tense [-Low] : Fa [ P xxvi operates to replace [e] by 6] (which is identified as aC+Coronal] low tense vowel, as opposed to the [-Coronal] low tense vowel [a]) in Akuapem and Asante, and to replace [e] by the [-Low] vowel [e] in Pante.

3.72 Vowel_22sgas.

The non-low front vowels ([I] and[E])are frequently replaced by their back-vowel counterparts (0] and[0] respectively) when preceded by a L+Round] non-vowel in a checked syllable (i.e., a syllable with a final nonm-vowel). This process shows considerable dialectal and sub4alectal variation. Thus, while the replacement of [I] by IA in the context in question is foun4 in some items in all dialects (e.g., Ak-As Ld4tiama]/Fa Ldjtfamba] 'business',from underlying /guib+bai, which has been changed to[dNim(b)a] by other P-rules), there are other items in which the replacement is dialectally restricted.Thus Akuapem has Lmuwn where Asante has [gin for 'dry up' (from underlying /wIp/, which has been changed to Ak [4x0]/As [mx?) by pther P-rules).On the otlaer hand Asante normally has Lpmel where Akuapem has Lptife] for 'weave' (from under- lyingwYd], which has been changed to [Inifn] by other P- rules, and later acquires a final vowel by still another P-rule--cf. Section 4.73). Similarly, while all three dialee.tts have Ld34o4kuj 'htp' (froq undeaying /guegisku/, which has been phanged to 44eqkuj by other P-rules), the replacement of 01 by [0] in the context in question is more widespread in Fante than j.t is in the other twodialects,Thus, wher9 Famte has Ltctionn, Akuapem and Asante have Lt94801 and itctor] respectively for 'wait' (from underlying /kusg/, whidh is changed to(tctickj in all dialects before undergoing certain dialect-specific rules respoisible for the dialect differences in the final phonetic forms). The changes we have been discussing may be stated in a rule of the following form: 3.72, 3.73 102

p xxv (P76):

+Vocalic r [Vocalic-]___L-Focalic]= -Back -(+Back]/+Ro und -Low [ to specify thatthe voyelsaffected (It is not necessgry L+Backis since, by by P xxii becomeL+HoundJ as well as established in Section0.51, the Segment the convention vowels.show the Structure conditionthat requires.that [Backsi and [RogndJ(g$0 20, same valuefor the features vowels Section 1.24) willautomatically applyto L+Backs, Oertred.by means ofP xxii, specifyingthem as redundantly t+RoundJ.) general one, and This statement ofthe rule is a diglectalrestrictions onthe replace- does not,iffoserany context in question. ment of LIJ byLIJ] or LEJ by LOJ in the be handled bydiacritic features an Sudh restrictions may 0.60), marking the individual lexicalitems (cf. Section to, or exemptfrom, applicationof the items as subject dialects. rule in th(appropriate instancesin the several also replaced In. Pante, thenon-low front vowels are before amorpheme7ftnal Dy theirback-vowel counterparts subdialects in whichmorpheme-final(id is lw] (in those of,whether the pre- retainedcf. Sectiop3.31), regardless is L+Roundl. Compare FaLtcow] and ceding non-vowel /kep/), Pa Ak[t9ew]/As [tce] 'hat'(from underlying Etsuw?] and AkLtxw?]/As Lt0] 'pic.0(from under1ying to Pante, mgythere- /txp/).The followingrule, specific fore be proposed: P xxvi (P77):

[+Vocalic : Fa -Back -L+Back)/ (+Round]+ -Low

3.73 PanteBackness Harmon:41. instances of thereplace- In additionto the various front vowels bytheir back-vowelcounter- ment of now-low such a replace- parts discussedin Section3.72, Pante shows cases inwhidh an under:1:Tingnon-low front ment in certain is followed by vowel occurs in agrammatical morpheme,tuat L+Back]. This a rootwhose first(or only) vowel is 3.73 103

"backness harmony" (or "roundnessharmony°) is similar to the "tenseness harmony" described in Section3.71, except that the range of morphemes affected by thechange is considerably smaller.Specifically, a[+High] front vowel (i.e., (ID occurring in a subject-copoord prefix or a possesspre pronoun is changed to its,L+BackJ(and redundantly 1.+Round]) counterpart (i.e.,LUD in the context in question, and a [-Low] front vowel(i.e., [I] or [ED occurripg in iqn aspectual or an ingrtssiv,prefix is changed to its L+Backj counterpart rUJor 1.0.1). (It dhould be noted that the change ofE_ to [0] is not found in subject-concord prefixes and possesstye prqnouns.Thus Pante, like Akuapem and Aaante, has /Ala Ljeboj'we break' rather than [jobo], and it has hen. pmaLEIn pen] 'our table' rather than [E5npen]--cf. Ak [jexi poki]/As [jar puna].) Illustrations of the effects of backnessharmony in Pante are provided by the following phoneticcontrasts between, an the one hand, Fante, and on theother, Akuapem and Asante: Akuapem-Asante Underlying Form Pante Form Form

/bY+dx+bo/ 'I am breaking' [meruboj taa(r)xbo] /b7+be+bo/ 'I come to break' (mabobol [mbebo] /je+be+bo/ 'we shall break' [jebobo] [jebeba] /00bu/ 'bls chest' [ne bu] in! by]

In the case of the ingressive prefix/ko/ (cf. klecticw 4.50), Fante shows a process which is the reverseof the process just illustrated.That is, the non-low back vowel LO] is replaced by its [-Back] (and redundantly[-Round]) counterpart [4 when it precedes aroot whose first (or only) vowel is [-Back].Thus from underlying /bY+ko+je/ 'I go to do', Fante derives(mTkeje] (cf. Ak-As (mYkoje]). Through the use of alpha notation,the fronting of the vowel of /ko/ can be combined withthe backing of the vowels of /bY/, /dx/, /be/9 etc. into a single rule, asfollows:

P xxvii (P 78)

+Vocalic +Vocalic aBack '.. [-aBack]/ X -aBack I : Fa -Low -G +4 [ [ (where X does not include a[+Yocalic,-G] segment) 104 3.73, 3.74, 3.80

Given this formulation, segments whj.c4 are exemptfrom the operation of thp mile, such as the lej of thesubject- concord prefix Ljej 'we', must be marked with adiacritic feature indicating this fact (cf. Section0.60).

3,74 Vowel Deletion and Addition.

There are a number of contexts in whichP-rules delete or add vowels. Certain of these P-rules are dependent upon the occurrence of particularmorphemes or morpheme classes andfor show dialect differences intheir application. (Those of such rules that are specific to constituentsof the finite verb are discussed in Chapter4.) There is, however, one very general vowel-deletion rule thpt maybe mentioned here. This is a rule which deletes al+High] vowel in a CV syllable when this vowel j.simmediately followed by a syllable with an initialIli]. This rule maY be stated:

P xxviii (P 80):

r+Vocalil zsr

The rule is responstble for suc4chwagesas the following : L bxrx -0I. bri J ' redden'9 biri -0 bri J 'blacken', (bum] [bru) 'beat', (buru(w)?) - [bru(w)7] 'demolish', [bxre] -[bre] 'bring', [bira] - [bra] 'come (imperative)'.

3.80Systematic-Phonetic Segmeall. Application of the P-rules mentioned inpreceding sections of this chapter to the set ofAkan systematic phonemes (cf. Table 1, Section1.10) results in a set of systematic-phonetic segments which are approximate representations of the sounds that occur in Akanutterances. Table 7 shows the cover symbols that we usefor the systematic-phonetic non-vowel segments: 3.80, 3.90 105

Underlying Segment jhwpbfk g tds n Nasalized p h 13w. m 4 As Final Consonant w m 7 33 r n Labialized m hw pw bw fw kw gwtw dw ow Labialized and Nasalized iw mw IOW nw Palatalized 9 4 pj bj fj tq djts dz sj p p Palatalized and Nasalized p4 4 Palatalized and. Labialized 91 tc4 44 V V

TABLE7

SYSTEMATIC-PHONETIC NON-VOWELSEGMENTS

The inventory of cover symbolsfor systematic-phonetic vowel segments includps,in 40.tion to the symbolsused in Table 1, the symbols Lüand LUJ for the frontedround vowels that result from,tbe applicationof 13 xxi (oft Section3.61) and the symbols Laj and Lijfor the L+Coronalj lowtense vowels that result from theapplication of P xxiv (cf. , Section 3.71). (At some pointp in the textthe symbols 1.3,g, q] are used as equivalents ofLpow,p4] respectively.)

3.90Tone Rules.

At the systematic-phonemiclevel, all Akan vowels are marked for on, of twocontrasting tones, low(which we represent asL-Tcne] and transerpe with a graveaccent) and high (which werepresent asL+Tone) and transcribe with an acute accent).At the phonetic level,however, even within asingle phrase uttered by asingle speaker, these two contrasting tones maybe manifested by a number of different pitch levels.Thus the pitch levels of Ak 01)636 Kliase &n51:4 yi 'He will go to Kumasethis morning' are approximately asis indicated below:

[

bfi k6 KA m4&sep4n5 yi 106 3.90, 3.91

Furthermore, at the phonetic level, we mayfind three-way pitch contrasts, such as:

Ak mfib6 r."1 'I will strike'

Ak me 1)6 ["1-] 'my stone'

Ak me tiO(-..) 'my chest'

(The vertical accent in the second exampleis used to transcribe the phonetic"mid" or"drop" tonecf. Section 3.92.)

The immediately followingsections discuss the general P-rules that arerelevant to the phonetic mani- festations of tone in Akan. (There are, in addition, many P-rules that affect the tones ofparticular types of morphemes or morpheme sequences.Those P-rules that bear particularly upon the tones of verbs arediseussed in Chapter 4, especially in Sections4.90 ff.)

3.91Downdrift.

In many so-called. "level"tone languages, the pitches that manifest different occurrencesof a particular contrastive tone--say high or low--within aphrase are more or less constant. In many others, however, thisis not the case. In Akan, for example, acontrastively low tone at the beginning of a phrase maybe manifested by a pitch that is as high as, or higherthan, the pitch that manifests a contrastively high tonetoward the end of the phrase (cf. the graph4.0 representationin,Section 3.90 of the pitches of Ak3bAk6 lamas& eln5DA yi, in which the pitches of the first and lastsyllablesrare approximately th, same, although the first syllable isL-Tonel and the lastL+ToneJ). This variation in thephonetic pitch associated withdif- ferent occurrences of aparticular contrastive tonewithin a phrase isaccounted for by a rule of"downdrift", which specifies that the pitch intervalbetween a low tone and a followinghigh tone is always less thanthat between a high tone and a followinglow tone. Thus, given a sequence of tones LHL (where L =low, H = high), the downdriftrule specifies that the pitch ofthe second L is lower than that of the first, since thepitch rise from the first L to the H is smaller than the pitchdrop from the H to the second L. Similarly, given a sequence oftones HIE, the downdrift 3.91 107 rule specifies that the pitch of the second H is lower than that of the first, since the pitch drop from the first H to the L is greater than the pitch rise from the L to the second H. To illustrate these points graphically:

LHL = (- HLH =

The downdrift rule does not affect the pitches of sequences of Ls in immediate succession or of sequences of Hs in immediate succession. In such sequences the pitch level of the first L or H is more or less maintained throughout the sequence. For example:

LLLHHH = [- OM =I

HELM = N O All of the phenomena associated with downdrift may be captured in a three-part rule whose parts are conjunctively ordered. (That is, any applicable parts of the rule are applied in order, and after all applicable parts of the rule have applied once, the resultant string meets the conditions for some part or parts of the rule, these parts are once more applied in order, and so on, until the string no longer meets the conditions for application of any part of the rale.) The first part of the rule adds a number representing a relative pitch level ("Pitchn") to the first tone-bearing segment or "a-tone segment sequence" that follows a phrase boundary. (By an"a-tone segment sequence" we mean a sequence of identically-valued tone- bearing segments separated only by non-tone-bearing seg-, ments.) The second part of the rule applies to a L+Tonej (high tone) segment or segment sequence that follows a L-Tone] (low tone) segment to which a relative pitch has alregdy been assigned. This part of the rule assigns to the L+Tone] segment or sequence a number that represents a rela4ve pitch two steps higher thanthat of the pre- 9eding 1-Tone] segment. (If the pitch of phe preceding L-Tone) segment is "Pitch n", that of the l+Tone) segment or sequence ip "Pitchn+2".) The third part of the rule gpplies to a l-Tone) segment or sequence that follows a L+Tone) segment to which a relativp pitch has already been assigned, and assigns to theL-Tone] segment or sequence a number that represents arelgtive pitch three steps lower than that of the prpceding L+ToneJ segment. (If the pitch of the preceding l+Tonej segment is "Pitch II", that of the [-Tone] segment or sequence is "Pitch n-3".) 108 3.91

The downdrift rule may be stated asfollows:

P xxix (P 86):

(a) [ *Tone)* -(+Pitch nj/[+PB]X Y (+Pitch (b) [+Tone]* - n+21/[3(it:h nil 7. y (c) (-Tone]* (+Pitchn-x)/L+r+TonePitch n

(where X and Y do not include tone-bearingsegments other than those affegted by the rule,and where X does not include al+PB))

(The condition to the effect that "X doesnot include a a phraseboundary--is relevant to parts(b) andc) of the rule, and is intended toblock application of the rule across phraseboundaries within a sentence. This condition expresses the factthat downd;ift ig restricted to the phrase, in the sensethat a L-Tonej segment after a phrape bou4dary may have apitch higher than that of the last LrTonej,segment thatprecedes the boundary, and, similarly, a L+Tonej segmentafter a phrasp boundary may have a pitch higher thanthat of the lastL+Tone] segment that precedes the boundary. In fact, however, there is also a kind of downdriftwithin tive senVenceras a whole, such that the pitch of thefirst L-Tonej orL+Tone] segment of each successivephrase is somewhat lower thanthat of the first similarly-valuedsegment of the preceding phrase. But we have not, in this text,attempted to formulate rules for intonationbeyond phrase boundaries.)

We may illustraVe,thewo;kplp,of t4e downdrift rule with the sentence Ak3bek5 Kumase anopa yi 'He will go to Kumase this morning.' First, then, part (a) of therule applies,to assign a pitch to theinitial tone-bearing segment, 0-. Let us say that "Pitch n" maybe given 'phe value 'IN when assignedby part (a) of P xxix to aL-Tonej segment. Then: 51:40 KamAse An5p6. yi

-0MAO Knn&s Slap& yi

Now part (b) of Vhe ruleapplies to assign "Pitch n+2" (or '5") to theL+Tone] segment sequence -bfiko. Thus: 3.91, 3.92 109

31)616 Kamitse fin5p& yi 3 5 5 51,616 KilmAse 11125pA yi

Now part (c) of the rule applies to assign"Pitch n-3" (or "2", since "n"here = the"5" of-k6) to the next L-Tone] segment, resulting in:

5 5 2 obfil6 KamAs6 in411 yi

And now part (b) applies to the[elone] sequence that fol- 2 lows Ki-, resulting in:

3 5 5 2 4 4 4 51)60 Kadise &nap&yi

Finally, after another application of part (c) of the rule, and another of part (b), we have:

5 5 2 4 4 4 1 3 3 31A1c6 lam&se An5p& yi

(In a model which included a level of"physical phonetics" the numbers assigned by P xxix could be replacedby, say, numbers representing fundamental frequencies.)

Although, because of its importance, we have presented and discussed the downdrift rule first among the tone rules with which we shall deal, it should be understood that the rule itself, since it applies only at the phraselevel, follows many other tone rules. In particular, all rules that change tonal values (i.e., replace a 1+ToneJ by a 1-Tone], or vice versa) or add tone-bearing segments must precede the downdrift rule.

3.92The "Drop" or "Mid" Tone. As was noted in Section 3.90, atthe phonetic level, three-way pitch contrasts are found inAkan in certain tonal contexts. Specifically, after acontrastively high 110 3.92 tone, the pitch of the nexttone-bearing segment(s) may be on the same level as theimmediately preceding material may be slightlylower than the precedingmaterial, or may be considerably lawer thanthe preceding material.To repeat the examples citedin Section 3.90, the pitchof the second syllable of Akmeb6 'I will strike' is on the same level as that of the firstsTllable, while the pitchesof the second syllables of Ak meb6 'my stone' and me bO 'my chest' are, respectively,gaaly lower and coinderably lower than the pitches of thefirst syllables.We may call the tone of the secondsyllable of mfib6 "high",that of the second syllable ofme b6 "drop" (or "mid"), and that ofthe second syllable ofa g "low". Drop tones in Akan occur onlyafter high tones or other drop tones. In initial position, orafter a low tone, only a two-way contrastis possible: between a high tone and a low tone. Languages that show thiskind of contextual restriction on(phonetic) tonal contrasts have been called (by Welmers(1959)) "terraced levellanguages". Stewart (1965) has pointed outthat, in the great majority of cases, the occurrenceof a high-drop(or drop- drop) pitch sequence may be tracedto a high-low-high sequtnce occurringat a deeper level.For example, Ak me bo 'mystone' is composed, in thedeep structure, of the high-tone possessive pronounme 'my' and the noun5b6 'stone' which may be analyzed intothe low-tone nominal prefix o-s and the high-tone nounroot b6.The segmental features of the NomP o-rarealways (leafed after a possessive pronoun. But the L-Tonej specificationof the prefix has its effect upon thesurface tone pattern. Let us say that there is a rule whiehdeletes all of the featuresof o- in the contextin question except thefeatures r+Segment,-Tone].This rule would have thefollowing effect:

If at this point thedowndrift rule (P xxix,Sectip 311) * applies, its result willbe something like: me + 0 + bo. (We shall assume that "Pitchn" may be given.pi value "5" when assigned by part(a) of P xxix to a+Tone] segment.)Now let us further proposethat there is a late rule that deletes allsegments specified only as 3.92, 3.93 111

[4.segment,-Tone] (or [+Segment,+Tone]). Atter this rule has applied, we are left with: ma + bo, which we may transcribe as me b6[rat bk]. The postulation of an ordered set of rules,then--(l) segmental feature deletion,(2) downdrift, (3) zero-segmentdeletion--permits us to account for the occurrence of a droptone on the phonetic level as arising from an underlyingstructure in which only high and low tones are specified.

3.93Tone Incorporation e, ToneSimplification. Actually, both from the pax,' ofview of phonetic plausibility and from the point ofview of the actual workings of the language, the accountof the origin of drop tones provided in Section3.92 is somewhat oversimplified. When segmental features are deletedbut a tonal specification retained, it is plausible tosuggest that the retained tone is added either at the endof the preceding syllable or at the beginningof the following syllable, andthat, if the syllable to whichthe tone is added has adifferent tonal specification from that ofthe added tone, the result will be a gliding tone. Thus, if we postulate that the rule that deletes the segmentalfeatures of o- after possessive pronoun adds theL-Tone] specification of 0- aV the end 9f the pronoun, wehave the following derivation: me + p- + bo -0 me +bo (where A represents a tone falling from high to low). While we in fact find that in somesubdialects, in some constructions,the deletion of segmental features from tone-bearingunits results in surface tone patterns in which gliding tones occur,usually gllding tones are not found in Akan.For example, me + bo is in most subdialects, obligatorilyrealized asra To account for this phoneticrealization, we may propose a rule of tonal simplification,which specifies that when a sequence of twotones is found in a singlesyllable, if either tone is high (or if bothare) the sequence is simplified to a single high tone. Thus, in a single syllable, HIVIMAM H. (The tonal simplification rule also specifies that a sequence of twolow tones on a single syllable is simplified to a singlelow tone: i.e., LL 1,--see below.) The tonal simplificatlon rule mustfollow Thus: me +5- + bo (by the downdrift rule; P xxix. 372 7 Nom P-deletion)me + -0 (bydowndrift) me + b6 - (by 5.33

4 tonal simpli/ication) lue; +136 (or ma INS)"

In the case with which we havebeen dealing, it does not,paTticularly watter whethef we saythat the low tone, o/ o- is added at tbe endof re nr at the beginning ofbo, since, given the rule oft(nal simplification andAthe fact t4at.this rule follow the dliwndriftrule, both me bo and me 114 (where ' represents a tone rising fromlow to high) resUlt in me 1)6. There are, however, Iasez in whichit is net:essm to specify whether atone is added at the end of one syllable qr at thebegianing of the next.For example, tbe tone of a hz,ro" a-abject-concordprefix (SC) (cf. discussion in Section 4.20) is ciddedat the end of the preced- iag syllable.Thus, given Ak-Fane + + 0 +bit 'His mother comes', this rule operates toderive ne + bit. The dowzdrift rule then applies tothis stringtcraerive, 5 :44 1 say, ne + 4 ba, and .tonalsimplification results in 5 4 1 ne +nit + bit (or LA A& bit), which is the occurring eurface pitch pattern. Now 1: the high tone of the SC hadbeen added at the beginning ts;* 1:4, ratherthan at the e#d of n nA, the derivation would hare been: Ak ne + ni + 0 + ba - 2 4-1 5 2 -4 ne + ni + ba - ne + n + b -0 ne +na + bat and the wrong surface pitch pattern would have beenpredicted.

On the other hand, there is anAkuapem-Asante rule (P 57, cf. discussion in Section 4.91,subsection (e)) which adds a high-tone unit at thebeginning of the perfect- aspect prefix /1/ when the immediatelypreceding tone is high. Thus, kt.'en Ak-As + a- kits& 'Kofi has spoken', this rule operates to deriveKOfi+ +kasa. To this the

5 5°.; downdrift rule appliest the resultbeing, say, Kofi + a- + 4 4 53 4 4 plat.Tonal simplification now results inKofi + a + kits& (or KOH akitat), which is the occurringsurface pitch pattern. -Ifis clear t4atsif a high-tone unit had been added at the etid ofKOfi, or at the end, rather thait. the besinning, of a-, the wrong surfactpitch pattern would have been predicted. (The (vacuous) addition of a high- tone unit at the end ofKofi would halo resulted in the 3.93 113

surface patternKOfi Akistl, while the addition ofa high- tone unit at the,end of 1- would have resulted in the sur- fact) pattern KOfi ikAsti.)

In most cases, a tone-bearing unit that lacks seg- ;Antal featuresz-i.e., p unit thst is specified only as L+Segmentt+Tonej or asL+Segment,-Tonel--is incorporated into the lerecedingv rather than thefoilowingsyllable. (This is tame whether the [4-Segment,uTone] unit is present in tae deep structure as a zero tone-bearing unit, or resOts from the epplication of a P-rule.) We shall take advantage of this gact in formulating all P-rules that result the occurrence of L+SegmentIaTone] units. These rules will not iu themmelves incorporate the units into either the preceding or the following syllable, but will instead have as outputs strings in which such units occur between syllable boundaries,. In those relaVively rare cases where the tone of the L+Segment,cTonej unitis ultimately to be incorporated into the following syllable, this fact will be indicated by a rightward-pointing arrow after the unit. Thus the Aimapem-Asante rule that adds a high tone before a perfect-aspect (PER) prefix that is preceded by a high tone will have some such form as the following:

/L+Tone:1(c)+ PER

This arrow All not appear in cases where the tone of the L+SegmentlaTone] unit is ultimately to be incorporated into the preceding syllable.Thus the rule that deletes VA' segmental features from the SG prefix 5- in possessive constructions will have some such form as the following:

raFj 0/[47ip3I/Possessive Noun Root

(In this formulation, [ur] refer4 to all specifications of segmental features, and l+NomP 3J 7afers to the nominal prefix 5- --cf. Section 2.52))

Iftek all rules that result in[4-SegmentoTone] units :Lave applied, ek tous-incorporationrule will move the unit into the followli_ syllable if it is marked with a rightward- pointing arrow and !Alto the preceding syllable if it is not. 114 3.93

The tone-incorporation rule maybe stated in two conjunctively-ordered parts as follows:

P xxx (P85):

(a) aTone =(C)-0 = (C) aTone [OP OP

+Segment (b) (C) =aTone -0 aTone (C) = [OP OP (where [OF] indicates that no segmental features are present)

(Part (a) of P xxx places the[+SegmentoTone] unit immediately before the vowel ofthe following syllable; part (b) places it immediatelyafter the vowel of the preceding syllable.) The tone-incorporation rule maybe assumed to precede the downdrift rule. In any evert, it precedesthe tone-simplificrition rule, which may bestated as follows:

P xxxi (P87): 1 +Tone [ None ] [gone1 whereaorY= +) f, 1( ... raTone][yTone] [-Tone](where a and y = ) 1OF OF 1LOF (where (021 indicates that segmentalfeatures are present and(Ø'I indicates that no segmental features arepresent)

P xxxi operates upon any sequenceof two adjoining tone- bearing units one of whidhis specified for segmentalfeatures and the other of whichis not. (Such a seqaence can result only from the applicationof the tone-incorporation rule, P xxx.) Its effect is to reducethe sequence of two unitb to a single unit.This single unit has thesegmental- feature specifications of theundeaying uuit that had sudh features specified, and has hightone (L+Tene)) except in cases where both of theunderlying units have iow tone (1-Tone]). 3.93, 3.94 115

We may now illustratethe operation of the tone- incorporation andtone-simplification rules in relation to the downdrift rule. Given an input string such as Ak-Fa ne + nit + 0 +IA., tone incorporation results inne + 5 21. 1 mg+ b downdrift in ne + +bit, and tone simplifi- 5 4 1 cation ia ne +nit + IA (or Ne n& IA 'His mothercomes°. Given an input string such asAk-As Kari + 014 + +kese. r+Segment}.9 (where44 ) tone incorporation resultsin 14.Tone 5 2 4 4 16.14, and tone KOfi+ IfOit +kesi, downdrift in Kofi + Oi + 5 47.-- simplification in Kofi +A + kis& (orlad &kits& 'Kofi has spoken°.

3.94Other Tone Rules. At the systematic-phonemiclevel, only vowels are specified for tone. At the systematic-phoneticlevel, however, all voiced segmentsmust be so specified. This specification is accomplishedby means of a tone-spreading rule, which copies the toneof a vowel onto adjoining voiced non-vowels in the samesyllable. This tone-spreading rule may be stated asfollows:

P xxxii (P 01): Laoiced] [aToneMaTone]

(This rule reapplies whenever theconditions for its application are met. The rule incorporates thenotational device mentioned inSection 3.26, in which theenvironment is to be interpreted aseither preceding or followingthe segment alf9cted bythe ru194) By means pf P xxxii, for example,Lam] 'meat' -0 [fiamJ, and Ak-Fa1.60] 'cultivate' -0 [d50]. (In phonetic transcriptions inthis text, the tones of voiced non-vowels arenot normally indicated unless the non-vowels aresyllabic.) There are various othergeneral tonal P-rulesthat would have to be includedin a more complete studyof Akan phonology than the present one.For example, Stewart(1962) notes that, in somesubdialects of Pante, a pre-pausehigh tone is in most casesrealized with high-rising,rather than high-level, pitch. In these same Fantesubdialects, 116 3.94 a pre-pause low tone that follows a high tone israised phonetically to the level of the preceding high tone. It is also the case that in Asante there is usually a pre- pause lowering of pitch below the levels predicted by the downdrift rule. These pre-pause intonational phenomena are not captured by our present set of rules.Nor, for example, have we attempted to go iu any detail into the intonation of yes-no questions, which appears to be marked by a slight raising of the general pitch levels of the sentence, together with a sentence-final interrogative morpheme that may be manifested as an extra-lnw tone. 117

Chapter 4

The Finite Verb

4,10Surface Structure ofthe Finite Verb. This chapter isconcerned with thesegmental patterns and tonepatterns found incertain parts of Akan that are daminatedby the node VP surface structures primarily to the (Verb Phrase). Attention is directed parts of such surfacestructures that aredaminated by other V (Verb).Akan VPs may includeparts dominated by nodes sudh as NP(Noun Phrase).The present chapter covers onlythose features of NPstructures--primarily associated uniquelywith the tonal features--that are For ex- occurrence ofthese structures asparts of VPs. ample attention isgiven here to tonalphenomena associated with certain NPsoccurring as verbalobjects (cf. Section of 4,95), On the other hand,the segmental structure nouns, and thetonal phenomena that areassociated with subparts of NPs in general,rather than specifically as VPs, are not discussed. The surface-structureverb phrases that are considered here are thosethat include afinite verb: i.e., is marked for aspect. (In addition to finite a verb that non-finite verb verbs, Akan surfacestructure has certain forms, comparable toinfinitives or gerundsin European These forms havetypically nominalsentence languages. subjects or objects.) functions, occurring,for example, as obligatorily in- The surface structureof a finite verb cludes nodes labeledASP (Aspect) andVR (Verb Root), Etcept in the case ofimperatives (cf. Section4035), it SC (Subject also obligatorilyincludes a node labeled surface structure of Concord), Optional nodes in the finite verbs are:NEG (Negative), ING(Ingressive), RED (Reduplication), and SUB(Subordinative). The usual order is shown in the of these nodesrelative to one another following tree diagram:

SUB SC ASP NEG ING RED VR 118 4,109 4.20

In the sections that immediately follow (Sections 4,20 through 4,81)9 the components of the finite verb are presented in the order in which they appear in the diagram above. Each presentation includes an account of the systematic-phonemic form(s) of the component under discussion. In addition, some of the presentations include an account of certain P-rules that apply to the component itself or to forms that include it. In most cases, however, these presentations do not include an account of tone-changing P-rules, discussion of the latter being deferred until Sections 4.90 through 4.96,

4,20The Subject-Concord Prefix,

The SC component of the finite verb is a subject- concord prefix. This prefix is not present in the deep structure of the verb, but is, instead, introduced by an obligatory transformation which inserts an SC as the leftmost constituent of any finite verb preceded by a subject, (The transformation fails to operate when the aspect of the following VP is imperative, since in such cases the subject has been deleted by a previous trans- formationcf. Section 4350 For example, given an input structure such as:

NP VP 4 1,rNP //// ASP VR

ono STA wo Kumase 'he' 'be -in"Kumase'

the SC-insertion transformation operates to yield the derived structure: 4.20 119

V NP

ASP VR

ono SC STA wo Kumase

If, as in the above example, thesubject NP is a personal pronoun, a further obligatorytransformation applies. This transformation has the effect of trans- ferring to the SC prefix certain syntacticfeatures of the pronominal subject, and then ofdeleting the pronoun itself from its original position.The features of the pronoun that are transferredto the SC prefix are person, number, and, in the case of third-person pronounsin Akuapem and Asante, animateness.For example, operating upon the derived structureabove, this transformation results in the structure:

7 VP

V NP

ASP V

I I SC STA wo Kumase [+3rd Person +Singular 4-Animate)

(The surface representation of the above structureis Owo Kumase 'He's in Kumase,' in whichthe initial 3- 77], represents the SC prefix.)

The systematic-phonemic forms of theSC prefixes that reflect the operation of thefeature-transfer transformation are shown in Table 8. 120 4,20

AkuaDem Asante Fante

Singular: 1st person /bi/ 4 2nd person MY/ 3rd person /5/ /5/ /5/ (animate)

Plural: 1st person

2nd person /bil/ /be/

3rd person /w5/ (animate)

Inanimate

TABLE 8

Systematic-Phonemic Forms of Subject-Concord Prefixes

(The following notes pertain toTable 8.

(1) The two forms shown forthe second-personsingu- lar in Fante aresubdialectal variants, as are the two forms shown forthe third-person plural in Asante,

(2) In some Fante subdialects asecond-person-plural subject may,be expressed insurface forms by the noun hom/hub/ 'you (pl,)', rather than, orin alteiEgEion with, the SC prefix/w5/ Rom is best regarded as a nounbecause of theMe patterns of the structuresin which it occurswhich show that, like other nounsin subject position I.;; is followed by anSC prefix--see below, 4.20 121

(3) As the chart indicates, Fantelacks a distinctive inanimate SC prefix, and uses the samethird- person forms foranimates and inanimates alike. In Akuapem and Asante, no numberdistinction is made in the case of the inanimateSC prefix.)

The forms shown in Table 8 ms,yundergo certain changes through the operation of P-rules. Some of these rules affect the segments of the SC prefixes,others their tones. P-rules affecting the segments of theSC prefixes include: the regressive nen-vowel,nasalizationrule, P 06, which, e.g., changes/bY/ to [TT]; the,tenseness-harmouy rule, P69, which, e.g., changes[m!] to [mI) before VRs with[+Tense] vowels; and the ?ante backness-harmonyrule, P 78, which, p.g., changesCm!) to [me] and bat] to (me] before VRs with L+Backj vowels. One rule affecting the tones of SC prefixes is P 52, which replaces the low toneof second-person Ses other than Fante /wo/ by high. (Low tone is postulated for the systematic-phonemic forms of thesecond-person SOB because certain P-rules thatnormally apply to VRs that follow a low tonee.g., theAkuapem-Asante rule, P 201 that changes the tone of a VR like/kiwi)/ 'speak' to Ak (kitsa)/As[kitsa] after a low tone--do apply when the VRfollows ;second- person SC.) If the subject noun phrase issomething other than a non-emphasized pronoun,the feature-transfer transformation discussed above does not apply. In such cases, the systematic7phonemic form of the SC prefixis /0/: i.e., a low-tone ([-Tone]) segment unspecifiedfor any other features.This zero SC prefix undergoes the sametone-changing P-rules as other SC prefixes(cf., for example, Section 4.91), which may result in a reppicement of theoriginal low tone on the prefix by high(L+Tone]). After these tone changing P-rules have applied, a low-tone zeroSC prefix is deleted in Akuapem and Asante,by means of P 64, but retained in Fante. (A high-tone zero SC prefix isretained in all dialects.) Thent the tone incorporation rule, P85, adds the tone of a retained zero SC prefixat the end of the immediately precedingsyllable (i.e., the last syllable of the subject NP). For example, P 85 changes:

Ak-Fa(ni ni Ina ] 'his"mother"comes'

to: Ak-Fa[nf ni 4.20, 4.30 122

(which, after the applicationof further P-rules,is real- Since a low- ized as Ak-Pa(ni ni bi] 'His mother comes9. tone zero SC prefixis deleted inAkuapem and Asante by available for toneincorporation in these P 61t, it is not available dialects. A low-tone zeroSG prefix is, however, for tone incorporationin Pante. Thus, given the following input to P 64:

[kbfi ibibi ] 'Kofi"will-come' this rule deletes theSC prefix in Akuapemand Asante, resulting in the surfacetone patterns

Ak-As[kbfi bibi] 'Kofi will come.'

