UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The RTR harmonic domain in two dialects of Yorùbá 2005

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
ubc_2005-0592.pdf [ 8.39MB ]
Metadata
JSON: 1.0092141.json
JSON-LD: 1.0092141+ld.json
RDF/XML (Pretty): 1.0092141.xml
RDF/JSON: 1.0092141+rdf.json
Turtle: 1.0092141+rdf-turtle.txt
N-Triples: 1.0092141+rdf-ntriples.txt
Citation
1.0092141.ris

Full Text

THE RTR HARMONIC DOMAIN IN TWO DIALECTS OF YORUBA by JEREMY PERKINS B.Sc, The University of British Columbia, 2003 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS in THE F A C U L T Y OF GRADUATE STUDIES LINGUISTICS THE UNIVERSITY OF BRITISH COLUMBIA September 2005 © Jeremy Perkins, 2005 Abstract The RTR Harmonic Domain in Two Dialects of Yoruba by Jeremy Perkins In this thesis, a process of vowel harmony is explored in two dialects of Yoruba where the tongue-root values of adjacent vowels generally agree. In Standard Yoruba, this process of tongue-root harmony affects only vowels within the prosodic word. However, in Moba Yoruba, tongue-root harmony affects vowels in the class of proclitics in addition to those contained in the prosodic word. It is argued that this difference in the domain of application of tongue-root harmony is captured by defining constraints that refer to different harmonic domains in each dialect. A prosodic domain that dominates the prosodic word, the clitic group, is posited in order to capture this dialectal difference. Three different optimality-theoretic accounts that deal with tongue-root harmony in Standard Yoruba are presented. The ability of these analyses to capture patterns within four dialects of Yoruba (Ekiti, Ife, Moba, and Standard Yoruba) and their general theoretical relevance are the main criteria for evaluation. An account utilizing alignment constraints (Pulleyblank 1996) succeeds in capturing the cross-dialectal patterns of tongue-root harmony in all four dialects of Yoruba, however it relies on the formulation of gradiently evaluated alignment constraints. This is a situation that is theoretically undesirable. An account enforcing stem-control (Bakovic 2000) succeeds in capturing the patterns seen in two of the four dialects. I argue against a basic assumption that this account relies on: that all V C V nouns are morphologically complex. It is shown that if at least some of these nouns are not analyzed as morphologically complex, the stem-control account cannot succeed in capturing the attested pattern of tongue-root harmony in any dialect of Yoruba. Finally, an account that utilizes markedness constraints prohibiting certain featural sequences (Pulleyblank 2002) can capture the pattern seen in Standard Yoruba. An adaptation of this account that includes positional faithfulness is offered to account for Ife, Ekiti and Moba Yoruba. This positional faithfulness account avoids the need to use gradiently evaluated constraints and it does not rely on morphological constituency. Instead, it uses prosodic constituents as domains of reference for OT constraints. ii Table Of Contents Abstract ii Table Of Contents iii Acknowledgments v Chapter 1 - Introduction 1 Chapter 2 - RTR Harmony in Standard Yoruba 3 2.1 RTR Harmony: The Basic Pattern 3 2.1.1 Roots in Yoruba 4 2.1.2 Harmony in V C V Nouns 6 2.1.3 Harmony in Derived V+CV Nouns 8 2.1.4 Disharmony in Compounds 9 2.1.5 Harmony in V C V C V Nouns 10 2.2 Harmony via Alignment 11 2.2.1 Basic Constraint Ranking for Alignment 11 2.2.2 Dialectal Variation: High Vowels in Ekiti Yoruba 13 2.2.3 Dialectal Variation: Relative Alignment in Ekiti and Ife Yoruba 14 2.2.4 Problems with Alignment: Gradient versus Categorical Constraints 15 2.2.5 Treatment of High Vowels: Opacity and Transparency 16 2.2.6 A n Alternative Alignment-Based Account: Prosodic Licensing 19 2.3 RTR Harmony via Stem-Control 19 2.3.1 Basic Stem-Control in Standard Yoruba 19 2.3.2 Treatment of High Vowels: Sympathy Theory 21 2.3.3 Problems with Stem-Control Theory: Dialectal Variation in the Behaviour of High Vowels 25 2.3.4 Problems with Stem-Control Theory: Morphological Structure 30 2.3.5 Towards a Prosodic Alternative 32 2.4 RTR Harmony via Prohibition 33 2.4.1 RTR Harmony via Prohibition in Standard Yoruba 33 2.4.2 RTR Harmony via Prohibition in Ife Yoruba 37 2.4.3 RTR Harmony via Prohibition in Ekiti Yoruba 42 2.5 Summary 44 Chapter 3 - RTR Harmony in Moba Yoruba 46 3.1 Moba Yoruba - Phonological Background 46 3.2 RTR Harmony in Moba Yoruba: The Basic Pattern 49 3.2.1 Harmony in V C V Nouns 49 3.2.2 Harmony in V C V C V Nouns 52 3.2.3 Disharmony in Compounds 53 3.2.4 Consonant-Deletion in V C V C V Nouns 54 3.3 RTR Harmony in Prefixes 56 3.4 RTR Harmony in Proclitics 57 iii 3.4.1 RTR Harmony with Single Proclitics 57 3.4.2 RTR Harmony with Multiple Proclitics 59 3.5 RTR Harmony in Enclitics 61 3.5.1 RTR Harmony in Enclitics 61 3.5.2 Tonal OCP in Enclitics 62 3.5.3 Implications for Domains 64 3.5.4 Implications for OT Accounts 68 3.6 RTR Harmony Outside the Verbal Domain 72 3.6.1 RTR Harmony and Adverbials 72 3.6.2 Root versus Non-Root Status 73 Chapter 4 - Analysis of Domain Size 76 4.1 Introduction 76 4.2 Prosodic Structure in Standard Yoruba 76 4.2.1 Syllable Structure in Standard Yoruba 77 4.2.2 Foot and PrWd Structure in Standard Yoruba 78 4.2.3 Prosodic Constituency in Moba Yoruba 80 4.2.4 On the Prosodic Status of Clitics 81 4.3 Nasal Harmony 85 4.3.1 Syllable-bound Nasal Harmony 85 4.3.2 Nasal Harmony across Syllable Boundaries in Moba 86 4.3.3 Nasal Harmony Beyond the Root 87 4.3.4 Summary of Nasal Harmony - Implications for Domain-Size 89 4.4 Clitics and Prefixes in Moba: Implications of a Domain Mismatch 90 4.5 Summary 93 Chapter 5 - An OT Account for RTR Harmony in Moba Yoruba 95 5.1 An OT Account for RTR Harmony in Moba Yoruba 95 5.2 Dialectal Variation: Ife, Ekiti, and Standard Yoruba 106 Chapter 6 - Conclusion 109 References I l l Appendix A - Constraint Definitions 113 Appendix B - Clitic-Aux-Verb Paradigm 117 iv Acknowledgments First, I would like to thank my consultant and fellow student, Oladiipo Ajiboye, without whose patience and help this thesis would never have been written. I am indebted to Doug Pulleyblank for his assistance and helpful advice along the way. Thank you also to Gunnar Hansson whose comments and assistance were appreciated immensely. I would also like to thank Rose-Marie Dechaine for introducing me to my thesis topic and for all the time and assistance she has offered me over the past two years. I would like to acknowledge Henry Davis, who was my first linguistics professor as an undergraduate student - thank you Henry for convincing me to continue studying Linguistics. I am indebted to Bryan Gick, whose help was invaluable during my first year as a graduate student and Martina Wiltschko whose teaching and help as both an undergraduate and as a graduate student were greatly appreciated. My fellow graduate students at the University of British Columbia have been wonderful friends and companions during this process and I have enjoyed many good times with all of them. I would especially like to thank Marion Caldecott, Kristin Johannsdottir, Leszek Barczak, Nahal Namdaran, Yoko Ikegami, Jeff Muehlbauer, Clare Cook, Add Ruangjaroon, and Christine Ravinski. I am also greatly indebted to the financial assistance of the SSHRC grant entitled, "Harmony and the phonetic basis of phonological features" for which Doug Pulleyblank is the principal investigator and Bryan Gick is the co-investigator. And finally, thank you to my parents, Ed and Karen, my sisters, Shannon and Emily, our dog, Molly (woof!) and all my good friends for everything you've done for me. I am lucky to have so many wonderful people in my life. Thank you all. v Chapter 1 - Introduction This thesis explores dialectal differences in tongue-root harmony based on a corpus of language data collected from a native speaker of both the Moba dialect of Yoruba and Standard Yoruba. The main dialectal difference that is explored in this thesis concerns the size of the harmonic domains in RTR harmony. While in Moba, the class of proclitics is included in the harmonic domain, it is not in Standard Yoruba. This is illustrated below in (1). The vowel in the 3SG proclitic harmonizes with the vowel in the verbal base in Moba but not in Standard Yoruba. (1) Proclitics in Moba and Standard Yoruba M B SY Gloss Meaning 3SG e s e 6 se 3SG='do' 's/he does/did' e j e o j e 3SG='eat' 's/he eat/ate' Assuming minimal differences in representations between Standard Yoruba and Moba, the domain-size difference could result either from a common prosodic domain that is mapped to different syntactic constituents in the two dialects or from a reference to two distinct prosodic domains. It is argued that the difference between the patterns seen in Moba and Standard Yoruba is only compatible with an account employing a reference to two distinct prosodic domains. Three recent accounts of RTR harmony in Yoruba utilize sets of OT-constraints that are unique to each account. One utilizes featural alignment constraints (Pulleyblank 1996), one utilizes constraints enforcing stem-control (Bakovic 2000), and a third utilizes prohibition constraints that ban featural sequences (Pulleyblank 2002). These accounts differ in subtle ways and the phonological and morphological behaviour of Yoruba at the word level does not always provide a satisfactory testing ground. The subtleties of these accounts are evaluated in light of the difference in harmonic behaviour of clitics in Moba and Standard Yoruba. Of these accounts, the stem control account is unable to capture the Moba pattern seen in the clitic domain with respect to RTR harmony. It also relies on morphological structure that is tenuously posited to hold across the board in all morphemes containing more than one vowel. This is contrary to the evidence in at least a few cases. I argue against an account utilizing stem-control as a result. Additionally, I argue against an alignment-based account for reasons independent of RTR harmony. Alignment of featural domains is problematic in general for an account that utilizes 1 gradiently evaluated constraints, as is the case for Yoruba. However, a categorical alignment constraint isn't able to account for RTR harmony in Yoruba either. Instead, a unique account that is based on the harmony-via-sequence-prohibition account is proposed. Rather than appeal to morphological constituency alone, as the stem-control account does, this account seeks to map prosodic structure onto morphological structure, and then use the prosodic categories defined as such, as domains of reference for OT constraints. This account captures not only the pattern of harmony in Moba and Standard Yoruba. It also extends typologically to the Ife and Ekiti dialects of Yoruba. This thesis is organized as follows. Chapter 2 begins with an outline of the basic pattern of RTR vowel harmony in Yoruba based on Archangeli and Pulleyblank (1989). Next, the three optimality-theoretic accounts of RTR harmony in Standard Yoruba are summarized in detail. A critical analysis is offered in areas where these accounts succeed and where they do not. The harmony-via-prohibition account is then extended to account for Ife and Ekiti Yoruba. Chapter 3 presents the crucial data in Moba Yoruba and exemplifies the differences between Standard Yoruba and Moba with respect to the domain for RTR harmony. A detailed discussion on the implications of the patterns seen (or not seen) in enclitics follows this. Chapter 4 offers the arguments for prosodic structure in Yoruba from Ola (1995). OT constraints are posited that formally define prosodic constituents. Two basic hypotheses are stated that could account for the status of clitics within this prosodic constituency. These hypotheses are then evaluated based on evidence from the RTR and nasal harmonic domains. Some implications of this result on domains in other processes are discussed. Chapter 5 presents an analysis based on references to prosodic structure for Moba RTR Harmony. This analysis is an extension of the harmony-via-prohibition account and is extendable to the other three dialects already considered (Ife, Ekiti, and Standard Yoruba). Chapter 6 is the conclusion. 2 Chapter 2 - RTR Harmony in Standard Yoruba Vowel harmony in Yoruba is seen in all words.1 The effect is that vowels are forced to agree with respect to their tongue-root orientation. Section 2.1 summarizes the RTR harmonic pattern in Standard Yoruba. Archangeli and Pulleyblank (1989) provide a basic description of this pattern. OT analyses based on this description have been posited by Pulleyblank (1996), Bakovic (2000), and Pulleyblank (2002). Sections 2.2, 2.3, and 2.4 provide summaries of these analyses. Discussion is included that highlights the basic strengths and weaknesses of each account. The harmony-via-prohibition account fares better than the other two accounts. An account that is based on the ideas of the prohibition account can derive RTR harmony in three dialects of Yoruba (Ife, Ekiti, and Standard Yoruba). 2.1 RTR Harmony: The Basic Pattern Many accounts and discussions exist surrounding RTR harmony in Standard Yoruba (Archangeli and Pulleyblank 1989, 1994; Bakovic 2000; Bakovic and Wilson 2000; Orie, 2001, 2003; Pulleyblank 1996). The basic pattern of Yoruba vowel harmony (Archangeli and Pulleyblank 1989) indicates that the active harmonic value is RTR (or - A T R ) 2 for reasons that will be outlined shortly. Standard Yoruba exhibits an RTR contrast only in mid vowels. High vowels are always produced as ATR and low vowels are always produced as RTR. Additionally, there is a distinction between nasal and oral vowels. While high and low vowels contrast for nasality, the mid ATR vowels, e and o, and the mid RTR vowel e (IPA s) are invariably oral. The back RTR vowel o (IPA o) can occur as nasal, but only following a labial sound: it is an allophone of the low nasal vowel, which occurs elsewhere.3 The vowel inventory with respect to place of articulation for 1 The lone monomorphemic exceptions are a limited class of C V C V nouns that are loan words. These loan words also do not conform to the VCV-templatic requirements of Yoruba nouns. I assume that by virtue of being loan words, these words are not subject to the same set of constraints as are the vast majority of native Yoruba lexical items. 2 The ATR/RTR distinction is variously handled via privative features or a binary feature, +/- ATR in previous accounts. I assume privative features, although this choice is completely arbitrary for the purposes of this thesis. 3 As will be discussed in section 3.1, Standard Yoruba and Moba differ in that no such allophonic variation is seen in Moba Yoruba. 3 Standard Yoruba as illustrated in Archangeli and Pulleyblank (1989) is shown below (tone is not included here).4 (2) Yoruba Vowel Inventory Oral Vowels front back ATT? i u high f\ 1 I V e o T?TT? e 0 mid K l K a low Nasal Vowels front back ATR in un high mid RTR (on) an low 2.1.1 Roots in Yoruba A root is generally defined as an element to which a morphological operation applies, such that that element cannot be analyzed further. The following two subsections explore tongue-root harmony in the class of nouns in Yoruba, which generally conform to a V C V template, minimally.5 Verbal roots, on the other hand, conform to a CV template. In this thesis, a distinction is made between V C V nouns that are analyzed as roots, and derived V C V nouns that consist of a prefix attached to a CV verbal root. In order for a V C V noun to be analyzed as morphologically complex, the following two criteria must be met: First, there must be a clear semantic relation between the CV verbal root and the V C V noun. Second, the CV verbal root must be independently attested as a bare root. When one of these two criteria is not met, I assume that the C V C noun in question is not morphologically complex and that it therefore constitutes a root. This results in a separation of V C V nominal roots and derived V C V nouns consisting of a prefix and a CV verbal base. This separation is illustrated in (3) below. 