Because P 64 does notapply in Pante, however,the SC prefix ofObibi] is available for applicationof the tone incorporation rule inthis dialect, resultingin: FaDaft bibi] which, after theapplication of furtherP -rules, is realized as: Pa[lad bbbi] 'Kofi will come.'

4.30ASPect. In the base componentof Akan grammar, theobliga- tory ASP constituentof a finite verbis rewritten in one of seven ways: (1) STAtive; (2) HABitual;(3) PROgressive; (4) PASt; (5) PERfect;(6) PUTure; (7) OPTative.These seven may becalled the basic aspectsof the Akan verb. In addition to these,there are two derivedaspects which result from the operationof transformationalrules upon one of thebasic aspects; these are: (8) Werative; (9) CONsecutive. STA (which is equivalentto what some otherauthors have called the'continuative') stands apartfrom the other eight aspects withregard to the verbroots with which it co-occurs. The other eight aspects maycollec- tively be called ACTive.With very few erleptions, any 4.30, 4.31 123 verb root that co-occurs with any one of them also co- occurs with each of theothers. STA, on the other hand, fails to co-occur with many of the verb roots that co- occur with ACT, and there are anumber of verb roots that %,o-occur with STA but not with ACT. Thus, for example, the verb root ba 'come' occurs only with the eight active aspects while THe verb root wo 'have, be in' occurs only with STA. On the other hand, Many verb roots occur with both STA and ACT: e.g., the root hn, which has the meaning 'be wearing' with STA, the meanidg 'put on! with ACT, In the lexicon, each verb root is marked as(+STA ) or L-STA ]: i.e., as co-occurring or failing to co-occur with STA:- Verb roots that are L-STA ] are automatically marked, by means of a redundancy rule, as[+ACT ]: i.e., as co-occurring with the eight activeaspects. -Thg root ba is, for example, [-STA J and therefore L+ACT J. NSTA ] verb roots are subdivided into two classes, [+ACT ] and [-ACT ], on the basis of whether or not they co-occur with-ICT in addition to STA.Thus wo is (+STA ] and [-ACT ] while hyc is [4aTi ] and (+ACT ).

The nine aspects of the finite verb will now be discussed, in the order in which they are listed above.

4,31STAtive and HABitual.

STA and HAB are alike in that they both have zero segmental forms.They must be distinguished from one another, however, because of their differing effects upon the tones of other elements of *Lie finite verb, as is illustrated by examples such as: STA HAB

Ak, Fa hy 6.11k 'he is 'he puts on' wearing'

Fa 'he isn't 'he doesn't put wearing' on'

As wOlihyL wonhyc 'you aren't 'you don't put wearing' on' 124 4.31, 4.32

STA HAB

As &el h6 mfi hS 'he isn't 'he doesn't from there' leave there'

In those cases in which formsthat include STA and those that include HAB differ tonally,the differences are accounted for on the basis ofdifferent P-rules that apply in the presence of the one orthe other aspectual morpheme (cf. Sections 4,91 and 4.92).Thus, although both STA and HAB are listed in thelexicon as segmentally zero, the difference between themawl; be preserved ininputs to the P-rules. This is easil, done,since they may be (0] and ;v73[0] (i.e., zero dominated by identified asSTA STA and zero dominated byHAB) respectively. (HAB is specified as having low tone inorder to account for the operation of certain P-rules. There does not seem to be a need tospecify a tone for STA.) 4,32 pROgressive.

The systematic-phonemicform of the PRO aspect is /d1/. The surface realizations ofPRO, however, have initial LrJ (except in those casesin which the CV form of the morpheme is replaced byanother form by means 9f P- rules--see below). As was noted in Section3,40, 1r] does not normally occurinitially in native Akan morphvmes, but is, instead, incomplementary distribution withLdJ, which occurs only morpheme4nitiall,. (That is, at the systematic-phonemic levelId] and Lr] are not distinct.) The fact that[r] occurs in the surface realization of the PRO morpheme hence suggeststhat this morpheme is, in some sense, functioning asif it were part of alamer unit that is responspre tothe P-rule. P 15,governing the distribution of[r] and [d]. In this conneaion, it may be notedthat, in inputs to thephonological component: PRO does, in fact, alwaysfollow a subject-concor4 prefix (cf. Section 4,20). Now if one pastulates arule elimin- ating morpheme boundariesbetween eubject-concordprefixes and aspectual morphemes,P 14, the /d/ of PRO, afterthe operation of this rule, occursin medial position. In this position, it is automaticallysusceptible to P 15. 4.32 125

Although there is good evidence that PRO has an underlying CV form, the surface forms that reflect the occurrence of PRO often have some other shape: a vowel or simply a tone pattern:. In fact, surface forms of PRO that derive directly from /di/ do not occur at all in colloquial Asante, and in both Akuapem and Pante occur principally as stylistic variants of other, perhaps more c/mmon, forms.

In Asante, verbs that include PRO are obligatorily subject to P 56. This rule operates in three steps: (a) first, it deletes all of the segmental features of the PRO morpheme, leaving only the tone intact; (b) secopd, if the morphepe immediately preceding PRO ends Dare. L+Low] vowel (i.e., LAD, j.t replaces this vowel by its L-Low] counterpart (i.e., LED; (c) third, if the morpheme imr mediately preceding PRO ends in a vowel (including an LE] that results from step (b) of the rule), it copies the segmental features of this vowel in the place originally occupied by the segmental features of the vowel of PRO. Some examples of the operation of the rule in Asante are:

Before P 56

Dd.& ribit] [Wm Obit] (Odum is- coming)

[Ati riba] Uttit Oba] littfi Obit] ibi]Cati (Ata is- coming) [an ribi] [kOfi Obi] [lad ibi] (Kofi iE- coming)

[Brig] [Bloat] [55bit] (he-is- coming)

(Ddfim ?Obi.] is ultimately realized as (OdUM IA) through application of subsequent P-rules.) (In spite of the obligatory operation of P 56 in colloquial Asante, the PRO morpheme is represented by re in conventional Asante orthography: Odum reba 'Odum is coming,' Ata reba 'Ata is cofing,' etc.) -Ur Pritris followed by the morpheme /N/--cf. Section 4.40there are several tone-changing P-rules that precede P 56, and only step (a) of P 56 applies.For examples, cf. Section 4.92, subsection (b).) 4,32, 4,33 126

In Akuapem and Pante,P 56 is optional,If P 56 is not applied, thevowel of the PROprefix may undergo certain changes inaccordance with thetenseness-harmony rule (P 69) andior the Pantebackness harmony rule(P 78), These rules account forforms such lig:, Ak.4aoreba [MIA] 'he is coming', Ak oredi[?ridi]/Fa oredzi [ridzi] 'he is eating', Fa oroko(5r6k5?] 'he is going', and Fa orodumI.Orfidle] 'he is extinguishing'. (PRO is normally represented by rein Akuapem orthographyand by either re or ro inPante orthography.Tenseness harmony is not reflected inthe spelling eitherof PRO or of a preceding SC.)

4,33 PASt and PERfect.

The PAS aspect is theonly aspect that isrepresented, in surface forms, by asuffix attached to theverb root, the other aspectsbeing representedeither by prefixes or by zero segmentalforms. Since all aspectualmorphemes precede the verb root indeep structure, theremust be a transformation that movesPAS to post-verb-rootposition. This transformation isroughly: I - PAS - (NEG)(ING) (RED) VR - Y.

SD: 1 - 2- 3 -4

SC: 1 - 0 - 3 + 2 -4

(This transformation followsthe transformationsthat replace PERfect by PASand PAS by FIR innegative verb forms--see below andSection 4.410

The systematic-phonemicform of the PAS suffixis /1/while that of the PERprefix is /a/, (In Asante but not in theother two dia1ects4the past suffix has an alternatedisyllabic, form,/NV.) In the presence of the NEGativemorpheme, transformationalrules apply which replace deep-structurePAS by surface-structurePER, and vice versa.That is, verb formsthat have a negative past meaninginvolve the prefix /a/,while those that have a negativeperfect meaning involvethe suffix/1/, 127 4.33

Thus the negativeof: Kofi 'Kofi removed it.' (Kofi removed) is: Ak-As Soft Aut. Fa Kofi pal. Kofi didn't removeit.' (Kofi didn't- remove) while the negativeof: Ak-As Kofi Fa Kofi 2,71. 9Eofi has removed it.' (Kofi has- removed)

Kofi Kofi hasn't removedit.' (Kofi hasn't- removed) negative pre- (The n in the negativeforms ropresents the 4,40, The e Le) in thePante negative fiA--cf. Section underlying past andaffirmative perfectis a reflex of the operation ofP 69 and P 84--see /a/9 and results from this switch- below. The transformationalrules that perform PER in negativeverbs are presentedin Sec- ing of PAS and devoted tion 4,416 The balance of thepresent section is of certain P-rulesthat affect thesegmental to a discussion This forms of Vs thatinclude the PAS orPER morphemes. discussion is arranged asfollows: (a) rules affecting "transitive" formationswith PAS; (b) rulesaffecting "intransitive" formationswith PAS; (c) rulesaffecting forms with PER. between (a) It is customaryto distinguish "intransitive" formationsthat include "transitive" and formations occur at the PAS suffix, The "intransitive" 128 4.33 the end of a sentence orclause; the "transitive"formations occur in mostother contexts: e.g.,immediately before a nominal serving as adirect or indirectobjects affecting the segmentalform of for- . The fir4!: rule mations that iucludePAS, P 34, appliesto both transi- length- tive and intransitiveformations. P 54 operates to point in the en anypre-PAS VR that endsin a vowel at the P-rules at which itapplies. The lengtheningconsists in the addition of alow-tone copy of theVR-final vowel to the VR.Thus ba[IA] 'come' -0/4baj+/1/, hwe[c46] 'look at'+/1/741)+/1/, bu [IA] 'break'Vi/41)411]+/1/, VRs bo [IA]4-/i/-41b6'5]+/V, etc. (P 34 does not affect that end in anon-vowel at the pointin the P-rules at acquire a VR- which it applies. Such VRs, however,nmy final vowel beforethe PAS suffix by meansof another rule, P 31--cf. Section4.73.) In the case oftransitive formationsthe PAS suffix by means itself is obligatorilydeleted in all dialects, remains of P 36. Since P 36 follows P34, the form that when the suffix isdeleted shows, inappropriate cases, the effects of P34: that is, it shows alengthening of a VR-finalvowel: e.g., Obaa Kumage. came toKumase.' Ohwec Kofi. 'He looked atKoff.' formations, see Sec- (For the tone patternsof the cited tion 4.91.) In the case of rootsthat have a VR-final non-vowel at the pointat which P 36applies, and which deletion of the have thus not beensubject to P 34, the transitive formations,by means of P 36 PAS suffix in identical to a leaves a formationthat is segmentally HABitual-aspect formation. In Asante, however,,;the seg- between PAS and HABformations is re- mental distinction step of P 36, stored through theoperation of a second which adds a low-tone copyof a VR-finalnon-vowel to a formation from whichthe PAS suffixhas been deletedby the first step ofthe rule.This seco4 stepof means of Lm] of the P 36 is, forexample, responsiblefor the final verb in: AsDfUli slid] 'He borrowedmoney.' 4 33 129

(cf. As (5fii sikid 'He borrowsmoneys') (The second step of P 36 does not apply inAkuapem and Fante, and thus habitual and transitive-pastformations in these dialects show only tonal distinctions: e.g., Ak-Fa [5fiisiki) 'He borrowed money.'

(cf, Ak-Fa [6fiisikfi] 'He borrows money.9)

Pante has one further rule thataffects the segmental forms of certaintransitive-past formations.This is P 35, which deletes thevowel that has been added by P 34 (see above) in transitive formationsinvolving disyllabic VRs with underlying voicedmedial consonants (cf. Section 4,75),This rule ip responsiblg for thesesmental dif-,, ference between FaL5tgIrc kOfil and Ak-As L5t9irii kofiJ 'He showed Kofi.' (The VR in these examples is kyere [tgiri] 'show', whose underlying form is/kid6/.)

(b) In intransitive formationsthat include PASI,the suffix vowel/i/ (but not the second vowel of As /i=j0) is made to agree in tenseness and nasalitywith the preceding vowel. The underlyipg[-Tenses-Nasal vowel of the suffix isrchanged toL+Tenpe] aftgr a 4-Tense] vowel by P 69, and toL+Nasal] after a L+Nasall vowel by P 73. Note the suffix vowels in thefollowing Asante intransitive-past forms: Dbititi(j1)] 'he came', [Obial(ji)] 'he broke',[mi(j1)] 'he gave', Dhatquin 'he saw'. P 37, There is a two-part vowel-assimilationrule9 that affects certainintransitive-past formations. The first part applies in all dialects. Tilts part of the rule, which changes certain occurrencesof LUJ toLI] before the PAS suffix, is responsiblefor the quality of theunderlined vowel in, e.g., Ak[5h6rwii]/As [512Urii(ji)] 'he washed" (of, the transitive-past formationsin Ak Dh6r0k(irfiwit as)/ As L5hilr?; kfirfiwAn6] 'He washed the cup.'). (For further discussion, cf. Section 4.74.) The second part of P37, which ip spectfic to Fante, affects any sequence of twoidentical L+BackJ vowels that precede the PAS suffix, (The vowels affected are a VR- final vowel and a low-tone copyof it that Oas been added by P 34.) The rule operates to changethe L+BackJ vowels to their [-Back] counterpartsbefore the suffix.Thus the 130 4.33

sequenceNu) is changed to [ii], [op] to [cc], etc. Examples are: Before P After P

Fa[Odwai] 'he arrived' -.0 [Odwiii] Fa [5bw65i] 'he beat (it)' [5bw) (The VRs in the above examples have the underlying forms /d4t/ and /b5/ respectively.The labialization of the root-initial consonants in the examples resultsfrom the operation of P 09, which precedes the ruleunder discussion. In phonetic transcriptions elsewhere inthis text, the effects of P 09 are sometimes notshown.) In both Akuapem and Fante, sequencesof three vowels at the end of intransitive-pastformations are normally subject to contraction by means ofP 38. In Akuapem, the contraction consists of thedeletion of the vowel that has been added by P 34 beforethe PAS suffix: e.g., Before P After P

Ak [hal) 'he came' [ ] Ak [Odfilli] 'he arrived' [Odfil] (The Akuapem surface forms in the right-hand column may e compared 4tht#e corresponding Asante surface forms: Obj0), lbdial(jc)).P 38 does not apply inAsante.) In Fante, the contraction(which may not occur in all subdialects) consists of the deletion of the PASsuffix: e.g., Before P After P

Fa[5bgai) 'he came' thiti] Fa [Odwiii] 'he arrived' [Odwil] Note that in Fante, the PAS suffixis generally deleted in both transitive andintransitive FAS formations. (In Akuapem and Asante, on the other hand,it is 4.33 131

deleted only in transitiveformations.) In some cases this results in homophony betweeLthe transitive and intransitive forms: e.g.,

Fa 15biakilmisf] 'He came to Kumase.' Fa[5bta] 'He came.' In other cases, however, the applicationof the vowel-assimilation rule, P 37, to theintransitive forms and its non-application to thetransitive forms results in contrasts such as:

Fa [Odfill kamisf] 'He arrived at Kumase.' Fa (bdwil] 'He arrived.' (The ordering of the P-rules is crucial inaecounting for such cases. In the transitive formations thePAS suffix has already been deleted at the pointin the rules at which vowel assimilation applies,and hence P 38 is inapplicable. In the intransitive formations, thePAS suffix is not deleted until after thevowel assimilation rule has applied.)

(c) There are two P-rules that affectthl segmental forms of verbs that includethe PER prefix A/. One of these, P 429 changes the segmentalform of SC prefixes that precede PER. (P 42 also affects formations that include the CONsecutive-aspectprefix /A/--cf. Sec- tion 4.36.) In general, the rule operates toreplace the vowel of SC by[a] (or LC, when the vowel of SC is [+Nasal]), before the [a] of PER, changing[mr] 'I' to [a], Ak-As [wo] 'you (sg)' to [wa],[js] 'we' to Lja], etc. tiowever, if the SC is Fa[s] [x] 'you (sg)1 or Ak-As Is] 'it, they (inanimate)1, P42 deletes it before PER, and if the SC islo] 'he, she, Fa it', P' 42 replace.: it by[wa] before PER. (In conventional orthography, the changes effected by P42 arp,accurately reflected in some cases, e.g., Ak abaLabaj partly reflected in other cases, e.g.,Ak waba Lwaabaj the has come', and ip W11 other casesnot reflected at all, e.g., Ak ysaba Ljaabaj 'we havecome'.)

The second P-rule that affectsthe segmental forms of verbs that include PER is thetenseness-harmony rule, P 69, which changes the PER prefix from[a] to [4]if the first vowel of the VR is[+Tme]. Then a later rule, P 849 changes [I] to Ak-As[4/Fa Lej in this context.The 132 4.33, 4.34

tensing rule affects the[A] of any SC prefix that precedes PER, as well as PERitself. Conventional Akuapem and Asante orthography donot represent thechange from [it] to bi], but convenUonal Pante orthographydoes represent the change from[i] to [eh compare, e.g., Ak-As mabisaLm&ibis6] and Fa mebisa[meebis&].)

4.34 FliTure.

The systematic-phonemicform of the FUTure prefix to is /bi/. It is probable,thatthis morpheme is related the INGressive prefix/be/ 'come (in order) to' andthus ultimately to the verb/bfi/ 'come' (cf. Section 4.50). In this connection,it is suggestive that thefuture and the ingressive prefixesfail to co-occur: i.e., that there are no futureingressive forms. However, future forms and ingressive forms are,at least in Akuapemand Asante, tonally distinct inmost cases, and this-- together with the ambiguityof those forms that arenod tonally distinct--seemssufficient grounds for distin- guishing the FUT and theING prefixes. (In Pante, the future is identical, bothsegmentally and tonally,with the habitual ingressive. Thus: l'He will do it.' Fa 513134 'He comes to do it.'

cf. Ak-As alzi 'He will do it.' Ak-As Seyg 'He comes to do it.9 In the presence of NEG, atransformational rule applies which replacesdeep-structure FUT by surfacestructure PRO (i.e., theprogressive-aspect prefix). The result of thistransformation is to make thenegative future and the negativeprogressive homophonous. Thus:

[STIP67] i'He won'tdo alau Ak-Fa[6(14)467] 'He isn'tdoing it.'

As an alternative tothis rule, Fantc has arule replacing the FUT prefixibt/ by the prefix /k5/ in thecontext of 4. 34 133 the negative morpheme. (AS/ isinterestingly, an ingressive prefix meaning 'go (inorder) to'--cf. Section 4.50.) The transformational rules thateffect these changes in the presence of the negativemorpheme are presented iaSection 4.41.

In all dialects, the vowel of/bg/ (together wJth the vowel of any SC prefix that mayprecede it) is subject to the tenseness-harmony rule, P69, and in Pante it is also subject to the backness-harmouyrule, P 78. Thus in Akuapem and Asante the prefix isrealized as[IA] before [-Tense7 vowels and (bALbefore 1+Tepse) vowels, yhile in Pante it is realized asDA] before L-Tense,-Backi vowels, [be] before [+Tense,-Back] vowels,Lb5] before [-Tenses+Back] vowels, and [IA] before[+Tense,+Back] vowels. The conventional orthographiesreflect this variation.Some examples are: obsba 'he will come' obefi 'he will leave, Ak-As obsbo 'he will beat' {Fa obobo

'he will bend' Fa obobu

In Akuapem and Asante, when theFUT prefix is preceded by the first-person-fingularC pretix /IA/ (whose usual surface forms areLe) and [IAD, the vowel of the SC and the consonant of FUT aredeleted, by means of P Some examples illustrating theeffect of P81 (which may be optional in somesubdialects) are:

Ak-As mEbe 'I will come' (cf. Fa pibEb&) Ak-As mefi 'I will leave' (cf. Famibifl) Christaller, in his Grammar (p. 59),identifies a verb form which he callstErriiEond future', a verb form marked by the prefix re plus theprefis be. This form, however, may beidentiried as the progressive ingressive, rather than a distinct aspect. (For further discussions cf. Section 4.50.) 134 4.35

4.35 OPTative and 1MPerative

The fully-specified morphophonemicform of the OFTative is/11/: i.e., a[+Nasal,+Tone] segment with other features unspecified. Phonetically /14/ is realized as a nasal consonantthat is homorganic with the conso- pant that follows it.Thus it is Lm] beforebilabials, III] before labiodentals,[n] before alveolars, etc. (cf. P 25). In Akuapem and Asante, but no.5in Pante, a [-Nasal] voiced non-vowel following /N/ is changed toits (+Nasal] counterpart by means of P 72: compare Ak-As L5165)/Fa.[5ib6] 'he should beat', Ak-As16find'i/Fa 16AdV] 'he should arrive', etc.

In the presence of BEG, atransformational rule applies whidh replaces deep-structureOPT by surfacestructure HAB. That is, the negative optativeis phonetically identical with the negativehabitual. (The negative optative and negative habitual are,however, differentiated orthographicallT: q.g., rlYs,[4W] 'he shouldn't do it' vs. 2111 Loppi9J 'he oesn't do it'. For a presentation of the pertinent transformation,and for further discussion and examples, cf.Section 4.41.) The Werative aspect, likeSTA and HAB (cf. Section 4,31) has a zero segmental form.Since IMP forms differ fromkA and HAB forms tonally (and in certain other ways), the difference between thesevarious zero forms must be preserved in inputs to P-rules.Thus, jugt as the STA and HAB forms are identified asincluding STALØ] and tbe IMP forms are id3ntified asincludingin[0] HAB°] As was noted previously,1MP is a derived aspect. It is introduced intoinputs to the phonological component as a replacementfor OPT when the deep-structuresubject is a second-person-singular pronoun. IMP and OPT are in complementary distribution, IMPoccurring only when the deep structure has asecond-person-singular-pronoun subject, OPT only when it does not. (OPT, rather than IMP, occurs when thesubject is a second-person-pluralpronoun.) The rule that replaces OPTby IMP also deletes thesubject pronoun itself. It has some such form as: 4.35 135

X - NP -OPT - Y Pro +2nd Person r+Singular SD: 1 - 2 - 3 .4

SC: 1 - 0 -IMP - 4 where the second element in the stracturaldescriptian represents Ak-As mo/Fa ewo 'you (singular)' . The rule operates, for example, on a tree such as:

NP--1*414b1P

V NP

ASP VR

(e)wo OPT hwe Koff 'you' 'look-ato 'Xofi'

to yield the derived tree:

V NP ASP VR

1 1 IMP hwe1 Kofi The phonological component converts this derived tree into the sentence: Ewe Koff. 'Look (you-sg.) at Kai.' 136 4.35

(This sentence may be compared with such sentences as:

Onhwe Kofi. 'He should look at ir.ofi.1

rAk-As Monhwe Kofi. 'Look (you-pl.) at Kai. iFa Wonhwe Kofi.

These sentences have optative verb forms.Note that the transformation that deletes the imperative subject precedes the subject-concord-insertion transformation.This ordering means that the imperative lacks a subject-concord prefix, since the structural description for the subject- concord insertion rule (cf. Section 4.20) requires a subject. Optative verb forms, on the other hand1 like all finite verb forms other than the imperative, include a subject-concord prefix.) Like OFT, IMP is transformationally replaced by HAB in the presence of NEG.That isthe negative imperative is, in surface formstidentical with the negative habitual (except that, as a result of the transformation discussed above, the imperative form lacks a subject-concordprefix).

There are two verb roots, ba/bra/ 'come' and d(ze /dx/ 'use, take', which show suppietive formsrespectively bra /bIda/ and fa(r(e)) /f&t/--in the OFT and IMP. (The notations d(z)e and fa(r(e)) indicate dialectal variation. De and dze are, respectively, the Akuapem-Asante and the Minte 4iIling that correspond to the systematic-phonemic representation /d1/. In the case of 4idal, the final /t/ of the systematic-phonemic form /1"Iti is always replaced by glottal stop (which is not represented in conventional spelling) in Akuapem and is sometimes replaced by glottal stop in pie other dialectas well. In certain 9o9texts, however, Lfarn occurs in Fante surface forms and lfarx] in Asante surface forms--cf. Sectiors 4.72 and 4.73.) The suppletive form of ba occurs only in the affirmative. Thus: Kofi mmra. 'Kofi should come.' Bra. 'Come (you-sg.).' 405, 4.36 137 vs: Ak-As Kofi mmma. 'Kofi shouldn't come.' Fa Kofi mmba. 1 .1 Ak-As Mma. 'Don't (you-sg.) come.' Fa Mba. i J (In the Akuapem and Asanteforms of the affirmativeOPT and of,the negativeOPT And IMP, the initial/b/ of /bidlt/ and /ba/ is replaced bylm] as a result of the operation of P 72, the rulethat,nasalizes voiced non-vowelsafter nasal consonants.)/fat/ replaces /da/ in bothaffirma- tive and negative OPTand IKP forms: Kofi mfa mma me, 'Kofi should give it to me.' (Kofi should-take should-give me) Fa ma me. 'Give (you-sg.) it to me.' Kofi mmfa mmma EA.'Kofi shouldn't give it to me.' Mfa mma me. 'Don't give it to me.'

Soft 41 ma me, (c.f.{Ak-As 'loft gives it to me.') Fa Kofi dze El. (Kofi takes gives me.)

In the case of the,negativeforms, however, thereplacement of /dx/ by/fat/ is part of a more generalsystem of suppletion, since the replacementoccurs in all negative forms. (Cf. Section 4.40, in which,also, there is a discussion of certain syntacticpeculiarities of d(z)e.)

4,36 CONsecutive. The second of the derivedaspects, the CONsecutive, is introduced into surfacestructures in one of two ways: as a replacementfor PROgressive or FUTurein non-initial verbs of serial-verbconstructions; and as a replacement for FUTure in purpose clauses.Examples of the first type are: 4.36 138

Oresan Elba. 1He is coming again.' (he-is-returning he-is-coming)

Obcsan aim. 'He will come again' (he-will-return he-will-come)

of a zerosubject-concord prefix, (The form aba consists An the CONpigrix a- and the verb rootba 'come'.) example of the secondtype is:

Ak-As !lefties no na waba. 'I called him so Fa Mefre that he would come.' (I-called him that he-will-come) PRO or FUT The transformationalrule that replaces by CON in non-initialverbs of serial-verbconstructions has some such form as: -Y X - VP -FUT

SD: 1 -2 - 3 -4 SC: 1 - 2 -CON -4

This rule would, forexample, operate on atree such as:

NP VP VP------"*VP

I I v V '/"%s .00//"ss\ ASP VP ArP VP

1 1 1 ono PRO san PRO ...ba 'come' 'bel 'return' 4.36 139 to yield the derivedtree:

VP V

A VR ASP'///Nh

1 1 ono PRO san CON ba which, after certain otheroperations, would be converted by the phonologicalcomponent into the sentenceOresan aba. (The original tree Ls itselfthe product of whatever transformations are involvedin the generation ofserial- verb constructions ofthis type. It is assumed herethat the rule that replacesPRO or FUT by CON precedesthe rules, mentioned in Section4.20, that introducesubject-concord prefixes and delete pronounsubjects.)

The transformationalrule that replacesFUT by CON in purpose clauseshas some such form as:

-FUT W- PUR[X 11PUR - 5 SD: 1 - 2 - 3 -4 - 5 SC: 1 - 2 - CON - 4

wherepuR[X-FUT YiPUR is a purposeclause containing a FUT verb. This rule would, forexample, operate on a tree such as: 140 4,36

NP VP ADV

V NP PUR

ASP VR

I I me PAS fre ono na ono FUT ba 'call' 'he"that"he' 'come' to yield the derived tree:

V NP PUR

ASP VR

NP VP V //N\ ASP VR

me PAS fre ono na ono CON ba

which, after certain otheroperations, would be converted by the phonological componentinto the sentenceMefre(e) no na wabao 4,36, 4,40 141

The systematic-phonemic form of the cm prefix /A/ is identical with that of the PER prefix /a/ which occurs in verbs with perfect-affirmativp and past-negative meanings (cf. Section 4,33), The tone patterns of verbs that include CON are, however, in some cases different from those of corresponding verbs that include PER. These differences result from the operation of certain tonal P- rules that apply to verbs that include CON or to verbs that include PER but not to both (cf. Section 4.91).

The two P-rules that affect the segmental forms of verbs that include PER (cf. Section 4.33, subsection (0), apply equally to verbs that include CON.These are P429 which operates to change the forms of SC prefixes before /7a/, and P 69, the tenseness-harmony rulewhich, in combination with P 84, operates to change any [a] in arprefix to Ak-As [a]/Fa [e] if the first vowel of the VR is L+Tense].

4,40NEGative.

The base rules that generate Akan deep structures permit the optional choice of one NEGative morpheme for every Verb Phrase, In surface structures this morpheme is immediately dominated by V, and occurs immediately after the ASPect morpheme,

While NEG occurs as a single morpheme in deep structures, its phonetic realization in surface structures typically consists of two or more discontinuous elements: a nasal prefix and one or more transformationally-introduced glottal suffixes, (The occurrence of glottal suffixes as exponents of NEG was first noted by Stewart, 1962, pp. 39-41 et passim.). The systematic-phonemic form of the nasal prefix is /N/: i.e., a [+Nasal,-Tone] segment with other features unspecifiea. Phonetically it is realized as a nasal consonant homorganic with the consonant that follows it--cf. P 02, The glottal suffixes are introduced by means of a transformation which, given a deep-structure occurrence of NEGI inserts a copy of NEG at one or more other points. These copies of NEG are realized phonetically as glottal stops,The NEG-copying transformation is not presented formally in this text. An informal account of the distribution of glottal suffixes in sentences that include NEG ishowever, presented immediately below. 142 4.44

Within a VP that includes NEG, then, aglottal suffix occurs, with certainstatable exceptions, after any word that is followed by a pause.Thus a glottal suffix normally occurs after the last word of theVP, before the pause that marks the end of the sentence, asin the following examples: Onhwe. [54467] (he-doesn't-look-at) 'He doesn't look at it.'

Onhwe Kofi. ki)fi?.1 (he-doesn't-look-at Kofi) 'He doesn't look at Kofi.' " Onhwe Kofi pg. [opcile kofipii?] (he-doesn't-look-at Kofi much) 'He doesn't look at Kofi much.'

(Note that in non-negative sentences thereis no glottal stop after hwe 'look at', Kofi'Kofi', or 211 'much'.) If one or more pauses occurwithin the VP (as a result of replacing word boundaries by pnraseboundaries), then a glottal suffix normally occurs after eachsuch pause, as well as at the end of the phrase. Thus, in a slow or emphatic rendering of Onhwe Kofi pii,with pauses inserted after each word, one would get

[5hcmg? kOfi? pi17]

(When adverbial clauses follow negated verbsthe clauses in same cases show pre-pausalglottal suffixation and in other cases do not.The occurrence or non-occurrence of glottal suffixes maybe taken as criterial for whether or not the clauses areconstituents of VP. That is, if a clause that follows anegated verb shows glottal suffixation, it is structurally a part of theVP, while if it fails to show glottalsuffixation, it is not.More investigation is needed of clause-typesthat do, and those that do not, show glottal suffixation. The former apparently include clauses introduced by ansa na'before', koptm se 'until', and na 'so that'; tnelater apparently Inc ude clausesintroduced by efi(ri) se 'because' and me se 'unless'. Clauses of this second type may be 4.40 143 called 'sentence adverbials': i.e., they are immediate constituents of S (Sentence) rather thanof VP.)

The only cases thus far discoveredin which a glottal stop fails to occur after aprepausal word within a VP that includes Nal are thefollowing: (a) when the prepausal word is a negativeperfect verb-form; (b) when the prepausal word occurs within aVP that ends in a SUBordinative morpheme(cf. Section 4.81); (c) when the prepausal word occurs within astructure that ends with the interrogative particle/A/. Examples are:

(a) Ak-AsOnhwee 3nhwee [49wEi]"[49wn] Fa Onhwee [5kwti] (hs-hasn't-looked-at) 'He hasn't looked at it, He hasn't looked at

(The first form given for Akuapem and Asanteis the form that occurs When the verb is notfollowed by a Noun Phrase; the second form is the one that occurswhen a Noun Phrase follows. In Pante, the same form occurs in Loth contexts.)

(b) {Ak-Fa ['LOA ni 67191 Kofi na onhwe. As [kOfi *IA] (Kofi EMPH he-doesn't-look-at) 'It's Kofi who doesn't look at it.'

(In Fante, the westion particle/a/ is reaiizecl phonetically either asLa]--in which case a preceding Lei is assimilated to it--or as a low-tonereduplication of the preceding segment. In Akuapem and Asante it has only the second of these realizations.) There are two Verb Roots, wo/la/ 'have, be inl and qz)e /di/ 'takeuse' that have suppletive allomorphs wnen they occur afterNEG. The suppIetive allomorph of 4.40 144 wo that occursin this c2ntextisi/dI/ (which is realized phonetically as Ak-As(ni), Pa (pi)); the suppletiveallo- morph of 4.(z)e is/fit/. (/fitt/ occurs as an allomorph well as in all of d(z)e in theaffirmative imperative as negative aspectscf.Section 4.35.) Thus the negative of: Kofi wo sika. 'Kofi has money.' (Kofi has money)

is: Ak-As Koff nnisika. Pa Koff nazi sika. 'Kofi doesn't have (Kofi doesn't-have( money.' money) 1. And the negatkfreof: Ak-As Kofi de in. Pa Kofi dze be.. Kofi brings it.' (Kofi takes comes

is: la-As Kofi mfa mma. Pa Koff mfa mba. 'Kofi doesn't bring it.' (Kofi doesn't- take doesn't- come)

The VR d(z)e hasunique syntactic, aswell as in which the morphological, properties. In those cases from the fully- occurring allomorphof d.(;)e, is derived /dt/z-i.e., when the specified morphophonemicorm Akuapem-Asante phoneticform isdx] and the Pante pho- [dzx]--the verb always occursin the sta- netic form is of tive aspect, butis followed byanother verb in one the active aspects: e.g., Od(z)e beba, 'He will bringit.' (he-takes will-come) 145 4,40

Ak-As Ode 'aae. Fa Odze baa, 'He brought it.' 1 (he-takes came)

In those cases,however, in which theoccurring allomorph is derivedfrom the fully-specifiedmorphophonemic form /fit/2 both theform of/fit/ and that of the following verb are in oneof the active aspects.

Ak-As Oremfa mma, 1Fa Oremfa mba. 'Re won't bring it.1 (he-won't-take won't-come) Ak-As Wamfa amma, Wamfa amba. 'He ddin't bring it,' (he-didn't-take 1Fa didn't-come)

The structuresillustrated above areall examples of serial-verbconstructions, correspondingto deep struc- 'tures in whichd(z)e is followed by anembedded sentence. In all cases ofserial-verb structuresexcept those that involve d(z)e as thefirst verb, there isalways agreement in aspectbetween the two (ormore) verbs involved. From this point of viewthe structures thatinvolve the allomorph of 1(z)ederived from /fat/ arenormal cases serializat on whilethose that involve theallomorph of gz)e derivedfrom /di/ are abnormal. It is therefore suggested here that theSTAtive forms of2.(z)e that occur in affirmative sentences aTtestrictly sur ace7structure prms and that in deepstructures d(z)e is a[+ACT ] ] verb (cf. Section 4.30)2 andheace may be in- serteronly after the activeaspects. A sentence like 3de beba would thereforehave as a deep structure some- 146 4.40, 4.41

ASP

Ono FUT d(z)e ono FUT ba 'he"will"take"he°'will"comel

There would then be a transformational rulethat replaces the ACT aspect before d(z)e(i.e., FUT in the above example) by STA in the appropriate cases.

4.41. Aspectual Changes in Negative Verbs. There are fewer aistinctive aspect-marked formsof negative verbs than there are of non-negativeverbs. That is, certain distinct aspectual meaningsthat are expressed by different forms in the case ofnon-negative verbs are expressed by a single form in the case ofnegative verbs. This collapsing of formal distinctions in negativeverbs results from the operation of transformationalrules that replace certain ASP morphemes by othersin the presence of NEG. Some of these replacements occur in allthree dialectst others are restricted to onedialect only. The aspectual changes that occur in all threedialects are: (a) FUT - PRO (i.e., the FUT morpheme is replacedby the PRO morpheme in the presence ofNEG); (b) OPT/IMP -0 HAB; (c) CON -.0 HAB; (d) PAS PERf (e) PER - PAS. A change that occurs only in Akuapem is: (f) PRO -0 HAB. A change that occurs only in Fante is: (g) FUT --,0HAB + INGk (where IliGk is the ingressive prefix/ko/--cf. Section 4.50).These changes are discussed in turnbelow.

(a) FUT - PRO.The negative future and the negative progra7Nrig homophonous inall three dialects, and the forms that express bothnegative-future and negative-progressive meanings involve the PROmorpheme /di/ 4,41 147 rather than the FUT morpheme /IA/ (cf. Sections 4.32 and 4.34)0 Compare, for example:

orekasa 'he is speaking' obekasa 'he will speak' and.:

oredkasa 'he isn't speaking, he won't speak'

The last example, in its future interpretation, results from the operation of a transformational rule that replaces FUT by PRO in the presence of NEG.This rule has some such form as:

X - FUT -NEG- Y

SD: 1 - 2 - 3 - 4 SC: 1 -PRO - 3 - 4

(b) OPT/IMP -0 HAB. The negativoptative and imperative iN-71785-7Brmally identical with the negative habitual. The following transformational rule, in which OPT or IMP is replaced by HAB in the presence of NEG, may therefore be proposed:

X - - NEG- Y

SD: 1 - 2 - 3 - 4 SC: 1 - HAB - 3 - 4

(c) CON -0 HAB. The negative consecutive, when it occurs in 11713517-Iiiffial verb of a serial-verb construction, is formally identical with the negative habitual. As was noted in Section 4.36, the CON morpheme in this graumatical context is a transformationally-introduced replacement for PRO or FUT. When NEG is pre:Ant in the verb, a further transformation operates to replace CO3 itself by HAB:

X -VP -CON- NEG- Y

SD: 1 - 2 - 3 - 4 - 5 SC: 1 - 2 -HAB - 4 - 5 4.41 148

The effect of this rule isseen in a sentencesuch as:

Kofi rensan nkasa. i'Kai isn't speaking again.' 'Kofi won't speak again.' (Kofi won't-return won't speak)

in which nkasa (Ak-As[hkaske)/Fa U)kAsi?]) is formally identical-ZEE the negative habitual formthat occurs in a sentence such as:

Kofi nsan nkasa. 'Kofi doesn't speak again.' (Kofi doesn't-return doesn't speak)

In the other contextin which CON occursui.e., in purpose clauses--theabove transformation does notapply, so that thenegative consecutive in thiscontext remains distinct from the negativehabitual. Thus in the purpose clause: na Kofi ankasa 'so that Kofi won't speak'

The form ankasa includes areflex of the CON morpbeme /A/, and 17-50s distinct from the negativehabitual form nkasa.

(d) PAS - PER and (e) PER -"PAS.Unlike the other transformational rules discussedin-Ms section, all of which have the effect ofobliterating deep-structure aspectual differences in the presenceof NEG (with the result that two aspectually-differentdeep structures have the same surface-structurerealization), the rules presented immediately below donot have this effect. These are the rules,referred to in Section 4.33, which replace PASt by PERfect, andvice versa, in the presence of NEG. The rules have the form: X - PAS -NEG - Y

SD: 1 2 3 -4 SC: 1 -PER- 3 4

X - PER -NEG -Y

SD: 1 - 2 3 . SC: 1 -PAS - 3 -4 4.41 149 I Since both rules are obligatory, thedistinction between deep-structure negative-past and negative-perfectverb forms is preserved, although in the surfacestructures a reflex of the PER morpheme is used to express apast mean- 1 ing while a reflex of the PAS morphemeis used to express a perfect meaning. For examples, cf. Section 4.33,

(f) PRO - HAB. In Akuapem, a negative-progressive meaning may be expressed by a form that ishomophonous with,the negative habitual.For example, the form onkasa [54kase]may, in Akuapem, mean'he isn't speaking' as well as 'he doesn't speak'. (In Asante and Fantev onkasa-- As [513kAsfifl, Fa[?jkasi?]--means only 'he doesn't 17417.) Thus an optional transformation likethe following may be proposed for Akuapem:

X - PRO - NEG Y

SD: 1 - 2 - 3 .4 SC: 1 -HAB - 3 - 4 (This rule must precede the rule given aoove which replaces FUT by PRO before NEG, since daep-structureFUT is never expressed by HAB.)

(g) FUT - H1B INGk. Fante shares with the other two dialects the rule presented above, inwhich FUT is replaced by PRO in the presence of NEG. In Fante, however, there is also analternative transformational rule that may apply to a deep-structure seauenceconsisting of FUT and NEG. In this transformation FUT is replaced by HAB rather than PRO, and the INGressiveprefix /Ito/ (symbolized as INGk--cf. Section 4.50) is inserted after NEG. The rule thus has the form:

X - FUT NEG Y

SD: 1 - 2 - 3 - 4 SC: 1 - HAB-3+INGk - 4 The products of this rule are homophonouswith the negative habitual ingressive formed with/ko/. For example, the Fante form onkskasa may meane4hqr 'he won't speak' or 'he doesn't go tospeak'. (The Is; of the ingressive prefix in onkskasa results from theoperation of the Fante backness-harmony rule, P 78.)Pante, then, has two 150 4.41, 4.50 negative future forms: one homophonouswith the negative progressivethe other homophonous with thenegative habitual ingressive formed with/16/. The second of these negative-future formations doesnot occur in Akuapem or Asante.

4.50INGressive. There are two INGressive prefixes, onehaving the meaning 'come (in order) to...1 or'come and ', the other having the meaning'go (in order) to ...1 or 'go and Their fully-specified morphophonemicforms are, respectively,/bc/ and /ko/. The ING prefixes do not occurin deep structures. They are introduced intosurface structures by means of transformatians that operate upon sentenceswith serial- verb constructions consistingof a VP whose VR is ba rea/ 'come' or ko r e /k6t/ 'go' plus another VP. (The notation kaaC)) for the VR 'go' indicates dialectal variat on--cf. Sections4.72 and 4.73.) These serial-verb constructions themselves aretransformation- ally derived from deep structuresin which sentential complements follow VPs that includeba or ko(r(e)). Thus, starting from a deep structuresuch as:

ASP VR NP VP

ASP VR

Kwame HAB baKwameHAB kasa

'Kwame' 'come"Kwame' 'speak' 151 4,50

the serial-verbtransformation operates toproduce the derived tree:

This structure, in turn,undergoes a transformationthat inserts the ING prefix/IA/ before the VR of the VP that follows be., (The ING prefix/bt/ will be symbolized here as INGb. The ING prefix/0/, which is transformationally inserted before the VRof a VP that followsko(r(e)) will be symbolized asINGko) The transformationalrule that inserts theING prefixes has some such form as:

ba -Y - ASP (NEG) -(RED) VR - Z X EB(r(e)) [ - -6 SD: 1 - 2 -3 . 4 5 -6 SC: 1 - 2 -3.4 +Ngibc1- 5

This rule operates,for example, upon thederived tree just presentd toproduce the tree:

ASP VR V ASP VR

Kwame HAB ba HAB INGbkasa 152 4.50

And. this tree, after appropriatespecification of the ASP and ING morphemes, is converted by thephonological component into the sentence:

wame ba bekasa. 1Kwame comes to speak.'

In cases like the above, wherenothing intervenes between ba (or ko r e ) and the verb that has the ING prefix, and where ASP does not equalFUT, a further, optional, transformation may operate todelete the V whose VR is ba (or ko(r(e)))itself. This transformation has thi-rorm: [ING1 - ASP(NEG) X - ASP (NEG) - ko( r( e) ) INGk Y

.5 SD: 1 - 2 _ 3 4 . 5 SC: 1 - 0 4

Condition: 2 FUT

If applied to the last tree presented,this transformation yields the tree:

NP /.6111%% ASP VR

KwRme HAB INGb kasa

The sentence that corresponds tothis tree is:

Kwame bekasa. 'Kwame comes to speak.'

Note that, as the transformationis stated, the deleted ba or ko(r(e)) must immediatelyprecede the verb that includes the ING prefix.Thus, for example, ba cannot be deleted from the structurethat underlies the sentence: 4.50 153

Kwame ba Kumase bekasa. 'Kwame comes to Kumase to speak.'

(There is no *Kwame Kumasebskasa.) The condition on the transformation totan-Mier-Mgt the element immediately preceding ba orko(r(e)) is not FUT is intendedto block, for example, thedeletion of beba fromthe structure that underlies the sentence:

Kwame bcba abskasa. 1Kwame will come tospeak.'

(There is no *Kwame abskasain the meaning'Kwame will The a- come tospeak''artE707-g no *Kwame bebekasa. of abskasa is therealization of theOria5Teme that oblWaIly replaces FUT innon-initial verbs ofserial- verb constructionscf.Section 4.36.) It should be noted, however, that thedeletion of the sequenceFUT + NEG + ba/ko(r(e)) is permitted by therule. Thus the rule can operate upon the structureunderlying, e.g.:

iAk-AsKwame remma mmskasa. 'Kwame won't come Fa Kwame remba mbskasa. to speak.'

to produce thestructure underlying: {Ak-AsKwame remmskasa. 'Kwame won't come to Fa Kwame rembekasa. speak.'

(The initial /b/s of/b1/ and /b6/ undergo assimilationto The a precedingnasal in Akuapem andAsante--cf. P 72. forms with re resultfrom the transformationpresented 4743., in which PRO replacesFUT in the presence in Section FUT of NEG. Note that thetransformation that replaces by CON in non-initialserial verbs (cf, Section4.36) fails to apply in caseslike the last examplecited. This means that the ruledeleting Vs whose VRs areba or ko(r(e)) precedes the rulethat replaces FUT byCON. example is not Given this ordering,the verb in the last part of a serial-verbconstruction at the pointin the rules at which theFUT CON rule would apply, sothe replacement of FUT byCON fails to occur.)

As was noted inSection 4.34, Christalleridentified the combination ofPRO and INGb as adistinct aspect, the 154 4.50

second future'. It is certainly the case that a form such as orebekasa may carrya future meaning, 'he Is about to speak', as well as a progressive-ingressive meaning, 'he is coming to speak'. It seems to us, however, that this meaning may simply be assigned, by the semantic component of the grammar,as a possible reading of the deep structure underlying thesequence PRO + INGb, without setting up a distinct second-future ASP morpheme. (The sequence PRO+ INGb allows of the 'second-future' interpretation only in the affirmative. Thus the negative forms Ak-As oremmckass/Fa orembekasacan not mean 'he isn't about to speak' but only 'he isn't comingto speak' or--in cases where re isa replacement for a deep- structure FUT--'he won't come to speak'.For further discussion of the 'second future',see Dolphyne, p. 37, which takes issue with the view adopted here.)

In all three dialects the vowels of the surface forms of the ING prefixes vary in accordance with the P- rule of tenseness harmony (P 69). In Fantethe vowels of the prefixes also vary in accordance with9 the Pante P-rule of backness harmony (P 78).The segmental forms of p/Gb that occur in Akuapem and Asarateare Lim] an4 lbeh Fante shows, additionally, the forms Lbo]and Lbo]. The segmen41 4borms pf INGk thatoccur in Akuapem and Asante are Lkoj and Lko]; Fante shows, additionally, the forms [kg] and [ke]. Examples are:

obeys 'he comes to do' obeyi 'he comes to remove' Fa oboto 'he comes to buy' (cf. Ak-As pinto 'he comes to buy') Fa obotu 'he comes to dig' (cf. Ak-As obetu 'he comes to dig') okoto 'he goes to buy' okotu 'he goes to dig' Fa okurc 'he goes to do' (cf. Ak-As okoyc 1.4ae goes to do') Fa pkeyi 'he goes to remove' (cf. Ak-Asokoyi 'he goes to remove') 4,50, 4,60 155

In Akuapem and Asante, theunderlying high tone on the ING prefixes is changed tolow by means of P 47, Compare, for example, the surfacetone patterns of Ak-As (BbijEj/Fa [5b6j6] 'he comes to do', While the surface tone of ING itself is low inAkuapem and Asante, certain other surface tonal phenomena inthese dialects give evidence of the underlying high toneof ING. For example, there is an Akuapem-Asante rule,P 20, which changes an underlying high-low tone pattern on adisyllabic VR to low-high in Akuapem, low-fallingin Asante except when the VR immediately follows ahigh-tone prefix, (The rule thus changes, e.g.,[Obisl.] 'he asks' to Ak [Obis:I'd/As[hia].) The fact that P 20 fails toapply to ING forms(cf. the surface tone pattern of Ak-As(Obebis& na] 'hecomes to ask him') indicates that atthe point in the rules at which P 20 applies, the INGprefix still has high tone.

4,60REDuplication, Reduplicating affixes that areconstituents of Verbs are in most casesintroduced into deep structure through the exercise of anoptional choice in the base rule for rewriting V(Verb), This rule has some such form as: V - ASP (NEG) (RED*) VR,where the parentheses around (RED*) indicate that the reduplicatingaffix is optional, and where the asterisk indicatesthat RED may be repeated an indefinitenumber of times (as inokasakasakasakasa- kasakasa... 'he talks and talksand talks and talks and talks and talks ) When RED is introduced by meansof an optionalchoice in the base rules, itadds a meaning of 'multiple activity'(or, in the case of stative verbs, 'multiple state') to the meaningof the VR. The multiple character of the activi4 mayinvolve its being performed by more than one actor,its being performed upon morethan one object,and/or its being performed more than onetime. In general, the more oftenRED is repeated, the greater the number of actors,objects, and/or repetitionsimplied.

In a small number of cases,RED is introduced into deep structures from thelexicon, where it occurs aspart of a VR. That is, there are g fewVRs which have the internal structure: RED + VR1 (whereVRI is itself, for the purposes of phonologicalrules, a VR). In such cases 156 4,60, 4,61

nepher RED nor VR1has any independent meaning,although YR' usually has some semantic relation to a homophonous VR that is listed in the lexicon without anobligatory RED Compare: d(z)id(z)i 'eat (intransitive)' and d(z i 'eat(transitive)1 (where the spellings with and wit cut z represent the Fante andAkuapem-Asante forms respectively), hwehwc 'look for' and hwe'look at', siesie 'arrange' apa sie 'put'. (There is at leapt one VR-BT-the RED + VR1 typo, sesa 'change', whose VR1 is not homophonous with a semanticirly related VR.This verb must nevertheless be analyzed as areduplicated monosyllabic VR9 rather than as a disyllabicVR, on tonal grounds. Compare, for example, the tones, in Fante of sest). 'he changes' and 5kisi Ihe speaks'--where kasa is an ordinary disyllabicVR7) The RED* that isintWauced as an optional constituentof V may co-occur with a VR which itself includes RED. This accounts for forms such asod(z)id(z)id(z)id(z)i 'he eats and eats' or ohwehwchwehwc 'he looks and looks for'.

The precise form of RED in any given caseis specified by P-rules. The rules themselves are presented in the Appendix.An informal account of the operation ofthese rules is given in the sections thatimmediately follow the present section. Section 4.61 discusses the general rule of monosyllabic RED; Section4,62, subsidiary rules affecting monosyllabic RED; Section4,63, the general rule of disyllabic RED; and Section4.64, subsidiary rules affecting disyllabic RED,

4,61General Rule of Monosyllabic RED, The general rule of monosyllabicreduplication (P 21) applies to a RED prefix that immediatelyprecedes a monosyllabic VR(or VR1),The rule operates to replace the RED symbol by a syllable thatis identical with the following VII (or, more specifically,identical with the representation of the VR at the point inthe P-rules at which the rule of monosyllabicreduplication applies), except that: (a) the tongue-height features, High and Low, of the vowel of yR are not cppiedin RED, but are instead specified asL+High] and L-Low] in all capes; (b) in certain specifiable cases, a nasalized(i.e., 1.+Nasa1.1) vowel in VR has a non-nasaliiedccunterpart in RED; (c) when VR has a final[-Nasal] non-vowel, this non-vowel does not appear in RED1(d) whatever the inherent tone of 4.61 15? the VR, RED has low tone and the VR that follows ithas high tone. These differences between RED and the following monosyllabic VR are discussed in turn below. Then, under (e), there is a discussion of theordering of the monosyllabic-RED rule in relation to the palatalization rule, P 11.

(a) Whatever the tongtw height of the vowel of VR, its reflex in RED is always a close--i.e., [+High,-Low]--vowel.This close vowel always agrees with the vowel of VR in tenseness. That is, if the vowel of VR is [+Tense], then RED has a[+Tense] vowel, while if the vowel of VR is [-Tense], RED has al-Tense] vowel. The vowel of RED normally also agrees with the vowel of VR with respect to frontness or backness; i.e., with respect to the feature[±Back]. (EXceptions to this generalization are accounted for by the operation of a subsequent P-rule--cf. Section 4.62subsectipn (a).) Thus if the vowel of VR ip l+Bak] Zi.e., is LIT] or LOD, -.11e vowel of RED is also L+Back 9 whi4e if the vowel of normally, VR is L-Back] (i.e., is CI], LE, orLAD, SO, is the vowel of RED. Exampleswith tone ignored) are: RED + VR [si7] 'stand' sisi [sisi7] LW] 'vomit' fefe [fifx7] [se] 'say' sise Lsise7) LW] 'resemble' sese base] [san 'cure' sesa [sisan [tclien 'cut' twitwa[tc4itc407] [bu(?)] 'bend' bubu [bubu(?)] [su(7)] 'carry on soso [susu(?)] the head' Ak so [son 'seize' SUSO [susoi] so [sot] 'light' SOSO [suso7]

(Forms with a final glottal stopin parentheses show dialect variationthe glottal stop occurring in some dialects but not In others.Fordetails, cf. Section 4.790 158 4,61

rft, (b) When the vowel of VR isLiL [7), :a], or1.0], the L+Nasal] feature of this vwel ioalso normally found in the vowel ofRED,Twae: RED + VR

'scratch' titi [titT7j 'hang on' sese [ea(?)] 'cry' susu [44] 'twist' totoLoom]

But when the vowel of VRis [1], the retention of the [+Nasal] feature inthe vowel of RED depends upon yhether the initial consonantof,VR is -.11.asal] or tlen the vowel 1-Nasal]. If this corsonant isL+Nasal of RED is[+Nasal], Ihus: VR RED + VR ma La(?)] 'give' mema [mYmg(?)] nambile] 'walk' nenamLenge]

If, however, this consonantis [-Nasal], then thevowel of RED is normally1-Nasal], Thus: VR RED + VR

ka [kg(?)] 'say' keka [kxkg(?)] sa Lsg?Y 'tie up' sesa [sisg9]

(For some speakers, and in the caseof certain specific items, nasalizationof the vowel of RED is optional when the vowel of VRis [+Nasal] and theinitial ponsonamt is[-Nasal]. Thus speakers are foundfor whom ltute(?)J, as w9ll asLtoto(?)J, is a9ceptable as a reduplication ofLte(?)], and for whomLsYsg?], as well as [sIsg,i, is acceptable as a reduplicationof La?). As far as we have been able todetermine, however, themly forms that are acceptable toall speakers are those atat are in accordancewith the generalizationspreviously made.) (;12 certain cases, the vowel ofRED, although it is [-Nasalj after the operation of the rule underdiscussion, may becomeL+Nasal] through the operation of a subsequent 4.61 159

P-rulesThus the reduplicated form of me[AO] 'meet' the present rule operates to is (pTmpgm?]. In this case9 produce a non-nasalized vowel in RED andthis vowel is nasalized by a later P-rule, P 74, whichnasalizes all close vowels that precede nasal consonantsin the same syllable. The fact that the nasalization of theclose vowel in COmpImfl does not result from theoperation of the monosyllabic-RED rule becomes obvious41 we compare it with the reduplicated form of Ramlpam?J 'sew', which is LpTmpam?). In the latter case, the close vowelof RED ;s nasalized even though thevowel of VR is itself i-Nasal].)

(c) In the case of those VRs which, atthe point of application of the monosyllabic-REDrule, have a final(-Nasal] non-vowel, this non-vowel is not copied in the formation of RED. The final non-v9wels that are thus dropped in,the formation of RED areLw], Lr] and P]. (Final Lwi, which corresponds to an underlying final /p/, is found in surface forms onlyin Akuapem and in some Fante subdialects. Final [I's], which corresponds to an underlying final /t/, is foundin surface forms only in Pante. Final PJ, which in many casescorresponds to an underlying finalA/2 is found in all dialects, but only in pre-pause position.It should be noted that, in the surface forms of VRs followedimmediately by pause, a 'final Lw]' or 'final[r]' is actually pre-final, since it is followed by a glottal stop in thisposition. It should also be noted that the Akuapemand Asante surface prms that correspond to Fante surfaceforms with final In) often show LrJ plus vowel (plus, in the caseof Akuapem, a glottal stop). The P-rules that account for these surface characteristics aresummarized in Section 4.79.) Some examples are: VR RED + VR hehawLhihaw7J lAk-Falhaw lhawn 'trouble' As-Fa2 ha Chan heha [lithe]

sow sosow(susown iAk-Fal isow?) 'catch' 2 As-Fa so Lso7J soso Isuson

iFa bar [barn bebarlbibarn 'cover' Ak -As bare lbari(/)] bebare LbIbarx(?)] 4.61 160

VR RED + VR hohor [huhue] Fa hor [hur?] Iwash' J [huhuru(?)] Ak-As .horo [huru(?)] hohoro 1

I fefar Ififarfl Fa far [far?] take' [fIfe] Ak-Asfa (fan fefa

'remove' pepa [pipafl Algi [sisifl 21 (sin 'stand'

1 2represen$ subdialectsof Fante: respectively, (Faand Fa in which those in whichfinal Lw) isretained, and those it is not.) those VRs which,at the point of In the case of final application of themonosyllab c-REDrule, have a LmI, En), or[x]--this [+Nasal] consonantia., an discussed, consonant, unlikethe L-Nasalnon-vowels just formation of RED. (Subsequent rules may, is copied in the of the however, apply tochange the pointof articulation delete it.) Some final nasal consonantof RED and/or to examples are: VR RED + VR [frmfee] fem [fee] 'lend' femfem 12210 [tOntue] 1Fa ton [tan?] 'forge' [tOntun5(?)] Ak-Astono [tun5(?)]} tontono [Ak kae] [Ak kYrjkae] kan [As kar] 'count' kenkan [As kifi3kaT) [Fa kYqkann [Fa kanfl1 of Ak-As tonois accounted for by (The 2 [5] at the end For an the operationof a P-rulecf.Section 4.73. explanation of thevariant pronunciationsof kan, cf. Section 474.) 161 4.61

(d) Monosyllabic RED isalways inserted with replaced by high tone low tone, althoughthis tone may be through the operationof subsequentP-rules. In Akuapem, consisting of RED plus a the tone patternof a sequence hwehwe 'lookforlis always monosyllabic VR--e.g., syl- identical with thatof a disyllabicVR whose second lable begins with avoiceless consonante.g.,bisa Thus:

AkMehwehwE nO. 'I look for him.' 4, % Ak Mibisa no. 'I ask him.' to This is also true inAsante, but inAsante, as opposed Akuapem, certain othersentence elements,such as object with RED-plus- pronouns, mayshow differenttonal behavior monosyllabic-VR verbs fromwhat they showwith verbs involving disyllabicVRs. For example:

As1'1ehwehw6 n6. 'I look forhim.' , % As Mibisa no. ask him.'

ofRED-plus-monosyllabic-VR verbsand In Fante, the tones e.g., verbs with disyllabicVRs are usuallydifferent: FaMehwehwg na. 'I look for him.' Fa Mibisa no. 'I ask him.' disyllabic VRs, (For a discussion of thetone patterns of cf. Section 4.75.) between low-toneand high-tone mono- The distinction well- syllabic VRs (whichis well-attestedin Fante, less attested in Asante,and completelyabsent in Akuapemcf. when the VR ispreceded by RED. Section 4.71) is lost VR like Thus a low-tone VRlike da 'sleep'and a high-tone ba 'give' show the sametone patterns intheir reduplicated forms. Compare: Non-reduplicated Reduplicated waided& Fa w5rica 'they aren'tsleeping' Fa w5Mb& 'they aren't coming' w5Mbebtl 162 4.61, 4,62

(e) The monosyllabic-RED rulefollows many of the Pi-rules that affect the shapeof the YR. In particular, it is important to note that itfollows the rule of palatalization P 11, Thus if the initial non-vowelof the VR is palatalized, so is itscounterpart in RED, and if the initial non-vowel of the VR is notpalatalized, neither is its counterpart in RED. Some examples illustrating this point are: RED + VR

gya [djen 'accompany' MiSIZALdjidjen hye LW] 'burn' 11/21a2 [çIW] I hehaw [hihawfl Ak-Fa haw Lhaw?] 'trouble' 2 As-Fa ha [120] heha [huhan

kyc [t9e(?)] 'divide' kyekys, [tgitqc(?)] ka Lka?) 'bite' keka [lake] twa[440] 'cut' twitwa[44it940]

The ordering of themonosyllabic-RED rule after the palatalization rule accounts forthe occurrence in surface forms of RED of certain sequencesthgt do notroccur in monosyllables of other types.Thus Liu] andLki] do not occur in thesurface forms of monosyllabicroots, since lh] in a root is alwgys palatalizedbefore a L-Nasal] non-low front vowel while aLk] in a root is always palatalized before any non-low front vowel.

4.62 Subsidiary RulesAffecting Monosyllabic RED.

After the general rule ofmonosyllabic reduplication has applied, the shape ofRED which it specifies is in some cases altered by theapplication of other P-rules.In this section thre9 such,P-rules arediscussed: (a) the monosyllabic-REDLI] LU] rule, P 22; (b) the homorganic- nasal rule, P 25;(c) the nasal-deletion rule, P 26.The discussion of these rules isfollowed by a discussion of a P-rulethat may alter the shape of aVR that follows monosyllabic RED: (d) the progressive voiced non-vowel nasalization rule, P 72, 4.62 163

(a) The general rule of monosyllabicreduplication specifies that theclose vowel occurring in RED agrees with thevowel of VR with respect to thefeature [±Back]. Thus if the vowel of VR isLA], which is [-Back], pe rule specifies that thevowel of RED should be the l-Back] close v9wel [I]. There 9.re, howeverl certain cases in which0], rather than LI], occurs in RED before a VR whose vowelis LA]. These cases reflect the opera- pion of a P-rule, P 22, whichchanges an [I] in RED to CU] under certain circumstances: viz., when the vowel of VR is LA], and the initial non-vowelof VR (and hence of RED) has certain features in common with[uj.

Three classes of initial non-vowels maytrigger the change of II] toLin in RED (wher the vowel of VR is LAD. These classes are: [4-Round,-Palatal], [4-Round,+Coronal], and[-Round9-Coronal]. In the case of the1+Round,-Palatal] non-vowels (where both of the mentioned feature-specifications are sharedby LUD, the change of LI] tolin is obligatory in most dialects, but optional in some Fante subdialects. Examples of the operation of the rule when theinitial non-vowel is L+Round,-Palatal] are:

Before LI] -- CU] After LI] - CU]

Lf(w)ufwew?] Ak-Fal [fwifwewfl 'chip' Fa2 Lfwifwe] lf(w)ufwen

Ak-Fa (gwigwen 'skin' [g(w)ugwa?] (cf. As [djmidjtion 'skin')

Ak -Fa 'scrape' lii(w)afiwg?) (cf. As 1.4414mV] 'scrape')

[k(w)ukwaw?] Ak-Fal lkwikwawn 'rub off' 2 As-Fa Lkwikwa?] lk(w)ukwe]

[swiswan 'be small' ls(w)uswOi

[wuwari(?)] Ak -As [wiwarI(?)] 'be tall' Fa Iwiwarn [wuwar?] 164 4.62

(The parenthesized '(w)'s in transcriptionsin the right- hand column represent a notationalvariation only.The initial consonants are labializedin all cases, but it is the usual practice in phonetictranscription in this text not to indicate labializationof consonants before rounded vowels.)