4 1 will use Yoruba orthographic conventions throughout: e = IPA [s], q = IPA [o], p = IPA [£p ] , s = IPA [J]; nasalized vowels are conventionalized as sequences of Vn - i.e. an = IPA [a] - there are no codas in Yoruba; There is a three-way tonal contrast: high tone = a, low tone = a, mid (unmarked) tone = a. Phonetic transcriptions will be placed inside square brackets when needed. 5 Cases of V C V C V nouns that are also analyzed as roots are discussed in section 2.1.5. 4 (3) Noun Complexity in Yoruba a. Morphologically Complex Deverbal nouns: V + C V de 'to hunt' ode 'hunter' ku 'to die' oku 'corpse of a person' b. Morphologically Non-Complex Nominal Roots: V C V ile 'house' le 'pursue' or 'drive away' or 'accompany'... (Delano 1969) ile 'land' or 'ground' le 'to be flexible' or 'stuck' or 'gummed' or 'to patch'... (Delano 1969) The alternative view, that all V C V nouns are morphologically complex, assumes that all V C V nouns consist of a CV verbal root with a prefix attached (Adetugbo 1967, Fresco 1970, Awoyale 1974, Akinkugbe 1978, Bakovic 2000). Under this view, the only kinds of roots are verbal CV roots. This assumes that language learners will generalize the pattern of V+CV derivation to form nouns, so that all nouns are composed this way, regardless of whether the CV verbal base is semantically related to an attested bare CV verb. On the other hand, if language learners interpret morpheme boundaries based on paradigm uniformity, it can be argued that at least some V C V nouns are not morphologically complex. Under this view, V C V nouns that are clearly related semantically to a given C V verbal root, as in (3a) above, would constitute evidence for a morpheme boundary. The pairing of a CV verb and a derived noun whose base is that CV verb constitutes a learnable paradigm. However, the V C V nouns in (3b) are not semantically related to any CV verb that could be posited as the base in an affixed form. Since there is no available semantically related base in the Yoruba lexicon, a learner must either posit an abstract base that is semantically related but that isn't attested elsewhere in the language, or a learner must simply conclude that the V C V noun constitutes an autonomous root. Under either of these two scenarios, one item must be introduced into the lexicon - either an abstract base or a V C V root. Since there is nothing to be gained by positing an otherwise unattested abstract base, I assume that a language learner would preferably analyze forms such as those in (3a) above as non-complex V C V roots. 5 2.1.2 Harmony in VCV Nouns The basic harmonic pattern concerning mid vowels in V C V nouns is outlined in this section. Agreement with respect to RTR is obligatory between adjacent mid vowels.6 This agreement is illustrated below in (4). (4) Mid Vowels: Contrastive ATR/RTR, harmonic triggers7 a. ewe ' leaf *ewe *ewe epo ' o i l ' *epo *epq ole 'thief *qle *ole owo 'money' *owo i *owo b. ese 'foot' *ese *ese eko 'pap' *ekq *eko obe i i 'soup' *obe *qbe oko i i 'vehicle' *okq *okd i While mid vowels exhibit a contrast for the feature RTR, high vowels are obligatorily ATR. They do not participate in RTR harmony, therefore. With respect to mid vowels, any mid vowel (ATR or RTR) can occur either preceding or following a high vowel.8 This results in both disharmonic sequences (5b and d), and in harmonic sequences (5a and c) of mid and high vowels. With respect to sequences of two or more high vowels, these are invariably ATR since Yoruba does not allow RTR high vowels. Therefore, any sequence of high vowels is necessarily harmonic (see (5e) below). (5) High Vowels: Obligatory ATR, no harmony a. ile 'house' igo 'bottle' 6 Throughout I refer to vowels separated only by consonants as adjacent. While this adjacency is not a case of absolute segmental adjacency, Gafos (1999), for example, argues that vowel gestures are actually articulatorily adjacent, even when intervening consonants are present. 7 The Standard Yoruba data in (4), (5), (6) and (7) are from Archangeli & Pulleyblank, 1989. 8 An independent constraint prohibiting word-initial u in Standard Yoruba is responsible for the absence of u-initial nouns in Standard Yoruba. This is used as an argument for prosodic structure in Ola (1995). 6 b. ile 'ground' itq 'saliva' c. eti 'ear' ori 'head' eku 'bush rat' oju 'eye' d. ebi 'guilt' okin i 'egret' ewu i 'clothing' orun • 'heaven' e. igi 'tree' inu 'stomach' isu 'yam' ilu 'town/city' As was the case for high vowels, there is no potential interaction between adjacent low vowels due to the obligatoriness of low vowels to also surface as RTR (see 6a below). In addition, there is no interaction between adjacent low and high vowels (in either order) with respect to RTR harmony (see (6b and c) below). This results in surface disharmonic sequences of low RTR vowels and high ATR vowels. Disharmony is tolerated in these cases in order to ensure that high vowels are invariably ATR and low vowels are invariably RTR. There are no ATR low vowels (represented below as a), as can be seen by the ungrammatical forms in (6b and c) below. (6) Low Vowels: Obligatory RTR a. aya 'chest' ara 'body' b. atu 'a type of cassava' ami 'sign' *smi c. iyan 'dispute' ika 'cruelty' *ika While these cases involving invariably ATR/RTR vowels do not exhibit harmony, the interaction of low vowels with mid vowels is one where harmony is seen. Low 7 vowels are unique in that they are obligatorily RTR 9 and they act as triggers of leftward (but not rightward) harmony. This pattern is illustrated below: (7) Low Vowels: Harmonic triggers a. ate 'hat' aro 'indigo' b. aje 'paddle' aso 'cloth' c. e p a ' groundnut oran i 'trouble' d. * e p a *6ran The above pattern exhibits the directionality of Yoruba RTR harmony. Since mid vowels are allowed to contrast for RTR following a low vowel but not preceding a low vowel, Archangeli and Pulleyblank argue that RTR harmony is strictly leftward. Example (7a) above would be ungrammatical if harmony were rightward. This leftward directionality also explains the ungrammaticality of (7d). In Archangeli and Pulleyblank (1989), RTR is seen as the active value because of the ban on ATR mid vowels preceding low vowels (7d). No such restriction exists on the distribution of A T R or RTR mid vowels with high vowels however. This indicates that while there is evidence for leftward spreading of RTR, there is no evidence for either leftward or rightward spreading of ATR. 2.1.3 Harmony in Derived V+ CV Nouns In the previous section, we saw that harmony applies within the root domain since it applies across adjacent vowels in a V C V root. However, harmony also applies across root-prefix boundaries in derived V+CV nouns. This is illustrated below for the agentive prefix in (8). This prefix is a mid, back, round vowel whose tongue-root value is determined via harmonic requirements. When this prefix attaches to a verbal base with an RTR vowel as in (8a), the prefix surfaces as RTR. However, when this prefix attaches to a verbal base with an ATR vowel as in (8b), the prefix surfaces as ATR. 9 One detail concerning the distribution of high and low vowels is not accounted for in the discussion in this thesis. No dialect of Yoruba allows a low ATR vowel to surface. In fact, this is a general property of the language family as a whole. However, there are numerous examples of high RTR vowels in the family. 8 (8) Derived V+CV Nouns a. de 'to hunt' i ode 'hunter' i i * o d e b. ku 'to die' 6ku 'corpse of a person' *6ku There are no cases where an ATR mid vowel prefix or an RTR mid vowel prefix can surface disharmonically. The tongue-root value of a prefix is determined completely as a result of vowel harmony, and not of faithfulness then: Therefore underlying values of mid-vowel prefixes are irrelevant. 2.1.4 Disharmony in Compounds Compound words are composed of more than one root. In Yoruba, compound words present disharmonic sequences of mid vowels. This is illustrated in (9) below. We find disharmonic sequences of both RTR mid vowels followed by A T R mid vowels and ATR mid vowels followed by RTR mid vowels. (9) Compounds are Disharmonic a- se i i 'to change' o w o 'money' s e w o i i 'to change money' b. e w e 'leaf o b e i i 'soup' e w e b e i 'any pot herb used for making soup' This disharmonic pattern in compounds is explained given that the domain for tongue root harmony is limited to either the root or the word. Since we have seen in the previous section that prefixes are included in the harmonic domain, this rules out the root. We can account for the lack of harmony across root-boundaries in compounds by positing that the root and word are aligned (via alignment constraints - this is discussed in more detail in chapter 4). By restricting the harmonic domain to the word, disharmony is allowed across root-root boundaries in compound words. 9 2.1.5 Harmony in VCVCV Nouns Archangeli and Pulleyblank (1989) further argued for a lexical specification of RTR that is linked to the right edge of a morpheme. This association convention derives the asymmetric pattern seen in the class of trisyllabic monomorphemic roots with medial high vowels, an example of which is shown below: (10) Harmonic in V C V C V Nouns a. od ide 'Grey Parrot' b. * qd ide c. * od ide As exemplified in (10) above, RTR mid vowels are aligned with the right edge of the morpheme, when possible. There are no monomorphemic cases where a high vowel is flanked by two RTR mid vowels, one on either side as in (10c). This is a case of opacity in vowel harmony: High vowels are opaque to RTR harmony in Standard Yoruba since they block transmission of the harmonic feature. Further, when a mid-high-mid root does carry an RTR value, it is realized only on the final mid vowel and never on the initial mid vowel, as is seen in (10b). Cases of imperfect right-alignment occur when a final vowel (or a sequence of final vowels) is high. In these cases, the RTR feature associates with the rightmost mid vowel instead. This is exemplified in (11) below, where the rightmost mid vowel is the only mid vowel. This is essentially the same pattern as was seen in (5d) above (repeated as (lib)) with the exception that the final two vowels are high. (11) Imperfect Right-Alignment of RTR a. ewir i 'bellows' b. ebi 'guilt' The above examples illustrate that the morpheme-level RTR feature is always associated with a non-high vowel, when present (Archangeli and Pulleyblank 1989). Additionally, there is a preference for orientation with the right edge rather than the left edge. The prediction is that RTR mid vowels that do not precede a low vowel bear root values of RTR. In Archangeli and Pulleyblank (1989), a distinction is made concerning two types of RTR association then. On the one hand, redundantly assigned RTR is found on low vowels (when they aren't associated with a root value of RTR) and on the other hand a root value for RTR is seen on rightmost non-high vowels in those roots whose lexical specification includes this RTR value. While both trigger harmony, only one of these is subject to right-alignment with the root - the root value for RTR. This root-value specific right-alignment allows redundantly assigned RTR values on low vowels to 10 escape the restrictions of right-alignment according to Archangeli and Pulleyblank (1989). Since these redundant values are not root-values, they are not subject to this right-alignment requirement. 2.2 Harmony via Alignment 2.2.7 Basic Constraint Ranking for Alignment Pulleyblank (1996) analyzes this alignment effect described in section 2.1 in Optimality Theory (Prince and Smolensky 1993) by using constraints of the form ALIGN(Catl, Edgel, Cat2, Edge2) (McCarthy and Prince 1993).10 In Pulleyblank's analysis, right- edge-alignment of RTR refers to the root domain, such that a single underlying root value of RTR aligns to the right edge of the root. This is accomplished via the constraint ALIGN(RTR, R, ROOT, R). In order to drive leftward harmony, a similar alignment constraint must be active that refers to the left edge of the RTR harmonic domain. The category that the left-edge- alignment constraint refers to is not the ROOT, however. Pulleyblank (1996) proposes the prosodic word as the category for left-alignment. The harmonic behaviour of the prefixes in Standard Yoruba as shown in section 2.1.3 suggests that the RTR-harmonic domain is the prosodic word and not the root. Recall from example (8) above, repeated as (12) below, that the agentive prefix harmonizes with the base it attaches to. (12) Prefixes Harmonize With Base a. de 'to hunt' o d e 'hunter' i i * o d e b. ku 'to die' 6ku 'corpse of a person' *6ku Left-edge-alignment of the RTR harmonic domain is relative to a category that includes this prefix then, the PrWd being one such category. The constraint, ALIGN(RTR, L, PrWd, L) enforces left-alignment of the RTR value with the prosodic word. In order to drive harmony, both alignment constraints must dominate DEPLINK- RTR and M A X - A T R . Additionally, MAX-RTR must dominate DEPLINK-RTR, so that For formal definitions for all constraints used in this thesis, see Appendix A. 11 alignment isn't satisfied by simply deleting the RTR value altogether. This is captured by the ranking in (13) below. (13) RTR Harmony via Alignment ALIGN(RTR, R, ROOT, R), ALIGN(RTR, L, PrWd, L), M A X - R T R » DEPLINK-RTR , M A X - A T R Tableau (14) below illustrates that this ranking enforces right-alignment of an RTR value. Candidate (14a) fatally violates ALIGN(RTR, R, ROOT, R) and candidate (14c) fatally violates MAX-RTR. Candidate (14d) fatally violates ALIGN(RTR, L, PrWd, L). The optimal candidate (14b) satisfies both alignment constraints and it retains the underlying RTR value. (14) Right-Alignment Enforced in V C V Nouns / o b e / ALIGN(RTR, R, ROOT, R) ALIGN(RTR, L, PrWd, L) MAX-RTR DEPLINK-RTR M A X - A T R a. obe i *! ®° b. obe ^^^^^^^^^^^ c. obe *! Z'K d. obe *! v ' 4\ *-, - This outcome is true regardless of where the RTR value is linked in the input. This is demonstrated in (15) below. Again, candidate (15b) is selected optimally for the same reasons as in (14) above. The alignment-based account posits a root-value of RTR that does not need to be linked anywhere in the input. Output constraints on alignment and faithfulness (and markedness in the case of low and high vowels) determine the location of the RTR value on the surface. (15) Left-Alignment Enforced in V C V Nouns / o b e / i ALIGN(RTR, R, ROOT, R) ALIGN(RTR, L, PrWd, L) MAX-RTR DEPLINK-RTR M A X - A T R a. obe *! * ^ b. obe • i c. obe *! d. obe *! i i 12 Finally, as mentioned above, low vowels are always RTR in Standard Yoruba. This fact is captured by ranking another grounding constraint, LO/RTR 1 1 above DEP-RTR. (16) Low Vowels are Invariably RTR / a t e / LO/RTR DEP-RTR a. ate *! «" b. ate This ranking forces low vowels to be produced as RTR, regardless of the underlying value for ATR/RTR. This ranking holds in all dialects of Yoruba - there are no known cases where advanced low vowels can occur.12 2.2.2 Dialectal Variation: High Vowels in Ekiti Yoritbd Pulleyblank's (1996) analysis is also able to account for other Yoruba dialects that differ in their patterns of RTR harmony. For example, the Ekiti dialect13 exhibits a harmonic system where high vowels are targeted in RTR harmony (Orie 2003). Pulleyblank accounts for high vowel opacity in Standard Yoruba by ranking HI/ATR 1 4 above ALIGN(RTR, L, PrWd, L) so that high vowels are always produced with ATR regardless of the pressure to left-align the RTR feature This is illustrated in (17) below. (17) High Vowels are Invariably ATR / i l e / HI/ATR ALIGN(RTR, L, PrWd, L) ®° a. ile i b. ile i i *! The Ekiti grammar would reverse this ranking such that ALIGN(RTR, L, PrWd, L) dominates HI/ATR, resulting in perfect left-alignment at the expense of incurring violations of the grounding constraint, HI/ATR. Additionally, M A X - R T R must dominate HI/ATR in order to rule out a candidate that satisfies ALIGN(RTR, L, PrWd, L) by deleting the RTR value. This is illustrated in the ranking in (18) below. 1 1 The constraint, LO/RTR is grounded in acoustic and articulatory enhancement relations (Archangeli & Pulleyblank, 1994). 1 2 This case is partially exhibited in Wolof, where short (but not long) low vowels can contrast for ATR/RTR (Pulleyblank, 1996). 1 3 Moba is actually a subdialect of Ekiti. 1 4 The constraint, HI/ATR is grounded both articulatorily and acoustically by enhancement relations between A T R and +high (Archangeli & Pulleyblank, 1994). 13 (18) RTR Harmony in Ekiti Yoruba / o d i d e / ALIGN(RTR, L, PrWd, L) MAX-RTR HI/ATR a. od ide * i * ®° b. od ide i i i c. od ide d. od ide *! 2.2.3 Dialectal Variation: Relative Alignment in Ekiti and Ife Yoruba Another feature of Pulleyblank's analysis is that there are two types of alignment effects seen in the various dialects of Yoruba. One type, absolute alignment, requires an underlying RTR feature to be aligned to the right edge of the root; otherwise, it will not surface at all. This case is exhibited in the Ife and Ekiti dialects of Yoruba (Orie 2001, 2003) as well as in Wolof (Pulleyblank 1996). On the other hand, relative alignment requires an underlying RTR feature to be aligned to the rightmost available segment. This is the case seen in Standard Yoruba. In relative alignment, it is more important to retain the underlying RTR feature at the expense of imperfect alignment. On the other hand, in cases of absolute alignment, an underlying RTR feature is deleted if the edgemost vowel is not a potential anchor due to higher-ranking constraints (such as HI/ATR). This situation is illustrated by Orie (2003) using the following examples: (19) Relative Alignment (SY) vs. Absolute Alignment (Ife and Ekiti) SY Ife and Ekiti a. ebi ebi 'guilt' ewir i ewir i 'bellows' b. eb i ebi 'hunger' ekuru ekuru 'food made of beans' The neutralization of an ATR/RTR contrast in mid vowels preceding a high vowel in dialects with absolute alignment is seen in (19a) above. On the other hand, this contrast is preserved in Standard Yoruba, where relative alignment is exhibited. Pulleyblank's analysis accounts for absolute alignment by ranking ALIGN(RTR, R, ROOT, R ) » MAX-RTR. This ranking states that alignment of the RTR feature to the right edge must be satisfied even if it means deleting a lexically specified RTR feature. The reverse case, relative alignment, is captured by reversing this ranking such that M A X - R T R » 14 ALIGN(RTR, R, ROOT, R). In this case, an underlying RTR feature is retained even though it might be impossible to align it perfectly with the right edge. The alignment violation is tolerated in order to preserve the underlying RTR feature. 2.2.4 Problems with Alignment: Gradient versus Categorical Constraints An important side issue here concerns the nature of evaluation of alignment constraints. Alignment must be gradiently evaluated in order for this account to succeed. First, consider a categorically evaluated alignment constraint. In this case, it is impossible to enforce harmony if the leftmost vowel is high. This is illustrated in (20) below. An RTR root of the form high - mid - mid would surface with the RTR right-aligned to the root (to satisfy the root alignment). However, perfect left-alignment is impossible due to undominated HI/ATR. Candidate (20d) is ruled out as a result. Additionally, since a categorical ALIGN(RTR, L , PrWd, L) constraint cannot discern between misaligned candidates (20a) and (20b), the former would be selected for faithfulness reasons. This is because it incurs one fewer DEPLINK-RTR violation (Archangeli and Pulleyblank 1994). Note also that M A X - R T R must outrank ALIGN(RTR, L, PrWd, L) in order to rule out candidate (20c), where the RTR value is deleted. (20) Categorical Alignment fails to Drive Harmony / iCeCe/ HI/ATR M A X - R T R (CAT)ALIGN(RTR, L, PrWd, L) DEPLINK-RTR i M A X - A T R «• a. iCeCe * b. iCeCe c. iCeCe *! : d. iCeCe i i i *'! llllilllR^ However, utilizing a gradient alignment constraint instead, it is possible to achieve leftward harmony without the requirement of perfect left-alignment. This is demonstrated in (21) below. Candidate (21a) fatally violates the gradient alignment constraint since the RTR value is misaligned by two segments. Candidate (21b) is selected optimally since the RTR value is misaligned by a single segment. 15 (21) Gradient Alignment Succeeds in Driving Harmony / i C e C e / i HI/ATR M A X - R T R (GRAD)ALIGN(RTR, L, PrWd, L) DEPLINK-RTR j M A X - A T R a. iCeCe i ®= b. iCeCe i i * c. iCeCe *! d. iCeCe i i i *! While a gradient constraint is able to capture the left-alignment effect, McCarthy (2003) offers arguments against gradient constraints based on cross-linguistic facts. McCarthy argues that all effects that have been analyzed using gradient constraints can be captured via some categorical alternative. Additionally, there are cases where gradient constraints are problematic. The proposal then, is to do away with gradient constraints completely, in favour of categorical alternatives. If we take McCarthy's point at face value, then this amounts to evidence against using left-alignment to drive RTR harmony in Yoruba. Likewise, this amounts to evidence against a gradient alignment constraint that drives relative right-alignment in Standard Yoruba. Since this alignment-based analysis relies on a gradiently evaluated constraint, it will be superceded by two analyses that will be presented in the following two sections that do not use gradiently evaluated constraints. 2.2.5 Treatment of High Vowels: Opacity and Transparency Opacity of high vowels, as outlined above, is captured in Pulleyblank's account first by ruling out gapped configurations where a harmonic value can skip a potential linking site Such a representation is argued to violate precedence relations because the medial vowel both follows and precedes an RTR mid-vowel (Archangeli and Pulleyblank 1994). This situation is illustrated in (22) below. (22) Gapped Configurations Violate Precedence Relations a dij j de k V precedes ' j ' which also precedes 'k' RTR^k ATRj ' j ' and 'k' both precede T Since these mid vowels are linked to a single RTR feature, this would imply that the medial vowel ('j') both precedes ('k') and follows ( V ) that RTR feature. This situation is argued to be phonetically uninterpretable and therefore linguistically ill 16 formed. GEN would not provide structures such as these for EVAL, and therefore they are ruled out as possible representations of surface forms.15 Once gapped configurations are ruled out, the high-ranking of HI/ATR ensures that a medial high vowel flanked by two mid vowels surfaces as ATR. One situation that could result in an RTR feature appearing on the initial mid vowel involves a single underlying RTR feature that is right aligned with the root (as would be expected) and a second RTR feature that is inserted onto the initial mid vowel. This case is ruled out in Pulleyblank's analysis by ranking DEP-RTR » ALIGN(RTR, L, PrWd, L), such that the insertion of an extra RTR feature incurs a violation of the higher-ranked DEP-RTR constraint. This ranking would be reversed in languages where 'transparency' of high vowels is observed, such that ALIGN(RTR, L , PrWd, L) » DEP-RTR. This situation is exemplified in Ife Yoruba (Orie 2001, 2003) and in Wolof (Pulleyblank 1996). However, given the formal definition of generalized alignment that is presented in McCarthy and Prince (1993), this ranking does not actually derive transparency since insertion of an RTR feature on the leftmost mid vowel does nothing to improve the alignment of the rightmost RTR feature with the left edge of the word. Both the opaque and the transparent forms would incur two violations (one for each of the two vowels that separates the rightmost RTR mid vowel from the left edge of the word). Therefore, ALIGN(RTR, L, PrWd, L) is unable to differentiate between these two candidates. DEP- RTR would then optimally select the opaque candidate since it incurs one less violation, regardless of its ranking with ALIGN. This situation is exemplified below for generalized alignment: (23) Generalized Alignment Fails to Derive Transparency / od ide / , RTR ALIGN(RTR, L, PrWd, L) DEP-RTR a. od ide i i ** *! ^ b. od ide ** Candidate (23b) would be selected optimally under either ranking of DEP-RTR and ALIGN(RTR, L , PrWd, L) since the rightmost mid vowel is misaligned by two syllables in both candidates (thus tying with respect to ALIGN(RTR, L, PrWd, L)). However, the insertion in candidate (23a) violates DEP-RTR fatally. There is a sense in which candidate (23a) is better aligned with the left edge though through insertion of RTR on the initial mid vowel. Pulleyblank (1996) captures this via a slightly refined definition of alignment. The concept of local alignment is 1 5 Although see Ito, Mester & Padgett (1995) where NO-GAP is assumed to be a violable constraint, and not a property of GEN. 17 defined as alignment of a feature over its own local domain. In this case, a local domain refers to the leftmost possible edge, assuming that other instances of RTR to the left of the RTR feature in question are not included in this local domain. Therefore, the rightmost mid vowel in candidate (23a) incurs only a single violation of local alignment (due to the medial high vowel) since the leftmost mid vowel is not included in the same local domain. With this definition of alignment, the typology of transparent and opaque neutral vowels is attainable. (24) Local Alignment Derives Transparency / od ide / , RTR LOCALIGN(RTR, L, PrWd, L) DEP-RTR ^ a . od ide i i * •. . * . - , . . • b. od ide On the other hand, re-ranking LOCALIGN(RTR, L, PrWd, L) and DEP-RTR so that DEP-RTR dominates LOCALIGN(RTR, L, PrWd, L), opacity of high vowels is captured, as is the case in Standard Yoruba. However, a second situation can result in a surface pattern of transparency with high vowels. Forms containing two underlying RTR values need to be considered given that the richness of the base holds as a property of Optimality Theory (Prince and Smolensky 1993). The richness of the base hypothesis states that every possible input representation should result in the optimal selection of some attested surface form. As it stands now, a form containing two underlying RTR values would actually result in the optimal selection of a transparent candidate like (24a) regardless of the ranking of LOCALIGN(RTR, L , PrWd, L) and DEP-RTR. This is because unlike the case where the leftmost RTR is inserted, there are two RTR features present underlyingly. There is now no DEP-RTR violation at all, and since DEP-RTR does not differentiate between candidates like (24a) and (24b), LOC-ALIGN(RTR, L , PrWd, L) will regardless of the mutual ranking. This situation is exemplified below in (25). (25) Transparency via Multiple Underlying RTR Features / od ide / , RTR, RTR LOC-ALIGN(RTR, L, PrWd, L) i DEP-RTR ^ a . od ide i i * b. od ide ** i | In order to overcome this potential roadblock, Pulleyblank utilizes a constraint that militates against having two RTR values present in the output: the OCR However, this constraint potentially militates against attested trisyllabic forms that contain a lexically specified RTR feature and contain a sequence of an initial low vowel, a medial high vowel, and a final mid vowel (i.e.: a-i-e, RTR). Since the high-ranked constraint 18 LO/RTR requires that the low vowel be RTR, and the OCP requires that only one RTR feature be present in the output, the optimal surface form should straightforwardly associate this RTR feature onto the low vowel, leaving the final mid vowel ATR. However, forms that violate the OCP with the pattern a-i-e are attested in Standard Yoruba. The fact that these contrast with forms like a-i-e can only be explained by positing an underlying root-RTR value in the former case and no such root-RTR value in the latter case. Pulleyblank's solution then is to restrict the OCP to the root-domain. In doing so, the OCP only applies to underlying root-values of RTR and not to those values inserted on low vowels in order to satisfy LO/RTR. 2.2.6 An Alternative Alignment-Based Account: Prosodic Licensing This alignment-based account is expanded upon and modified in Orie (2003) in her analysis of Ebira vowel harmony and three dialects of Yoruba (Standard Yoruba, Ekiti and Ife). Orie adopts the basic account that Pulleyblank (1996) proposes with one notable exception: Right-edge alignment is replaced by a constraint that refers to prosodic licensing instead. The rightmost syllable is analyzed as the prosodic head in Standard Yoruba (Ola, 1995). A constraint is then formulated that licenses a single harmonic root value (an underlying RTR value) on the prosodic head. This constraint, LIC-PH, duplicates the effect of ALIGN(RTR, R, ROOT, R) since all roots have final vowels that are prosodic heads. However, it has the advantage that it is necessarily categorical, and it therefore avoids the problems that gradient constraints have. On the other hand, this account does use the gradient constraint, ALIGN(RTR, L, PrWd, L) and so it only partially addresses this issue. In conclusion, whether we use ALIGN(RTR, R, ROOT, R) or LIC-PH, we obtain the same result - the root value of RTR is at the right edge of the root. The account Orie proposes utilizes prosodic constituency instead of morphological constituency to capture the harmonic effects. 2.3 R T R Harmony via Stem-Control 2.3.1 Basic Stem-Control in Standard Yoruba An account of vowel harmony systems has been proposed by Bakovic (2000) that does not refer to featural alignment. In this account, harmony systems are of two types both of which are driven by a constraint16 that incurs violations of adjacent disharmonic segments with no inherent directionality (AGREE). First, there are stem-controlled systems where the harmonic value in the vowel of the stem of affixation controls the harmonic values of an affixed form. This is enforced by setting up an output-output correspondence relation 1 6 Here and elsewhere, see Appendix A for formal constraint definitions. 19 between the segments in a stem of affixation and those in its corresponding affixed form. Faithfulness constraints (SA-IDENT-F) that refer to this correspondence enforce identity between corresponding segments in the base and the affixed form. These are constraints enforcing identity between two related output forms (Benua 1995, McCarthy 1995, Burzio 1996). The second type of language is of the dominant-recessive type. In this type of language, the unmarked value of the feature is dominant in the sense that regardless of morphological constituency, this value is the trigger for harmony. This dominant-recessive type is not relevant in the discussion of Yoruba RTR harmony, and will therefore not be discussed further. Bakovic's account of Standard Yoruba is based on identity between stems and their corresponding affixed forms, where it is assumed that all words with multiple vowels are morphologically complex. In these cases, only the final vowel is considered a stem vowel and all preceding vowels are assumed to be prefixal vowels. In this account, directionality is ultimately controlled by the stem vowel, resulting in leftward RTR harmony since there are prefixes but no suffixes in Yoruba. Stem-affixed form identity is enforced more strictly than the AGREE(ATR) constraint. This ranking accounts for the attested disharmonic low-mid vowel sequences, as is seen in tableau (26) below.17 Candidate (26a) is selected optimally since it satisfies the higher-ranking constraint, SA- IDENT(ATR). Candidate (26b) fatally violates SA-IDENT(ATR) even though it fares better with respect to the lower-ranked constraint, AGREE(ATR). (26) Stem-Affixed Form Identity Dominates AGREE: Low-Mid Sequences Stem: [Ce] /aCe/ SA-ID(ATR) AGREE(ATR) «• a. aCe b. aCe *! On the other hand, mid-low sequences differ in Yoruba in that they must agree. This follows assuming the same ranking shown in tableau (26) above. Since the stem vowel is a low vowel, it is required to be RTR due to both high-ranking markedness constraints and SA-IDENT(ATR). However, AGREE(ATR) would always optimally select the harmonic candidate, regardless of input specifications. This is seen in tableau (27) below, where candidate (27b) is optimally selected since it satisfies both SA- IDENT(ATR) and AGREE(ATR). While candidate (27a) satisfies SA-IDENT(ATR), it incurs a single fatal violation of the lower-ranked constraint, AGREE(ATR). Note that low vowels are required to be RTR due to high-ranking markedness constraints. Therefore, I do not consider a candidate with an initial low, ATR vowel in tableau (26). 