Some explanation of certainof the above examples may be in order. At the point in the operationof the P- rules at which the monosyllabic-REDrule applies, the [+Round,-Palatal] non-vowels [fw], [gw], and[Sw] occur before LA] in Akuapem and Fantebut not in Asante. In the case of[fwA], the sequence is attested in only a few Akuapem-Fante items which apparentlyhave no Asante counterparts. In the case of[gwA] and [RwA], the situation is more complex. The underlying forms of theVRs for 'skin' andscrape' are, respectively,/guak/ and AM/. In all three dialects,/guak/ - [guan (by P 03), and /hink/ - [haV] (by P 03) -,[ETV] (by P 06). In Akuapem and Fante, the forms[guafl and [Eagfl undergo, before the application of,t4emonosyllabic-RED rule, only labialization (P 09) and LUJ-deletion(P 10). Thus in these dialects,[gua?] - [gwuen - (gwa?], and[i;a5?]-[Ewag?] -[Rwg?]. In Asante, however, the Asante[I]-insertion rule (P 07) applies toLguan and [IM], yielding [guien and [Rarp]. Thereafter, these forms undergo not only labi 14.zation andLUJ-deletion but also palataliza- tioR and IJ-deletion(r9spectively, P 11 and P 13)4Thus As Lguia?c-[gwuian - Lgwia?] - LOIlian -[dawn, and As [WV] -- [Ewa3-57] --'[fiwIr6 -- [4101] --P [445?]. The forms to which the monosyllabic-REDrule applies are therefore [gwafl and[65?) in Akuapem and Fante,[dNe?] and [W] in Asante and the resultant reduplicatedforms are 1 , . Ak-Fa [gwigwafl andLfiwIhwg?], As [djmidjge?) and Since the Akuapem-Fante formsbegin withL+Round,-Palatalj non-vowels, the [I] -,CU] rule obligatorily applj.es to them. The Asante forms, on the otherhand,rbegin with L+Round, +Palatal] non-vowels, and the [I] LU] rule does not apply in such cases in Asante(see below)... The development of deep-structure RED +/guek/ and RED + /hirek/ in the different dialects is summarizedbelow: Ak -Fa As Ak -Fa As /k/Underlying form 14?] REDRED + + guekgue? RED + gue?guek REDRED ++ ha5khag? RED + haghaft NasalizationLabializationAsante [I] insertion RED + gwue? REDRED + +guie? gwuie? REDRED ++ EUEwag? REDRED + + Rwaig?Eaig?hag Palatalization[U] deletion RED + gwe? RED + gwie?djmie? RED + fiwg? REDRED + ++ 4117.5? 611? Monosyllabic[I][I] deletion--0 [U] RED g(w)ugwe?gwigwe? REDda4idale? + djle? w ? 111445? 166 4.62

In the case of the [4-Roundi+Coconal] and [-Round, -Coronal] initial non-vowels (where l+Round] and [-Coronal] are the features shared withIUD, the chanee of I] to [U] in RED isnever obligatory, and occurs (optionally) only in Akuapem and Fante. Since the application of P 22 is optional, the forms that result from its operation vary freely with the forms that result when it is rot applied. In the examples of surface forms shown below, the first form of each verb that appears in the right-hand column is the form that results if P 22 has not applied, the second the form that results if it has.

VR RED + VR

21 ban 'remove' [pipa?] [pupan ba [WI] 'come' [biba]"[buba] ma [mg?] 'give' [mTmi?] [mama?] fa (fan 'take' [fifa?] [fufa7] ka [ka?] 'bite' [lake] " [kukan haw [ia(w)?] 'trouble' [llIha(w)?]"[huha(w)?] twa Ltclefl 'cut' [tclit910] [tclutclien dwa[0d410] 'report' [dtitlidjlefl [djudilen

For some Akuapem and Fante speakers, certain verbs which otherwise meet the requirements for the optional application of P 22 are specifically marked as being exempt from the application of the rule. Sowe speakers, for example, have only[kIkg?], and not lkukg?], as the reduplicated form of[kg?] 'say'.

(b) As is noted in Section 4.61, subsection (c), if VR ends in a nasal consonant, this consonant is copied in the formation of monosyllabic RED. In such cases, a subsequent rule (P 25) applies, obligatorily in some dialects, optionally in others, to make the final nasal consonant of RED homorganic with the initial non-vowel of VR. The homorganic-nasal rule is obligatory in Akuapem, Asante, and some Fante dialectsloptional in other Fante dialects. Some examples of its operation are: 4.62 167

Before After Homorganic-Nasal Rule Homorganic-Nasal Rule

[sims5mfl 'strip' [sinsge] Chrmem?] 'blow the nose' [hrKihrm?] benmin?] 'swallow' (mTmen?] [kunkwann 'cackle' lktmkwan?] [PIL200] 'leave a space' LPImpaig] (ttgitolg] 'sell' (tuntoV]

(Various subsequent rules may alterthe forms shown in the right-hand columnabove. For example, in those cases where the vowel of REDis not nasalized atthis point in the rules, it issubsequently nasalized byP 74. Subse- quent rules that mayalter the forms of theVRs are discussed in Sections4.73 and 4.74.)

(c) In Akuapem (and perhapsin some subdialects of the otherdialects) a nasal consonant atth9 end of monosyllabic RED is normallydeleted before aI+Nasal] non-vowel at the beginningof VR. The rule that deletes the nasal consonant(P 26) is obligatory in some cases and optional in others. That is, there are some casesin which the only occurringsurface form of RED showsdele- 4 tion of the final nasalconsonant, while there areothers in which surfaceforms of RED with andwithout final nasal consonants occur as freevariants. (Apparently the obligatoriness or optionalityof the rule does notdepend upon generalizablephonological characteristicsof the VRs, but is a matter,instead, of idiosyncraticrule features associated with individualVRs.) In most of those cases in which free variation occurs,the form without anasal consonant in RED ispreferred.

The nasal-deletion ruleoperates to delete thefinal nasal consonant of REDnot only before thenasal consonants lm] and In] and[p] (the nasal consonant11)) does not occur root-initially at the pointin the P-rules atwhich this rule applies), but alsobefore theLNasali glides [0], IA], and LE]. Examples are: 168 4.62

Before After Nasal-Deletion Rule Nasal-Deletion Rule AkLememn 'sink' LmYmTej Ak[nannamn 'suck' Lainamfl Ak411)1140] 'grow' LpIPIeJ Ak[MAO] 'scrape out' 12005n?] AkblYpgYn?j 'weave' Ak 'blow the uost° CRY5Ym?] (The forms listed in theleft-hand column assume that the homorganic-nasal rule discussedin the preceding subsection precedes the nasal-deletionrule.)

(d) In Akuapem and Asante, a[-Nasal,+VO.ced] consonant at the be4nningof VR is changed toits L+Nasal] counterpart after aL+Nasal] consonant occurring at the end of a monosyllabic RED.The progressive-nasalizationrule which effects this change(P 72) is not limited to the context of RED + VR, but,rather, opergtes whenever an appropriate consonant sequence(i.e., L+Nasal] Lollowed by [-Nasal,+Voiced]) occurs within word boundariesat the point in the P-rules atwhich progressive nasalization applies. Thus changes of the samekind occur in VRs immediately after the negativenasal prefix (cf. Section 4,40) or the optative nasal prefix(cf. Section 4.35) as occur after aRED that ends in a nasalconsonant. Examples of the operationof the progressive-nasalization rule in RED + VR structures are:

Before Progressive After Progressive Nasalization Nasalization

Ak-As [bumbumfl 'spread' [bummue] Ak-As [dundolg] 'soak' [dunno0] Ak-As [dtlapdjam?] 'bewail' [daappam?]

Ak (gulnwa0) [gu13gwel)?] 'wither' As [dSiiipdjApig] [4:144iPPlee]

(For an account of the dialectdifference in the last example, cf. subsection(a), above.) 4.62, 4.63 169

The progressive-nasalizationrule does not operate in most Fante subdialects,which account& for the contrast between the Fante sureaceforms[bambam?J,,etc. on the one hand and the Akuapem-Asantesurface formsLbammam?], etc. on the other. (For the nasalization of the vowelsin the surface forms just cited, cf.P 74.)

Note that the progressive-nasalizationrule must follow the Akuapem nasal-deletionrule discussed in subsection (c), above, since anasal consonant at the endof RED is not deleted before a uasalconsonant that is introduced by means of theprogressive-nasall.zation rule. This fact accounts for contrastssvah as Ak inangm] 'suck' vs. Ak idannam] 'extinguish', which show the following derivations: Before monosyllabic-RED RED + nam? RED + dum? Monosyllabic-RED namnam? dumdum? Homorganic-Nasal nannam? dundum? Nasal-Deletion nanam? Progressive-Nasalization dunnum? Close-Vowel Nasalization dannam?

4.63General Rule of Disyllabic RED. The general rule ofdisyllabic reduplication, P23, applies to a RED prefix thatimmediately precedes either a VR with two or moresyllables or another RED prefix,the shape of which has alreadybeen specified. The rule operates to replace the REDsymbol to which it applies by a sequence of twosyllables. These two syllables are identical to the two syllablesthat immediately followRED, 9xcept Oat, if the secondsyllable aft9r RED ends in a L-Nasal.1 non-vowel(i.e., [rj, hi], orin), this non-vowel is not copied in the formationof RED. (If the second pyl101e after RED ends in a[+Nasal] non-vowel (i.e., ln], or CIA), this non-vowel is copied inthe formation of RED, although it maybe optionally deleted by a subsequent P-rule--cf. discussionin Section 4.64, subsection (a).) Some examples of theoperation of the disyllabic-RED rule, when RED occursimmediately before a VR withtwo or more syllables, are: 170 4.63

VR RED + VR

kasa 'speak' kasakasa kamfo 'praise' kamfokamfo Ak nantew Aknantenantew Asnante 'walk' As nantenante Fanantsenantse(w) 1Fa nantse(w) yeramyeram 'yawn' haram haramharam 1 kanyan 'stir up' kanyankanyan Ak patiriw 'slide' pat ipat iriw

(Yeram and haram are dialectal variants.)

As is noted in Section 4.60 anindefinitely long sequence of RED symbols maybe generated by the base rules. In such sequences, the monosyllabic-REDrule applies only to an occurrence of REDimmediately before a monosyllabicVR, and the disyllabic-RED rule applies in all other cases.Thus the disyllabic-RED rule applies to an occurrence of RED that precedes anyother occurrence of RED whether the latter is itselfmonosyllabic or disyllabic.For example, given the sequence RED +RED + 'buy'l first the monosyllabic-RED ruleapplies to the RED immediately preceding to to convert the sequenceto RED + toto, then thedisyllalac-RED rule applies to convert the sequence to totototo.Further examples of structures of this type are: RED + VR RED + RED + VR

Ifoforo iforofoforo foro 'climb' jra for fofor fofofofor

san 'return' sensan sensansensan gya(w) 'leave' gyigya(w) gyigyagyiga(w) kum kunkum kunkumkunkum nam 'walk' ne(n)nam ne (A) namnsin) am 4.63, 464 171

(The parenthesized '(w)'s in thethird example represent dialect variation. They are present in Akuapem and some Fante subdialects, but arereplaced by [?) in Asante and other Fante subdialects. The parenthesized'(n)'s in the last example also representdialect variation. In Akuapem, they are usually absent, as aresult of the applicationof the nasal-deletion rule(cf. Section 4.62, subsection(c)), but they are retained in theother dialects.) Similarly, given the sequence RED +RED + Ak nantew 'walk', the disyllabic-RED ruleapplies first to the RED immediately preceding nantew, toyield RED + nantenantew, and then again, to yieldnantenantenantew.

4.64Subsidia Rules AffectinDis llabic RED.

After the general disyllabic-REDrule has applied, there are a number of other rulesthat may or rxxat apply to alter the forms generatedby it. Among these rules are two affectingthe tones of such forms, which arediscussed separatoly in Section4.94. The present section discusses six rules which affectthe segmental structure of forms that result fromapplication of the disyllabic- RED rule. These rules, which arediscussed in turn below, are: (a) nasal deletion in disyllabic REDP 24; (b) homorganic nasal, P 25;(c) progressive voiced non-vowel nasalization, P 72; (d) vowel geminationin disyllabic RED, P 274,(e) vpwelpominationin RED + RED structures, P 28; (f) le] P 84.

(a) If the second syllable afterRED ends in a (+Nasal] consonant, the disyllabic-RED rule copies this final consonant. There is, however, a subsequentrule (P 24) that optionally deletes afinal nasal consonant from disyllabic 2ED if the consonantis preceded by a vowel in the same syllable.(Where the nasal consonant ia not preceded by a vowel inthe same syllOile--i.e,,,yhere it is syllabic, as in man'turn' (Ak [m50], As LBW], Fa[millj)--it is never deleted.) Examples of the operation of this rule are: 172 4.64

Before Final-Nasal After Final-Nasal Deletion Deletion yerayeram (yeramyeram 'yawn' liaramharam haraharam 1 kanyankanyan 'stir up' kanyakanyan pempampempam 'sew' pempapempam pempanpempan 'leave a space' vempapempan sensensensen 'hang' senscsensen

In those cases in whichthe RED from which thefinal nasal is deleted immediatelyprecedes a disyllabic VR(or a reduplicationof such a VR), the nasalityof the vowel before the deleted nasal isnot affected by thedeletion. If the vowel wasI.-Nasal] before deletion of the consonant, it remains[-Nasal] (c.g.,Ak (jirfimjiri.0] Wirtoiramfl), and if it gas L+Nasallbefore deletion of the consonant, r IP it remains(+Nasal] (e.g. AkLkiipikkarlig] laPakaPROJ). (For an explanation of the tonepatterns of the forms cited inthis section, see Section4.94.) In those cases, cn the otherhand, in which the REDprecedes RED plus a monosyllabic VR(or a reduplication of such a sequence), the final-nasal-deletionrule specifies that the preceding vowel be-zomesLi-Nasal] in all casesThis p4rt of the rule applies vacuouslyif the vowel was L+Nasalj be:ore deletion of theconsonant. Thas Ak1.1)1mpg/wimple] I 4 % r LZi LpImpaprmpIe)). But if the -4- LpImpapImpaigj (ultimately, vowel was[-Nasal] before deletion of the consogant, the final-nasal-deletion rule changes itto 1.+Nasaljo Thus AkLpImpimpimpitmfl [pimpipimplm?J (ultimately, Lpimpipimpam?), and Ak tsinser3sinse] LsinsEsinsinfl (ultimately,isinsisinsW]). (Akuapem forms have been cited in order to avoidirrelevant dialect variations. Apparently, also, use of thenasal-deletion rule is more widespread in Akuapem than itis in the other dialects. Abante speakers, in particular, seemto use the rule in only a limited ;Amber of cases,and although in these cases forms with and withoutfinal nasals in RED alternatefreely, in many other casesonly forms in which thefinal nasal is retained are acceptable.) 4.64 173

(b) As was noted inSection 4.62, subsection 25) to (b) application of thehomorganic-nasal rule (P a finalt nasalconsonant in a monosyllabicreduplicating prefix is obligatory inmost dialects. In the case of a disyllabic reduplicatingprefix, on the othcr helyi,application of thehomorganic-nasal rule is neverobligitory when the final nasal ofRED is preceded by a vowelin the same syllable. If a RED of this typeis generated, and if the optional nasal-deletionrule discussed insubsection (a) above is not applied, Akuapemand Fante speakers may optionallyapply the homorganic-nasalrule in certain cases, while Asantespeakers apparently neverapply it. (Even in those dialects in whichthe rule may be applied, its use is rather erratic,and it does not seem tobe possible to predict on anygeneral basis in just which cases applicationof the rule ispossible.) For those speakers who use the rule, aform like[kalakiim1cal3kam?] (REA; + RED + kum 'kill') mayoptionally be changed to ikkkazAkkriej. Similarly, AkLaynt6i3tilnt5/3?) (RED + RED + toq'sell') may be changed to[aniOntisnt50?] (ultimately (tgnOntOnt5e3) andFa(kii3kRakirjk&n?] (RED + RED + kan 'count') toliclk&ijkinkan?) (ultimately (k7/31613kM3kan?]). In Akuapem, but notin Asante or Fante,application of the homorganic-nasalrule iE obligatory inreduplicated forms of disyllabic VRslike man 'turn aside' whose second syllable is a syllabic nasal. Thus Ak[mahmAV] -0 (m&Mmai37).

(c) The Akuapem-Asanteprogressive voiced non-vowel nasalization rule,P 72, changes a[-Nasal, +Voiced] non-vowel to its Li-Nasal;counterpart after a (+Nasal consonant in the sameword. This rule follows the homorganic-nasal rule,and is inapplicable wherethe homorganic-nasal rule has notapplied.

As was noted in subsection(b), above, the homorganic-nasal rule may applyin Alcuapem, butapparently not in Asante to a nasal occurringat the end of adisyllabic RED. Where the nasal issyllabic, as in the caseof mall 'turn aside', applicationof the homorganic-nasalrule in Akuapem is obligatory. In other cases, the rule maybe either optional orinapplicable. 174 4,64

In those cases in which the homorganic-nasal rule has appliedt either obligatorily or optionally, to a nasal that immediately precedes a L-Nasal,+Voiced] non- vowel, this non-vowel is obligatorily changed, by means of the progressive-nasalization rule, to its L+Nasal] counterpart. Thus if the homorganic-nasal rule has optionally applied to change Ak [dinnfimainniamn[RED+ RED + dum 'extinguish') to [dannfindannamfl, the latter is obligatorily changed, by means of the progressive- nasalization rule, to [dinnannAnAm] (ultimately, [dannandimalam?]). Similarly, after the homorganic nasal rule has operated obligatorily to change Ak[dila5d1337] (RED + dan 'turn') to IdaAdahn, the latter is in turn obligatorily changed to [dnaig] by the progressive- nasalization rule. Since Asante apparently never permits application of the homorganic-nasal rule to a final nasal in disyllabic RED, the progressive-nasalization rule cannot apply to the consonant that immediately follows the RED.Thus Asante does not have forms like[danannannamflor [diulnW).In Asante, RED + RED + dum is realized either asjdInnamnnimfl,or, if nasal-deletion has pplied, [dannUAnnamfl.RED + dan(e) (As [all [dial),hasa varietysof realizations inAsante--[dUdii], [dlaSeanY]--none of which shows progressive nasalization.

In some cases, the homorganic-nasal rule applies vacuously: that is, the nasal consonant to which the rule applies happens already to be homorganic with the following non-vowel, and thus is not changed by application of the rule. For example, when the Lomorganic-nasal rule applies obligatorily to the Akuapem formation [gwitzwihT] (RED + [gwaiet] 'flee% the product of the rule is still [gw615gw4ig]. But even where the homorganic- nasal rule has applied vacuously, the progressive- nasalization rule that is contingent upon it still applies, and changes [cwgwW] to[gw1154w4ig].

(d) Fante has two optional vowel-gemination rules that apply to structures that include disyllabic 4.64 175

RED. Both rules operate to replace part of the structures to which they apply by a sequence of two identical vowels. The first of these rules, which is discussed in the present subsection, is P 27, the rule of vowel gemination in disyllabic RED; the second of the rules is discussed in subsection (e), below.

The rule of vowel gemination in disyllabic RED applies optionally to a RED whose second sylla4e consists either of a syllabic non-vowel or of the vowel LI].Thp rule operates to replace the syllabic non-vowel or the LI] by a copy (except for tone) of the vowel of the first syllgble. If it is a nasal consonant that is replaced, its Li-Nasal] feature is transferred to the replacing vowel and to the vowel of the first pyllable. (If the vowel of the first syllable is already L+Nasal], the feature-transfer applies vacuously.) Some examples are:

Before Vowel After Vowel Gemination Gemination

Fa [dAilda?] 'turn' [diida?] Fa[Alma ?] 'turn aside' [martm&A?] Fa(fikii?] 'borrow' [fiifii?] Fa 'waste' Fa[palpa] 'break' CpUpa] Fa[dSZidgfl 'cease' [d4;6cAZI]

(For an account of th:% dErivation of VRs of the types involved in these examp5. as well as an explanation of the apparently anomalous tone patterns of the first two examples, cf. Section 4.77.)

In the case of the last two examples listed, a second part of the vowel-geminatIon rule obliWorily changes the [A] of the VE to rEl.Thus Lpaapai] - [p&Apn] and [ddja] Ldje6dj41]. Furthermore, [dj&6djel] undergoes the censeness-harmony rulecf. subsection (f), below--which results in the form [c4460.41].

(e) The second of the Fante vowel-gemination rules is P 28. This rule (which may be found in subdialects of other dialects as well) applies to structures 176 4,64 of the shape RED + RED +VR9 where the VR ismonosyllabic. As is explained in Section4.63, the first RED of such a structure is realized as adisyllabic reduplicating prefix, the second as amonosyllabic reduplicatingprefix. Thus, given a si,ructuresuch as RED 1- RED + ba'come', the realization is bebabebalbibAbiba), which has the in- RETITTIT+ [bI] The op- ternal structure RED VREbai. tional Fante rule underdiscussion in some casesaffects both the second andthird syllables of thestructures in question (i.e., both thesecond syllable of thedisyllabic RED and the monosyllabicRED) and in other cases just the third syllable(i.e., the monosyllabicRED). The second syllable is affected onlywhen its vowel is[A] and/or when it picludes afinal consonant.The rule operates to replace LA] byLE] as the vowel of the second syllable* (More precisely, the rule changesthe vowel fromL+Lowj to [-Low], leaving allits other featuresintact.) If the second syllableincludes a final consonant,which is necessarilyL+Nasal], this consonant is deleted, butits L+Nasal] feature is trangerred to thepreceding vowel. (If the vowel is alreadyL+Nasal], the feature-transfer applies vacuously.) The third syllable is inall cases replaced by a single vowel,which is (except fortone) a copy of thevowel of the secondsyllable. (Where LE] has replacedLA] in the second syllable, it isLE] that is copied in the thirdsyllable.) Some examples of the 1 operation of the rule are:

Before Vowel After Vowel Gemination Gemination

[bibAbibkj 'come' Cbibaba] IpimpAmpimpaml) 'sew' [pimpiip&m?) [t?mitclitgt,Iitgman 'cut' Ltvgit?Ileetcyl?] [sinams1ns50] 'strip' Isins5500] [41t9itgitcir'i] 'catch' (tcitclitcir?) [dzidzidzidzi) 'eat' [dzidziidzi] 'look for' [91119116k0]

LfZiftlfZIfitr?) 'climb' Cff(ifisr?) IcadfidadArn 'arrive' ldadadhrn Calnd5nasn6n?) 'soak' laindiid50) 4.64, 4.70 177

(By the application of a subseTztent,rule,P74, [Ampiipie], LpYmpitpamn, Lsinsresge]-0 lsinselsgm?j, andLaindiid5nn-0[dondiid50.1.)

rt.!, (f) The [e] -*[ale) rule, P 84, replaces anLei that is followedby a+Tense] vowel in the next syllablit,by ank] or LA] (according to whether the underlying 19] [-Nasal] or [+Nasal]) in Akuapem and Asante, and by antl]orla] in Fante. The rule applies obligatorily in some cases, optionally in others.One of the cases where application ofthe rule is obligatory in most dialects is the case in which the disyllabic-REDruht,has generated a sequence that includes a REDending in Le] before a VR or RED whose (first) vowel istense. Some examples(with tone ignored) are:

Defore,P ail After P 84 lAk-As[bisubisel [bisebise] 'ask' Fa [bisebise]

{Ak-As [dainilidaing]) Wing] 'stand' Fa [dj-inadian5]

(In some Fante dialects, application of P84 is apparently optional in the context underdiscussion. In these dialects, Ibisebise], etc. need not bechanged to [bisebise], etc.) The [e] -- [e]/[e] rule does pa apply in cases in which a disyllabic RED endsinrriaij plus a nasal consonant. Thus the reduplication ofLbirem] 'beat' is [birembirem]. If, however, the optional rule of nasal deletion in disnlabic RED (P24),applios, [birembirem] to Lbirobirem], the Lai Lai/ ej rule applies to the latter form, yieldingAk-Asbirebiremj, Fa [birebirem]. (For a discussion of the rule of nasal deletion in disyllabic RED, cf. subsection(a), above.)

4.70Verb Roots. The great majority of Akan VRshave underlying forms with either one or twosyllables.There are also 178 4.70, 4.71 a few VRs withunderlying trisyllabic forms.The underlying structure of monosyllabic,disyllabic, and trisyllabic morphemes that function as VRsis essentially the same as that of other rootmorphemes with the same number of syllables.This means that VRs conform to all of the morpheme-structureconditions presented in Chapter 2, and that, in the present chapter,it will be necessary to specify only suchadditional morpheme-structure conditions as apply to VRs to theexclusion of root morphemes of other types. Similarly? VRs are subject to all of the general P-rules(nasalization, labializationpalataliza- . tion, etc.) discussed inChapter 3, and It will be necessary here todiscuss only those additional P-rulesthat apply to VRs exclusively. The discussion is organized as follows: monosyllabic VRs are discussed in Sections4.71 through 4.74, disyllabic VRsin Sections 4.75 through4.77, and trisyllabic VRs inSection 4.78. Section 4.79 is devoted to a discussion offinal glottal stops in VRs.

4.71Tone Raising in Monos llabic VRs. In the case of monosyllabic VRs,there are apparently no specialmorpheme-structure conditions thatdistinguish the VRs from monosyllabic rootmorphemes of other types. That is, apparently all andonly those sequence-structure and segment-structure conditionsthat apply to monosyllabic root morphemes in general applyto monosyllabic VRs. (For a discussionof sequence-structure conditionsthat apply exclusively to disyllabic VRs,cf. Section 4.75.) There are however, some P-rules that applyexclusively or pri- manly to monosyllabic Vits. One of these7 the rule of tone raising in monosyllabicVRs (P19), Is the subject of the present section.

In their underlying forms,VRs, like root morphemes of other types, may,have eitherhigh or low tone on their vowels. Thus mg. /jc/ 'do, insult'is a high-tone monosyllabic VR while 21 /pc/'seek, gather' is a low-tone monosyllabic VR. In surface forms, on theother hand, the contrast between high-toneand low-tone monosyllabic VRs is often lost. The contrast is best preservedin Fante, where we find surfacetone patterns such as: 4.71 179

High=Tone VR Low-Tone VR

Fa 51)634 'he will do' 5b6pi 'he will seek' Fa Agyi 'he doesn't dol AM2i 'he doesn't seek'

Akuapem and Asante do not showsurface tonal contrasts in these cases. In these two dialectsobeyc and obcps have the same tone pattern(Ak-As 5b634,51=6), as do ?nye and 2321(Ak-As Agyi, Aig). (In some Fante subdialects, 5b6.06 varies freelywith 5b6pi andAIRI varies freely with Aig. For adiscussion of this variation, seebelow.) In Asante, while the contrastscited above for Fante do not occur, there are afew cases in which surface tone patterns do reflect theunderlying contrast between high-tone and low-tone monosyllabicVRs. Thus Asante shows tonal differences likethe following:

High-Tone VR Low-Tone VR

As 5r4y6 [5576] 'he is doing' 5repl 155pi] 'he is seeking'

As 5b6y6, nO 'he will 5bta Ei Ihe will insult him' seek him' (literally: 'he will do him')

(In the second example, theunderlying tonal contrast between the VRs is reflectedin the surface forms by the tone on the object pronouns,rather than by the tone on the VRs themselves--seediscussion below.) In Akuapem, on the otherhand, there are never any surface tonal differences thatreflect the underlying contrast between high-toneand low-tone monosyllabicVRs. For example, the Akuapemequivalents of the above Asante forms are:

Ak5reyi 'he is doing' k4S1A 'he is seeking' Ak heye nO 'he will insult 5bEpt nO 'he will seek him' him' 180 4.71

The explanation for dialect differences like those cited above is to be found in a P-rule, P 19, that changes low tone on a monosyllabic VR to high tone. In Akuapem the rule is obligatory in all cases, so that all underlying low tones on VRs are changed to high. (Subsequent P-rules may, however, result in surface forms in which VRs have low tone.) In Asante the rule is obligatory except where the VR immediately follows the PROgressive-aspect morpheme (cf. Section 4.32), in which case it does not apply. (The non-application of the rule to affirmative- progressive verbs in Asans accounts for the above-cited tonal contrast bevoreen zen and 22m. The tonal contrast between ii034 n6 and 5bEDE nO results from an Asante P- rule, P , that changes the underlying low tone of sentence-final object pronouns to high tone in certain cases. One such case is when the object pronoun occurs after a high-tone ASP marker plus a high-tone monosyllabic VR. Since the object-pronoun tone-raising rule precedes the rule,that raises the tone of monosyllabic VRs, obeys and hens are still tonally distinct at the point IFTEe P-rules at which the object-pronoun rule applies, and the rule operates to raise the tone of a sentence- final pronoun after the former but not after the latter. For further discussion of Asante object-pronoun tone- raising, cf. Section 4.95.)

In Fante, the monosyllabic-VR tone-raising rule is never obligatory. In some subdialects the rule does not occur at allwhile in others it is optional. Those subdialects in: which tone-raising does not occur always show,snrface tonal contrasts such as that between 5btyp and obeps or onye and mune. Thoseosubdialects in which tones:raisin ls optional have obspe,yarying freely with obeDe and 2m2E varying freely WiTE-2E2A, and the variant forms that reflect the application of the tone-raising rule are tonally identical with similarly-constituted forms that have an underlying high-tone VR.

(Since in Akuapem the distinction between high-tone and low-tone monosyllabic VRs is always obliterated, in a phonology devoted to this dialect alone, it would be possible to say that the underlying tone of the vowels of all monosyllabic VRs is high, and to state a special morpheme-structure condition to this effect. Given the comparative orientation of the present study, however, it 4.71, 4.72 181 seems preferableto account for theabsence of tonal contrasts betweenmonosyllabic VRs in Akuapemin the way discussed above.) (It should be noted that,in some cases, P-rules other than themonosyllabic-VR tone-raisingrule may account for a surfacehigh tone on a monosyllabicVR with underlying low tone.For example, when amonosyllabic VR immediately follows theOPTative-aspect morpheme(cf. Section 4.91, subsection(g)), P49 requires that itstone be high, so that inthis context even thoseFante dialects in which the monosyllabic-VRton-raising rule never applies show syrface hightone c- VRs withunderlying low there are tone: e.g.,Ima 'he should se'. Similarly, P-rules that may change anunder.:ing high tone on aVR to 1'50 low: e.g., theAkuapem-Fante tons-reversal,rule, that accounts f9r Ak-Fa 2zA'he does', Tye 'he isdoing', etc.--cf. As as 'he doe;Aki 'he ts oing'.) 4.72 Re lacement of FinalNon-Vowel in Monos llabic VRs by Glottal Stop,

There are significantdialect differences in the P-rules that apply tomonosyllabic VRs whoseunderlying forms end in a non-vowel.As is noted inChapter 29 the final non-vowels that occurin underlying forms arethe six stop (i.e.,L-ContinuantJ) consonants /p,t,k,b,d,g/,which are changedby P-rules toLw,r,?,m,n,n) respectively. After these changes haveoccurred, various otherP-rules may apply tomake further changes informs with final non- vowels. The further changes arediscussed in the present section and the two thatimmediately follow it.

As was noted above,underlying final/k/ is changed to a glottal stopby an early P-rule. But a final1.?) in a surfaceform may also correspondto an underlying final /p/9 /t/, /b/, or /di(or may have one of variousother sources--cf. Section4,79), This is because, afterearly ;0-rul's have changedfinal /p/, /t/,/b/, ana /d/ to lw), lr], Du], and [n] respectively, alater rule may in tur replace one of the latter byl?)

The replacement offinal [w] by[?] is dialect- dependent, but is independentof the grammaticalfunction of the root, (The replacement is obligatoryin Asante and in some subdialects ofFante, while it does not occurin 182 4.72 other subdialects of Fante orin Akuapem--cf. P29.) The replacement of finalEr], (m], or [n] by[?],on the other hand, is not only dial9ct-dependentbut is confined largely to cases in which theLr] 19 or[n] occurs at the end of a monosyllabic VR. The P-rule that replaces afinal (r],(m],or[n] by [?]in a monosyllabic VR(1) 30) depends for its operation on the presence of aspecial rule feature on theVR. That is, it apparently cannot bepredicted on any general basis just which VRs with final[117 [m], or En] permit or require the replacement, and whichdo not. Furthermore, there are many VRs that arespecified positively for this rule feature in Asante butnegatively in Akuapem and/or Fante, and there are differencesin detail in the way in which the rule operates in theseveral dialects.

In Asante, the rule appliesto the majority of mono- pyllabic VRs with finalEr) or [n] and to many with final Lm], but is almost always optional. (When the option of glottal-stop replacement is notexercised, further rules obligatorily apply to the forms inquestioncf. Section 4.73.) In the case of at least one VR,[16r] (underlying form: /kSt/) 'go', the rule is optional only when theVR is followed by the PASt-aspectsuffix (cf. Section4.33), but is obligatory in all other cases.Thus,,while 'he went' is, in Asante, either[5k5rii] or[5ko51], 'he goes' is only[516?]. (The absence of a glottal stop in 1.5kni) results from the operationof P 04, which deletes all glottal stops that arenot f4lowed by a pause.)For most other VRs with final[r] or In], and many with final [m], glottal-ster.4 replacement is optional inAsante, regardless of c(ext. Someexamples (with tone ignored) are:

Before After [r,m,n] [?)