20 (27) Stem-Affixed Form Identity Dominates AGREE: Mid-Low Sequences Stem: [Ca] / e C a / SA-ID(ATR) AGREE(ATR) a. e C a *! ^ b . e C a The above tableaux (26) and (27) exhibit a system of harmony that is stem- controlled. The right-to-left direction of RTR harmony in Yoruba results due to the morphological structure of V C V nouns in Yoruba. The final vowel is the stem vowel and is therefore subject to the high-ranked faithfulness constraint, SA-IDENT(ATR), while initial 'prefixal' vowels are not subject to this constraint. AGREE forces 'prefixal vowels' to harmonize with 'stem vowels.' Note that if there were instances in Yoruba where suffixes occurred, we would expect to see left-to-right tongue-root harmony triggered by the stem vowel onto the suffix vowel in these cases. However, since Yoruba is a strictly prefixing language, harmony appears to be invariably right-to-left. 2.3.2 Treatment of High Vowels: Sympathy Theory The constraints forcing low vowels to be RTR and high vowels to be A T R are undominated. This breaks down however when considering words with final high vowels (root high vowels) that can never be produced with RTR due to the high-ranking restriction on [+high] co-occurring with RTR. The data in (19) above demonstrate that vowels other than the root vowel can introduce RTR into a morpheme. It is possible for RTR to occur on a prefixal mid vowel that precedes the high root vowel in violation of the high-ranked constraints, SA-IDENT(ATR) and AGREE(ATR). Given an input with an RTR feature anywhere, Bakovic points out that the above ranking for stem-control would optimally select the fully harmonic ATR candidate over the actual surfacing form that preserves this input RTR. The following tableaux from Bakovic (2000) illustrate this problem (the stem referred to in the SA-ID(ATR) constraint is given in the line above the tableau). 21 (28) Stem-Control Fails to Allow Relative Alignment Stem: [Ci] /eCi/ HI/ATR SA-ID(ATR) AGREE(ATR) IO-ID(ATR) ^ a. eCi #4b. eCi i *! c. eCi *! Stem: [Ci] /eCi / i HI/ATR SA-ID(ATR) AGREE(ATR) IO-ID(ATR) ^ a. eCi «*b. eCi *! * ' in" c. eCi *! No possible input could result in a surface form with relative alignment of the RTR feature (candidate 28b). However, this pattern is attested in Standard Yoruba. In order to solve this problem, Bakovic uses constraints posited in McCarthy's (1999) sympathy theory. The general idea here is that candidate (28b) is more faithful to a sympathy candidate where the high vowel is actually specified as RTR. This sympathy candidate is chosen by a single selector constraint, which is undominated only for the purposes of selecting a sympathy candidate. The selector constraint in this case is ROOT-IDENT(ATR). The sympathy candidate is defined as the optimal candidate with the additional requirement that it must satisfy the selector constraint. The following tableaux (179 and 189 from Bakovic 2000) illustrate how a sympathy candidate is selected.18 Following McCarthy (1999), the sympathy candidate is denoted with the symbol, ©. The selector constraint is denoted with the symbol, * . Those candidates that obey the selector constraints are denoted by the symbol, S. 22 (29) Selection of the Sympathy Candidate (actual output = [qmu]) Stem: [mu | / omu / 10- ID(HI) HI/ATR 1 9 SA- ID(ATR) • RT-ID (ATR) AGR(ATR) IO-ID(ATR) © a. o m u i * * b. o m u * *! c. o m o i i *! Eî BSiiiSIII • d. o m u *! . . . . . ... e. o m u *! Stem: [mu / o m u / IO-ID(HI) HI/ATR SA-ID(ATR) • RT-ID (ATR) AGR(ATR) IO-ID(ATR) © a. o m u * b. o m u * c. o m o • i *! d. o m u *! e. o m u i *! ** Note the ranking of IO-ID(HI) above HI/ATR. This ranking is crucial in preventing high vowels from satisfying the featural co-occurrence constraint, HI/ATR by changing the value of [+HI] rather than the ATR value. Thus, candidate (29c) which violates IO-ID(HI) is non-optimal as the sympathy candidate. Candidates (29d) and (29e) are only ruled out because of the role of the selector constraint, *RT-ID(ATR). Candidate (29d) would otherwise be selected optimally since it does not incur any violations of the higher ranked constraints, IO-ID(HI), HI/ATR and SA-ID(ATR). However, since the sympathy candidate must satisfy the selector constraint (*RT- ID(ATR)), candidate (29d) is ruled out. The sympathetic candidate is then selected from those candidates that do satisfy *RT-ID(ATR). Since candidate (29c) violates the high- ranked constraint, IO-ID(HI), it is ruled out. Candidates (29a) and (29b) tie on all of the higher-ranked constraints. It is AGREE(ATR) that militates against candidate (29b) in favour of the sympathy candidate (29a). Having defined how a sympathy candidate is selected, a correspondence between the segments in the sympathy candidate and those in the actual output can be referred to. Bakovic uses the constraint ©-IDENT(ATR) as the faithfulness constraint that enforces Bakovic (2000) denotes HI/ATR as *[+Hl, -ATR] in his account. 23 identity between segments in the sympathy candidate and the actual output. Bakovic's ranking argument, shown in tableau (30) below, for ©-IDENT(ATR) is as follows: The actual output, qmu, is unfaithful to the sympathetic candidate only in that the high vowel is ATR, and not RTR. Therefore, IO-ID(HI) must outrank ©-IDENT(ATR) in order to rule out candidate (30b). The sympathetic candidate (30a) violates both HI/ATR and SA- IDENT(ATR). Therefore, at least one of these constraints must dominate ©- IDENT(ATR), in order to rule out candidate (30a). On the other hand, AGREE(ATR) is violated in the optimal candidate (30c) in order to avoid a second violation of ©- IDENT(ATR). Candidate (30d) satisfies AGREE(ATR), but incurs an extra violation of ©-IDENT(ATR). ©-IDENT(ATR) must then dominate AGREE(ATR). The extra violation of ©-IDENT(ATR) in candidate (30d) is thus fatal, and candidate (30c) is selected optimally. This ranking results in the attested surface pattern in Standard Yoruba as is shown in the tableaux below (from 181 and 182 in Bakovic 2000). (30) Sympathetic Faithfulness Succeeds Stem: [mu] /omu/ i IO- ID(HI) HI/ A T R SA- ID(ATR) • RT-ID j ©- (ATR) i ID(ATR) AGR (ATR) IO- i n r v n o © a. omu i *! b. omu *! • c. omo i i *! V d. omu * I ** | ## lllllllltyî Siiyiflt ^ e. omu i * I * Stem: [mu] /omu/ IO-ID(HI) HI/ A T R SA- ID(ATR) • RT-ID ! ©- (ATR) j ID(ATR) AGR (ATR) IO- ID(ATR) © a. omu i *! * b. omu *! * ^ ^ ^ ^ ^ ^ ^ c. omo i r *! d. omu * , * * 1 ^ e. omu i •i- ; * In both cases above, the correct candidate, (30e), is selected once sympathetic faithfulness is introduced. The stem-control theoretic account successfully captures the surface pattern seen in Standard Yoruba RTR harmony (in disyllabic words). 24 2.3.3 Problems with Stem-Control Theory: Dialectal Variation in the Behaviour of High Vowels While it can account for the facts of Standard Yoruba RTR harmony, one major problem with stem-control theory is its inability to extend to accounts for the patterns in Ife and Ekiti Yoruba. First, it is shown that opacity in Standard Yoruba is captured by the ranking presented in the previous section. Second, it is demonstrated that a simple re- ranking can account for the difference between a language where absolute right- alignment is seen (i.e. Ife Yoruba) and one where relative right-alignment is seen (i.e. Standard Yoruba). Next, it is demonstrated that the ranking for absolute right-alignment results in neutralization of an ATR/RTR contrast non-finally, a situation that does not allow for transparency of high vowels. Further, it is demonstrated that there is no possible ranking that would result in transparency of high vowels. Additionally, in Ekiti Yoruba where high vowels participate in RTR harmony, there are no cases where a high vowel actually triggers RTR harmony. Therefore, a successful analysis of these facts should not posit a high RTR vowel that triggers harmony in of Ekiti Yoruba. The stem- control account, however, posits a sympathy candidate containing a final high RTR vowel that is responsible for 'triggering' leftward RTR harmony on the surface, thus predicting the presence of a dialect where a final high vowel can trigger RTR harmony. The fact that there is no such dialect remains unexplained then under a stem-control account. Standard Yoruba exemplifies a situation where high vowels are opaque. The stem-control account captures this effect given the above constraint ranking. Tableau (31) below illustrates this using a hypothetical input where all three vowels are underlyingly RTR. In this case, since the input contains a root with a single RTR vowel, the selector constraint, *ROOT-lDENT(ATR) is only satisfied by a candidate that retains this RTR value. The sympathetic candidate is chosen from the four candidates (31a, 31b, 31c, and 31e) that satisfy the selector constraint. Two of these candidates (31a and 31e) violate the high-ranked markedness constraint, HI/ATR, and are thus ruled out in the selection of the sympathy candidate. Of the remaining two candidates, candidate (31c) fares better with respect to AGREE(ATR) and is thus selected as the sympathy candidate. In the actual selection of an optimal candidate in (31) below, candidates (31a), (3 Id) and (31e) are ruled out due to fatal violations of the high-ranked constraints HI/ATR and SA-IDENT(ATR). The remaining two candidates, (31b) and (31c) differ in that (31c) satisfies ©-IDENT(ATR) while (31b) incurs a single violation of ©- IDENT(ATR). Therefore, the opaque candidate (31c) is selected optimally. 25 (31) Opacity of High Vowels: ©-IDENT(ATR) » AGREE(ATR) Stem: [dide] / od ide / • t IO- ID(HI) HI/ATR SA- ID(ATR) • RT-ID i ©- (ATR) i ID(ATR) AGR (ATR) IO-ID (ATR) a. od fde i i *! ' ^ V -v ; '"' * * b. od ide • i S : *! ®° • c. od ide S : fcis|lf!!llli|l d. od ide *! e. od ide i *! '€ ' 'vf Note that re-ranking AGREE(ATR) above ©-IDENT(ATR) does not affect the result in tableau (31) above. The same opaque candidate ((31c) and (32c)) is optimal regardless of the mutual ranking of AGREE(ATR) and ©-IDENT(ATR). This is demonstrated in (32) below. (32) Opacity of High Vowels: AGREE(ATR) » ©-IDENT(ATR) Stem: [dide] / od ide / i i IO- ID(HI) HI/ATR SA- ID(ATR) • RT-ID (ATR) AGR (ATR) ©- ID(ATR) IO-ID (ATR) a. od ide i i *! • b. od ide * * i 4 it. ^ © c. od ide d. od ide *! e. od ide *! f> • &\' -"Si? However, this ranking affects the pattern seen in disyllabic words. Recall that within the stem-control account, sympathy theory was invoked to account for disyllabic words where a final high vowel was preceded by an RTR mid-vowel (see (30) above). A sympathy candidate is posited where the high vowel triggers leftward RTR harmony and the faithfulness constraint, ©-IDENT(ATR) enforces identity to this sympathy candidate. The high-ranked constraints ©-IDENT(ATR) and HI/ATR then enforce the optimal selection of a candidate where an RTR mid vowel precedes an A T R high vowel. If, however, AGREE(ATR) is ranked above ©-IDENT(ATR), the pressure to be faithful to the sympathy candidate with RTR harmony is replaced by the pressure to have perfect harmony. Since high vowels are invariably A T R due to the high-ranking of HI/ATR, the initial vowel is then forced to agree with the ATR value of a following high vowel. This is essentially a situation of neutralization of RTR non-finally. This situation is exemplified in (33) below. 26 (33) Neutralization of RTR Predicted Non-Finally: AGREE(ATR)» ©-IDENT(ATR) Stem: [mu] / o m u / 10- ID(HI) HI/ATR SA- ID(ATR) • RT-ID (ATR) AGR (ATR) •- ID(ATR) IO-ID (ATR) © a. o m u *! • b. o m u • *! c. o m u *! • •* • 18111111 ^ d. o m u * Candidates (33a) and (33c) are ruled out since they both violate the high-ranking constraint, HI/ATR. Candidate (33b) fatally violates AGREE(ATR) though, meaning the ATR harmonic candidate (33d) is selected optimally even though both vowels are underlyingly RTR. This is essentially a situation of positional neutralization where an ATR/RTR contrast can exist in a final non-high vowel. This situation is attested in Ife Yoruba (a situation of absolute alignment, where an underlying RTR feature is perfectly right-aligned or else it does not surface). However, Ife Yoruba exhibits transparent high vowels rather than the opaque high vowels that were predicted in (32) above.20 In order to allow transparency in Ife Yoruba, a constraint that favours candidate (32b) over candidate (32c) must dominate AGREE(ATR). The only constraint that does this in the stem-control account is IO- IDENT(ATR). However, we cannot rank this constraint above AGREE(ATR) without also preserving unattested non-harmonic sequences of mid-vowels. Therefore, the constraint set in the stem-control account is unable to extend to cases of transparent high vowels.21 The stem-control account then accounts for the situation of relative alignment with high-vowel opacity in Standard Yoruba. However it predicts a situation that is unattested in any dialect of Yoruba and fails to account for the pattern of transparency 2 0 The Ife form, odide (meaning 'Gray Parrot') corresponds to the Standard Yoruba form, odide. This exemplifies the difference between the two dialects: In Ife the initial mid vowel in such a sequence agrees with the final mid vowel (transparency), whereas in Standard Yoruba, the initial vowel agrees with the adjacent high vowel and is thus invariably ATR (opacity). However, note that in both cases the tongue-root value of the initial vowel is predictable, not contrastive. 2 1 Bakovic & Wilson (2000) use a targeted constraint, ->NO(+Hl, -ATR) to derive transparency under a stem-control theoretic framework. 27 that is seen in Ife Yoruba. The unattested situation, represented by the ranking of AGREE(ATR)» O-IDENT(ATR), is one where an ATR/RTR contrast can exist only in a final non-high vowel. In this situation, high vowels appear to be 'opaque' only because this contrast is neutralized elsewhere. While Ife Yoruba is similar in that it neutralizes an ATR/RTR contrast non-finally, it exhibits a pattern of transparency in high vowels. The stem-control account does not have the tools to enforce transparency however without the introduction of a targeted constraint (Bakovic and Wilson 2000). This is summarized in (34) below. (34) Summary of Predictions of Stem-Control Theory Language-Type / o m u / / o d i d e / • i Stem-Control: ®-ID[ATR] » AGREE(ATR) (Standard Yoruba) o m u od ide i Stem-Control: AGREE(ATR)» ©-ID[ATR1 o m u od ide Ife Yoruba o m u od ide One further dialectal issue is raised by Orie (2003), who argued for an alignment- based account as opposed to a stem-theoretic account. Of the problems raised by Orie with the stem-theoretic account in Bakovic (2000), the most relevant argument in the current discussion concerns the fact that no dialect of Yoruba allows RTR high vowels in root-final position. Even in Ekiti Yoruba, where high vowels can occur as RTR, this can only happen preceding an RTR mid or low vowel. 