[dur] 'reach' [dufl [fir] 'leave' [fin [war] 'be long' [we] [him] 'blow the nose' [RV] [kam] [kV] [lam] 'know' 'see' [k?) ['An] 'pinch' [pie?] [On] 'forge' [tv] 4.72 183

In Akuapem and Fante the rule replacing final Lr,m,n] ina monosyllabic VR by[?] applies inmany fewer cases than it does in Asantel but in those cases where it does apply, its application Is usually obligatory. In Akuapem the application of the rule is pidependent of the grammatical context. Thus in Akuapem, Ldurj 'reach' -0 ld10] and [Icor] 'go' -0 [ko?] j.n all 9ontextp. (The rule j.s obligatory in..Akuapem for Ldur], Lkor], Lfarj 'take', lfir] 'leave', [haff] 'see', and perhaps some other Vils. The rule is opti9nal for at least one VR--underlying /kuat/, surface Ltclian or LW/9=n 'cut'--but in this case as well its application does not depend upon the grammatical context.) In Pante, according to Stewart, 1962 (pp. 139-140), the rule applies to only five VRs, and ist furtherpore, context-dependent. The five VRs are: tdurj, [far], Lhanj, [Icor], and Lt9ifer]. (The last of these is the form of underlying /kuet/ at the point in the rules at which the rule under discussion applies.) The contexts in which the rule applies are (in the subdialect of Pante described by Stewart): (1) when the VR is immediately followed by an object; (2) when it is immediately followed by another VR (as pgrt of a serial-verb construction)/ (3) in the case of Lfar] and [tclor], but not the other three VRs) when it is followed by pile past suffix. EX- amples of surface forms involving Lfarj are:

Fa[5farfl 'He takes (it).' Fa [5fa nfim] 'He takes meat.' Fa [5fa ma k6fi] 'He takes (it) for Kofi.' (literally: 'he-takes gives Kofi') Fa [5111] 'He took (it).' (For an explanation of the final [?] in,(5farfl, cf. Sec- tion 474. In the other examples, theP] that replaces the [rj of (far) is deleted by the general rule, PO4, deleting all glottal stops not immediately followed by pause,) (Fante subdialects other than those investigated by Stewart may show differences in detail from the subdialect herdescribes, with respect to the operation of theFr,manj l?) rule. Thus for at least some Fante speakerskamj 'kill' may undergo the rule, and the rule may apply to forms in grammatical contexts other than those listed: e.g., in absolute-final position.) 4,73 184

4.73VoI/elAdditionii.WaicVRst The rule just discussedin Section 4.72replaces a final Er],bid, or [n] in a monosyllabic VRby N. If this rule does notapply to a particularVR of the appropriate form, or if itapplies optionally andthe option is not exercised, alater rule2 P 31,obligatorily applies to add a vowel afterthe [rl,IA], or In]. This same rule applies to add avowel after a final(w] (which corresponds to an underlying/pV) or a final[1.1] (which corresponds to an underlying/g/) in a monosyllabicVR. (The effects of thevowel-addition rule arefrequently reversed, in whole or in part,by a still laterrule that replaces the added vowel byLfl--cf. Section 4,74, subsection(a).) The vowel that is addedby P 31 is aclose--i.e.2C+High, -Low]--vowel that is specified forthe features1±Back), [±Nasal] according to certain featuresof [tTense], and vowel and/or fipalnon-vowel of the VR.Specifically, the added vowel isl+Baclq if the vowel of the VRis of the l+Back,+High] (i.e., 11]) or if the final non-vowel VR is Lm] or(w]; otherwise it is[-Back]. The added vowel [+Tensel-Low], and is[+Tense] if the vowel of the VR is L+Nasal); it is[+Nasal] if the final non-vowel of theVR is otherwise it is[-Tense,-Nasal]. Some examples are: Before Vowel Addition After Vowel Addition

[jilt] 'overflow' [jiri] [stir] 'hinder' lazi] [s5r] 'worship'

lin(ir) 'thatch' Cam] 'blow the nose' (himid LfEm] 'borrow' [fEmi] [dAm] 'extinguish' Ln]'weave' [sfin] 'go beside' 1.t6n) 'forge' LOno] IfEw) 'thrive' Lf614;] Lam] 'trouble' [htlwid [paw] 'throw out' [pilwa) 4.73, 4.74 185

Before Vowel Addition After Vowel Addition

1s613] 'hang' [864i) [kid 'count' [khi] Lt6d 'sell' [t613i]

(The forms listed above reflect the operation of the tone- raising rule discussed iu Section 4.71. In those dialects in which tone raising does not apply, VRs with underlying low tone still have low tone at the point in the rules at which vowel addition occurs., In these dialects, for example, the low-tone roots /tad/ 'forge' and/t5g/ 'sells have the forms [ten] and[t533] respectively before vowel addition. Voyel addition changes [On) to (One] and [t54] to [tht].)

4.74 Other Rules AffectinA Monosyllabic VRs.

This section discusses four further rules that affect certain monosyllabic VRs.The discussion is arranged os follows; (a),tone ppreading (P 01) and added vowel - 1?) (r 33); (b) in] (P 70) and le] [i] (1) 71).

(a) The vowel that is added by the vowel-addition rule dtwaased in Section 4.73 is in many instances later lost tgh application of a rule, P 33, that replaces the added vowel by In in certain specifiable contexts. ,In such caseshowever, there is often, in addition to the L?), another trace left of the effects of the vowel- addition rule: namely, a low tone on the non-vowel that precedes the glottal step. This low tone reflects the application of tone-spreading rule, P 011 which copies the tone of a vowel onto voiced non-vowels in the same syllable. (The tone-spreading rule has a condition on it to the effect that it reapplies immediately vhenever the conditions for its application aro met.)For example, yhen,the vowel-addition rule changes Idum] 'extinguish' to[dfima], this latter form is automatically resyllabified as[6=ma] (by there- syllabification rule) 12,32), and the tone-spreading rule awlies to change[du=mt] to [dfiluit]2 (Actually, the form is lau=1113, sjpnce the high tone of the Lit] is spread to the preceding LC. But in transcriptions in this text we do not ordinarily show the gfects of ppe spreading in such culst) When the final ia] of[dfimma] is replaced by 0j, the [1] still retains its low tone, which accounts for the tone pattern found in, e.g., Ak-As 186 4.74

Dbedfill kanii) 'He will extinguish a lamp.' (In the example, there is no(?) after 14fidiaj because of the effectsof P 04, which deletes anyP] not followed by a pause, and which reapplies whenever theconditions for it aremet.)

The rule replacing an addedvowel by[?] is most re- steicted in its operation in Asanteless restricted in Akuapem, and still less restrictedtn Pante. In Asante the rule applies obligatorilyto aU] that has been added after Lm], and, in spme cases,either optionally or obligatorily1 to an LIthat has been added after(13]. (Those VRs with final 13] to which the rule applies are marked with a special ruleOature.) The rule operates to replace the added vowel byL?) except in forms of the so- called "intransitive" past(or "intransitive° negative per- feet): i.e., past (ornegative-perfect) forms that occur stt the end of a sentence orclause (cf. Section 4.33). Thus the rule operates tochange[Obeetia] 'he will extinguish (it)' to[Obedilin and [5bfitir 'he will sell (it)' to L5bit6137], but leaves[Wel 'he extinguished (it)' and 15t6iti] 'he sold(it)' intact. (The final tone of [Obedia?] and15b6t6V) is changed from low to high immediately before a pause, by meansof P 68, and the final [V] of L5bit6h?] is replaced by[0] in some Asante subdialects by means of P 71: cf. subsection (b)1 below. Asante[Wag]-+[Milan (ultimately, Damti]) through operation of a vowel-assimilationrule, P 37 (cf. Section 4.33, subsection (b)). In some Akuapem subdialectsforms like[Milanare submittedto the labializationfull% P 09, before undergoing vowelassimilation: cf. Ak[Odamwii] 'he extinguished (it)', and discussionbelow. Asante15t6zit] 5t6DYi] (ultimately,[5t5nii]) through operation of the mn - ln] rule, P 70: cf. subsection (b),below.) The operation of the rule inAkuapem is similar to its operation in Asante, theonlyrdtfferences being that in Akuapem the rule appltesto a LUJ that has beenadded ofter Lw] (as well asImp and applies obligatorilyto an LIJ that has been addedafter my VR-finalLd. The first of these differencesresults from the fact that inAsante VR-final [w] is obligatorilyreplaced by glottal stop by means of P291 so that the properconditions for the vowel- addition and vowel-replacementrules are never met. Thus in Asante[hiw] 'trouble' -0 DIV], while in Akuapem[him] -0 [hiwi] [h&O] [WO]. The second difference accounts 4.74 187 r.s. for distinctions such asthat between AkLheigj and As [SW] 'swell'. In both dialectsCOQ] is changed to [gib] by the vowel-addition andtone-spreading rul.es. In Akuapem, therule replacing the addedvowel by,L7] applies obligatorilyto11160], changing it to bleig]. In Asante,application of this ruleto[Eogi] is optional. If,the rule isapplied, then the variantAsante form [hilV], identical with theAkuapem form, results. But if the rule is notappl/ed,TO] remains intact until it is changed bz theLW - [n rulecf.subsection (b), below--to(gab As in Asante, therule that replaces anadded vowel by [fl fails to applyin Akuapem when theVR is part of an intransitive-past orintransitive-negative-perfect verb. Thus the rule does notapply to Ak(Odfimat] 'he extinguished (it)' or AkDthtt] 'he sold (it)'. In the case Akuapem is of 'he,extinguished(it)', the surface form in [amwII], while, as noted above, thecorresponding Asante form isDdamill.The labialization of the[m] in the Akuapem form and theabsence of labializationin the Asante form reflect the factthat in Akuapem(at least in some subdialects), but not in Asante(or Fante), intransitive-past forms that have undergonevowel addition aresubmitted, when the added vowel is[UJ, to the labialization rule,P 09. By means of this rule,Ak[OdUmat] Ddilmwat].The vowel- assimil(ition rule, P37, the#,changes Ak[almwd] to 1dlimw11](ultimatelyjoamwil]).Since labialization does not apply to As[?dilmat], the vowel-assimilation rule operafts to changethis form to As[Odfimti] (ultimately, [Mazda)). (For the tonal differencesbetween the past forms of the twodialects, cf. Section4.91, subsection (d) In Pante, the rulethat replaces an addedvowel by [P] applies regardless of whether theadded vowel follows (r], Liu], [n], Cw], or(d. The rule fails to apply,how- another ever, if theadded vowel isimmediately followed by vowel, whether thisfollowing vowel is inthe same word (in which case it representsthe intransitive-pastsuffix) is retained, or in thefollowing word. When the added vowel it always undergoesassimilation to thefollowing vowel. (As in Asante, labializationdoes not apply.) Some 188 4.74 examples of Pante purface forms to whichthe rule replacing an added vowel by1.4?] have applied (with the glottal stop having been deleted if no pausefollows) are: Main 'He thatches (it).' lOka dtin] 'He thatched a house.'

16dilia?] 'He extinguishes (it).' [Oda landziit] 'He extinguished a lamp.' 'He weaves (it).' 1.6hitw kfiJ 'He bothers Kofi.' Dtg nim] 'He sold meat.'

(In the last example, pe fipal(23] of [t6i3] has been changed to In] by theLQ] Ln] rule.) Examples of surface forms in which the addedvowel is retained, but is assimilated to the following vowel, are: Detrii] 'He thatched (it).' Mar& glean] 1He thatches houses.' [OdAmit] 'He extinguished (it).' [(Sarni idandzilt] 'He extinguishes lamps.' 'He wove (it).'

1.6himit &MIA] 'He bothers Amba.' Loton edfir] 'He sells medicine.'

(b) As has already been noted,there is au Asante-Fante rule (noV restrictedto monosyllabic VRs) that replacesN] by ln] in certain contexts. In Asante this rule, P709 applies onlywhen thQ,yowel is intervocalic within a word. Thus As 1.5tkinj 'he sold it' - 1.5t5nii], but the 1.13] of As [5b6t6V] 'he will sell (it)' remains intact. In FaDte, the replacement ofL4] by Ln] occurs both when theLQ] is intervocalic within a word and when it is word-final(or is followed by IJP. Thus not only does Fa1.5t64ii] 'He sold it' - [ot6niT] but Fa [5b6t5e] 'He will sellit' - 1.51:6tn?]. Asante also has a rule (once more notrestricted to monosyllabic VRs) that replaces aword-final sequence consisting of a velar nasal plus aglottal stopi.e., 189 4.74, 4.75

two by a nasalizedclose vowel.The rule operates in ptages, firstdeleting the final[?], then replacing the rule are not 14] by a vowel. (The two stages of the necessarily interdependent.That is, the[?]may be, deleted and theLQ] left intact, and, similarly, anLg] that has never beenfollowed by1?]--e.g., a final C4J of a noun--may bereplaced by avowel.) The vowel that results from operationof this rule agreesin backneps LI] and tenseness withthe preceding vowel,and is thus after a17,Back,+Tense] vowel, after aL-B-gck,-Tensej vowel, 111.1 after aL+Back,+Tensei vowel, andLe] after a l+Back -Tense] vowel. The rule, P operates, for U ti I fl 1 example, to chgean L3t137] 'grow' tojii, Ij Lkaej'count' to[kki] (ultimatelyLat.], after application of the nasal- vowel-agreement rule,P73), and1t613?] 'sell' toLt0] this rule (ultimately,Lti6]). ;11 some subdial9cts, applies only whenthe Lig] follows aL-Back] voweli. (These subdialects have thechange of[31:0] to [ail], and [kg] toLka], but not the change of1t5b?] to Lt501.) (The rule replacingCO] by a nasalized vow4 does not apply in Fante,and neither thisrule nor the1.4]- mid rule applies inAkuapem.)

4.75Structure of DisyllabicVRs. In addition to thesequence-structure constraints on disyllabic morphemespresented in Chapter21 the underlying forms ofdisyllabic VRs show certainfurther con- (a) canonical forms;(b) the vowel qualities straints on: medial that may occur inthe successivesyllablef (e) the non-vowels that may occur;(d) the tone sequencesthat may occur on thevowels.These constraints arediscussed in turn below.

Section 2.21, there is a ph (a) As was noted in positive condition(PC) to the effect thatthe canonical form of Akan rootsyllables is one of thefollowing: /CV/9 /CVV/, /CVC/1 or/CVVC/. If, however? thesyllable is the secondsyllable of a disyllabicVR, it is further constrained by thefollowing condition: [-Vocalic] [Vocalic]([-Vocalic])] PC Vl. VR[IC= 4.75 190

That is, the second syllable of adisyllabic VR is always of the shape /CV/ or /CVC/. (Christaller's dictionary lists at least one apparentexception to this generalization: hodwo(4) 'become slack' whic9h in our analysis, would have the underlyingform9/11%oguept.)1 1

(b) The vowels that occur in the successive syllables of disyllabic VRsconform to the tenseness- harmony constraint stated inSqS0 5, Section 2.p.. (There are one or twoexceptions, such as n(y)insen/d/dsig/ 'become pregnant'.) In addition they conform toat least two other constraints, which maybe stated as morphemestructure conditions that applyspecifically to disyllabic VRs: viz.9

-High v.-High PC V2: aBack ' *Back Low VR ctLow ocalil I-TC Vl: VR[X +Vocalic] X= X +Back Xi +High

[+Back +High

PC V2 says that if neither ofthe vowels of the successive syllables of a disyllabicVR is close (i.e., if both areL-Highp, then the vowels of both syllables must be j,dentical with respect totongue-frontnesp (they must be laBack]) andtongue-height (they must belaLow]). The condition thus permits bothvowels to be /E/, /A/9 or/0/, but does not permit anynon-identical succession of these vowels. Examples of VRs permitted by thecondition are: mg.!). /p6s4/ 'worry' pete(w) /p6t4/ 'bruise' kasa /ktIsA/ 'speak' kata /kIttg./ 'cover' foto(w) /f5t6p/ 'knead' Doto(w) /p5t5p/ 'corrupt'

The condition excludesdisyllabic VRs of the shapes /CECA/, /CECO/, /CACE/, /CACO/,/COCE/, /COCA/. 191 4.75

syllable I-TC V1 saysthat i; the vowelQf the second L+Back,+Highj), then thevowel of of a VR is/U/ (i.e., is must also be/U/. Thus the condition the first syllable butu(w) /bUtap/'upset', permits boto(w)/b6t44/ 'decrease', not permitdisyllabic VRs of theshapes, etc., but does (There is at /CICU/, /CECU/,/CACU;, or /cop/. e.g., which must belisted least one VR kamfo/klibeik/ "praisro'. 9 exceptiOn-frthe condition.)Note that the converse as an That is, if thevowel of this conditiondoes not obtain. syllable is /U/,the vowel ofthe second syl- of the first VRs such as lable need notalso be /U/. Thus there are /brake/ 'open', dua/dig/ tumiAfibt/ 'be able', bue 'plant', suro/sad/ 'be afraid', etc. (It is possible thatboth of theconditions stated disyllabic above as specificto disyllabicVRs apply to We have notbeen able, however,to morphemes in general. condi- investigate a numberof apparentexceptions to the conditions tions that wouldhave to beexplained were the to be generalizedin this way.) (c) In Chapter 2,only one generalmorpheme- structure condition waspresented whichapplies exclusively non-vowels: viz., 8(1SC 39wLich speci- to morpheme-medial consonants which arenot fies the shapeof syllable-final (In combination withSq6C 2, whidh also morpheme-final. in general,SW 3 applies tosyllable-final consonants syllable-final consonantthat is not mori- specifies that a with pheme-final must be avoiced stop thatis homorganic non-vowel of thefollowing syllable.The voiced the initial in all cases stops that conformto this condition are changed by a P-rule,P 03, to thecorresponding nasal subsection is concernedwith fur- consonants.) The present morpheme-medialnon-ilrowels: specifi- ther constraints upon apply to morpheme- cally, those furtherconstraints that in disyllabicVRs, (Once more, it is medial nonsavowels stated here asspeci- possible that someof the constraints fic to VRs apply moregenerally.) Three such constraints areformalized anddiscussed below.

I-TO V2: VR[X= [+Consonantal] 4.75 192

This I-TO, in combinationwith SqSO 2(Section 2.21), specifies that theinitial segment oZthe second VR musV be a trueconsonant syllable pf a disyllabic Sq$0 2 speci- (i.e., a L-Vocalic,+Oonsonantaljsegment). itial seentsgm inroots are non- fies that all pyllable- permitsrthem to be vowels (i.e., L-Vocalicsegments), but either true consonantsInin which case thgy areL+Oonsgnan- tal]) or glides (in which casethey areL-Oonsonantalj). I-TO V2 thus furthurrestricts the classof syllable- initial segments that canoccur in rootsin those cases where tha root is aVR and the syllableis the second excludes VRs of the syllable.Specifically, the condition shape /0VwV/,OVW, OVW, etc., inwhich the initial segment of the secondsyllable is a glide. (Obristaller's dictionary includes atleast one apparentexception to this condition, kwaha 'tiearound'. Informants with Wham we havecheckamiR, however, unfamiliarwith this item.)

I-TO V3: ya[x glItticrallx] 4 [3:=4"1

This condttion statesthat the only labial(i.e., (-Ooronal,-Backj) consonant that can occurat the beginning o: the segondsyllgble of a disyllabicVR is the voiced (If, gs is uflually (i+Voicedi) stop(L-Oontinuant)) /b/. tha case, the vowelthat follows the/b/ is L+Nasalj, the /b/ is changed to Lmjby I' 06.) The condition Vhfspermits 'stick VRs such as lorima/lab*/ 'turn, twist', none /p6b4/ to', home/hbbt/ 'rest', or seber(e)/8igi7 'turn inside out1,70 excludes VRs that have/p/ or /f/ as the initial consonant of the secondsyllable. (There are at least two exceptions to thecondition, seps(w)/sfiplp/ 'become free-- which may be metathesizedfrom =WO,/pisip/ 'tug at'-- exceptional Ln and kamfo/kibf6k/ 'praise'--whicli is also its ;Fars: cf. subsection(b), above.) The MS conditionsthus far discussedin this subsection limit theinitial segment of thasecond syllable of a disyllabic VR to onaof the followinr: sixsystematic immediately phonemes: /b,t,d,s,k,g/. When this segment is 193 4..75 preceded by a consonant(i.e., when the first syllableof the VR is of the shape/C(V)VC/), there is a furtherlimi- tati( i to the followingeffect: [-Vocalic] = [ I-TC V4 : VRDE

+Coronal -Contimant -Voiced

That is, only thevoiceless alveolar stop/t/ may occur in the contextin question. Thus VRs such as or hinti(w)/hfdtip/ 'stumble°, kantan/kidtig/ 'sprawl', Denten/pidtig/ /jerk' are permitted, but VRsof the Shape /CVCbV(C)/, etc. are excluded. (There are a few VRs whose underlying structure to be /CVCsV(C)/ cr /CVCkV(C)/--e.g., danso 'be unusual', pinkvs'approach'. It is likely, however, hat these VRs arecompoundsi.e., that there are two morphemesin their underlyingstructure.) (d) Of the four possible tone sequeucesthat may occur in theunderlying forms ofdisyllabic morphemes, only two actually do occurwhen the morphemes are VRs,and furthermore, these two are incomplementary distribution, according to whether theinitial consonant of thesecond syllable of the VR isvoiced or voiceless. When this consonant is voiced, the VRhas an underlying low-hightone pattern, and when thisconsonant is voiceless,the VR has an underlyinghigh-low tone pattern.These ficts are sum- marized in the followingmorpheme-structure condition: r +Vocalic +Vocalic PC.V3: LF-Consonantal X =(aVoieFsl -Consonantalxi aTone VR [-aTone J The following are someexamples of VRs that conform to PC V3 where a = +: 194 4,75

home /hebi/ 'rest' 201 /Ail 'stick to' huru(w) /hisidistp/ 'jump' fra /fiat/ 'mix' me /OW 'agree' mina/gidir'stand' /1:46/ 'help' wie . /wigek/ 'finish' boa Ak-Fa ,...ann -7a V-21 /magi/'flee' /digi/'melt' s Amen(e) s nan(e), 1Lklic 1 The following are some examples of VRs that conform to PC V3 where a =

looto(w), /p6t5p/ 'corrupt' kata /kilti/ 'cover' poso(w) /p6s4/ 'tremble' bisa /bisV 'ask'

-As tie 1 /bike/ 'listen' kae /kiki/ 'remem- tsie ber'

i.k-Fa {Ak-Faman anI/diski/ 'turn' }/biki/'turn As man(e) lAS darttel aside'

It may be noted from some of the above examples that in the surface segmental form of the VR(whiah is reflected in the comventional spelling) the distinction between under4ing voiced and voiceless medial consonants if some4 times lost.Thus wie ([0.0]) and:twit(Ak-As [tio]/ Fa [tsie]) do not show any medial consonant at all, while As &rano(djOnf) andAs dane[dinf]show the same medial consonant. This loss of Mairlying segmental differemes results from the operation of certain P-rules discussed in Section 4:76. The underlying contrast between the voiced and voiceless medial consonants is, however,frequently reflected in the tonal behavior of the VRs.Compare the tone patterns of the following surface forms: Fa [NIP] 'He finishes.' Fa[hole] 'He listens.' As[obedini nO] 'He will flee to him.' AsDbfidini 'He will turn him.' 4.75, 4.76 195

There are at least fourVRs for which we postulate underlying voiced medialconsonants, but which fail to show the expectedunderlying low-high tone pattern.These VRs, bre /bid/ 'bring',tumi /tfibI/ 'be able,overcome' kaa w/kitetp/ 'wither's and pagya(w)/pitgitp/ 'raise', const tute exceptions toPC V3.-(It would be possible to postulate underlying formsfor bre and tumi withmedial /t/ and /p/ respectively, but inthat case a ppecial account would have to begiven of the medialLr] in the surface form of bre while in the caseof tumi not only would a,special account haveto be givenZrEhe change of /P/ to imj, but thepostulated underlying formwould be an exception to I-TC V3--cf.subsection (c), above.)

4.76 Tone Reversal in Disyllabic VRs.

There are various P-rules thatchange the underlying tones of disyllabicVRs. Most of these rules are syntactically conditioned: e.g., theydepend upon the occurrence of the VRwith some particular aspectmarker, the occurrence of the VRin a dependent clause, etc. r 4.es of these types arediscussed in Sections 4.91 ff. tehe rule to be discussedin the present section, onthe other hand, is not syntacticallyconditioned, but is, instead, context-independentin some of its applications, phonologically conditioned inothers. This is the Akua- pem-Asante rule which changes ahigh-low tone pattern on a disyllabicVR to Ak low-high, Aslow-falling.This rule distinguishes between twoclasses of VRs with underlying high-low tone patterns: those in which the underlying medial consonant of theVR is /k/, and those in whichthis consonant is /t/ or/s/. As is noted in Section4.77, subpection (a), a medial/k/ in a disyllabic VR is changed to Lgj by P 05 ana isthen either nasalized by P 06 or deleted by 08, while a medial/t/ or /s/ remains intact. The tone-changing ruleunder discussion here followsthe rules which are responsiblefor the realization of aVR- medial /k/ as[13] or [0], and distinguishes, among VRs with an underlying 4gh-lowtone pattern, between, thope j.n. which there is aL+Voiced] medial consonant (i2e., `- US] /k/) or no medial consonpt(i.e2, 10) 4- Lg.! 4- /k/) and those in which thereis a L-Voicedj medialconsonant (i.e.s /t/ or /s/). In those cases where theunderlying form has medial /k/, therule is completely 196 4.76 context-free. In the case of VRs withmedial /t/ or /s/, the rule fails to apply if theVR is immediately preceded by a high tone (e.g., by afuture-aspect prefix or an ingressive prefixcf. Sections 4.34and 4.50 respectively), but applies in all other cases. Some examples of the operation of tba rule are:

Before P 20 After P 20

'he turns' Ak[di4f]/As Ddicf] 'he doesn't turn' Ak[?idi:ab]/AsDitaien (ridlgi) 'he is turning' Ak[rida.73f]/As(ria4cf) [Bach] the turned* AkDdiyAs (Bbigvi] the vill turn' Ak(bidivr)/As [Ibidie] [516dhi] 'be gots and turns Ak[116divi]/As (616divi] [Masi] 'he speaks' Ak [5kisi]/As[Masi] Niciskt] 'he doesn't speak' Ak(Bilkisifl/As (513kisin Driktiaij the is speaking' AkDrikis&VAs Drikasa] Natsii] 'he spoke* Ak Dkisia]/AsDkisai]

(The rulp,f4ils,Vo apply to, e.g.,(lbfikftsi) 'be will speak', Lok5k&saj 'be goes andspeaks'. In the case of the latter, note that the rule mustprecede the Ak-As rule, P47, which changes high tone on an INprefix to low, resulttng in, e.g.,(5k5kisi], the occurring Ak-As surface form ia non-pre-pause position.) (For subsequent rules that affect forms likeDdigf], resulting in surface forms like Ak[Malin/As Della] [5d1f], of. Section 4.77, subsections (6) and (0.) The effect of P 20 is to make thesurface tone patterns of Vic with underlyinghigh-low tone patterns identical with those of rRs withundeaying low-high tone patterns. Thus the surface torks patternof, e.g., Ak-As [bisi] 'be asks' (cf. the underlying VR/1514/ 'ask') is the same as that of Ak-As[5t9irl] :Ito shows' (cf. the underlying VR /kid6/ 'show°. (In Asanto ? 20 changes RobisidtoDbisa], which is changed to[bisi] by P 87.) Apart, however, from the surface tonepatterns of certain 1 4.76, 4.77 197

forms in which the VR is precededby a high-tone prefix, there ars--in Asante, but notia Akuapem--sometimes other surface indications Of the underlyingtonal difference between such VRs.For example, Asante has an object- pronoun tone-raising rule,1'17, which operates to change the tone of an object pronounfrom low to high when the preceding verb ends in a high tone.It operates, for example, to change the tone of an object pronounafter a VR like /kidk/ to high, but does noV,affect the tone of an object pronounafter a VR like bisi/. Thus we find in Asante surface tonal contrasts such as:

As [5t9irtdi] 'He shows him.' As [Obis&nil] 'He asks him.'

in which the tones of the object pronounsreflect the underlying tonal difference between theVRs, even though the VRs themselves no longershow this difference. (Akua- pem does not have theobjt-pronoun tone-raising rule in qu3stion, and there is no surfacetonal contrast between Ak [5t9irtni]and Ak(Obis& no]. For further discussion of Asante object-pronoun tone-raising,cf. Section 4.95.) P 20 does not apply in Pante,and therefore Pante surface forms usually show atonal-contrast between VRs with underlying high-low tone patterns andthcose with underlying low-high tone patterns: e.g., Fa1.6kelsli] 'he oeaks' vs. Dtgirt] 'he shows',[513ktis&?] 'he doesn't speak' vs. Dirt9irtfl 'be doesn'tshow', etc.

4.77 Other Rules Affecting Dis llabic VRs.

The following discussion ofcertain P-rules affecting the segmental forms of disyllabicVRs is arranged as follows: (a) rules affecting medial consonants ofdisyllabic VRsf (b) rules affectingVRs with final non- yowels# (c) rules affecting disyllabicVRs of the Shape LCVr/m/n/a]. (a) Of the six systematic phonemes which regu- larly occur at the beginningof the second gyllable of a disyllabic VR, /b,t,d,s,k,g/(cf. Section 4.75, subsection (c)), only two, /t/ and /8/, are always leftintact by the 198 4.77

P -rules. The other four are altered byP-rulGs in some or all cases. The alterations that occur arsummarised in the following chart:

UndeT:ying Defore pefore Cont9rant L+Nasall Vowel I-Nasal] Vowel

/b/ [m] No change

/d/ [11] [r] CIO 0 /k/ [g] [13] [g]- 0

Some of the modifications summarised onthe chart are not specific tomedial consonants in disyllabic VRs, but, rather, reflect the operaVionof more general P-rules. Thps the change of /bld,g/ toLm,n,q] respectively before aL+Nasall vowel is independent of the grammatical function of the morpheme and occurs inmorpheae-initial as well as nonAtorpheme-initial yllables (cf.P06). Similarly, /d/ is always chapged toLr] intervocalically when the following vowl is L-Nasal] (cf. P15).Some examples of these changes in disyllabic VRs are:

Before P 06 After P 06

/pebi/ 'stick to' [Ai] /pYdt/ 'agree' [pint] /ditgf/ 'melt' [niqf]

Before P 13, After P 15

/fid.1/ ' call I (firi]

(For subsequent P-rules that affect formslike(pini]and [niqi],cf. subsection (c), below.) As the chart indicates, underlyingmedial /b/ i a VR remains intact when the followingvowel is L-Nasalj. Thus corresponding to the underlying form/sibit/, lurn insi.de pit!, we have the surface forms Ak-Assibxrij, Fa Lsxbxr?j. (It may be noted, however, that in the great majority of cases,the voyel following a medial/b/ in a disyllabic VR is L+Nasali.) 4.77 199

Underlying medial /g/ in a disyllabicVR (and possibly in m)rphemes of other types) isdeleted before a single 1,4lasal] vowel by means of P 08, Some examples of the operation of this rule are: Befqre 43 Deletion After tg) Deletion

/*711/ 'help' [bias.] /wieik/ 'finish' (wield

(Ekamples like Ak pagyaw[pid4;0] (underlying form: /p6g1p/) 'raise' show that the[g)-deletion rule must follow the [I]-insertion rule, P 07,144ich inserts an [I] between a [fLand a low tense vowel. LI] imertion, by changing begew] (derived from/pe6p/ by P 03) to [phiaw], protects the medial [g] from deletion by P 08, since [gj is not deleted before a[VV) sequence. Underlyiag medial /k/ in a disyllabic VR isfirst changed to Lg] by means of P 05. Thereafter it is treated like an underlying /g/ Qccurring in thisposj.tion: that is, it is changed to LW when followed by aL+Napal] vowel (bSr P 06), and it is deleted when followed by aL-Nasal] vowel (by P 08). Some examples are:

Before After After er Nasalizational-Deletion

/ad/ 'turn' 1414] [dki] 'turn [big!) [mia] aside' /kald/ 'remember' Lkagz [ka) [bike/ 'listen' Itige] [tie] (In Akuapem and Asante, the forms in the last twocolumns would, in most contexts, haTe a low-high,rather than a high-low, tone pattern, as the result of theoperation of the tone-reversal rule-discussed in Section4.76. For further rules affecting forms like[d11131] and Cmil)i], cf. subsection (c) below.) (There are several exceptions to the postulated /k/ Lg] rule: e.g.;buka(w) 'bend', se /sflap/ 'waste'. It is possible, however, that further analysiswill show that these are compounds.) 200 4.77

(b) Final non-vowels in disyllabic VRs trigger some of the same P-rules asfinal non-vowels in monosyllabic VRs (cf. Sections 4.72 through4.74). In particular, the following rules may apply todisyllabic, as to monosyllabic, VRs: (1) P 29, which, in Asante and somer Pante dialects, replaces a YR-final[w] by [7]; (2) P 31, which adds(i)or[U] after a VR-final non-vowel/ (3) P01, which spreads the low tone of the added[i]or[U] to the preceding non-vowel;(4) P 33, which replages the added vowel by L?] in certain cases:viz, after Lm] or LA] inrall dialects (except that in Asante the replacement afterLQJ is optional), after Lw] iAkuapem and those Pante dialacts inrwhich a VR-finalLw] has been retained, and after Ir] orLni in Pante only; (5) P 70, which changes ;inal or intervocalic(1 to (u] in Pante, and interyocglic 1.13] to Ln] in Asante; (6 P 71, which changes finalLeJ to a nasalized close vowel in isante.Some examples of the operation of these rules on disyllabic VRs are shown onthe following page (p. 201?. (The two Asante forms that develop ;ma the input form(ern4) result from the fact that the 1.11 P] rule ig op4onal in Asante when the consonant preceding the (1/1 is 14). If the option is not exercised, then the [Q] - [4 rule operates upon(eritig] to yield B(rai).If the option of applying,the[V]-(?) rule is exercised, the form chat results,(Wirtoil?), undergoes the (0] [I] rule, which results in the fora [rif]; Various subsequent rules may apply to the formslisted-- e.c., the pre-pausetone-raising,rIlle, P68, which accounts for the final high tone on Ak/FaiLburn4?), etc. in pre- pause position, or thenasalization-spreading rule, P73, which changes As[tifra]to[Orli].) (c) Disyllabic VRs whose secondrsyllable,rin the gourse of,the P-rules, assumes theshape Irk, LmJ,Ln] or IA) plus LI), undergo certain subsequent dialect-specific rules which account for surface dialect differences such as: Ak-As [firfl/Ta[fifl 'respect' (underlying fora: /fidi/) Ak-As (hifig]/Fa [OP] 'breathe' (underlying form: /11011/); Ak-As fprni]/Fa [pin?) 'agree' (undeaying form: /pte/); Ak (n165?]/As [nint] (nit]/Fa (nIA/J 'melt' (underlying form:/digf/).These dialect differences result (0) Input Fora 'demolith;iburawi °turn [sibfr] inside ;yawn; (hiris] (1) Cw) - [7] Fa2As rwilra7] out; scratch'Dario] (2) Vowel Addition "Fe [bizrfiwa]r % 1 (sibiri)r Chitriimi] Ak , Lsibiii] (3)(4) Tone[v] Spreading PaFaAk 1 Lburuwuj[baran Fa [sibii?) CharSiti][hitritul?] (5)(6) [g][13^] - Ei] - [211 PaAsAs [(thrill][Orfila]Diria] 4.77 202 from the applicationof the samedialect-specifjpc rules that opply tc otherVR-final sequences ofLr], Lig], [I] has or 1.4j plusLI] (i.e., sequencesin which the been introduced bythe vowel-additionrule--cf. subsection (b), above, andSection. 4.73). These rules are: [q] -0 [n] rule (P 70), and the [V] -0[71 rule (P 33), the the 0] -0Di rule (P 71).