2 2 In other words, high vowels never act as harmonic triggers of RTR harmony in any dialect even when they can occur as RTR, implying that high vowels only ever are retracted by harmonic requirements, and not because of faithfulness. If we were to posit a candidate that has a final RTR high vowel that triggers harmony at any level (opaque levels included) in Standard Yoruba, we would expect that this candidate should surface in some dialect. The fact that Ekiti avoids this suggests that there is no level at which a high RTR vowel can trigger harmony. The cases in Standard Yoruba presented by Bakovic as evidence for underlying RTR high vowels (cases like qmu above where sympathy theory is invoked) are, however, consistent with the alternative analysis, where an RTR feature (wherever it is found on the input) cannot right-align due to the general ban on high RTR vowels. Imperfect or relative alignment as derived by Pulleyblank (1996) via ALIGN constraints would not have this problem. The constraint M A X - R T R is high-ranked That high vowels harmonize preceding RTR vowels, but not following them in Ekiti, lends further evidence to the fact that RTR harmony in Yoruba is strictly regressive. 28 enough in this account to prevent deletion of an underlying RTR feature.23 Rather than appeal to derivational opacity and sympathy theory to solve this problem, the surface form of o m u is simply avoiding a MAX-RTR violation. In fact, this account allows for an underlying RTR feature that is associated with a high vowel (it can either delete completely, yielding omu, or it can re-associate yielding omu). This is actually independently necessary due to cases of derived V C V forms where the final root vowel is a high vowel, and yet the initial prefixal vowel surfaces as RTR. This is illustrated in (35) below. (35) Disharmony in Derived V C V Forms with Final Vowels a. ku 'to die' 6ku 'corpse of a person' b. m u 'to drink' o m u 'drinker' In (35a), we see a harmonic sequence of an ATR prefixal vowel and ATR root vowel. However, in (35b), a disharmonic sequence of an RTR prefixal vowel and an ATR root vowel is seen. In both of these cases, it is clear that the V C V noun is a derived noun with the agentive prefix. However, recall from section 5.1.3 that prefixal vowels are only ever allowed to vary via harmonic requirements - they cannot contrast for ATR/RTR. The variation of the ATR/RTR value in (35) can be viewed as a variation not in the underlying values of the prefixes, but instead in the underlying values of the verbal roots. Since the verbal roots are high vowels, an underlying RTR feature cannot surface due the highly ranked markedness constraint, HI/ATR. If M A X - R T R dominates ALIGN(RTR, R, ROOT, R) though, it is possible to satisfy M A X - R T R by re-associating the RTR value onto the prefixal vowel. The main point here is that accounts that use M A X - R T R need not posit that RTR high vowels trigger harmony at some opaque level while the stem-theoretic account must. The cross-dialectal facts of Yoruba suggest that no high vowel triggers harmony, regardless of its underlying ATR value, thus suggesting that M A X - R T R is more appropriate in this case than a constraint enforcing identity with an unattested sympathy candidate. This assumes that there is only one RTR feature for M A X - R T R to preserve. Pulleyblank (1996) uses a version of the OCP, which outranks M A X - R T R in order to rule out surface forms with two distinct RTR features. 29 2.3.4 Problems with Stem-Control Theory: Morphological Structure An additional problem with the stem-control theoretic account is that it relies on the assumption that all V C V nouns are morphologically complex. Bakovic argues that since Yoruba is strictly prefixing, the initial V in these words is seen as a prefix and the stem- value of the rightmost vowel is dominant and triggers leftward harmony. The apparent leftward directionality follows from the morphological character of Yoruba rather than some arbitrary setting of a left/right directionality parameter. A strong argument can be made that there is no such morphological complexity to at least some nouns. Once a separation is established between derived V C V nouns and non-complex V C V nouns, it is impossible to account for the harmonic pattern seen in the latter case. As outlined in section 2.1.1, there are clear cases of morphologically complex agentive forms where an agentive prefix modifies a CV verbal base resulting in a derived noun (see (12) above). These cases are clearly applicable in the stem-theoretic framework. However, there is a clear division between this subset of derived nouns and the general class of V C V nouns in Yoruba. This division is illustrated in (36) below. (36) Noun Complexity in Yoruba a. Deverbal nouns: V + CV (repeated from (12) above) de 'to hunt' i o d e 'hunter' ku 'to die' 6ku 'corpse of a person' b. Nouns (general case): V C V ile 'house' le 'pursue' or 'drive away' or 'accompany'... (Delano 1969) ile 'land' or 'ground' le 'to be flexible' or 'stuck' or 'gummed' or 'to patch'... (Delano 1969) If we are to posit that both the nouns in (36a) and (36b) are complex, we must also posit a high vowel prefix i - in (36b) that derives 'house' from a verb to which it seems to have no semantic relation to. The differences between (36a) and (36b) are straightforwardly accounted for if we assume that forms in (36a) are morphologically complex, involving a 30 prefix that derives an agentive form from a CV verb. On the other hand, I assume that forms in (36b) are not morphologically complex since there is no semantic relationship between the CV verbal base and a potentially derived noun with the hypothetical prefix V . Additionally, the situation in (36b) introduces a major problem with respect to learnability. An output-output correspondence is capitalized on in order for a language learner to use a constraint like SA-IDENT in the first place. In cases such as (36a), there is language data available for both independently occurring stems and their corresponding affixed forms. The language learner simply pairs the stem with the affixed form and an output-output correspondence is set up so that SA-IDENT can now apply. However, there is no such data available in cases such as (36b) since there is no semantically related stem that is available for an output-output correspondence to be set up. In order to set such a correspondence up, a language learner must posit an abstract stem that does not actually occur independently. This is a problem since in formulations of output-output correspondence, the stem must be an independently occurring output form in order to enable a language learner to extract the necessary morphological pieces to set up the correspondence in the first place (Benua 1995, McCarthy 1995, Burzio 1996). Contrary to this, the stem-control account must posit that learners can set up correspondence relations between unattested stems and hypothetical affixed forms. This is a dangerous situation since it affords the language learner a large amount of freedom to posit abstract morphemes without overt evidence. Even if these arguments are ignored, stem-control theory would still not be distinguishing between the cases in (36a) and (36b). Moreover, at the very least, some distinction must exist between these types of nouns.24 The above discussion argues against morphological complexity in the V C V nouns in (36b). Taken seriously, this would amount to counter-evidence to the stem-control account in Yoruba. The pattern where an ATR/RTR contrast is permitted following a low vowel but only RTR vowels are found preceding a low vowel (see (7) above: ate, *eta) relies on this morphological complexity that is assumed in Bakovic (2000). Since the harmony-driving constraint, AGREE is non-directional, it is impossible to account for this directional effect with the stem-control-theoretic constraint set, once morphological constituency is removed. The prediction is that roots with a low vowel should not contain any ATR vowels as this would violate AGREE (if we are to have harmony at all 2 4 There are, however, cases where the prefix vowel is not predictable but the semantic relation is clear. For example, the word erg, which means 'machine' is semantically related to the verb rq, which means 'to make or manufacture.' The prefix, e-, has the same derivational function as the q- / o- prefix in (36a) above: it is a nominalizing prefix that gives the reading 'one who/that Xs, ' where X is the verbal base. 31 within a root). Once we do away with the morphological complexity, we are left with a dominant-recessive type system, where low vowels are essentially the dominant RTR harmonic triggers due to markedness considerations. However, this predicts that directionality should follow from this dominant/recessive relationship. This is clearly not the case, as was seen in (7) above: Directionality is fixed and is leftward (for whatever reason). Additionally, there is nothing guaranteeing the rightward orientation of RTR in mid-high-mid trisyllabic forms (as outlined in section 2.1.5 above, only the rightmost mid vowel exhibits an ATR/RTR contrast; the initial mid vowel in this configuration is invariably ATR). Therefore, if not all nouns in Yoruba are morphologically complex, we cannot maintain either a stem-control account or a dominant/recessive account of the facts of RTR harmony in Yoruba. 2.3.5 Towards a Prosodic Alternative One possible solution that 'repairs' this potential problem that stem theory encounters is to replace the reference to morphological constituency in favour of prosodic constituency. Ola (1995) argues for prosodic structure where patterns of deletion, truncation and reduplication are sensitive to prosodic constituents including syllables, feet and prosodic words. Rather than appeal to morphological structure, it would not be surprising for speakers to capitalize on independently defined prosodic constituents in processes like vowel harmony. The findings of Ola (1995) suggest that Standard Yoruba words are parsed into binary iambic feet. Additionally, it is concluded that onset-less vowels are not actually syllabified. Instead, these vowels are represented as nuclear moras that are licensed directly by the PrWd (they are not parsed into feet or syllables). This type of structure results in exactly the same type of right-headed inside-out structure that Bakovic proposed for the morphology. Therefore, given the problems stated in the previous section with the morphological approach, and given the inherent similarity between the morphological and prosodic structures in Yoruba, the natural solution for a language learner might be to capitalize on prosodic constituency instead. 32 (37) Parallelism of Morphological and Prosodic Structure in Yoruba Prosodic Structure (Ola 1995) Morphological Structure (Bakovic 2000) PrWd A.F. 2 N|uc V C V C V Pfx Pfx 1 K.F.-H(Stem to A.F. 2) V CV CV The fact that Yoruba is strictly prefixing was argued as evidence for analyzing all V C V nouns in Yoruba as containing a root and a prefix (Adetugbo 1967, Fresco 1970, Awoyale 1974, Akinkugbe 1978, Bakovic 2000). However, given that prosodic constituency is enforcing binary feet with a prosodic head on the right, and given that a root CV verb would tend to occupy this prosodic head position rather than an affix, we might expect Yoruba to be prefixing and not suffixing for prosodic reasons. Adding a prefix would fit the independent requirements of right-headed binary feet, whereas adding a suffix would either shift the prosodic head from the vowel onto the suffix or violate the binary footing requirement. A language learner might capitalize on prosodic constituency rather than morphological constituency in V C V C V words (and any word with more than one vowel). The rightmost vowel is analyzed as the prosodic head and could therefore be singled out in an OT analysis by a constraint referring to this position. An alternative account that utilizes prosodic domains is superior to the stem-control theoretic account. This is true because the morphological constituency that the stem- control account relies on must be posited at an abstract level for at least some nouns. However, prosodic constituency, of which there is independent evidence, would not involve any abstraction. The learnability issue can thus be addressed by focusing on prosodic constituency rather than morphological constituency. The possibility of an account using prosodic domains will be revisited in chapters four and five. 2.4 R T R Harmony via Prohibition 2.4.1 RTR Harmony via Prohibition in Standard Yoruba There is however another account based on agreement-type constraints proposed by Pulleyblank (2002). This account utilizes constraints of the form *FG. Recall that AGREE(F) as posited by Bakovic (2000) is inherently symmetric. Any sequence of 33 segments that do not agree for the feature F, incurs a violation. By postulating a constraint that takes into account the ordering of the features, on the other hand, a separation can be made between two types of AGREE violations: *FG and *GF. 2 5 This allows for a potential directionality effect depending on interactions with faithfulness and markedness constraints: Sequences of FG can be avoided, whereas sequences of GF might be maintained, if need be. The directionality can then be defined by the relative rankings of markedness constraints and faithfulness constraints. With respect to Yoruba RTR harmony, this is good news since the pattern seen in low vowels can now be accounted for without referring to morphological structure. The following ranking (from Pulleyblank 2002) is offered to account for RTR harmony in Standard Yoruba: (38) RTR Harmony via Prohibition LO/RTR, HI/ATR » [MAX-RTR] R X » *RTR-C 0 -ATR » *ATR 2 6 -C 0 -RTR » [MAX-ATR] R T This ranking will preserve RTR root values whenever possible (when there is a non-high vowel present) since [MAX-RTR] R T is highly ranked. As in the Alignment-based account (Pulleyblank 1996), this account utilizes M A X - F to ensure the preservation of a root RTR value.27 This is illustrated in (39) below. Candidate (39c) is ruled out because it violates the high-ranked constraint, HI/ATR. Candidate (39b) is ruled out because the root RTR feature is deleted in violation of [MAX-RTR] R T . The disharmonic candidate, (39a) is optimally selected then. Note that there is no need to specify the location of the underlying root RTR value. (39) [MAX-RTR] R T Preserves Root RTR Values /eb i / , [RTR] HI/ATR [MAX-RTR] R T *RTR-C 0 -ATR *ATR-C 0 -RTR ®*a. ebi i ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ b. ebi *! c. ebi *! 2 5 Constraints are formally defined in Appendix A. 2 6 Underlined features denote which feature is the locus of evaluation for the given prohibition constraint. One violation is incurred for every occurrence of the underlined feature value that meets the constraint's sequence condition. For example, the hypothetical sequence, e-e-e, incurs two violations of *RTR-o°-ATR, one for each ATR vowel. However, the constraint, *RTR-°°-ATR would only result in a single violation in an e-e-e sequence, since there is only a single RTR vowel. 2 7 M A X - F enforces retention of an underlying autosegment, F, in the output, but not necessarily retention of an underlying link to F in the output. It allows for re-association of an autosegment. 34 By ranking [MAX-RTR] R T and *ATR-C 0 -RTR above [ M A X - A T R ] R T , this enables leftward RTR harmony (as opposed to rightward ATR harmony, which is unattested since high vowels do not trigger spreading). In fact, the only way a root ATR value could surface on a mid vowel in a disyllabic word is if there is no root RTR value underlyingly. Recall that in Standard Yoruba there is a pattern of relative right-alignment of RTR values that is enforced. Sequences of mid vowels separated by a medial high vowel only allow an ATR/RTR contrast to occur in the final mid vowel. Since high vowels are opaque to harmony, the initial mid vowel must be ATR. This pattern of relative alignment is captured by ranking *RTR-C 0 -ATR above *ATR-C 0 -RTR and [MAX- A T R ] R T . This is illustrated in (40) below. Candidate (40a) fatally violates the constraint *RTR-C 0 -ATR. The optimal candidate (40b) re-associates the underlying RTR value to the right-edge in order to avoid a violation of *RTR-C 0 -ATR. This introduces violations of both *ATR-C 0 -RTR and [MAX-ATR] R T , but violations are tolerated since these constraints are lower ranked than *RTR-C 0 -ATR. (40) Relative Right-Alignment of RTR / o d i d e / *RTR-C 0 -ATR *ATR-C 0 -RTR [ M A X - A T R ] R T a. od ide *! ®° b. od ide * In order to account for the asymmetric pattern concerning low vowels, an additional condition needs to be introduced. The constraint *RTR-C 0 -ATR would drive rightward harmony from low vowels onto mid vowels if the ranking in (38) were left as is. As shown in (41) below, the harmonic candidate, (41b), is incorrectly selected as the optimal candidate since the constraint, *RTR-C 0 -ATR militates against the attested disharmonic candidate, (41a). Candidate (41c) is straightforwardly ruled out by the high- ranking LO/RTR constraint. (41) Low Vowels Incorrectly Predicted to Trigger Rightward RTR Harmony / a C e / LO/RTR * R T R - C - A T R [ M A X - A T R ] R T a. a C e *! b. a C e i c. a C e *! This situation is repaired by adding the condition that *RTR-C 0 -ATR only applies to pairs of non-low vowels. This is motivated insofar as the class of vowels that can 35 occur as ATR is exactly the class of non-low vowels. Additionally, Pulleyblank notes that two parameters seem to determine which segments are targeted both in OCP effects and in harmony, cross-linguistically. These two parameters are proximity and similarity. Proximity is built into the above constraint by allowing only intervening consonants (C 0; i.e. - adjacent vowels are targeted). The condition added to *RTR-C 0 -ATR is just one of similarity. The new constraint, *[RTR, NONLOl-Q-rATR. NONLOl would replace *RTR-C 0 -ATR in the above ranking and this would ensure that low vowels do not trigger rightward spreading of RTR. This is illustrated in tableau (42) below. (42) Low vowels prevented from triggering rightward RTR harmony / a C e / LO/RTR *[RTR, NONLO]-Q-TATR. NONLOl [ M A X - A T R ] R T ®° a. a C e b. a C e *! Another familiar problem concerns an input trisyllabic form with a medial high vowel flanked by mid vowels with two RTR root values. As in the alignment-based account, a form like this is ruled out by referring to the OCP. In the prohibition account, the OCP is unified with a *FG constraint, the only difference being that in an OCP- version, F=G. The constraint *RTR-oo-RTR 2 8 is invoked to prevent these OCP violations. As is illustrated in (43) below, the ranking, *RTR-QQ-RTR » [MAX-RTR] R X is needed to rule out candidate (43a) from surfacing faithfully. Recall from (41) above that the ranking of *[RTR, NONLOl-q- rATR. N O N L O l » *AT_R-C 0-RTR is needed to enforce right- alignment of the RTR feature, as in candidate (43b). Candidate (43c) is ruled out since the medial high vowel is preceded by an RTR mid vowel, thus fatally violating *[RTR, NONL01-G.-IATR. NONLOl. (43) OCP Prevents Multiple RTR Features From Surfacing / o d i d e / i i HI/ATR *RTR -oo-RTR [MAX-RTR] R T *[RTR, NONLO]- Cn-rATR. NONLOl *ATR-C 0 -RTR a. od ide • i *! ^ b. od ide i * liiiiiiiillii^^Bii c. od ide t * *! It is worth noting that this is the first point of departure from the stem-control account offered by Bakovic (2000), where the constraint set used, did not refer to autosegmental representations. The decision not to use M A X - F type constraints is what As in section 2.2, the OCP must refer to root-RTR values only, so that a-i-e sequences (for example) are allowed to surface. 36 buys stem-control this broader theoretical compatibility. AGREE refers to segmental adjacency and requires only that adjacent segments agree with respect to feature, F. There is no reference to any level of autosegmental representation. However, *RTR-oo- RTR crucially refers to RTR autosegments. If the OCP constraint were to apply segmentally, it would actually militate against RTR harmony (by virtue of its dominance of [MAX-RTR] R T ) . Taking this argument to its logical conclusion, we find in Yoruba a case where we need to refer to an autosegmental version of a non-local OCP. The constraint *X-oo-X is ranked above [MAX-RTR] R T to enforce opacity of high vowels in Standard Yoruba and Moba. This prevents forms with two RTR root values from surfacing on two mid vowels that flank a medial high vowel (i.e. - dwurq, *qwurq). However, it must not apply to adjacent RTR vowels since this would militate against RTR harmony. In order to avoid this, it must be the case that *X-oo-X (at least in Yoruba) refers only to autosegmental occurrences of X . By allowing *X-oo-X to evaluate segmental occurrences of X , this essentially becomes a segmental markedness constraint with the proviso that one value of X is allowed (in whatever domain *X-oo-X applies over). By ranking this markedness constraint above [MAX -RTR] R T , there is no motivation for harmony to occur. The proviso that one value of X can occur rescues us though if we refer to an autosegmental evaluation of *X-oo-X. This would allow a single autosegment (required to be as right- aligned as possible via other constraints) to spread onto adjacent non-high vowels, thus achieving harmony. As soon as a high vowel is reached, opacity is enforced since it is not possible to spread over this high vowel due to a NO-GAP condition that is either enforced in GEN (Archangeli and Pulleyblank 1994, Pulleyblank 1996, Gafos 1996; Nf Chiosain and Padgett 2001) or in CON (Ito et al. 1995). Therefore, it is necessary to refer to an autosegmentally defined OCP constraint in Yoruba. While there is no problem with either a segmentally defined or an autosegmentally-defined version of the OCP per se, it is worth noting that the spirit of the stem-control account (to be autosegmentally-free) is violated by the autosegmental version of the OCP that is needed in the prohibition account. The segmentally enforced OCP would be more in line with stem-control, but would fail in doing its intended job in Yoruba. Stem-control does not need to refer to any OCP-type constraint since the job of preventing forms like qwuro from surfacing is handled instead by virtue of the fact that the constraints SA-IDENT and AGREE dominate all of the IO-IDENT constraints. 2.4.2 RTR Harmony via Prohibition in Ife Yoruba While the prohibition account can explain the pattern of RTR harmony seen in Standard Yoruba, it remains to be seen how it would account for the patterns seen in other dialects. The following accounts are based on Pulleyblank (2002) with one exception concerning 37 the formulation of the general constraints enforcing faithfulness. I utilize MAXLINK-F rather than M A X - F constraints as the general constraints enforcing faithfulness in the analysis here and elsewhere. MAXLINK-F is posited to incur violations for every segment that has an underlying link to F but that does not surface with such a link. Evaluation of this constraint requires the learner to set up a correspondence between the segments in an abstract underlying form and the surface form. It does not require the learner to set up any autosegmental representation at all. MAX-F , on the other hand, requires both an abstract underlying form and an autosegmental representation, since M A X - F incurs violations of every underlying autosegment, F that is deleted from the surface representation. Note that M A X - F is satisfied by candidates that re-associate the F feature. Therefore, the language learner must track each autosegmental occurrence of F independent of the segmental tier, in order to evaluate this constraint. This constraint adds one more layer of complexity in terms of processing then. I assume that the kind of added complexity such as the type needed to evaluate the constraint M A X - F is avoided by language learners, if possible. In cases where MAXLINK-F will suffice to enforce faithfulness, then, I assume it is the active faithfulness constraint. However, it will become apparent that in some cases it is necessary to refer to the autosegmental tier in order to capture the attested surface pattern. Only in these cases must the added complexity introduced with M A X - F be tolerated by a language learner. Turning now to the analysis for Ife Yoruba, transparency of high vowels is attested and the prohibition account should be able to explain this. In order to allow transparency in the first place, it is necessary to demote the OCP constraint, *RTR-oo- RTR, so that at the very least, it is ranked below [MAX-RTR] R T . This would allow the faithful candidate, (43a) to surface in tableau (43) above. While this in itself does not drive transparency, it is essential to lower the ranking of any constraint, such as the OCP, which would militate against transparency. In order to actually drive transparency, it is necessary to introduce a constraint that is able to actually enforce harmony across a disharmonic high vowel. Unlike the underlying form in (43) above, where the initial vowel is already RTR before harmony has applied, underlying forms with an initial A T R vowel must be forced to surface with an initial RTR vowel. The constraint I propose is *IATR. NONHII-QQ-IRTR . NONHI]. Every non-high A T R vowel that precedes a non-high RTR vowel anywhere in the word incurs a violation of this constraint. The non-high condition is introduced so that the neutral high vowels do not incur violations of *[ATR. NONHll-oo-rRTR. NONHI]. Otherwise, this constraint would always favour deletion of the final RTR vowel and not the initial ATR vowel in mid, A T R - high, ATR - mid, RTR sequences. Given the non- high condition, two solutions would satisfy this constraint. One solution would involve deletion of the initial ATR feature, so that it could be replaced by an RTR feature (this is the attested pattern of transparency in Ife). The other solution would involve deletion of 38 the final RTR feature, so that it could be replaced by an ATR feature (this is an unattested pattern of neutralization). In order to ensure that transparency and not neutralization is optimal, another constraint that militates against deletion of RTR must dominate a constraint that militates against deletion of ATR. This can be achieved by ranking *[ATR. NONHl1-oo-[RTR. NONHI] and MAXLINK-RTR above MAXLINK-ATR. As in Standard Yoruba, HI/ATR is assumed to be undominated since there are no RTR high vowels in either dialect (this is not shown in the tableau below). Tableau (44) below illustrates how transparency is enforced in Ife Yoruba. (44) Transparency Enforced in Ife Yoruba: *TATR. NONHll-oo-[RTR. NONHI], MAXLINK-RTR» MAXLINK-ATR / o d i d e / i *TATR. NONHl]-oo-fRTR. NONHI] MAXLINK-RTR MAXLINK-ATR a. od ide *! ^ b. od ide c. od ide *! However, another feature of Ife Yoruba, absolute alignment, must also be accounted for. Recall that non-final RTR is neutralized in Ife Yoruba; the only occurrences of non-final RTR are in fact due to harmonic requirements. This can be alternatively viewed as leftward A T R harmony, triggered by a final A T R vowel, affecting all preceding vowels. The constraint, *rRTR. NONLOl-oo- r ATR. NONLO] would incur violations for every RTR non-low vowel that is followed by an A T R vowel anywhere in a word. Note that this constraint could enforce either rightward RTR harmony or leftward ATR harmony (both of which are attested, in principle).29 An additional concern in enforcing transparency of high vowels must be dealt with though. The constraint, *rRTR. NONLOl-oo-rATR. NONLO] can potentially drive leftward A T R harmony, a situation that must be avoided in Ife, where high vowels are transparent. Since the effect of leftward RTR harmony at a distance overrides local leftward ATR harmony, *fRTR. NONLOl-oo-rATR. NONLO] must be outranked by * |ATR. NONHl1-oo- rRTR, NONHI]. The reverse ranking would derive opacity and not 2 9 The non-low condition is necessary in order to allow low vowels to precede ATR vowels. Otherwise, the incorrect prediction is made that low vowels should trigger rightward RTR harmony. Additionally, it is assumed that LO/RTR and M A X - L O dominate *IATR. NONHll -oo-rRTR. NONHI] in order to force all underlyingly low vowels to surface as low RTR vowels. Likewise, high vowels are forced to surface as ATR via the undominated constraints, HI/ATR and MAX-HI. 39 transparency. This is illustrated in (45) below. The candidate with an opaque high vowel (45a) fatally violates *1ATR. NONHll-oo-fRTR. NONHI]. The optimal candidate (45b) satisfies this constraint by inserting an RTR value onto the initial mid vowel. This results in a pattern of high vowel transparency. (45) Transparency Enforced in Ife Yoruba: *IATR. NONHll-oo-IRTR. NONHI] » *IRTR. NONLO |- oo-[ATR, NONLO] / o d i d e / *IATR. NONHI1- <x>-[RTR, NONHI] *TRTR. NONLOl- y [ \TR. \ T n \ i o | a. od ide *! ^ b. od ide i i *! Since RTR is neutralized non-finally though in Ife, MAXLINK-RTR must in turn be outranked by *TRTR. NONLOl-oo-1 ATR. NONLO]. This is demonstrated in tableau (46) below. The faithful candidate (46a) fatally violates *IRTR. NONLOI-oo-rATR. NONLO]. In order to satisfy *TRTR. NONLOI-oo-rATR. NONLO], the optimal candidate (46b) deletes the underlying RTR value at the expense of violating MAXLINK-RTR. (46) Absolute Alignment in Ife Yoruba / ew i r i / i *TATR. NONHI 1- oo-[RTR, NONHI] *TRTR. NONLOl- oo-[ATR, NONLO] MAXLINK-RTR MAXLINK-ATR a. ewir i i *! : ®" b. ewir i While absolute alignment is achieved by the ranking in (46) above, this ranking actually yields neutralization of RTR and not harmony with transparency in words with final RTR vowels. This is demonstrated in (47) below. (47) Neutralization of RTR Predicted / o d i d e / *IATR. N0NHI1- oo-[RTR, NONHI] * rRTR. NONLOl- oo-[ATR, NONLO] MAXLINK-RTR MAXLINK-ATR a. od ide *! iiiiilllllp b. od ide i i *i ^ c. od ide 40 The constraint, *IRTR. NONLO|-oo-[ATR. NONLO] militates against the transparent candidate, (47b), in favour of candidate (47c), which neutralizes the RTR contrast completely. In fact, this neutralization does not occur word-finally though. In order to preserve an ATR/RTR contrast word-finally, a positional faithfulness constraint that refers to the final vowel in a root must be introduced. This constraint, which will be left unformalised until Chapter 5, is [MAX-RTR] R t F i n a l . 3 0 By ranking this positional faithfulness constraint above the harmony driving constraint, *[RTR, NONLO]-oo-[ATR. NONLO], it is possible to ensure that a root-final RTR vowel surfaces faithfully, while ensuring that an ATR/RTR contrast is neutralized elsewhere. Importantly, [MAX-RTR] R t F i n a ] cannot dominate HI/ATR, otherwise this would preserve RTR values on high vowels root-finally, a situation that is not attested. The ranking shown in (48) below captures the RTR harmonic pattern in Ife Yoruba. (48) RTR Harmony in Ife Yoruba: HI/ATR » [MAX-RTR] R t F i n a | , * |ATR. NONHll-oo-|RTR. NONHI] » * I RTR. NONLO l-oo-l ATR. NONLO] » MAXLINK-RTR » M A X L I N K - A T R / o d i d e / i [MAX- RTR] R t F i n a , *TATR. NONHI]- oo-[RTR, NONHI] *IRTR. NONLOl- oo-[ATR, NONLO] MAXLINK -RTR MAXLINK -ATR a. od ide *i ^ b. od ide i i * • c. od ide *! d. od ide i * *! Candidate (48c) fatally violates the positional faithfulness constraint, [MAX- RTR] R t F i n a l , allowing the transparent candidate, (48b) to surface instead. Candidate (48d) satisfies [MAX-RTR] R t F i n a l by re-associating the root-final RTR value. However, it fatally violates MAXLINK-RTR. Note that it is necessary for at least one of these RTR- faithfulness constraints to be of the MAXLINK-F type in order to rule out candidate (48d). A summary of the ranking proposed for RTR harmony in Ife Yoruba is given in (49) below. M A X - R T R R T , the constraint used in Pulleyblank (2002), is replaced by this reference to root-final position. What was originally viewed as a root-value RTR feature, is now viewed as a segment-level RTR feature. The reference to root-final position will be expanded upon and a formal constraint will be built in the following chapters. 