;n Pante, the V -0[?] rule Oltgatorily replaces a final LI] in apolysyllabic VR byL?J after any one of In tit, other two the non-vowels[r], [m], [n], or[4]. dialects, the replicementoccurs onlyafter Lx1), obliga- (Note-that, while torily 1.4 Akuapem,optionally in Asante2 leaves intact afinal LI] that tollows an V -0 L?] rule by L?j a final IAin Akuapem andAsante it does replace [m]inthese dialects. In those cases [11 that follows an the where the vowelfollowing[m] has been introducgd by vowel-addition rule, theadded vowel isalways LU], and Thuso in Akua, hence is replaced14 Lfl in alldialects. pem andAsante, as in Pante,[pis] 'sew' - [p&MO] -0 Similarly, in the caseof a disyllabic VRwith (p&i:?]. L?] in all a final[me] sequence, the La] isreplaced by (underlying form: dialects. Thus[p] 'drive away' Lpl not only in Pante, butin Akuapam and /pibk)-0 the 4santerag well. This difference inthe oprmation of Pi rule accountsfor such dialectdifferences as LVJ Nemi] and that between Ak-As[fIff] and. Pa [fif.?], Ak-As Fa[ap], and Ak-As[gni] and Fa (piA?]. It is in thetreatment of[e] sequences (in the second syllable ofdisyllabic VRs) thatthe three dl.alects the LV] show the greatestdiversity. In Akuapem, once [?] rule has applied to, e.g.,[ilia] to yield [nai?], no the option ofapplying further rulesapply. In Asante, if [rAi] is not exercised, the [V] [?] rule to a form like yielding[Wa]t the [g] [n] rule obligatorily applies, while if the optionof applying the[V] - (?] rule is exercised, then theresultant form[ni67], is obligatorily [V] rule. %In Pante, the changed to[naf] by the [e] -0 [n157], yielding (13] - In] rule obligatorilyapplies to 203 4.78, 4.79

4.78 Trisylletic VRs.

The list of Akantrisyllabic VRs is a veryshort one. It consists of loam'do on purpose', bransam 'embrace', brantam 'end',Datiri(w)/watiri(w) 'slide', Ak sakra/As sakyera(cf. Fa satyer) 'change',Ak samana (cf. As-Fa semen) 'summon',and perhaps a fewother derived items. It seems likely thatall of these VRs are formations or borrowings(e,g., samana is borrowedfrom Until further informationabout the internal English). how- structure or provenienceof those forms isobtained, ever, it seemsbest to simply listthem in the lexicon as exceptional. Such a treatment of thetrisyllabic VRs permits one to state apositive morpheme-structurecondition to the followingeffect:

V4: [X= )] PC VR

(where X does not include asyllable-boundary (m), and where the parenthesosindicate an optionalelement). That is, all regular VRshave either one ortwo syllables.

4.79 Final Glottal StoDin VRs.

There are many VRs, bothmonosyllabic and disyllabic, whtch, in theirsurface forms, have afinstl glottal stop when they occurbefore a pause. In most cases, these final glottAl stops areintroduced by P-rulesin which the glottal stop replacesthe final segmentof the iaput fora, which may be either exvowel or a non-vowel. Thre are, however, certain occuregneesof final glottal stopwhich apparently cannot be explainedin this way.The discussion of VR-l'inal glottalstops below is arranfeils fol- L?J re- lows: (a) 1.7] replacing aVR-final vowel; (b) placing a VR-final non-vowel;(c) other VR-final fls. (The material presentedunler (a) ane(b) is for the most part a summary ofmaterial presented morefully in preceding sections.) (a) When bhe surfaceform cf a VR ends, in pre- pause position,in a non-vowel plus aglottal stop, the glottal stop always represents areplacement for a final yowel.The rule that effectsthis replacement is the VI] - 0] rule, P 33. The replaced vowel isalways one of the close (i.e.[+High]) vowels [I] and EU], and the non-vowel that pmedes it (at thepoint in the rules at which the [V] -0 1.'0 rule applies) isalways one of the following: Mt Ir],[m], fnj, or W. The [I] or[11] that is replaced by the glottal stop may or maynot have been present in the underlyingfora of the VR.If this vowel comes from the underlyingform, then the underlying form is a disyllabic VR whose secondsyllable is of the shape /CV/. If the vowel does not come fromthe underlying form, then the undeaying formis either a monosyllabic VR of the shape /CVO/ or adisyllabic VR with a second syllable ofthis shape, and the vowelthat is replaced by the glottal stop hasbean added by the vowel- addition rule, P 31.

The [V] -0[7] rule applies least widely in Asante, somewhat more widely in Akuapam,ami most widely in Pante. In Asante, the rule appliesobligatorily only to a VR whose final syllable (aV thefppropripte point in the rules) is of the shape LAU] (Lmaj orLm01), it also applies optionally to a VR whose final syllablehasae dhapeLa] (1.0] or La]). In Akaapam, the rule appliesobligatorily to VRs with final syllables ofthe shape(211], [41], and (w141). In Pante, the rule appliesobligatorily to any VR- final [CV] syllable in which[C]=[w], [r], [m], (n), or [12] and [V] = [I] or DJ], (except that[C] A [w] in those Pante subdialects in which thehi] -0 [7] rule is obliga- torycf. subsection (b)below). Some examples are:

Underlying Form Before [V] [?] After [V] -0[?]

/hip/ 'trouble' Ak/Fal [hUni] Ak/Fal [Wei] (cf.As/Fa2 [1410])

/jit/ 'overflow' WA] Fa (cf. Ak/As

/AD/ 'drink' [jib] [216ifl

/hiftli/ 'breathe' [gig] Fa [gi?] (cf. Ak/As[gif])

Aild/ 'forge' Fa (ti,f0] Fa [to:10] Ak/As [toile] (cf. Ak/As Ettlan)

/wedig/ 'scratch' [gerrulli) As - Derii3Y]) 4.79 205

(Various sUbsequentrules may apply tothe forms shown in the third column.For details, see thepreceding sections.) (b) When the surfaceform of a VR ends, in pre- the glot- pauseposition, in a vowelplus a glottal stop, tal stop usuallyrepresents a replacementfor a final non- below.) vowel. (For exceptions, cf.subsection (c), There are three P-ruleswhich ryplace VR-finalnonvowels the /k/ -0 PJ rule(P 03); the 1103" These are: Cw] -0 (?] rule (P 29); and the[r,m,n] -0 P] rule (P 30).

The /k/ -0[?] rule applles in all dialects. The rule rates to replace afinal /k/ in anunderlying form by Some examples of theoperation of the rule are:

Before /k/ [?] After /k/ -0[?] (pik] 'remove' [p40] [sik] 'stand' [A] [pOsik] 'rub' [p6si?] [fdtak] 'mix' [ffita?)

(Since the /k/ -0[?] rule applies in all cases,surface forms for which wepostulate a final/k/ do not actually provide evidence as tothe nature of theunderlying final under- consonant.Our reason forpostulating that this lying consonant is/k/ is that doing sopermits us to state a morphemestructure condition(SqS0 29 Section 2.21) to the effect that,amorpheme-final non-vowel may be any one of thestop(L-Oontinuaut]) consonants /p/, / p ,t,k,b,d,g/ andonly these. Since morpheme-final /t/, /b/, /d/, and/g/ are 00E04 ingur:aqe form; tn at least somedialects (as LwJ, LrJ,LaJ, LnJ, and 1.4.1 respectively), /k/ is the mostplausible source forpost- [pal, [si?], vocalic final(?] in surface forms such as etc.) Pante The [1111 -[7] rule applies in Asante and some Pa2).The rule operatesto replace a final pulgialects /p/) by bid (which isderived from anunderlying final [1. Some examples are: 236 4.79

p.efore [w] [7] Alter NJ - 1?

'weed' As/Fa2 'trouble' As/Fa2 'touch lightly' As/Fa2 'laugh° As/Fa2

(In those dialects--Akuapem andFal--inwhidh the Lw] .--16 [fl rule does not apply, forms with figs [I/ cbligatorily undergo the vowel-additionand LV -0b] rules, whidh result in surface formsin which w] is followed by N--cf. subsection(a), above.)

The operation of theErom.,nj -PP] rule has been explained in detail in Section4.72.In Akuapsm and Pante, the rule applies obligaVorily 'to a smallset of monosyllabic VRs with final Ir. or !Jib (The detailed workings of the rule differ in the twodialects.) In. Asante, the rule apparently may apply to anymonosyllabic YR with final [r]orIn], and to cite majority of monosyllabic VRs with final (m], but its application is in most casesoptional. Some examples of the operation of therule ares

Before Cr m n] -4.[?] Atter [r,m,n] -1?.1

(k6r] 'go' [kV] Ch6r) 'wash' As (120] [Rim] 'blow the nose' As bli?) [gni 'see' [On [pin] 'pinch' As[pb]

(c) Thorp are a good many VRs whosesurface forms nd invowel-plus-Lfl in Akuapsm, but simply end in a vowel in Asante and Fante, e.g., Ak[b6?]/As-Fa [IA] 'beat' Ak [tcl?]/As-Fafatal] 'divide' Ak [kin/As-Fa[ki] 'say' Ak [se?]iAs-Fa[so] 'be big'. 4,79, 4,80 207

The origin of the glottal stop in theseAkuapem forms is not clear.Were an underlying final /k/ tc bepostulated for them (cf. subsection (a), above), onewould expect that the surface forms would show a finalP] not only in Akuapam but in the other two dialects as well.

Our tentative assumption is thatAkuapem ip tn the process of generalizing the useof a pre-pausal Pj as a marker of the entire set of VRs. (In this connection, it may be noted that veryfew nouns in Akuapem--or in the other dialects--have afins1777.) One Akuapam informant with wham we have worked, for example, uses afinal glottal stop in all but about a dozen of a listof over two hundred monosyllabic VRs.However, since there clearly existVRs in whiph,Akuapem sppaktqs do notuseraotinal glottal stop-- 04.,Lba) 'come', Lcmej 'look at', Lwuj 'die'--it is not popsible to postulate an entirely generalrule that adds aL?) at the end of a VR with a final vowel. Instead, it seems to be necessary to.assign arule feature to those VRs that show the addedIn in Akuapem (or, alternatively, to assign a negative rule feature tothose that do not show the added Lfl), and to postulate anAkuapem rule (P82) that is dependent upon this feature.

4.80SUBordinative. The SUBordinative (which is identicalto what Stewart calls the 'subjunctive', and whichis associated with what Christaller calls the 'connected forms'of verbs) does not occur in deep structures.It is introduced into surface structures by means of transformationsthat apply to certain clauses that are syntacticallyand/or semantically sub- ordinate to some preceding sentence element: e.g., relative clauses and clauses followingthe emphatic marker na (which emphasizes what precedes it).These transformainns operate to introduce the formative SUBat each of two points in the subordinated clause: (1) as a constituent (in our formulation, the finalconstituent) of V; and (2) as the finalconstituent of the VP.For example, the na- emphasis transformation (together with thesubject-concord transformations discussed in Section4.20) would operate upon a deep structure such as: 208 .80

ASP VR

Koff MPH STA wo Enyiresi 'Kofi' 'be inl 'England' to yield the derived structure:

SC ASP VR

Koff na o-I STAI wo1 SUB Enyiresi SUB ('It's Kofi who's in England.°

The SUB that occurs as a consti uentof V never has any segmentalrealization. Its presence in a V does,however, trigger certaintone-changing P-rules, such that a subordinated V is in most cases tonallydifferent from the same V in an independentclause. For example, in an independent clause, the configuration: V

SC ASP VR

o- STA wo

is realized with low tone on boththe subject-concord prefix o- and the VR wooThus: 209 4.80, 4.81

Owo Enyiresi 'He is in England'

When SUB is present in theV, on the otherhand, the tones of the subject-concordprefix (in Akuapemand Fante) and the VR (in alldialects) are changed tohigh. Note the tones on owo in:

Ak4raKbfi ni 6w5 inyiAsi 'It's Kofi who's in As KU/ ni51611nYirtisil England' inyiresi 1 As the last example011t1ws, the SUB that occursas the final constituentof the Vi has(in some cases) different realizations in, onthe me hand, Akuapemand Pante, and, on the other,Asante. In Akuapem andPante, a VP- final SUB, like a V-finalSUB, lacks asegmental realisa- added tion.It is realized, instead, as afinal high tone to the last syllableof the VP: compare thetones on Enyiresi in the independentand subordinateclauses above. (For an account of how theadded high tone inthe subordi- nate clayseresults in afinal mid tone in thesurface In form ofEnyiresi, see the discussion inSection 4.81. Amite, a VP-final SUB isin some casesrealized, either obligatoray or optionally, as ahigh-tone vowel suffixed to the last syllableof the V: compareEnyirbsi in the independent clause withEnTiresi6 at the end of the sub- ordinate clause above.

The P-rules that changethe tones of Vs thatinclude SUB follow various otherP-rules that affectthe tones of Vs, and are discussedin detail togetherwith these other P-rules in Section 4.96.The P-rules thataccount for the surface realizations ox theVP-final SUB arediscussed in the immediatclyfollowing section.

4.81Rules Affectin VP-FinalSUB. The various surface formsof VP-final SUB areall derived, by means ofP-rules, from a suffixwhich is listed unspecified as to in the lexicon as ahigh-tone mid. vowel frontness, tenseness, ornasality.That is, the vowel is [-Highl-Low,+Tone] but is unspecified forthe features [tBackj, (tTense], and(illasal]. The suffix attachesto the last word ofthe VP.The P-rules that applyto the suffixed vowel (and, in somecases, to thesyllable 210 4.81 preceding the suffixed vowel) are: (a) SUB-deletion (1'39); (b) vowel harmony in SUB (P40);(c) tone- raising before SUB (I'41). Of these rules, only (a) operates in all dialects(although the details of operation show dialect differenees--seebelow); (b) and (c) operate in Asante only. (a) The SUB-deletion rule, 1'39, deletesall of the segmental features of theVR-final SUB suffix, and adds the high tone of the suffix at theend of the immediately preceding syllable. If this syllable had high tone before the addition of the hightone from SUB its surface tone is unaffected by the added tone. Compare the final tones in: Kafi w5 /mint& Mon is in Aunts.'

Ak-ll'a Kai na owoisinte. 'It's Kofi who's in Asante.° As KUL ni 65 isinte. 1 (Yor an account of the raising of verb tonesin the presence of SUBcf. Section4.96.The tones of noun prefixes, such as;he A- of Asante, are normally raised after high tones.) If, awever77Ersyllable precedingSUB had low tone before the addition of thehigh tone from SUB, this low tone becomes a rising tone,which, in surface forms, is realized either as a hightone (if the tone before the rising tone is low) or as a aidtone (if the tone before the rising tone is high). (For details of the replacement of a rising tone by high oraid tone, of. the discussion in Section 3.930 Compare the final tones in:

% Kofi wo Nkran. °Kofi is in Accra.°

KielnawInman 6 6 "-.-- 'Itos Kofi who's in Accra.' As Kolfl ni 66 India. {Alc-Fa and in: KOZI wl inyiresi. 'Kan is in England.'

. . , e "4".lag1 n1-1 51--it .W.Pal 'It ' s Kofi who's in England.' {As Kafia 1115Enyirlisi. 211 4,81

In Akuapem and nate,the SUB-deletionrule obligatorily applies in all cases,whPtever the fors of thesyllable preceding SUB. In Asante theSUB-deletion rule obligatorily applies inall oases in whichthe syllable before SUB does notend in'a close vowel(i.e., whenaver this syllable ends in/A/9 /g/, /0/1 or in'anon-vowel). ,When the syllable beforeSUB ends in a closevowel (/I/ or /U/), theSUB-deletion rule fails toapply in some cases, appliesobligatorily in others,and in still others applies optionally. There is a good dealof subdialectalvariation within Asante governing theapplicability ornon-applicability of the SUB-deletion rulein cases where thesyllable before SUB ends in a closevowel. There is alsosubdialectal variation, in those caseswhere the rule doesapply, with regard to whetherapplication is optional orobligatory. In the case of theAsante speaker whose usagein this area these: we have studiedmost closely, thefacts appear to be (a.1) SUB-deletion neverapplies when the preceding close vowel isin a verb, an objectpronoun, or one of the words bi'some', no 'the/that',zi 'this'. Thus, for this speaker,SUB-deletion does not occurin cases such as thefollowing: it.' As Kofina obetoo, 'It's Kofi who will throw it,') (cf.As Kofi Into. 'Kofi will throw

As Kofina obisaa see. 'It's Kofi who askedme.1 (a,As Kofi bisaa me, 'Kofi asked me.')

As Kaina odii bie, 'It's Kofi who atesome.' (of,As Kofi dii bi, 'Kofi ate some.°

'It's Kofi who saw As Koffna ohuu baaZil this woman.' the/that} (of, As Kofi huu baa11-2 'Kofi saw this iz1. woman.') 4.81 212

(a.2) SUB deletion isobligatory when the preceding close vowelis in the name of a person: e.g., 'It's Kofi who sawKwaku.' As Koff na ohuu Kwaku. (cf. As Koff huuKwaku. 'Kofi sawKwaku.9) rather than: *Kofi na ohuu Kwakuo.

(a.3) In most other cases,SUB deletion is apparently optional. Thus: pmase. 'It's Koff who's inKumase.1 As Koff na owo

'It's Kofi whoate something.' As Kofi na odii ibiribie. (cf. Kofi dii biribi. 'Kofi atesomething.°

above, surfaceforms often providetonal As was noted dif- evidence of a deletedVW-filial SUB. In some cases, a be apparent ferent type ofevidence of thedeleted SUB may of a clause-final in surface forms: viz., the absence glottal stop. Final glottal stops arenever presentin the surface forms forms of clausesthat include SUB0 In the surface do not includeSUB, final of clauses(or sentences) that they may glottal stopsgenerally have oneof two origins: particular morphemessuch as theNEGative be reflexes of replacives for (cf. Section4.40); or they may serve as VR-final segmentsof various types(cf. Section4.79). For example, in: Dad Ifi, amisin Kumase.1 As[lad MA kamisffl 'Kofi isn't from r ° Fa ikon mfikamisffl

the final glottalstop is a reflexof NEG (cf. Dafi fi kamisi] 'Kofi is fromSumase.9 4.81 213

(For an account of the dialectdifferences in the toneof the negative-stative verbmfi, cf. Section4.92, subsection (a).) On the otheraia, in:

[kOfi 616?] 'Kofi should go.' the final glottal stopis a replacive forthe final consonant present in theunderlying form of aVR. (The underlying form of theVR in this case is/kSt/ 'go'. This is changed to[k6r] by P 03, and then to(ko?] by P 300 Whatever the origin of aglottal stop, it may occur only in a positionimmediately before a pause.For example, the glottal stop atthe end of the surfaceform (15k5?j 'should go' in pre-pause positionis not found in:

[lad 1516 khmisf] 1Kofi should go to Kumase.'

The reason f9r this isthat there is a P-rule,r Q4$ that deletes anyL?) that is not followed by[9]or L.J (i.e., phrase or sentenceboundary).Now it ig preciselythis rule which accountsfor the absence ofL?J at the end of clauses that includeSUB. At the point in theP-rules at which the[?]-deletion rule applies, a VR-finalSUB has the form of a suffixedvowel, and any glottalstop that precedeo-it is thereforedeleted. The SUB-deletion rule itself comes at a laOrpoint in the rules: i.e., after the deletion of anyL?) that may have preceded Xt4 This ordering is apparent in theabsence of a finalPJ in, e.g.

[kOfi ni 6616] 'It's Kofi who shouldgo.' (b) As was noted in subsection(a), above, a VR-final SUB is alwaysdeleted in Akuapemand Pante, but is sometimes retainedin Asante when itimmediately follows a close vowel. When SUB is retained,it has the form of a suffXxed vowel: orianally, a vowel-withthe and with otherfeatures un- features L-Righ,-Low,ToneJ and specified. In surface forms, thebackness, tenseness, nasality features of theSUB suffix always agreewith those of the precedingvowel.This agreement is accom- plisthed by means of P40,which copies theappropriate features of the precedingvowel onto the vowelof SyB By means of thisrule, the SUB suffixis specified as Lei 4.81 2l4

[o] after after[i], [a] after [i],[o] after [u], and [4-NasalL P 40 first Ey]. When the precedingvowel is [5] specifies the suffixedvowel as[I], [I], [5], and respectively, but asecond after[I], [f], [0], and (C) changes the suffixedvowel from a aid part of the rule form the suf- to a close vowel, sothat in the surface 'fixed vowel isidentical with thepreceding vowel. Examples of the operationof the P40follows The 9WITB" in the firstColumn representsin eadh case a [-Highl-Low,+Tone] vowel unspecifiedfor other features. After P40, After P40, Before P40 Part l Part 2

As (bidi) + SUB (b4dii] 'will-eat' [bit9ii] A. (bitgi] + SUB [bitgii] 'will-squeeze'

AsDAff] + SUB [bif16] 'will-vomit' (b6sif] As(b6f) + SUB (bisii] 'will-stick-to'

As [14134] + SUB (b6bU8] 'will-bend' (bihtli) As (behfi] + SUB [bh] 'will-see'

As (bitó) + SUB DiAt65] 'will-throw' (bAte6) As [1:44] + sus (Atoi) 'will-twist'

(e) In those caseswhere the Va.-finalSUB suffix Asante, the toneof the precedingclose is not deleted in of the vowel is in some casesaffected by the presence before the SUBsuffix is suffix. Specifically, a low tone to high by meansof 24 -a, Apparently in some oases changed where the retention thiu change occursonly in those cases obligatory (of,sUbsection (a,l),above): of the suffix is before SUB is a i.e., when thsconstituent immediately 4.81, 4.90 215 verb, an object pronoun, or one of a limitedrustier of other morphemes such as 121 'some'. In those cases, on the other hand, where the retention ofthe SUB suffix is optional (cf. subsection (a.3), above), a low tone preceding SUB is not changed to high.Thus there is a change of tone on the object pronoun in:

A.[kafi ni Obisikmif] 'It's Kofi who asked me.' (cf. As[kali bistat mi] 'Koff asked me1)

and on in: As(kOfi ni Odii bif) 'It's Kofi who ate some.' (cf. As[kd dil bi] 'Koff ate some.° But the low tone preceding the SUB suffixremains low in: . A. Daft nit 5w6epireniC 'It's Kofi who is in England.' (cf. As[lad wb iyarsi] 'Kofi is in England.°

As[koldi ni 50,;46] 'It's Koff who has gone to sea.' (cf. As Nadi k6 p0] 'Kofi has gone to sea.°

4.90 Tone-Changing P-Rulbs. The remaining sections of this chapter aredevoted to a discussion of certain P-rulesthat change the tone patterns of verbs (and, in some cases, otherVP constituents). Attention is focused primarily uponP-rules whose operation depends upon certain sequencesof constituents of the VP: e.g., some particularaspect prefix plus some particular type of Verb Root. (Many of the tone-changing P-rules that apply to individual constituents,rather than to constituent sequences, have beendiscussed in preceding sections.)The discussion is arranged as follows: tone changes in affirmative simple Vs(Section 4.91); tone changes in negative simple Vs(Section 4.92); tone changes in ingressive Vs (Section 4.93); tonechanges in reduplicated Vs (Section 4.94); tone changesin verb-plus-pronaun constructions (Section 4.95); tone changesin subordinative Vs (Section, 4.96). 4.91 216

4.91Tons Chan es in AffirmativeSimple Vs.

For the purposes of this andthe following section, a "simple" V is onethat does not include aningressive, reduplicative, or subordinativemorpheme.A simple affirmative V thus normally has theshape SC + ASP + VRi a simple negativeV the shape SC + ASP +NEG + VR. (In the case of the affirmative pastand the negative perfect, which are formed with an aspectualsuffix rather than a prefix (cf. Sections 4.33 and4.41), the simple Vs have the shapes SC + VR + ASP andSC + BEG + VR + ASP respectively.) The following discussion oftone-changing P- rules that apply to affirmativesimple Vs is arranged according to the particularaffix involved: (a) STAtive (b) HABitual; (c) PROgressive;(d) PASt; (e) PERfect#(f) FUTure; (g) OPTative; (h) Werative;(i) CONsecutive.

(a) STA. As is noted in Section4.31, the STAtive-aspect morpheme has nosegmental surface form. STA does, however, trigger certainP-rules that may change the tone of other elementsof the V. In the case of affirmative-stative forms,only one such rule applies. This is P43, which makes thetone(s) on the vowel(s) of a VR thatfollows STA low. The effect of P 43 isillus- trated by the surface tonepatterns of the following affirmative-stative forms: OH 'he is from' Okhr& 'he is holding'

.4tizi 'he is wearing' 5.fiti 'he deserves' (Before application of P43,,the VRsin these examples show the following tones: fi, fAti.) (There are two stative verbs, Ak-AsRe/Fa am 'be' and Ak-Fa ze/Asn'be good', that are tonallyirregular. Both these verbs require hightone on :he VR, and0/1112 requires high tone on apreceding SC as biell.)

(b) HAB.Like STA, the HABitual-aspectmorpheme has nosegmentirsurface form. In the lexicon, however, HAB is marked as carryinglow tone. (The low tone on HAB is postulated in orderto account for theapplication of P 16 and P 50 to formsthat include it--seebelow.) While HAB does not, ingeneral, trigger P-rules(that is, HAB need not, in general, bementioned in P-rules), there 4.91 217

include HAB are certainP -rules that applyto forms that other AST as well asto forms thatinclude a number of morphemes.These are: P19, P50, P16,and P20. P19 is the rule thatchanges a low tone on a monosyllabic VR to high(cf. discussion inSection 4.71). The rule appliesobligatorily in Akuapemand Asante, and optionally in certainPante subdialects. (In other Pante subdialects P19 does notapply.) Examples of the appli- zation of P 19 toaffirmative forms thatinclude HAB are: Wore P After P 19 % 5fi 'he takes' 5fi 516 'be fights' 61E6 Vs P 50 changes certainlow-high tone sequences on It operates that includemonosyllabic VRs tohigh-low. It applies when- in Akuapem andPante but not in Asante. low-tone aspectualprefix is immediatelyfollowed ever a ('I the VR is not, in by a monosyllabicYR with high tone turn, followed by asuffix). The high tone onthe VR in the input to P50 maybe inherent(as in the case of ba 'come', hw6 'lookat') or may, in Akuapem(and some Pante subdialects), result fromthe prior applicationof P19 application of P 50 toforms (see above). Examples of the that include HAB are: Before P 50 After P 50

51)6. 'he comes' Ak-Fa 6bi 5bw6 'he looks at' Ak-la 6hwi Ak 5f& 'he takes' Ak 6fi Ak 5k6 'be fights' Ak 6k6 at P 16 applies onlyin Asante. It adds a low tone the end of a high-tonemonosyllabic VR that isimmediately preceded by an aspectualprefix with low tone. Examples of its applicationto affirmativehabitual forms arts: Before P16 After P 16

51)1 'he comes' Asha hitt 'he looks at' As 51.twg 218 4.91

(The falling tones that result fromthe application of P 16 are later changed back tohigh tones by P 87. The occurrence of falling tones onthe VRs before the application of P 87 is, however,sometimes attested by the surface tone pattern of the wordthat follows the yeti).For example, if this item hasunderlying initial high tone, its surface initial tone is aidafter a verb to which P 16 has applied. Consider, for example, thefollowing Asante derivation:,514 Gain& 'he comes to Ghana' 5bg Gani (by P 16) - 3bi Ghini (by P 86) -5b& Ghini (by P 87)0 P 20 applies in Akuapem andAgents, but not in Pante, to disyllabic VRs whoseinherent tone pattern ishigh-low (cf. discussion in Section4.76).It operates to change this high-low pattern to low-highin Akuapem and low- falling in Asante, except when thesyllable immediately preceding the VR has high tone.Examples of its application to forms that include HAB are:

Before P 20 After P 20 AkObis& 'he asks' s A abisi As obisa

Ak&tiro 'he listens' btik As&tie

(The tone patterns of the forms in theright-hand column may be comparedwith the surfoco tone patternsof the corresponding Fante forms--obisa,otsio--to which P 20 has not applied.) (In Asalifi7fhe-Triil falling tone that results from applicationof P20 is later changed to high by P87. The fact thatP20 resulti in a low-falling tone sequence in Asante but alow-high tone sequence in Akuapem is, however, sometimesattested ty the surface tone pattern of the word thatfoliows the verb.Compare the surface tone patterns ofAkObisi 104/As Obi.s& 'HA asks a man.'. The Akuarm derivation iseiLl ikaa obis& aka& (by P 20) -6, Oasts. Aka& (by a P-rule that raises the tone of a noun prefixaster a high tone). The Asante derivation is Obisi ikai (by P20) -obis& ik;i1 (by P87). Th: P-rule that raises the toneof a noun 4.91 219 prefix after a high tone follows P 20 but precedes P87, and therefore does not apply in the cited Asante derivation.)

(c) PRO. All of the dialect-specific tonal rules that apply to affirmative forms that include HAB (cf. subsection (b), above) also apply to affirmatiTe forms that include the PROgressive-aspect prefix /di/, except that the Akuapem-Asante rule, P 19 (the rule that changes a low tone on a monosyllabic VR to high), does not apply in Asante to an affirmative verb.that includes PRO.Thus, for examplet the low-tone VR fa 'take' retains its low tone in As 5refa[55110] 'he is taking'. (For a discussion of P-rules that affect the segmental form of the PRO prefix2 cf. Section 4.32.) In Akuapem, on the other hand, kbef& kik& through application of P 19, and then P 50, the Ahuapem-Fante rule that reverses the tones on a low-tone aspectual prefix followed by a high-tone monosyllabic VR, changes orefil to the Ak surface form 5refil. Other examples of the operation of rules P 19 and P 50 on Vs that include PRO are:

Before P 19 After P 19 After P 50

5reki) 'he is fighting' Ak kbek6 Ak kbela 5r4b1 'he is coming' Ak-Fa 5rebit

Examples of the operation of P 16 (the Asante rule that adds a low tone at the end of a high-tone monosyllabic VR that is preceded by an aspectual prefix with low tone) and P 20 (the Akuapem-Asante rule that changes the tones of disyllabic VRs with an underlying high-low tone pattern) in affirmative-progressive verbs are: 5rebti 'he is coming' -I. As 5rebil (55ba] (by P 16) andkbektisa 'he is speaking' Ak kbekits4/As kbelasi[55kitsi] (by P 20).

(d) PAS. Affirmative formations involving the PASt-aspect suffix /1/ (cf. Section 4.33 for discussion of the suffix itself), undergo some of the same dialect- specific tonal rules as do HAB and PRO formations (cf. subsections (b) and (0, above). Thus 5fil + PAS 'he took' Ak-As 5fti + PAS (by P19),Nazi). + PAS 'he spoke' - Ak 501s& + PAS/As 5kasti + PAS (by P 20). (The effect of P 117747Bn affirmative-past forms in Asante is later nulli- fied by the operation of P 45--see below.) 4,91 220

In Asante and Pante,there are further rulesthat apply to affirmative-pastforms, In. Asante, P 45 changes the tone of monosyllabicVRs in affirmative-pastforms to low in all cases. Compare the tones of: Akuapem Asante % 5.05 AhErmil, 'He bought a book,' 3too nhoma 5t6e. 'He bought it,' at55e(71).