41 (49) Final Constraint Ranking for Ife Yoruba RTR Harmony HI/ATR MAX-HI LO/RTR M A X - L O ^ ^ M A X ^ R ] P r H d *ATR. N O N l T l ^ R r R ^ N H I RTR. N O N L O - ^ A T R . NONLO MAXLINK-RTR MAXLINK-ATR 2.4.3 RTR Harmony via Prohibition in Ekiti Yoruba A second test of the constraint set used in the prohibition account can be conducted with the pattern attested in Ekiti Yoruba. In this dialect, high vowels are not only transparent to RTR harmony, they also actively participate in it. This is illustrated below in (50). In order to allow high vowels to participate in harmony, the constraint, HI/ATR must be ranked below the harmony-driving constraint, *ATR-C 0 -RTR. 3 1 This rules out candidate (50a). Additionally, the constraint, | M A X - R T R ] R t F i n a l must dominate HI/ATR in order to prevent high - mid, RTR vowel sequences from satisfying *ATR-C 0 -RTR by changing the root-final RTR value to ATR. This rules out candidate (50b). The optimal candidate (50c)is one where the root-final RTR value is retained and the initial high vowel is changed to RTR. 3 1 Note that the non-high condition that was present in the account for Ife Yoruba has been removed. Since both high and mid ATR vowels are targeted in leftward RTR harmony, this condition needs to be removed to allow A T R high vowels to be targeted in RTR harmony. Note also that the proximity relation for this constraint is set to adjacent vowels (C 0 ) . It could also conceivably be set to non-local (oo) without any negative effects on the analysis. However, since there are no neutral vowels in this dialect, there is no reason to use non-local relations; local relations will suffice. 42 (50) High Vowels Participate in RTR Harmony / iCe/ i *ATR-C 0 -RTR [MAX-RTR] R t F i n a l HI/ATR a. iCe *! b. iCe *! <*" c. iCe However, while high vowels participate in RTR harmony, they do not actually act as triggers of it. Specifically, there are no occurrences of root-final high RTR vowels; all final high vowels are ATR. Given the ranking in (50) above, however, an underlying form with a final high RTR vowel will remain RTR and will trigger leftward RTR harmony. In order to prevent this from happening, it is necessary to rank the constraint [HI/ATR] R,Finai32 above [MAX-RTR] R t F i n a l . This will ensure that all root-final high vowels surface as ATR. This is illustrated in tableau (51) below. (51) Neutralization of ATR/RTR Contrast in Root-Final High Vowels / e C i / *ATR-Q-RTR [HI/ATR] R t F i n a , [MAX-RTR l R t F j n a , a. eCi i *! «" b. eCi In fact, by ranking HI/ATR above MAXLINK-RTR generally, we account for the fact that there is no ATR/RTR contrast in high vowels: Their tongue-root values are predictable either due to harmonic requirements or due to the markedness constraint, HI/ATR (as in (51) above). One final consideration is that we must rule out a candidate like eCi in tableau (51) above; like Ife Yoruba, Ekiti exhibits absolute alignment of RTR with the right edge of the root. This candidate fares as well as the optimal candidate on the constraint ranking that has been put forward so far. However, by ranking *[RTR. NONLO] 3 3-C 0-[ATR, NONLO] above MAXLINK-RTR, leftward A T R harmony is enforced. This rules out the disharmonic candidate in question and the pattern of RTR harmony seen in Ekiti Yoruba is achieved. Tableau (52) below illustrates this. 3 2 This is essentially a positional markedness constraint. Given that certain privileged positions allow a greater degree of variation to occur (positional faithfulness), this type of positional markedness constraint would effectively undo positional faithfulness constraints. This is an undesirable situation. 3 3 The non-low condition is necessary in order to allow low vowels to precede ATR mid vowels without triggering rightward harmony. LO/RTR and M A X - L O are assumed to dominate M A X - A T R in order to ensure that all underlying low vowels surface as low, RTR vowels. 43 (52) Leftward A T R Harmony Neutralizes Non-Final RTR in Non-Low Vowels / e C i / i [HI/ATR] R t F i n a l [MAX- ^TR] R t F i n a , HI/ATR * r RTR. NONLOl- C 0 -[ATR, NONLO] MAXLINK -RTR a. e d *! ĤllPISllllll «" b. eCi c. eCi * *! i iHl i l l i l^Mll i The final ranking for Ekiti Yoruba is given below in (53). (53) Final Constraint Ranking for Ekiti Yoruba RTR Harmony With the addition of positional faithfulness and positional markedness, the prohibition account can account for three dialects of Yoruba. It fares better than the stem-control account in terms of its ability to extend typologically and account for dialectal variation with respect to the transparency in Ife Yoruba and does not need to refer to any gradiently evaluated constraints as the alignment-based account does. It does however fail to offer a satisfactory solution to account for the pattern where final high vowels cannot surface as RTR in Ekiti, despite the fact that they can participate in RTR harmony. A positional markedness constraint must be posited to account for this. 2.5 Summary The basic pattern of RTR harmony has been outlined for Standard Yoruba. First, the distinction between morphologically complex V C V nouns and non-complex V C V roots is made. V C V nouns with mid vowels exhibit complete harmony for ATR/RTR. High vowels were shown to be invariably ATR and to be opaque to harmony. This provided evidence on the right-alignment effect where root values of RTR are right 44 aligned with the root, whenever possible. In cases where high vowels occur at the right- edge of the root, a root RTR value is as right aligned as possible. Low vowels differ in that they are invariably RTR and trigger strictly leftward RTR harmony. A featural contrast for tongue-root value in mid vowels can occur following a low vowel, but not preceding it - only RTR mid vowels can occur preceding a low vowel. This pattern was shown to extend to prefixes, in the cases of agentive nominalized forms. This implies that the domain of harmony extends beyond the root to the word. Compound words exhibit disharmonic sequences of mid vowels in some cases. This is explained by restricting the domain of RTR harmony to the domain of the prosodic word. This pattern is accounted for in three optimality-theoretic accounts that use different sets of constraints. These three accounts were summarized and discussed in comparison with each other. Of these three accounts, various problems with an alignment-based account and with a stem-control based account were raised. The alignment-based account succeeds in deriving the patterns seen in three dialects of Yoruba, but it relies on a gradiently evaluated version of alignment. This is undesirable for theoretical reasons external to the facts of Yoruba. Stem-control is also able to account for the pattern in two of the three dialects of Yoruba that were explored here, but it relies on the assumption that all V C V nouns are morphologically complex. Given evidence that in at least some cases, V C V nouns are morphologically non-complex, this stem-control account cannot hold up. An account utilizing prohibition-type constraints is superior in its ability to account for the harmonic patterns that are seen in three dialects of Yoruba. There is no need to refer to gradient constraints in an account utilizing prohibition-type constraints. Additionally, there is no need to rely on the fact that all V C V nouns are morphologically complex. Instead, it is suggested that a reference to prosodic categories might provide an alternative way to capture the facts of RTR harmony in Yoruba. In Yoruba, morphological structure and prosodic structure are often co-extensive and therefore it can be difficult to tell which kind of constituency is being referenced in processes that are domain-restricted. The next chapter illustrates the pattern of RTR harmony seen in a fourth dialect, Moba Yoruba, where tongue root harmony extends over a larger domain than in Standard Yoruba. 45 Chapter 3 - RTR Harmony in Moba Yoruba Up until this point, we have dealt almost exclusively with Standard Yoruba. This dialect is the standardized dialect that is taught in schools. It is most closely related to the Oyo dialect of Yoruba and therefore the Yoruba orthography is based on this dialect. Oyo Yoruba is spoken in the region around Oyo, Ogbomosho, and Ibadan. Of the other two dialects that we have seen, Ekiti is spoken in and around Ado Ekiti and Ifaki while Ife Yoruba is spoken in the town of Ife. The dialect that is featured in this chapter, Moba Yoruba, is spoken in a region north of the Niger delta on the west side of the Niger River. The villages it is spoken in are Ulale, Ekan, Ayedun, Ilofa, Odo-owa, Obo Erimope, Otun and Igogo (Ajiboye p.c). Moba Yoruba exhibits a similar pattern of RTR harmony as is seen in Standard Yoruba, with one notable exception concerning the apparent size of the domain. Moba RTR harmony appears to also affect proclitics while in Standard Yoruba proclitics are clearly outside the domain for RTR harmony. This chapter lays out the crucial Moba data, which, when compared to the corresponding data in Standard Yoruba, point towards an analysis for Moba that is identical to Standard Yoruba except for the size of the harmonic domain. 3.1 Moba Yoruba - Phonological Background Moba Yoruba differs from Standard Yoruba with respect to the general segmental inventory as well as the distributional facts of certain phonemes. With respect to the consonantal inventory, the contrast that exists in Standard Yoruba between / s / and / s / (phonetically [J]) is neutralized to / s / in Moba. This is illustrated below in (54). While there is a contrast in Standard Yoruba between the words in (54a) and (54b) with respect to the S / s distinction, there is no such contrast existing in Moba Yoruba. (54) Neutralization of the / s / - / s / Contrast in Moba Yoruba M B SY Gloss a. se se i 'to do' use ise i i 'message a s o aso i i 'cloth' 46 osi osi 'poverty' SU SU 'to make into a ball' osunan osunwon i i 'measuring container' so i so 'to speak' esuro esuo 'Redflanked Duiker' ese i i ese 'foot' egusi egunsi 'melon / a food made from melon seeds' With respect to vowels, the first difference between Moba and Standard Yoruba is regarding the neutralization of word-initial lul in Standard Yoruba. As can be seen in (54a) above in the word for 'message', an initial lul in Moba corresponds to an initial HI in Standard Yoruba. This neutralization is only seen in word-initial position however, since lul does occur elsewhere in Standard Yoruba. This is illustrated in (55) below. While in Moba, (55a) and (55b) exhibit a contrast between initial lul and initial HI, this contrast is neutralized in Standard Yoruba, where only l\l and not lul can occur word- initially. In (55c), we have examples where the Standard Yoruba forms neutralize the initial lul, but a non-initial lul is allowed to surface faithfully. Finally, in (55d), we see that the initial neutralization is strictly word-initial, lul is allowed to surface in Standard Yoruba words where it is the first vowel as long as a consonant precedes it (as long as it is not the initial segment). (55) Neutralization of initial lul in Standard Yoruba M B SY Gloss a. use i ise i i 'message' urq iro i 'falsehood' ule ile 'house' b. ito ito i 'saliva' iye iye 'intelligence' ile i ile i 'land/ground' c. ulu ilu 'town/city' usu isu i 'yam' 47 d. ku ku 'to die' mu mu 'to drink' ru ru 'to carry a load' Ola (1995) uses this pattern of neutralization in initial position as evidence supporting the hypothesis that onsetless vowels are not syllabified (as will be discussed in more detail in section 4.2). This amounts to requiring that all syllables must contain onsets; otherwise, the vowel is not a syllable head. This neutralization then could refer to the vowel's prosodic status as a syllable head. The / i / - / u / contrast might then only occur in syllabic nuclei in Standard Yoruba. This difference between the dialects could then depend on whether or not onsetless vowels are well-formed syllable nuclei. In Moba, on the other hand, if onsetless vowels could constitute well-formed syllable nuclei, this would explain the fact that the / i / - / u / contrast can occur in any position. This would follow since any vowel is a well-formed syllable nucleus in Moba. Alternatively, if the requirement for a well-formed syllable to contain an onset were in place in both dialects, it could be attributed directly to a difference in the positions that this contrast can exist: in Moba, / u / could occur in any position; in Standard Yoruba, it can only occur in syllable nuclei. There is one final difference between Moba and Standard Yoruba, concerning the distribution of the nasal vowels, [a] and [5]. In both dialects, only high and low vowels exhibit a contrast for the nasal/oral distinction. However, while Moba does not allow any mid nasal vowels, Standard Yoruba exhibits an allophonic variation between the mid nasal vowel, [5] and the low nasal vowel, [a]. This variation is conditioned by an immediately preceding labial consonant, which triggers progressive labial harmony. This is illustrated below in (56). 48 (56) Allophonic Variation of Low Nasal Vowel in Standard Yoruba M B SY Gloss b. a. moto ogbo lepo gbo ti 'to be wise' 'to deceive' 'to hammer' 'thirty' 'tray' 'to draw water' 'car The pattern above illustrates that the low nasal vowel, / a / surfaces as [5] when it is preceded by a labial consonant in Standard Yoruba, as is seen in (56a). However, the low nasal vowel, [a] surfaces faithfully elsewhere as is seen in (56b). This section has summarized the differences between the segmental inventories and their contextual behaviour. The most important difference for the purposes of this thesis, tongue-root harmony, is examined in the next section. 