In Pante, P 46 placeshigh toneon anymonosyllabicVR that precedes the PAS suffix,and reverses the tonepattern of an, disyllabic VR thatprecedes the suffix, changing a low-high pattern to high-lowand vice versa. Thus 3di + PAS 'he slept' -4. Fa5dtt + PAS [5dtad, 4711, + PAS 'he showed' -4. Pa + PAS[1t9fri], and 5kasi + PAS 'he spoke' Bkisa + PAS [51disfai], (P46 also Lpplies to negative foriFIFFante thatinvolve the PAS suffix--cf,Sec- tion 4,92, subsection(c),) (e) PER, In affirmative formationsthat involve the PERfect-aspectprefix /i/ (cf. Section403), P 19, P 50, P 16, andP 20 apply as they doin various other formations alreadydiscussed.P 19 changes, eAre, [widifit?) 'he has taken' to Ak-As(wfiflf P 50 changes, e,g,,[wilabit] 'he has come' to Ak-Pa[wAtibi]i P.16 changes to As(wilibi]; and P 20 changes, e.g.,baikitsi] 'he has spoken' to Ak[witakitsti]/As In addition, there arecertain'tonal rules that are specifically triggered by the occurrenceof PER in an affirmative formation.The first of these, P48v places a high-tone onthe first (oronly) syllable that immediately follows PER, In Akuapem and Asante, thisrule operates vacuously in the case ofmonosyllabic VRs (which azealready high by virtue of theoperation of P 19), In Pantehow- ever, it isP48 that changes a low-tonemonosyllabic VR to high immediately afterPER. In all dialects, afterP 48 has applied, anymonosyllabic VR that immediatelyfollows PER has high-tone,either because the underlyingtone of the VR is high orbecause the underlying low tone of theVR has been replaced byhigh (by means of P 19 in the caseof Akuapam and Asante, orby 4.91 221 means of1)48 in the case ofPante). It will be recalled that Akuapem and Pantehave a rule, P50, which reverses the tones on a low-toneaspectual prefix followedby a high-tons monosyllabic VR.Since PER has low tone, and since all monosyllabic VRsthat follow it havehigh tone in both Akuapem andPante, P50 always appliesto such formations in these dialects.Thus the rule applies inboth dialects to affirmative PERforms who Vis have underlying low tone--note the surfacetone pattern of Ak-2a 'he has taken' (cf./fit/ 'take')--as well as to those whose VRs have underlyinghigh tone.

In addition to raisingthe tone of low-tone monosyllabic VRs in Pante,1,48 raises the tone of thefirst ayllable of all disyllabicVis in Akuapes and Asante(since in these dialects, afterthe application of P 20,all disyllabic VRs have initiallow tone), and ofdisyllabic Vis with underlyinginitial low tone inPante. kmaples of the operation of P48 on disyllabic Vis are:

Before P 48 Atter P 48 iAk fik 'he has spoken' AA [wiikisit] Asiwik&sa]i

(wilt9irg] 'hs has shown' (wiitOri]

No further tone rulesapply to forms like[-lit9Ir4] in P 51, Akuapem and Asantebut in Pante thereis a rale9 which changes a low tone onPER to high immediatelybefore a high tone,producing the Pante surfacetone pattern [willtqfrib This rule is alsoresponsible for the high tone on PER in a surfacefora such as Pa[wittkisi].

In Akuapea and Asantewhen the tons of thesyllable immediately preceding PER ishigh (in affirmativeformations), a high tone is added, by meansof P 57, at the beginning of PER itself. For example, through theoperation of P571AAk[wttikfts&VAs [welikitsi] 'you (sg.) havespoken' - Ak[waakftstO/As [wilk&si], and As[Alba] 'you (nib) have come' -0Cwfabil]. (By later rules Ak[wriakils&]/ As(wAlkisi] CwIdikisid and As (wabt] badibi], the results being the ocearringsurface tonepatterns.) 4.91 222

(f) FUT. There are only two tonerules that apply to yRsthat...Immediately follow theFUTure-aspect prefix /bc/ (cf. Section4.34), and neither of theserules is limited to formationsthat include FUT. One of the rules is the generalAsante-Akuapem rule, P19, which changes the tone oflow-tone monosyllabicVRs to high(and which is not, of course,dependent upon the occurrenceof between, FUT).This rule accountsfor the,tonal difference The e.g., Ak-As[Obefin and Fa Dbefifl 'he will leave'. second rulel P 79,changes certainhigh-low-high sequences to high-rising-high(ultimately realized ashig12?-drop- high--cf. Section3.93). Among the sequences towhich P 79 applies are a sequenceconsisting of the FUTprefix followed by a VR with alow-high tone pattern. By means of P 79,[bfit911.6] 'he will show' -Dbfitvrfi] (ultimately, [bfitOrfi], etc.

It may be noted thatthe Akuapem-Asanterule changing a high-lowtone pattern on aVR to low-high, P20, does not apply to VRswith medial /t/ or/s/ that immeaiately follows a high tone.Thus Akuapem and Asantesinw,t4e same tone pattern asdoes Fante forverbs like Lobefitej 'he will fan' andDballsit] 'he will speak'.

(g) OPT.Mopsyllabic VRs thatfollow the OPTative-aspect prefix/N/ (cf. Section4.35) have high surface tone in alldialects,regardlesp,of,whether their TAnderlying tone is high orlow: e.g.,LoOco?] 'he should so' cer. the underlyingform of the VR,/k6t/ 'go'), of the 1.6ifin 'he should leave'(cf. the underlying form follow OPT VR,/flt/ 'leave'). (INGressive prefixes that also show high tonein all dialectscf.Section 4.93.) Disyllabic VRs tliat followOPT have a high-hightone pattern in Akuapem andAsante, a high-lowtone pattern in Fante--once more, regardlessof the tonerpftVer4of the underlying form of theVR: e.g., Ak-AsLoKikasaj/Fa (66kilsa] 'he should speak'(cf. the underlying VR,/kitsi/ 'speak% Ak-As(tiviirfi]/Fa [56.4frE] 'he shouldshow' (cf. the underlying VR,/kid6/ 'show'). While it would be possible to give somecther account ofthe surface tones that occur on VRs informs that involveOPT, the simplest analysis seems to be toposit a singleP-rule, P 49, which directly produces theoccurring surface tones. 223 4.91

the (h) IMP.As is explainedin Section 4.35, tors) IMPerative-aspect prefix(which has a zero segmental the environment is a transformationalreplacement of OPT in (The second- of an underlyingsecond-person-singular SC. person-singular SC isitself deleted by meansof the same transformation that replacesOPT with IMP.) Like OPT, 1KP is replaced by HABin negative forms, sothat verbs with negative-imperative meanings are(except for the fact that they lack an SC)formally identicalwith negative-habitual verbs. Imperative forms show nodialect variation intone. Following IMP, allmonosyllabic VRs havehigh tone before VRs show a pause,low tone elsewhere,while all disyllabic tone pattern bothbefore a pause andelsewhere. a low-low IMP These tonal characteristicsof verbs that include result from theoperation of P 61 andP 66. (P 61 assigns low tone to 1KPformations.P 66 changes thelow tone of monosyllabic VR that fol- a sentence-finalor clause-final lows IKP to high. Some examples of therurface tone patterns of IKP formt! are:

KS. K; Ktmise. 'Go to Kumase.' 'Sleep.' Di hi. 'Sleep here.' Iasi (Kati). 'Ask (Kofi).' Kyerl 'Show (Kofi).'

(i) CON.The underlying formsof affirmative- consecutive verbs areidentical with thoseof affirmative- 4.36) perfect verbs, sincethe CON prefix/a/ (cf. Section The surface tone pat- is homophonous withthe PER prefix. consecutive and perfectverbs, how- terns of affirmative certain tone- ever, arefrequently distinct,since there are changing P-rules thatapply to theformations that include CON but not to thosethat include PER,and certain other tone-changing P-rules ofwhich the reverseis true.

The presence of CONin affirmative verbswith monosyllabic VRs triggerstwo tone-changingP-rules, P 43 and P 43 P 66, that do notapply to verbs thatinclude PER. places low tone on anymonosyllabic VR thatimmediately in those follows CON.P 66 reversesthe effect of 1'43 224 4.91 cases where the verbis sentence-final or clause-final, changing a low tone on a monosyllabicVR that follows CON to high tone in these contexts. (1'66 also applies to imperativescf. subsection(h),,above.) Thus we find low tone on the VR in, e.g.,Una) waibi hi] '(and) he will come here', whichreflects the operation of 1'43, but to which P 66 has not applied (since the YR ba/b4/ 'come' is not in sentence- or clause-finalposition7: and high tone on the VR in, e.g.,((21) Aloft] qand) he will come' which reflects the operation of both P43 and P 66. tIt may be noted that we postulatethe application of 1'43 in all cases and the reversal of its effectby P 66 in sentence- or clause-final position, ratherthan adopting the apparently simpler alternative ofrestricting the application of 1'43 to cases in which thentis non-final.The reason for this decision isthat if a fora such as sentence- final waaba had always had a high tone on theVR, one would expect the Ak-Fa rule, P 50(which changes a low-high tone pattern on an aspoctual prefix followedby monosyllabic VR to high-low), to have applied:That is, one would expect [witibi) to have become Ak-Fa [Alibi], which is, in fact, the tone pattern found in the perfect inthese dialects. If, however, the consecutive form is[waaba] at the point in thb rules at which P50 applies, theconditions for the application of this rule are not met, andits failure to apply is explained.)

In Akuapem and Agent*, when the toneof the syllable immediately preceding CON is high(in affirmative formations), P 57 (which also applies to forms thatinclude PER) adds a high tone at the beginning of theaspectual prefix*. This results in a falling tone on theprefix, which, by later rules, becomes a hightone. Thus AkAs [(ni) witibi,h1] '(and) you will come bare'is changed by P 57 to [(ni)wiabi hi] (ultimately,Uni) wiftbi hij) and Ak-As [(11),wiabij qand) youwill come' is changed by P 57 to [(ni) wiabit] (ultimately,[(ni) wilbi]). Apart from P 57, the onlytone-changing rule that affirmative formations with CON andPER have in common is P 20, the Akuapem-Asante rule thatchanges a high-low tone pattern on a disyllabic VR to Aklow-highlAs low-falling. (This rule is responsible for differences such asthat between Ak[(ni) waikasi]/Ls Uni) waakasi) and FaUnt wiakisa] qand) he will speak'.)The other tone-changing 225 491, 4.92

(c), above, apply to P-rules discussedin subsection include PER but notto those thatinclude formations that low tone on the CON. For example,P48, which changes a after PER, does notapply first syllableof a VR to high tone on the firstsyllable of a VR to high to change a low Ak after CON: compare Ak[walkoisfij 'he has spoken' and will speak'. Similarly, P 51 [(ni) witak&sti] '(and) he the PER prefix which is responsiblefor the high tone on not apply to theCON in Fa[wilitOr6],'he has shown', does prefix in Fa((ni) witatOrti '(and) hewill show'.

4.92 Tone Changesin NegativeSimple Vs. in Section 4.41,only six different As is noted include the NEGa- aspectual morphemes occurin verbs that HABitual PRO- These are theSTAtive, tive prefix/11/. (The gressive, PASt,PERfect, andCONsecutive morphemes. negative-habitual formis used to expressnegative-opta- meanings as well asnegative- tive andnegative-imperative form is used habitual meanings. The negative-progressive well asnegative-progressive to expressnegative-future as in In negativeforms, the PASsuffix /I/ ocotrs meanings. while the PERprefix verbs withnegative-perfect meanings with negative-pastmeanings.) Of the /a/ occurs in verbs simple verbs, there six aspect-markedforms of negative involving the HAB,PER, and CON arethree--the forms specific to morphames--to which notone-changing P-rules That is, the onlytone- negative verbsnormally apply. that normallyapply to theseverbs are changing P-rules that also applicableAkuapem-Asante rules two rather widely (However, see affect variousnon-negative verbforms. account of Asantetone-lowering in Section 4.95 for an These two Akuapem- negative verbsfollowed bypronouns.) P 19, therule that replaceslow tone Asante rules are: low-tome prefix;and on amonosyllabic VR byhigh after a rule that replaces ahigh-low tone pattern on a P 20, the low-falling in Asante disyllabic VR bylow-high in Akuapem, before the VRdoes not have hightone. when the syllable rules to negative Examples of theapplication of these and PER morphemes are: simple verbsinvolving the HAB 4.92 226

After P 20 Before P 19, Lfter P 19 Ak-As Wit Esidoesnl ttake' {AkHaas& nkasa AsBikkAl RerCasnltspeak' Ak-As waif& take' iAk wlikisi Aswilkisa. didn'tspeak'

(Since the rules in question donot apply in Pante,Pante has the forms listed inthe leftnost column assurface forms.Forms involving the CON affix arenot listed separately above, but wouldbe homophonous with thelisted forms involving the PERaffix.)

In the case of negativesimple verbs involving the remaining three aspectualmorphemes--(a) STA, (b) PRO, (c) PAS--certain specific tone-changingP-rules do apply. These rules are discussedbelow.

(a) STA. A. is noted in Section4.91, subsection (a), thereIra rule, P43, that makes thetone(s) on the vowel(s) of a YR thatfollows STA low:This rule applies to negative as toaffirmativestative verbs. In the case of negative-stativeverbs, however, P44, which follows P43, changes thelow tone in the syllableimmediately after NEG to high. In. !Ante, 1'44 alsochanges the tone of NEG itself tohigh. The operation of the rule may be illustrated asfollows: Before P 44 After P 44

iisf3. 'he isn't from' Ak-Asiigl/Fa ,5fih7 6 ,3121271, he ion' t wearing ' Ak-As ,Uhyi/Pa ?Akar& 'he isn't holding' Ak-As CALkArkPaOfikfiri aidittia. 'he doesn't deserve' Ak-AsNetttiVila Whit 227 4.92

In Akuapem, nofurther rules apply tonegative- stative verbs. In Asante, a furtherrule, P 57, applies tn those caseswhere the syllableimmediately before NEG haz high tone--e.g.,where it is the SO(subject-concord prefix) wo- 'you(81;0'. In such emu:, ahigh tone is added at-The beginningof NEG, so that NEGacquires a falling tone. For example: Before P 57 After P 57

wtithy6 'you aren't wearing' As wtahyl, wafiti 'you dou't deserve' As w&ifitti

(By the operation of subsequentP-rulesthe sequence of a fallingtone plus a high toneis realized as asurface sequence consistingof a high tone plus amid tons.Thus ve findnegative-stative surfaceforms such as AsWeak,I Ak wahyll (P 57 also applies in the case ofaffirmative-perfect andaffirmative-consecutiv, verbs where PER andCON are preceded by ahigh-tone syllable--cf. Section 4.91,subsections (e) and(i).)

In Fante, a furtherrule, P 55, applies toall negative-stative verbs thathave an overt subject-concordpre- before a fix. This rule replaces a lowtone on SC by high high tone on. NEG,changing, e.g.,4hyfi, to SA.hyl. Some further examples are:

Before P 55 After P 55

II . aid 'he isn't from' Fa omits Ofikfiri 'he isn't holding' Fa Ofiketri

(b) PRO. Two of the tone-changingP-rules that apply toaTTEmative sinple verbs thatinclude PRO (cf. Section 4.91, subsection(0) also apply to negative simple verbs thatinclude this morpheme. These are the Akuapem-Asante rules P 19and P 20. (In the case of P 19, which changes a lowtone on a monosyllabicVR to high, the rule fails to applyto affirmative-progressiveverbs in Asante, but appliesto negative verbz inboth Asante and to Akuapem.) Examples of theapplication of these rules negative simple verbsthat include PRO are: 228 4.92

Before P 19 After P 19 After P 20 Ak-As oreafa t taking, he won't take' Ak 5ritikiet Breik&si speaking, As Brifikiei he won't speak'

In addition to these rules,all dialects have two further rules that apply tonegeive verbs that include PRO, but not to theiraffirmative counterparts.These are: P 53, which replaces thelow tone on PRO by highbefore NEG, and P 55, whidh r;places alow tone on SC by high before a high-tone PRO. (In Pante, P 55 is also responsible for high tones on Ses innegative-stative verbs--cf. subsection (a), above.) Examples of the operationof these rules are: Before P 53 After P 51 ,After P 55

iAk-As 40iiff: Ak-As 5r&ifi Ak-As Srilifit Pa 5reiti Fa BAIA Pa SAW.

Ak Ak orenkasa As 5rakisaAs 5rttliklisa As 5rtoilkisti Fa 5reAkiltat Pa Breliklisi Pa Sreiktisi Aggli ' he isn' t dividing9 he won't divide' AEgkig orenyere .5sigthal 'he tshowing9 he won't show'

(In Asante, step (1) of P 56obligatorily applies to delete the segmental featuresof PRO--cf. Section 4.32.Thus Sr&ifft Urfida] .4. AsMVP], Srakisa Driikisa?] 1717341san, etc. In Akuapem and Pante,application of 4.92 229

P 56 is optional, and thusthese dialects show freely varying surface forms such as Ak[5rlifl?] [WV] and Fa[Srlifit?)

(c) PAS. Negative formations involvingthe PAS suffix(i.e.77ormations with negative-perfect meanings) undergo most of the samedialect-specific tonal rules as do affirmativeformations involving the suffix. Specifically, in Akuapem and Asante thenegative formations undergo P 19 if they have alow-tone monosyllabic VR and P 20 if they have adisyllabic VR with a high-low tone pattern, while in Fante the negativeformations all undergo P 46. Thus P 19 changes + NEG + da + PAS'he slept' toAk-As+NEG++PAS,P20changes+NEG+kasa + PAS to Ak5 + NEG + ks + PAS/As 5 + NEG +kisa+ PAS, and P 46 changes5 + NEG + dit + PAS, 5 + NEG17irta+ PAS, and 5 + NEG +15yerE + PAS 'he showed' to, respectively, Fa 5 + NEG + da + PAS, Fa5 + NEG + as& + PAS, and Fa5 + NEG +kyeri + PAS, (The Asante rule, P45, that changes the tone of a monosyllabic VR tolow in affirmative formations involving PAS does not applyto negative formations.)

After the above rules have appliedto negative formations that involve the PAS suffix,certain further dialect- specific tonal rules may srply. Akuapem has no further tonal rules that apply when the VRis monosyllabic, but has two rules that apply when it isdisyllabic: P 57, which adds a high tone at the beginningof NEG when the preceding syllable has high tone, and P 58,which changes a low tone on the first syllable ofthe VR to high after alow-tone NEG. Examples of the effect of theserules are:

Before P 57 After P 57 After P 58

Ak wOlikasa walkastie 'you haven't spoken' Ak 512k1A4 512k&stte 'he hasn't spoken' (Note that, as a result of the applicationof P 57, wakasta does not meet the conditions required for the application of P 58, The falling tone on the NEGprefix inwakista is later replaced by a high tone by P 87, resulting in the surface tonepatternwakitsiti.) 4.92 2303

Asante has these sametwo P-rules applyingto negative-perfect Vs with disyllabicVRs, but the rulesapply in the opposite orderin this dialect*Thus Asante dhows derivations like thefollowing:

Before P 58 After P 58 After P 57

Aswakitsaie waktisaie waktisaii 'you haven't spoken' As 5alasiiiit 5fikiisalte 'he hasn t spoken'

(Through the operation oflater P-rules,wakislie is changed tow6fikitsitie and 5fikasaite is changedto in 5Akisate.) (For a discussion ofdialect differences the segmental formsof verbs involvingthe PAS suffix, cf. Section 4.33,subsections (a) and(b).) Unlike Akuapem, Asantehas a tone-changingrule that certain negativeformations that applies specifically to This is P62, include monosyllabicVRs and the PASsuffix. which applies to suchformations whenever ahigh tone pre- tone cedes NEG. The effect of the ruleis to change the of NEG to high andthat of the VR tolow. Compare the tones of: AkuaDem Asante watg AhOmi. 'You haven't Wit BB ihOmit. bought a book.' 'You haven't wEaSt5B4(yi). bought it.'

In. Ninte, inaddition to P46(see above), two further rules apply tonegative formationsthat include tone on the PAS suffix: P 54 and P 55.P 54 places high NEG and on thesyllable that immediatelyfollows NEG. (Since the syllable thatimmediately followsNEG is frequently already high,either inherently or as aresult of the prior operationof P44.6, this partof P 54 often applies vacuously.)P 55 (which is notlimited in its operation the PAS suffix) placeshigh tone on an to verbs involving EXamples of the SC prefix thatprecedes a high-toneNEG. operation of theserules are: 4.92, 4.93 231

Before P 4 After P 54, After P 55

Fa 5idAlt gal& 'he hasn't slept' Fa 5ikista 5fikftsgi 6fikiasta 'he hasn't spoken' Fa Blakyeri 5fikyeri 5fikyeri 'he hasn't shown'

4,93 Tone Changes in Ingressive Vs.

Verbs that include one of the twoAkan INGressive prefixes, /b6/ and/kV, may undergo, in addition to the vowel-harmony P-rules discussed inSection 4.50, certain special tone-changing P-rules. As was mentioned in Sec- tion 4.50, Akuapem and Asantehave a rule, P47, that changes the tone of ING prefixesfrom high to low.This rule accounts for tonaldifferences such as that between Ak-As :opyl and Fa6b6y6 'he comes to do'.The ordering of P 4 in relation to otherP-rules that affect tone is significant.For example, P 47 follows P 20,the Akuapem- Asante rule that changes a high-lowtone pattern on a disyllabic VR to low-high exceptimmediately after a high tone. For this reason P 20 fails toapply to forms that include INGt and the high-lowtone pattern on the VR is retained in forms such as Ak-AsObebisit na.'he comes to ask him' and Ak-AsOkObisitn 'he goes to ask him'. (Cf. Ak-As Obis& 116 'he asks him'. This form, which lacks the ING prefix, reflects the operationof P 20.) On the other hand, P 47 precedes certain P-ruleswhich assign high tone to syllables that followspecific aspectual prefixes: e.g., P 48 and P 49,which apply respectively toPERfect- affirmative and OPTative-affirmativeforms. Thus we find high tone on the ING prefixesof, e.g., Ak-Aswibbi 'he has come to do(it)' and Ak-As alski 'he should go todo (it)'. There are several othertone-changing P-rules that affect certain forms that includeING but do not affect otherwise identical forms thatlack ING. One of these is P 79a rule thatchanges certain high-low-high sequences to high-rising-high(ultimately realized as high-mid-high). 4.93 232

Among the sequences towhich P 79 applies is asequence consisting of a high-toneING plus a YE with alow-high tone pattern. Through the applicationof P 79, e.g., Fa 5bkesrfi 'he comesto show' -4,,5bEkyerk (ultimately, 4;kyere,--cf. Fa ,5kyerg 'heshows'). Since P 79 follows P 47, the Akuapem-Asanterule mentioned abovethat changes the tone of ING tolow, Akuapem andAsante ingressive forms that includelow-high VRs do notusually include an appropriate high-low sequenceat the point inthe rules in- at which P 79applies, and the ruleis thus usually applicable in thesedialects. ,Thus,theAkuapem-Asante counterpart of Fapbekyere is obe ere. In those cases mentioned above, however,where the e feet ofP 47 has been reversed by arule that assignshigh tone to ING in Akuapem after certain aspectualaffixes, P 79 may apply and Asante as yell asin Tante. Thus in a perfect- affirmative-ingressive formedwith a low-high VR,the Akuapem-Asante surface,fomrpflectsthe operation of lwaabetcire] 'he has come to P 79: e.g.; Lk-Aswabekyere show'. (The origial tone patternof this form is this is wabayeri [willbEtcirfi]. In Akuapem and Asante, first chanfed to[ulabitgirk] by P 47, then restored to [wailbEt9ire] by P 48. In Pante, P 47 doesnot apply but the Pante ruleplacintOip tone onpe,PER prefix, P 51, applies,changing[wasbetgIre] to [waabet9Fe]. The effect of P79 is to chanv Ak-As[willibit9Ire] to [wAabitcIrg] (ultimately[wiabitqlrE]) and Fa [witabgt9irE] to [watibfit9IrE](ultimatelyhaibitOri]).

In Akuapem andAsante, ;-..here is anothertone-changing P-rule, P67, which applies tocertain ING formations that includedisyllabic VRs. In this case, the rule applies to VRs thathave a high-low tonepattern. When a prefix plus such a VR occurs sequenceconsisting of an ING immediately before a pause,P 67 operates to add ahigh final low tone at the endof the VR: i.e., to change the tone. The rule applies, forexample, to tone to a rising Ak- change 9 sentence-finalor clause-fipaloccurrence of As ObibisZ 'he comesto,ask' toObebisI (which is ulti- maegiTiTilized asObebisi). P 4:7753Ties only toforms pause:MU-it does not apply whenthe verb that precede pattern of is followed. by anobject: cf. the surface tone Since P 67 is an Ak-As Obebisit no'he comes to ask him'. 493, 4.94 233

Akuapem-Asante rule, it has no effect uponFantefforms. For example, the Fante counterpartof Ak-As obebisi is Obebisa. It should be noted thatP 61 and P 66, the rules that account for the specialtone patterns of imperative verbs (of. Section 4.91, subsection(h)) never apply when ING is present as well as IMP. We thus find contrasts such as the following betweenthe surface tone patterns of imperatives that do and donot include ING:

IMP + ING IMP

{Ak-As bcda 'come sleep' citt 'sleep' Fa bfidi

iAFk-As bld& h& 'come sleep dix htx 'sleep here' a bgd& 116.1 here'

iAk-Asbebis&.} 'come ask' bisa 'ask' Fa bgbisia

Ak-Asbebisl.KOfi 'come ask bisa Kofi 'ask Kofi' {Fa bebiAKofi Kofi'

4.94Tone Changes in ReduplicatedVs.

This section discusses certaintone-changing P-rules that apply to forms that resultfrom application of the general rule of disyllabicreduplication. As is explained in Section 4.63, thedisyllabic-RED rule, P 23, operates to replace a RED symbol by a copyof the two syllables that immediately follow RED. (When there are two or more occurrences of REDbefore a disyllabic VR, orbefore a reduplicating syllable plus amonosyllabic VR, P 23 reapplies until all REDs have beenspecified. In certain cases, further rules mayalter the segmental structureof forms that result from applicationof the disyllabic-RED rule--cf. Section 4.64.) After the disyllabic-RED rulehas applied, a tone-changing rule,P59, operates to replace any high tones on VRs orREDs that follow the first(or 234 4.94 only) occurrence of disyllabicRED by low tones. Some examples of the operationof P 59 are: Before pa Atter P 59,

5kyerEkyirt 4yerfikylri 'he shows and shows'

lkylorfikyerikyerfi 5kyerEkyertkyeri 'he shows and shows andshows'

5hwehw6hwehwt 5hwehwihwithwi 'he looks and looksfor'

Ak 5kitatkits& Ak 5kits&kisi As5kliakitsa As5kAsakisi Fa 5k1sitkfisit Fa 5k&sitkasi 'he speaks and speaks'

(It will be observed from thelast set of examples that the disyllabic-RED rulefollows P 20, the Ak-As rulethat changes k&si to Akkassi/As,kisil etc. Asantelkisakits& is later changed toBkisilkasa by P 87, and may be further changed by P 65,diagaTI-Bilow.) In Pante, in addition tolowering tones that follow the first occurrenceof disyllabic RED, P 59lowers the tone of a nasal consonantat the end of thefirst disyllabic RED itself.Thus the rule operates tochange Pa hOnt5ntimt6n 'he sells and sells and sells'to Fa 5tOnt6ant5n.In Akuapem and Asante, thetone of a nasal consonant at the end of thefirst disyllabic RED is not lowered by P 59. Thus P 59 changes AY-As5tOnt8tanat5 (the Ak-As equivalent of Fahant5ntOnt5n) to AY-As BtimithltOntBil.

In Akuapem and Pante,after P 59 has applied, no further tone rules aretriggered by the occurrenceof disyllabic RED. In Asante, however,there is a further rule, P 65, that appliesin certain specifiable cases: viz., whenever anaffirmative verb formationthat includes disyllabic RED occurssentence-finally or clause-finally. 4,94 235

The rule operates upon theproducts of P 59, which, in Asante, always shov a low-high tonepattern on the first (or only) occurrence of disyllabicRED, and low tone on any following occurrencesof RED and on VR. P 65 affects all syllables that followthe first (or only) occurrence of RED. It adds a high tone immediatelyafter the low tone of the first such syllable, sothat this syllable shows a rising tone, and replacesthe low tone of all subsequent syllables affected by therule to high. Examples of the operation of the rule are:

Before P 65 After P 65 As BkyerEkyer ATIAtZiEi As 5kyerEkyerikyeri 5kyer6kyerfikyfiri As 5hwehlighwehwi hwothwEhwehwfi As 5kisakasi 501.4114s&

After P 65 has applied, subsequentP-rules have the effect of replacing the rising tone onthe syllable follawing RED by a mid tone.Thus we have surface tone patterns such as:

As5kits&kitsli. 'He speaks and speaks.' As3klis&kbosiikfts1. 'He speaks and speaks andspeaks.'

Since P 65 applies only tosentence-final or clause- final verbs, Asante shows tonalcontrasts such as the follawing:

As5hwihwthwehwt. 'He looks and looks(for it).' As5hwehwfihwithwi KOJI. 'He looks and looks for Kofi.'

Akuapem and Fante do not show tonalcontrasts of this kind:

Ak-Fa5hwiphwEhwehwi (lad). 'He looks and looks for Kofi.,

P 65 does not apply to negativeverbs. Thus:

As5nhwehwEhwehwi. 'le doesn't look and look(for it).' 236 4,95

4,95 Tone Changes inVerb-Plus-Pronoun Constructions.

This section is concerned withtwo Asante P-rulea: P 17, which changes lowtones on pronouns to high after certain Vs; and P 18, whiehchanges high tone on certain Vs to low before pronouns. The pronouns whose tone ischanged to high by P 17 may be eitherobject pronouns or possessive pronouns. (Prior to the operation of P 17, allobject pronouns have low tone in Asante. Asante possessive pronouns, anthe other hand, have low tone onlyin certain cases, the occurring tone depending upon the nounthat follows the pronoun. The possessive pronoun has lowtone if the root of the following noun has two or moresyllables or is a monosyllable belonging to arelatively small set of 6inalien- ables"; it has high tone if theroot of the following noun is a monosyllable notbelonging to the above-mentionedset of.inalienables: thus,,As nisfide6 'his,pamade' As ni b& 'his child' but As neki-ThiraiSt's no din 'his roon'.) P 17 operates to replace alow tons on an object or possessive pronoun by a high tonewhen the following conditions are met: (a) the pronoun immediatelyfollows a verb with a final hightone; (b) if the pronoun is anobject pronoun, it is immediately followedby a pause. Examples of the operationof P 17 are:

Before P 17

la). 1 him.' } AsAkatiln;1:44 'BS shows his son.'

a). him.' } AsObehanil bit, 'He will see son., i

', 1 126. ' 'He should see As ----Oahu illa?°ne va. ,o} hisson.1}Ligine bi.

,..._. ELI. , him.' 'He doesn't see As ahftne ba. his son.°itaalni131.1 a. him.' 126. { 2 f AsWAIL,Intlat,/'He goes tosee/his ()kW ne be.. 4,95 237

(In the case of the first three examples above, the tone pattern shown in the right-hand column is the surface tone pattern. In the case of the last two examples, subsequent rules%change the tone patterns on the Vs:As gigs:Lai b&7, Ohi nO/neb& by P 18--see discussion below--while As ok6hfi n6/ne ba - 660. n6/ne b4 by P 47--cf, Section 4.93.) It should be noted that pronoun tones are raised only after those disyllabic VRs that end in a high tone at the point in the P-rules at which P 17 applies. Thus the raising does not occur after disyllabic VRs that acquire a low-high tone pattern as a result of the operation of P- rules, but whose underlying tone pattern is high-low.Com- pare, for example, the surface tone patterns of:

As Okyers no. 'He shows him.' and:

As Obisa no. 'He asks him.'

Unlike kyerc, whose underlying tone pattern is low-high (/kidt/), bisa acquires its low-high surface tone pattern as aresulf77 the operation of P 20 and P 87, The underlying high-low tone pattern(/bisi/)has not yet been changed at the point in the P-rules at which P 17 applies, and hence the conditions required for the application of the rule are not met. Similarly, in order for P 17 to raise the tone of a pronoun that follows a monosyllabic VR(in the appropriate contexts), the VR must have high tone at the point in the P-rules at which P 17 applies. Thus P 17 does not apply, for example, to a pronoun that follows FUT plus amonosyllabic VR such as fa 'take' (underlying form /fat/) since at the point in the P-rules at which P 17 applies, fa still has low tone, and this low tone is replaced by EIgh only later, through the application of P 19.Thus we find a contrast between the surface tone patterns of:

As Obfihil 226. 'He will see him.' and:

As Obefa no. 'He will take him.' 238 4.95

(On the other hand, P 16, the Asante rule thatadds a low tone at the end oa monosyllabicVR that immediately follows a law-tone aspectual prefix,precedes P 17.Thus As Wadi xi; 'He has seen him' Wadi nil, by P 16, and no longer meets the requirements forpronoun-tone-raiiing.)

As was noted above, P 17 changesthe tone of an object pronoun only when the pronounis immediately followed by a pause: i.e., when it is in sentence-final orclause- final position. Thus the tone of the object pronoun remains low in:

As 51çyerfi xi; ditt. 'He always shows him.' AsOath& n; inE. 'He will see him today.' The second rule to be discussed here,P 18, changes to low a high tone on a monosyllabicVR that is immediately preceded by NEG when the VR isimmediately followed by a pronoun with high tone.In, many cases, the high tone on the pronoun has been introduced as aresult of P 17.Thus we have derivations such asthe following: Before.L17 Atter P 12 After P 18

, Oihfir. 0hI11S Oilhari./ As nebee} ne ne him.' 'He doesn't see ihis son.

The high tone on the pronoun may,hawever, have a different origin, as is true in the case of the pronounsof place, hi 'here' and h6 'there', where the high tone isinherent. Fertinent exaiges of surface tone patterns are:

As5Ak5 h6. 'He doesn't go there.' As Alm& hi. 'He hasn't come here.'

In cases where the pronoun followingthe VR is low, P 18 does not apply. Note the surface tone pattern of:

AsOnhu no da. 'He never sees him.'

In the derivation of this sentence,P 17 has not applied to raise the tone of the object pronounno (because the 239 4.95, 4.96

not immediately pre-pause--seeabove), object pronoun is VR. and hence P 18 doesnot apply to lowerthe tone of the

4.96 Tone Changesin SubordinativeVs

The SUBordinative occursat two points insurface structures: (1) as a constituent(in our formulation, the final constituent)of Vf and (2) as thefinal constituent P-rules triggered bythe SUB that occurs asthe of VP. As final constituent ofVP are discussed inSection 4.81. section VP-final SUBis listed in the is noted in that result of the lexicon as a high-tonemid vowel, but, as a operation of P-rules,usually lacks asegmental realization (More specifically,VP-final SUB never in surface forms. and has a segmentalrealization in surfaceforms in Akuapem Fante, and in Asantehas a segmentalrealization only in a cases.) The present sectionis concerned limited number of certain with the SUB thatis a constituentoi V, and with interactions betweenthis SUB and VP-finalSUB. The SUB that is aconstituent of V has nounderlying systematic-phonemic form.Its presence in aV does, how- the application of atone-changing P-rule, ever, trigger certain low tones P 60.P 60 has theeffect of replacing within the V by hightones or by fallingtones. (The low themselves have been tones that areaffected by P 60 may iutroduced by earlierP-rules.) of P 60 There is one differencebetween the operation other, in; on the onehand, Akuapem andFante and, on the in Akuapem andFante the rule appliesto replace Asante: by a high a lowtone on SC (thesubject-concord prefix) while in Asante(at least in the sub- tone in all cases, SC is dialect we havestudied mostclosely) a low tone on replaced by a hightone only when theSC is immediately NEGative. followed by. PROgressive,PERfect, CONsecutive, or (o-) of a V like For example, thelow-tone on the SC in which the SC isimmediately aAkyerl 'he doesn't show'9 is replaced by ahigh tone in the presence followed by NEG in the of SUB in alldialects; Thus in Asante as well as (For other two dialectsP 60 changes,Ukyer6 to 4kifferi. However, the the change of theNEG to n-, seebelow.) SC of a Vlas kyke, 'he shows',in which low tone on the PRO, PER, CON or NEG, the SC is notimmediately followed y 240 4.96 is not affected by P 60 inAsants, but is affected by this rule in Akuapam and Pante. Thus P 60 changesBlueri to As 51(yerilik4laSkArk. (For the change of theVR &irkto kyirft,, see below.) EIcept for the difference justdiscussed, P 60 may be stated as applying to the sametypes of constituents in all three dialects. In addition to its effect on a low-tone SG, P 60 operates,without any dialectal rP,stric- tions, to make the followingchanges in the presence of SUB: (1) a low tone is replaced by ahigh tone on: (a) any aspectualprefix; (b) an INGressive prefix;(c) a monosyllabic Verb Root that isimmediately preceded by STAtive or NEGative orimmediately followed by PASt;(d) the first syllable of a disyllabic VR thatis not immediately preceded by NEG, FUT, or ING;(2) a high tone is added at the beginning of a low-tone NEG, the regultbeing a falling tone an NEG.Some examples illustrating theabove changes in Vs that include SUB arelisted below.

BeforeP 60 After

(1.a) As lakli 'he has done' was As ;riyi [550] 'he is doing' 6rgyg[56j11 5rekyerfi 'he is showing'

rAk ,6,6bb613;16 (1.b) Ak-As 5baY6 'he comes to do' Its 6k5kitail Ak-As Makisi 'he goes to speak'1AkAs 5k6klait

(1.0 118 'he has' A.

Fa 'he doesn't sleep' gdi As 'he disi'

iAk-PaSla (1.d) 4621 'be shows' As Blaiori

51E14I Ak-As Wm& 'he speaks' iAkAs ;kW 241

Before P 60 After P 60

(2) onyc 'he doesn't do' onyc Uikyerl 'he doesn't show' lAgzflEk

2 1 A he doesn't go 61116k&si Fa Bik6kisi to speak' i (Through the operation of later P-rules,sfia,SY akyitri and OilkOkitsi are changed to, respectively,whit AlikyeA717117-gEk3kisi--the occurring surface forms. With this exception, the tone patterns shown inthe right- hand column are themselves those of the surfaceforms that occur in non-pre-pauseposition. In pre-pause position, certain other tone patterns are found in some casesbecause of the effect of the VP-final SUB--cf.Section 4.81 and below.) P 60 follows most of the other P-rules thatchange the tones of constituents of Vs. There are, however, certain tone-changing P-rules that follow P 60. One such rule is P 61the rule that places low tones on VRs inthe imperative (cf. Section4.91, subsection(h)). As a result of this ordering, imperative forms(at least, those that do not immediately precedepause) do not differ in tone according to the presence or absence ofSUB. Compare:

% Ak-AsBisa Kai IAsk Kofi. Ak-As WO nibisi lofi. You ask Kofi.'

(When the imperative V immediately precedes pause,however, there is sometimes a tonal differencebetween -forms that include SUB and those that do not.This difference results from P-rules triggered by theVP-final SUB--cf. Section 4.81rather than from the operation ofP60. An example is the final high tone on: Ak-As WO abisi. 'You ask.' (cf. Ak-AsBisi. 'Ask.')) Another rule that must follow P60 is P 62, the Asante rule that, after a high tone on SC,rtssigns a high 242 4.96 tone to NEG and a low tone to a monosyllabic VR, in a negative-perfect formation (cf. Section 4.92, subsection (c)). Thus for the tone pattern of the verb of:

As 5116 nit 6fikn h6. 'He haan't gone there.' (cf. As 3k65 h6. 'He hasn't gone there.° we assume the following derivation:

Before P 60 After P 60 After P 62

5165 Stik65 6fik56

The SUB that is a constituent of V may occur immediately before the SUB that is the final constituent of VP. (This is the case whenever the subordinated V is not followed by an object, an adverb, etc.)Under such circumstances, the surface tone pattern of the verb usually reflects the regular application of P 60 plus the regular application of those P-rules that apply to the VP-final SUB or are cortingent on its occurrence. For example, in: Ak-Fa Kfi at 61474. 'It's Koff who's doing the surface tone pattern of,6reyt, reflects the following dertvation: the underlying tone pattern oreye + V (where V represents VP-final SUB) is, by means of P 39 and P 50, changed to Breyi +0, and this, in turn is changed to 61434 + by P 60, Then 6reya + 0 -0. 61404, by P 85, and 6104y1 6reyi by P 86 and P 87.

There arehowever, certain cases in which subordinated Vs occurring in VP-final position do not show the expected tone patterns. For example, in Asante the SUB forms of affirmative HABitual, PROgressive and PERfeoct verbs formed with monosyllabic VRs show, in VP-final posi- tiona tone pattern inwhich there is high tone an the SC prefix and (in the PRO and PER forms) the ASPect prefix, and drop tone on the VR. Note the surface tone patterns of the verbs in:

(does Aslafi A ,6reyk (66j6). 'It's Koff wholisdoingit.' lat.wilyk. has done 243 4.96

In non-final positionsuch verbs show thetone patterns that are predicted byP60: i.e low-high in the caseof the HAB forms,high-high-high in the caseof the PRO and PER forms: does AsKOfi na:re; idAmtt. 'It's Kofi whois doing work.' has done 1w:Eki71 (The tone patterns that occurin the input forms toP60 are low-highfor HAB and low-low-highfor PRO and PER: cf. the non-subordinatedverbs As 2li 'he does', Asbreyt, 'he is doing', As4144 'be has done'.)Since the underlying verbs end witnhigh tone, and since the rulestrig- gered by VP-final SUBusually leave intact ahigh tone an the constituent thatprecedes the SUB, one wouldexpect that the subordinated verbswould show the same surface tone patterns in VP-finalposition that they show in non- final position. It may be noted thatthe irregular VP-finalsurface tone patterns that occurin Asante in the above caseswould be perfectly regularin Akuapem or Fante. (In the latter dialects the non-subordinatedverbs end in a high-lowl, rather than a low-high,tone pattern: e.g., Ak4PaLTA, 514yk, wilTI. For the derivation ofthe VP-final form Ak-Fa Arm from5reyi, see above.The derivation of the subordinated VP-final formsAk-FaLAandwitkvk follows similar lines.)Possibly, then, the Asantetone patterns reflect some influence from aneof the other dialects. In any case, it seemsthat a special "adjustment"rule is necessary in orderto account for theArsftabe forms, and we have postulated ruleP63, whieh simply takes asinput the forms that result fromthe application of theusual rules and converts them tothe forms that actually occur. Another such adjustmentrule may be required to account for certainsurface tone patterns that occurin Pante. According to data fromStewart (personalcommunication), the subordinatedVP-final forms of certainverbs formed with monosyllabic VRswith underlying low tone(e.g., /da/ 'sleep') would seem to beirregular.Consider, for example: 244

Fa 5rildi 'He is sleeping(there).' FaKafi ni Sredi 115. 'It's Kofi who is sleeping there.' FaKafi ni 'It's Kai who issleeping.'

For the verb in tbelast sentence, the usualrules would predictSredi rather than 6r&di.Since we have not been able to confira Stewart'sdata, however, we havenot attempted to foraulate tbeadjustment rule that wouldbe necessary todertve the surface formsthat he reports. 245

Appendix

Phonological Rules

The following is a linearly-orderedlisting of all the P-rules discussed in the text. The ordering of the rules should beregarded as highly tentative. That is, while in many casesthere is clear motivation for the ordering assumedhere, in many others the ordering is arbitrary and is subject torevision.

Each of the eighty-seven ruleslisted is identified by a two-digit Arabic numeral: P 01, P 02, ...P 87. In the case of those rules which were given an interimnumbering in Chapter 3 of the text,the Roman numerals used for the interimnumbering appear in parentheses afterthe Arabic numerals. For example, the notation"P 01 (P xxxii) means that P 01 was given the interimnumbering Pxxxii in Chapter 3.

All but the last five ruleslisted (P 83-P 87) are rules of morpheme-level and/or word levelphonology. When a "variable" symbol-- X, Y, etc.--appears in one of therules P 01 - P 82, therefore, it is to be understood that thevariable does not include a wordboundary, and this condition is left unstatedin the rule itself.

P 01 (P xxxii) TONE SPREADING

[+Voiced[-Vocalic ] +Consonantaa} [aToneWaTone]

(Note: This rule reappliesimmediately whenever the conditions for its application aremet.) I.

P 02 (P v) NASAL PREFIXES

, -Vocalic r +Consonantal 1+ +Nasal + low>

<+Nasal> I I+G

P 03 (P ii1) SYLLABLE-FINAL CONSONANTS

[+Nasal] / [Oloiced ]

......

(+Voiced] / +Anterior [-Nasal (+Consonantal] [-Consonantal, -Coronal <+Glottal> f [-Nasal 111 (+Voca).ic] / +Coronal -Nasal

P 04 (P xi) GLOTTALSTOP DELETION

(+Glottal Constriction] -* 0/ X

(where X is not a (+PB]

(Note: This rule reapplies immediately whenever the conditions for its application are met.) 247

P 05 MEDIAL [k] [g]

[-Voiced] -* [+Voiced]/ [X st WO] yR +Back

P 06 (P REGRESSIVE NON-VOWEL NASALIZATION

[4.0onsonantad +Voiced

[-Vocalic] .4 (+Nasa1]/ r i /----[+Zaer] IL-Consonantalj

P 07 (Pxix1) [I] INSERTION

+Back $4, +Loy 0 +High +Tense

i[k][+Back] : As [+Vocalic] : Ak-Fa

P 08 (p xiv) MEDIAL [g] DELETION

US] 4 0/V V(C).12

P 09 (P xv) LABIALIZATION

[-Vocalic] [+Round]/m [+Round]

(Note: For a discussion of theapplication of this rule to certain products of P 31in same Akuapem subdialects, cf. Section)e.74, subsection (a).) 248

P 10 (P xvi) [U] DELET/ON

[U] -o 0/ (+Vocalic]

P 11 (P xvii) PALATALIZATION: 1

[-Nasal]

/ +Vocalicv r-Vocalic1-0(+Palatal]/ L+Back j [VPalatall" +Nasal +Roundil

(where X (V)(C)Ist/s X)

P 12 (P xxi) FRONTED ROUND VOWELS

[-Back [+Segment] -0 r +Round], mJlIe

P 13 (P xx) [I] DELETION

(I] -0 0/ (+Vocalic]

P 14 (P xiii) FORMATIVE-BOUNDARY DELETION

[+FB] -0 0/[+S0] (+ASP]

P 15 (Pxii1) [d] -0 [r]

+Consonantal +Coronal -0(+Vocalic]/V V +Voiced [-Nasal 249

P 16 LOW-TONE ADDITION TO MONOSYLLABIC VERBROOT

0 .01+Seg1eni1A-Tbne1 (0[+Tone](C)]...,#: As L+Tone j L+ASP JVR

(Note: This rule reappliesio outputs of P 19)

P 17 PRONOUN TONE RAISISG

[-Tone] -0 (+Tone]4R[X [+Tone](C)] +

E+PRONOUN IX Possessive

P 18 TONE LOWERING BEFORE HIGH-TONE PRONOUN

1+Tone4 [+Tone] -o [..Tone]/[+NEG]+ [C (C) As VR j +PRONOUN

P 19 TONE RAISING IN MONOSYLLABIC VR

(+Vocalic] 4 (+Tone]/XvR(C (C)1: Ak-As

(where X 0 X[+PRO] in Asante)

P 20DISYLLABIC VR TONE REVERSAL

(+Tone]X ([.+Consonantall )(-Tone] X] VR[X <-Voiced>1

[-Tone]x ([.+Consonanta]l)1+ToneJ41+Segmenti> -Tone VR[x <-Voiced). 2

<[-Tone]>3. (C) (RED) : Ak-(As; 250

P 21 GENERAL RULE OF MONOSYLLABICREDUPLICATION

air +High lraF (a) RED > C1 4ONasa1>j>2} 3/---+YR[LCLasaq . -Nasal k+High-Nasal>2 [-Tone

(P2 1 LONasalN.1)] 3

(where Cmg CV C2 = C and D2F] = all unmentionedfeatures) 1 2'

(b) [-Tone] 4 (+Tonel/+RED01. [C (C)] VR

P 22 (I] [U] IN MONOSYLLABIC REDUPLICATION

+Round -Palatal (+Back]/C[4.BED](C)+ 1I [4.Low ](C)] [-Back] (1[i+Coronal FRound irCoronal1j

(Note: For dialectal restrictions,cf. Section4.62, subsection (a).)

P 23 GENERAL RULE OF DISYLLABICREDUPLICATION

RED >seX1=C1V1/ +XI=COAC2 "4Nasal>1)31

(where Xi=X1, C1=C1, V1=V1,C2=C2) 253.

P 24 NASAL DELETION IN DISYLLABICREDUPLICATION

(a) +Nasal -0 0/11(V +vR[X m X] +RED

rasal] rasall I +Nasal (b) +RED +RED °I +RED[CVC] +RED

(optional)

(Note: For dialectalrestry.qons,cf.Section 4.64, subsection (a).)

P 25 (P vi) HOMORGANICNASAL

-Vocalic *Corona' OBauk / (..) +Nasal OBack yHigh I.yHigh

(Note: For dialectal restrictionsin verbs involving disyllabic reduplication, cf. Section4.64, subsection (b).)

P 26 NASAL DELETION IN MONOSYLLABICREDUPLICATION

rrocalic IV( C ) : Ak +Nasal -.0 +Nasal +RED VRL 252

P 27 VOWEL GEMINATION IN DISYLLABIC REDUPLICATION

{alEl] +Voca:Ic +Vocalic aF r-Vocil raT aF (a) ONasal =V+Nasal>j + ..) {0Nasall =i<+Nasall+ :Fa (optional) yTone 6Tone ONasal ONasal {+RED +RED yTone óTone +RED J L+RED

(where [iF]= all unmentioned features)

(b) (+Low] -0 [-Low]/[+RED]+ [C =I] VR (where part (a) of rule has applied)

P 28 VOWEL GEMINATION IN RE-DUPLICATED MONOSYLLABICVERBS

[Vocalic] [4:Vocalil (a) (a) +RED -0 Og+RED1+ (C)+ (CV(C)]: Fa(optional) +RED VR

4+Vocalil a .4+Nasal 0/ diasal +Nasl. +RED +RED +RED +RED

I [+Low] ->[-Low]/EVocalic]+[+RED +RED

.... [a] .4 LOF]/r [+Vocalic / OF + +RED +RED

(where [a] and [OF] = all unmentionedfeatures)

P 29 x) FINAL [w] 4 [7]

[w] [fl/ + :As,Fa2 253

P 30 FINAL [r,m,n] -0 [7]

(1114- [fl/EVR I+ [n] +P 30

(Note: For dialect-specific conditions onthis rule, cf. Section 4.72)

P 31 VOWEL ADDITION TO NON-VOWEL-FINAL VERBROOT

4sr I+Vocalic +High +Vocalic rtVocalic rB+Back>> if+HighN. +Consonantal} -ack 14 s+Backd/i /4.4C:crnal > i+TenseS$ ` erior il:::: 44w /3 Olisal)4 IONasal>11 +VR +VR 1-Nasal f [ J OM.

P 32 RESYLLABIFICATION

CVC=V## CV=CV##

(Note: This rule reapplies immediately wheneverthe conditions for its application aremet.)

P 33 FINAL VOWEL -4 Pt]

[+Vocalic] 0//7[30+ " +High O: (+Voiced]

(where: (a) affected segment 0 (+PAS] (b) various dialect-specific conditions are met (c;f: Section 4.79) 254

P 34 PAST VOWEL ADDITION

+Segment (+PAS] 0 -3N aF / [X[aF ([?])] VR -Tone j $Tone

(where F = all segmental features,and a = + for [Vocalic], - for [Consonantal])

P 35 PAST CONTRACTION: 1

: Fa (+Segment] [+PAS] CII/VR[X[+Tone]X[-Tone]X]

P 36 PAST SUFFIX DELETION ANDCONSONANT LENGTHENING

(a) (+PAS] i+ 0/ X

(where X # (+PB])

+Segment' rocalil] : As (b) al,-0[-Vocalic/ [X VR aF aF .1

(where the input to (b) has undergone(a))

P 37 ASSIMILATION OF VOWELS TO PASTSUFFIi

um[X[-Vocalic]] +[+PAS] (+Back] -4 [-Back]/ "1 tin[X[ --- ](?)] +[+PAS] : Fa 1vil +Vocalic P

P 38 PAST CONTRACTION:2 (+Pas] (+Segment] -0 0/ (X(+Vocalic]((?1)] VR (+Vocal1c][7 255

P 39 SUBORDINATIVE. AND NOMINAL-SUFFIX DELETION +Segment (aF) -0 0/ +Tone i+SU4 [t+NS j

(where SUB is the last IC of VP; optional or inapplicable in Asante in certain cases--cf.Section 4.81, subsection (a))

P 40 VOWEL HARMONY IN SUBORDINATIVE ANDNOMINAL SUFFIX

(a) rense / fiTense+ : As +NS yNasal yNasal

+Nasal (b) (-High) -o (+High)/tSUal : As +NS I

(where SUB is the last IC of VP)

P 41 TON! RAISING BEFORE SUBORDINATIVE ANDNOMINAL SUFFIX

(-Tone) (+Tone)/ + i+SUB : As +NSfJ

(where (aF) indicates that segmental features are present; for restrictions on application ofthis rule to segments preceding SUB, cf. Section4.81, subsection (c))

et P 42 SUBJECT-CONCORD ASSIMILATION IN PERFECTAND CONSECUTIVE

(a) [3] -0 (wilitSj[a]

(b) (+Vocalic)

P 43 STATIVE/CONSECUTIVE LOW TONE

[X x] (+STA]((+NEG])VR (+Voca].ic] [-Tone]/ (+CON]vR(C (C)]

P 44 NEGATIVE-STATIVE HIGH TONE

: (+Segment] -0[+Tone]/[+STA]+[<+NEG1+C

P 45 PAST LOW TONE

(+Vocalic] -0 (+Low]/XvR(C (C)] +[+PAS] : As

(where X # X+(+NEG]+X)

P 46 PAST TONE CHANGES

(a) [-Tone] -> [+Tone]/vR[C (C)]+[+PAS] : Fa

: Fa (b) vR(X[-Tone]X(+Tone]X] lim[X(+Tone]X(-Tone]X]/ +[+PAS]

+[+PAS] : Fa (c) vil[X(+Tone]X[-Tone]X] ->vii[X(-Tone]X(+Tone]X]/

(where (b) and (c) are disjunctivelyordered)

P 47 INGRESSIVE LOW TONE

4 [-Tone] : Ak-As +ING 257

P 48 HIGH TONE AFTER PERFECT

X] [+Vocalic] VR ING

P 49 OPTATIVE TONE

])[X X] (a) (+Vocalic] -.0 [+Tbne]/I+OPTHING[X

: Fa (b) [+Vocalic] [-Tone]/[+0PT]((+ING])11RDC = C X]

P 50 MONOSYLLABIC VR TONE REVERSAL

Tone +Tone : Ak-Fa i±SC} +w,[C[+Tone](C)] i+SCp+VR[C[-Tone](C)]/ +ASP "' +AS

P 51 HIGH TONE ON PERFECT

(+PER] 4 (+Tone)/ +C(+Tone] : Fa

P 52 HIGH-TONE SECOND-PERSONSUBJECT CONCORD

[-Tone] -> (+Tbne]/ [ +SC +2nd Person

(where +SC I0 pa [wan 1+2nd Person 258

P 53 HIGH TONE ON PROGRESSIVE BEFORE NEGATIVE

[-Tone] -0 (+Tone]/[4:174+[+NEG]

P 54 NEGATIVE HIGH TONE BEFORE PAST

: pa (aTone] -o [+Tone]/[441E7i+C[;Voca1iC]/...2([4.PAS]

P 55 HIGH TONE SUBJECT CONCORD BEFOREHIGH-TONE NEGATIVE, PROGRESSIVE, OR OPTATIVE

[-Tone] -* (+Tone]/Lsc I OPTI

(+NEG] +PRO

P 56 PROGRESSIVE-ASPECT FORM

(a) [r][e] 0/[4:Segmen] : As(obligatory), Ak-Fa (optional) OTone +PRO

(where [aF] = all unmentioned feature specificationsof the vowel of PRO)

egment (b) [+Vocalic] (-Low]/ X +Low +PRO

(wheri X 0 +[+NEG]X)

+Segment (c) [OF] [aF],[0]+ OTone +PRO

(where a = + for [Vocalic] and - for[ConsonantLl], and where X 0 +[+NEG]X) 259

P 57 PREFIX HIGH-TONE AGREEMENT

[+PER] : Ak-As ÷[.++TSeongmeent .. }/l+Toae](C)(+) [+CON] Ak-As (+NEW+vR(X=X1+(+PAS] : Ak-As

[+STA]+ +[+NEG] : As

(Note: In Asante, this rule follows P 58)

P 58 HIGH TONE AFTER NEGATIVE BEFORE PAST

e (+Vocalic] -0 (+Tone]/[ 1+11,R(X XIBX]+[+PAS] : Ak-As

P 59 TONE LOWERING IN DISYLLABIC REDUPLICATION

(a) [+Tone] [-Tone]/RED[X=4+X X

(b) (+Tone] -.0 [-Tone]/,,[X=X [ +Consonantal ] : Fa Lwau +Nasal

P 6o VERB-FINAL SUBORD INATIVE TONE CHANGES

[tSC I +AS4 +ING STAll 1 tWEG4' I x+(+sus] (a) [-Tone] -o (+ToneVVR(C (C)]/ v

k -NEG [C X=90/-FUT1+ VR -ING

(Note: In Asante applies to[]only if following formative is +SC PRO, PER, CON, or NEG) 260

P 60 Continuation of VERB-FINAL SUBORDINATIVE TONECHANGES

(b) 0 -0.[113456m51/' ' v[+SUB] +Tone /---A0E0IX

P 61. IleERATIVE LOW TONE

+STA] (+Vocalic] -.). [-Tbne]/ [a+NEGll 1 [X X1Y (+IMP] vR .

1r, 1 (If environment is (+STA]((+NEGDvR[X.-X]Y, Y does not includev[+SUB])

P 62 HIGH NEGATIVE, LOW VERB ROOT BEFORE PAST

[..+Tsoenmeentt\r-Tone f)+NE G VR[C[+Tone](C)]"4'[:17121+VRI:C[-Tone](C)]/

[+Tone](C)+ +[+PAS] : As

P 63 SUBORDINATIVE TONE ADJUSTMENT

Tne] :PER vo[C[OTone](C)] +PRO

/r+Tonei :pER 14+suBli+suB] : As 's.LtSC [C[-Tone](0)11 orm I I VR

P 64 LOW-TONE SUBJECT CONCORD DELETION

OF - 0 : Ak-As +SC 26].

P 65 TONE RAISING IN DISYLLABIC REDUPLICATION

regment] //,/ ( X=X ] CV( C ) X(+PB] : As +Tone RED

(where W 0 X+RED)

(b) [-Tone] -0 [+Tone]/ +Tone X Xf+PB] : As OF

(where+Tone has been introduced by part(a) of the rule) OF

P 66 CONSECUTIVE/IMPERATIVE HIGH TONE BEFORE PHRASE BOUNDARY

(+CON] (+Vocalic] . (+Tone]/ 1(+Imp]va(C (C)][+PB]

P 67 HIGH TONE AFTER INGRESSIVE DISYLLABIC VRBEFORE PHRASE BOUNDARY

0 rSegment] (+PB] : Ak -As +Tone VR

P 68 HIGH TONE ON VERB-FINAL NON-VOWELBEFORE PHRASE BOUNDARY

[4:Vocalic]. [-Tone] f+Tione1/ r tVocalic ([?])[+PB] 'Iv` +Tone +Consonant

P 69 (P xxii) VOWEL TENSING (WITHIN WORDBOUNDARIES)

[+Vocalic] .0 (+Tense)/ X-I" Y +G +Tense -G

(where: (a) if the affected segment precedes X, X does not include a (+Vocalic, -G] segment;(b) if the affected segment follows Y, Y does not include a(+Vocalic, -G3 segment) 262

P 70 (P iii) [0] [n]

i(.+)11 : As-Fa fol -* inl/V : Fa

P 71 (P iv) [OM] VOCALIZATION

(a) [1] -.0 0/[0] ## : AS

rVocalici (b) [0] -0aBack / aBack ## : As [BTense BTense

P 72 (P vii) PROGRESSIVE NON-VOWEL NASALIZATION

-Vocalic] r p[-Vbcalic1(0 t+Nasal :-Fan +Nasal

P 73 (P viii) NASALIZED VOWEL SEQUENCES

[+Vocalicwi (+Vocalic] +Nasal

P 74 (P ix) REGRESSIVE VOWEL NASALIZATION

[+Vocaiic +Conso I (+Nasal]/ [nantal]. +Nasal

P 75 (P xviii) PALATALIZATION: 2

-Vocalic [+Vocalic]: pa +Consonantal -0, [+Palatal]/ +Palatal -G 263

BACKING P 76 (P xxv) PROGRESSIVE VOWEL

+Vocalic ]/[-Vocalic] (+Bcka (-Vocalic]= [-Back +Round -Low

P 77 (P xxvi) REGRESSIVE VOWEL BACKING

Back (+Back]/ [+Round]+ : Fa .Low

P 78 (P BACKNESS HARMONY

*Back (-aBack]/ X -aBack : Fa -Low .G +o

(where X does not include a(+Vocalic,-G] segmea)

IN ENVIRONMENTHIGH-LOW HIGH P 79 HIGH-TONE ADDITION

Vocalic (+Tone] [+Segmeni, /(+Tone]=C(-Tone]s im+Consonantal} 0 +Tbne +Voiced

BEFORE (r) P 80 (P radii) VOWE L DELETION

[+Vocalic] mr +High .10 0/C

P 81 FUTURE:mebe me

4111111IMID +Vbcalic : Ak-As (+Segment] -.* 0/. +SC +Consonantal +1st Person +FUT [+Singula_ 264

P 82 GLOTTAL-STOP ADDITION

+Vocalic + : Ak [+ P 82]

P 83 (Pxxiii) VOWEL TENSING ACROSS WORD BOUNDARIES

i+Vocalic] (+TenseY ## C -4411 -Tense +Tense G

P 84 (P xxiv) (a] -0 (10/(e]

1(+Coronall r+Vocalic] : Ak-As +Lou -0 1.+Tense +Tense [...Low] : Fa

P 85 (P xxx) TONE INCORPORATION

+Segment (a) aTone =(C) -0 =(C)aTone OF

(b) (C)= aTone -0 aTone [OF OF

(where OF in4icates that no segmental features are present)

P 86 (P xxix) DOWNDRIFT

(a) (aTonerl -10 (+Pitch n]/(+PB]X Y

rone ix y (b) (+Tone]ii -0 (+Pitch n+z]/ +Pitch nj

rTone (-Tone]il -0 (+Pitch n-11/ (c) +Pitch n

(wh-re X and Y do not incluls tone-bearing segments other thun those affected by the rule, and where X does not include a (+PB]) 265

P 87 (P xxxi) TONE SIMPLIFICATION

row.] ronl rone] (where a or y = +) OF OF OF 401) fame' ryTonel ronl/(where a and y = -) j OF Ihp j LOF present (where [OF] indicates thatsegmental features are features arepresent) and [OF] indicatesthat no segmental

4( 267 Bibliography

Bach, E. (1968) "Two Proposals Concerning the Simplicity Metric in Phonology." To appear in Glossa.

Bailey, C.-J. and J.-C. Milner (1967) "The Major Class Features 'Sonorant' and 'Vocalic' and the Problem of Syllabicity in Generative Phonology." Unpublished paper.

Balmer, W.T. and F.C.F. Grant (1929) A Grammar of the Fante-Akan Language. London: Atlantis Press.

Berry, J. (1957) "Vowel Harmony in Twi." BSOAS 19.124-130.

Berry, j. (1960)English, Twi, Asante, Fante Dictionary. London: Macmillan & Co., Ltd.

Boadi, L. (1963) "Palatality As a Factor in Twi Vowel Harmony."JAL 2.133-139.

Chamsky, N. (1964) "Current Issues in Linguistic Theory." The Structure of Language, ed. J. Fodor and J. Katz, pp. 50-118. Englewood Cliffs, N.J.: Prentice-Hall.

Chomsky, N. and M. Halle (1968) The Sound Pattern of English. New York: Harper and Raw.

Christaller, J.G. (1875)A Grammar of the Asante and Fante Language Called Tshi (Chwee, MT. Basel.

Christaller, J.G. (1881)Dictionary of the Asante and Pante Language Called Tshi (TW1). (Second edition published 1913)% Basel.

Doke, C.M. (1935) Bantu Linguistic Terminology. London: Longmans.

Dolphyne, F. (1965) "The Phonetics and Phonology of the Verbal Piece in the Asante Dialect of TWi." University of London Ph.D. thesis.

Fromkin, V. (1965) "On System-Structure Phonology." Igasum. 41.601-609.

Koelle, S.W. (1854)Polyglotta Africana. London.

Ladefoged, P. (1964) A Phonetic Study of West African Languages. Cambridge: University Press.

Peterson, T.H. (1967) "Ordered Rules in Terraced Level TOne Systems." Paper presented at the LSA summer meeting.

Redden, J.E. and N. Owusu (1963) Twi Basic Course. Washington, D.C. 268

Schachter, P. (1961) "Phonetic Similarity in Tonemic Analysis, with Notes on the Tone System of Akwapim Twi."Language 37.231-238.

Schachter, P. (1962) Teaching English Pronunciation to the TVi-Speak- la, Student. Legon: University of Ghana Press.

Schachter, P. (1964) "Some Comments on J.M. Stewart's 'The Typology of the TVi Tone System'." Legon: Institute of African Studies.

Stanley, R. (1967) "Redundancy Rules in Phonology."ItampagE. 43.393-437.

Stewart, J.M. (1962) "An Analysis of the Structure of the Fante Verb With Special Reference to Tone and Glottalisation."University of London Ph.D. thesis.

Stewart, J.M. (1964) "The Typology of the Tvi Tone System." Legon: Institute of African Studies:

Stewart, J.M. (1966a) "Asante Twi in the Polyglotta Africana." Sierra Leone Language Review 5.

Stewart, J.M. (1966b) "A Deep Phonology of the Akan Monosyllabic Stem." Paper presented atthe Seventh West African Languages Congress.

Stewart, J.M. (1966c) "Some Suggestions For a Unified Akan Orthography." Unpublished paper.

Stewart, J.M. (1967) "Tongue Root Position in Akan Vowel Harmony." Phonetica 16.185-204.

Ward, I.E. (1939) The Pronunciation of TVi. Cambridge: Heffer.

Welmers, W.E. (1946) A Descriptive Grammar of Fanti. Suniement to Language 22.3.

Welmers, W.E. (1959) "Tonemics, Morphotonemics, and Tonal Morphemes." General Linguistics 4.1-9.