3.2 R T R Harmony in Moba Yoruba: The Basic Pattern 3.2.1 Harmony in VCV Nouns Within roots, we find that both Moba and Standard Yoruba exhibit the same pattern of RTR harmony. Mid vowels in disyllabic words are required to have identical tongue-root values. This pattern was illustrated in (4) above for Standard Yoruba; it is illustrated for Moba below in (57); Standard Yoruba forms are given to allow comparison. Disharmony is tolerated in monomorphemic loan words in Yoruba, such as this one. 3 5 In Standard Yoruba, ogbo can also be produced with an initial low tone (dgbb). In Moba, however, the initial vowel must be produced with a mid tone and not with a low tone. 49 (57) M i d Vowels in Moba M B S Y Gloss e w e e w e ' lea f * e w e i * e w e i e p o epo 'o i l ' * e p o * e p q o le ole ' thief *q le *o le eo / e w o o w o 'money' * q w o * o w q e s e ese i i 'foot' * e s e i * e s e eko i i eko i i 'pap' *eko i *eko o b e • i obe i i 'soup' * o b e i * q b e oko i i oko i i 'vehicle' * o k o i *oko i With respect to disyllabic words, we find that high vowels do not participate in R T R harmony and are invariably A T R . This is seen in (58) below. (58) High Vowels in Moba M B SY Gloss a. ule ile 'house' igo igo 'bottle' b. ile i ile i 'ground' ito ito i 'sal iva' c. eti eti 'ear' er i or i 'head' eku eku 'bush rat' oju oju 'eye' d. ebi i ebi 'guilt ' ok in okin i 'egret' e w u e w u i 'clothing' o run orun 'heaven' 50 igi inu usu ulu undin igi inu isu i flu ad i 'tree' 'stomach' 'yam' 'town/city' 'palm nut oi l ' Again, low vowels pattern similarly in both Moba and Standard Yoruba, they are invariably RTR and they allow an RTR contrast in mid vowels following a low vowel (59a and b), but this contrast is neutralized preceding a low vowel (59c and d). Roots with two low vowels (59e) or a low vowel and a high vowel (59f and g) are attested as well; low vowels and high vowels have predictable tongue-root values regardless of their position in the word, RTR for the former and ATR for the latter. (59) Low Vowels in Moba M B SY Gloss a. ate ate 'hat' aro aro 'indigo' b. aje aje 'paddle' a s o i aso i i 'cloth' c. e p a epa 'groundnut' oran oran 'trouble' d. * e p a * e p a *6 ran *6 ran e. aya aya 'chest' ara ara 'body' f. atu atu 'type of cassava am i ami 'sign' g- inyan iyan 'dispute' ika ika 'cruelty' uya fya 'punishment' uja ija 'fight' una ina 'fire' intan itan 'story' ila ila 'okra' 51 3.2.2 Harmony in VCVCV Nouns Trisyllabic words also pattern similarly in Moba and Standard Yoruba. As was outlined above in chapter 2, the pattern seen in disyllabic roots can be extended to trisyllabic roots that contain only low and/or mid vowels, however high vowels exhibit opaque harmonic behaviour.36 Trisyllabic roots with a high medial vowel flanked by two mid vowels illustrate the opaque status of high vowels. This opacity is seen in Moba as well as in Standard Yoruba. (60) Opaque High Vowels in Moba M B SY Gloss er ipe erupe 'earth' e w u re i e w u re 'she-goat' en ibo i elubo 'yam flour' ekuro ekuro 'palm kernel' od ide i od ide 'Grey Parrot' o w u r o i owuro 'morning' oruko oruko 'name' *o ruko *q rukq *o ruko i *oruko In all of the trisyllabic forms in (60a) above, the final mid vowel is RTR. Crucially, the initial mid vowel could never be RTR regardless of the RTR value of the final mid vowel as is seen in (60b). This opacity of high vowels in both Moba and Standard Yoruba is not simply a case where the A T R feature of the high vowel is triggering leftward spreading of ATR since we have seen that an ATR/RTR contrast exists both preceding and following high vowels in (58) above. This pattern is true in Moba in trisyllabic forms with mid-high- 3 6 Regarding the harmonic status of high vowels in Moba, it appears that they do not harmonize. However, given that in other dialects (Ekiti, for example) we have evidence of high vowel participation in tongue-root harmony, phonetic testing should be done to confirm that these high vowels do not in fact show (phonetic or phonological) retraction effects in Moba. 3 7 The Standard Yoruba form offered here for 'palm kernel' differs from the Standard Yoruba form in Archangeli & Pulleyblank (1989) which is dkurq. 52 high vowel sequences as well. In (61a), we find that RTR can surface on a mid vowel preceding two high vowels. In (61b), we find that ATR mid vowels can also surface in this same position. (61) Mid-High-High Trisyllabic Sequences in Moba M B SY Gloss a. eliri eliri 'a type of rat' ewir i i ewir i i 'bellows' eburu i eburu • 'shortcut' b. ekuru ekuru 'food made of beans' obunr in obinrin 'woman' 3.2.3 Disharmony in Compounds As was the case in Standard Yoruba, we find disharmonic sequences of both RTR mid vowels followed by A T R mid vowels and ATR mid vowels followed by RTR mid vowels in compounds. (62) Disharmony in Compounds M B SY Gloss se se 'to change' e(w)6 o w o 'money' se(w)6 s e w o 'to change money' e w e e w e 'leaf obe • i obe i i 'soup' e w e b e i e w e b e 'any pot herb used for making soup' This disharmonic sequence is tolerated if we restrict the domain of leftward RTR harmony to the prosodic word. Each root in the above compounds would constitute a separate occurrence of a prosodic word. This would allow disharmony to exist in compounds such as the ones in (62) above. 53 3.2.4 Consonant-Deletion in VCVCV Nouns There are processes of consonant-deletion that occur in Standard Yoruba and Moba Yoruba that potentially interact with tongue-root harmony. When a consonant is deleted intervocalically, it is usually accompanied by a process of assimilation in order to repair the vowel hiatus that results. With trisyllabic words, we find that the first consonant (Cl) deletes in Standard Yoruba, but this is not the case in Moba Yoruba, where consonant deletion is avoided altogether.38 The fact that the initial consonant (rather than the final consonant) deletes in Standard Yoruba, is cited by Ola (1995) as evidence that the final syllable is the prosodic head (this is discussed further in section 4.2.1). Faithfulness constraints referring to the final position are then able to prohibit deletion of material in this final head syllable. (63) Cl-Deletion in Standard Yoruba M B SY Gloss a. er ipe ~ ' * e e p e erupe ~ - eepe 'earth' od ide <• - * o o d e od ide ~ ' o o d e i 'Grey Parrot' oru jq -- * 66jo *6rujq - 6 6 j o 'daily / same day b. ewure i ~ *eere ewure > - *ee re 'she-goat' en ibo -- * e e b o i elubo ~ * e e b o 'yam flour' ekuro <• - * e e r o i ekuro ~ *ee ro i 'palm kernel' In (63a) above, it is apparent that Cl-deletion is active in Standard Yoruba but not in Moba Yoruba. (63b) shows that this process of Cl-deletion is not allowed.in some cases though. Instead, it applies idiosyncratically to some words and not to others. After Cl-deletion has occurred in the Standard Yoruba forms, progressive vowel assimilation follows. For example, the medial high back round vowel in erupe assimilates to the preceding mid front vowel, e (with the retention of the low-tone). This yields the form eepe . This form is disharmonic on the surface. The same disharmony is seen in all of the other forms with Cl-deletion and assimilation in (63a). If this assimilation were to be occurring at the same level of derivation as the harmony rule, then we would expect that these forms with Cl-deletion should exhibit RTR-harmony. In derivational terms, the 3 8 An exception to this rule of Cl-deletion is dwurq which means 'morning.' In Moba, the first consonant can be deleted, yielding either dorq or durq. In Standard Yoruba, only dorq is licit; *duro is illicit in Standard Yoruba. 3 9 The Standard Yoruba word, ddjq, is exceptional in that the full form, *orujq, is illicit. 54 fact that they don't, implies that the harmony rule must apply before assimilation does. In an optimality-theoretic framework, this derivational opacity is usually dealt with via faithfulness, much in the same way sympathy theory dealt with misaligned RTR-values in the stem-control theoretic account above. In this case, there must be a pressure to be faithful to the corresponding full form (before Cl-deletion has applied). This would ensure that the initial and medial vowels in eepe, for example, remain ATR, in agreement with their A T R values in the full form, erupe. An interesting twist to the pattern mentioned above concerns a pattern of optional w-deletion that is seen in Moba Yoruba, and not in Standard Yoruba. Unlike the C l - deletion pattern mentioned above, this deletion process applies to any 'w' , including those occupying the onset of the head syllable. In addition, unlike the Cl-deletion pattern mentioned above, this w-deletion does not apply with progressive vowel assimilation following it. Therefore, there is no potential interaction with RTR-harmony, since all underlying vowels are preserved, including opaque medial high vowels. This pattern is illustrated in (64) below. (64) w-Deletion in Moba Yoruba M B SY Gloss a. e w o ~ eo o w o ~ * o o 'money' o w u ~ ou o w u ~ * o u 'jealousy' w a n ~ an w o n ~ *6n 'to measure' b. *un r inwo ~ unrino i r i n w o ~ * ir ino 'four-hundred' * i y a w o ~ iyao i yawo ~ * i yao 'wife' * a w o ~ ao a w o ~ * a o 'plate' * a d a w o l e ~ adao le a d a w o l e ~ * a d a o l e i i 'a beginning' c. iwe ~ * ie iwe ~ * ie 'book' ewir i ~ *eir i i i ewir i ~ *eir i • i 'bellows' w e ~ * e w e ~ * e 'swim' In (64a) above, free variation exists in Moba between the full forms on the left and the forms with w-deletion on the right. In (64b), the full form is ungrammatical, while the form with w-deletion is grammatical. In (64c), only the full form is grammatical, while the form with w-deletion is ungrammatical. However, the forms with w-deletion are ungrammatical in Standard Yoruba in all of the cases in (64). This pattern of w-deletion does not interact with RTR harmony, since there is no vowel assimilation accompanying it. 55 3.3 R T R Harmony in Prefixes In section 2.2 above, we saw that the agentive prefix in Standard Yoruba is included in the harmonic domain as was shown in (12), where the tongue-root value of the prefix was a function of the tongue-root value of the root to which it was attached. This situation is also found in Moba. (65) Harmonic Behaviour of Prefixes in Moba M B SY Gloss a. de i de 'to hunt' o d e ode 'hunter' * o d e * o d e i b. j ou j o w u 'to be jealous' o jou o jowu 'a jealous person' *o jou *6 j 6wu The agentive prefix is included in the RTR harmonic domain in both Moba and Standard Yoruba, then. This harmonic behaviour is used as a diagnostic in defining a prefix as such. Prefixes are those elements that are automatically forced to harmonize with the tongue-root values of the base to which they attach. As it was shown in section 2.3.3, the agentive prefix can contrast for ATR/RTR when it is added onto a high-vowel verbal base. In this case, the prefix can contrast for ATR/RTR. This is repeated in (66) below. (66) Harmonic Behaviour of Prefixes with a High-Vowel Verbal Base M B SY Gloss a. ku ku 'to die' oku oku 'corpse of a person' b. m u mu 'to drink' q m u q m u 'drinker' The RTR feature that shows up in the nominalizing prefix in (66b) is assumed to be due to a root RTR feature that is associated with the verbal base, m u , 'to drink.' Since the verbal base has a high vowel, this RTR feature cannot surface without violating the undominated constraint militating against high RTR vowels. However, once a mid- 56 vowel is added on as a prefix, this RTR feature can surface on the prefix. According to this hypothesis, the difference between the verbal bases in (66a) and (66b) is that the base in (66a) does not contain an underlying RTR value, while the base in (66b) does. 3.4 RTR Harmony in Proclitics 3.4.1 RTR Harmony with Single Proclitics The crucial data that show differences in domain size between Standard Yoruba and Moba are in the class of proclitics, which attach to verbal hosts. Since RTR harmonic behaviour is restricted to mid vowels, only those clitics with mid vowels could potentially exhibit harmonic behaviour in (67) below. The clitics are attached in turn to a pair of verbs, one with an A T R vowel, and the other with an RTR vowel. 4 0 (67) Proclitics - Differences in Domain-Size in Moba and Standard Yoruba Clitic M B SY Gloss Meaning 1SG m e / m l de m o de lSG='arrive' Tarrive(d)' m e / mi lo i i m o lo • lSG='go' T go/went' 1PL a de a de lPL='arrive' 'we arrive(d)' a lo a lo lPL='go' 'we go/went' 2SG 6 de i o de 2SG='arrive' 'you(sg.) arrive(d)' 6 lo • i o lo i 2SG='go' 'you(sg.) go/went' 2PL in de e de 2PL='arrive' 'you(pl.) arrive(d)' in lo • e lo i i 2PL='go' 'you(pb) go/went' 3SG e de 6 de 3SG='arrive' 's/he arrive(s/d)' e lo i i 6 lo i 3SG='go' 's/he goes/went' For a complete paradigm of proclitics with verbal root vowels varying for both tongue- root value and tone, see appendix B. 57 3PL an de w o n de 3PL='arrive' 'they arrive(d)' an lo w o n lo 3PL='go' 'they go/went NEG ke de ko de (3SG)=NEG='arrive' 's/he didn't arrive' ke lo ko lo (3SG)=NEG='go' 's/he didn't go' FUT e de m a a de FUT=' arrive' 'will arrive' e lo i i m a a lo FUT='go' 'will go' A summary of the complete subject proclitic paradigm (including the NEG and FUT proclitics) is given below in (68). (68) Summary: Harmonic Behaviour of Proclitics Clitic M B SY 1SG harmonic alternation invariably A T R 1PL N/A (low V) N/A (low V) 2SG invariably RTR invariably A T R 2PL N/A (high V) invariably RTR 3SG harmonic alternation invariably A T R 3 PL N/A (low V) invariably RTR NEG harmonic alternation invariably ATR FUT harmonic alternation N/A (low V) While all Standard Yoruba proclitics surface invariably as A T R or RTR, all but one of the Moba mid-vowel proclitics exhibits harmonic alternation with the host verb. The four clitics that do harmonize in Moba correspond to Standard Yoruba cognates that are invariably ATR. The lone non-harmonic clitic in Moba is invariably RTR (despite its Standard Yoruba cognate being invariably ATR). Crucially, there is no evidence of an ATR clitic in Moba that is non-harmonic. I will assume then that an underlying ATR clitic is subject to harmony,41 while an underlying RTR clitic is not, as expected assuming right-to-left RTR harmony. In the above presentation of the clitics in Yoruba, I am assuming that the negative and future auxiliaries are in fact clitics. Their harmonic behaviour itself provides Assuming the richness of the base hypothesis, an underlyingly unspecified proclitic must be considered. Such an unspecified clitic would be predicted to participate in harmony as well. 58 evidence for this. I am defining the class of clitics as those particles that participate in leftward RTR harmony, but not leftward ATR harmony. A distinction can be made between clitics and prefixes then. While prefixes never exhibit contrastive behaviour for ATR/RTR, clitics that are underlyingly RTR will surface as such, even though a disharmonic sequence will be introduced. Additionally, the phonological shape of the FUT marker (V) matches the shape of five of the six proclitics. A verbal root on the other hand, must minimally contain at least a C V syllable (an onset is obligatory). Finally, a distinction can be made concerning the grammatical functions of clitics and prefixes. While the clitics listed above are all inflectional morphemes, the agentive prefix (and, it is assumed, other prefixes) have a derivational function. These facts are all consistent with a treatment of the subject proclitics, negative, and future markers as clitics and of the agentive marker as a prefix. 3.4.2 RTR Harmony with Multiple Proclitics While single clitics were shown to be included in the harmonic domain in Moba, it remains to be seen whether sequences of clitics can occur, and if so, whether the outermost clitic is included in the harmonic domain. The answer to both of these questions turns out to be 'yes.' 4 2 In (67) above, it was shown that the Moba negative marker, ke I ke can occur attached to a verbal base. This results in a third person reading, perhaps due to a null 3SG marker (the 3SG clitic e I e cannot precede the negative marker in either dialect). However, it is possible to get sequences of the NEG marker with the other subject clitics. Since we cannot force the 3SG morpheme to surface with the NEG marker, we have only two mid vowel candidates that are testable; of these two, only the 1SG form is potentially harmonic, since the 2SG is invariably RTR without the negative marker (and we wouldn't expect it to suddenly exhibit harmonic behaviour with the NEG marker). Standard Yoruba data is included for completeness. (69) Harmony With Two Clitics: SUBJECT + NEG Subj. Clitic M B SY Gloss Meaning 1SG mi ke de n ko de lSG=NEG='arrive' T don't/didn't arrive mi ke je n ko je lSG=NEG='eat' T don't/didn't eat' * m e ke je * m e ke je A complete paradigm of three auxiliaries, the NEG, FUT and PROG markers, with all six subject proclitics, is given in appendix B. The vowel in the verbal base is allowed to vary with respect to its tone and tongue-root value. 59 2SG o ke de o ke de 2SG=NEG='arrive' 'You don't/didn't arrive' o ke lo o ko lo 2SG=NEG='go' 'You don't/didn't go' As expected, the 2SG form is invariably RTR in Moba and invariably ATR in Standard Yoruba. However, the 1SG form in Moba, which alternated between a high and mid vowel in (67) above, does not alternate when the negative marker is present. Instead, the form with the high vowel is grammatical, and the form with the mid vowel is ungrammatical. However, the future marker e I e comes to the rescue here. Combining future and subject marking, we have another opportunity to test whether multiple clitics exhibit harmonic behaviour. This time, the 3SG form turns up overtly and the 1SG form surfaces with the mid vowel. (70) Harmony With Two Clitics: SUBJECT + FUT Subj. Clitic M B S Y 4 3 Gloss Meaning 1SG m e e de m a a de 1SG=FUT= 'arrive' 'I will arrive m e e lo m a a lo 1SG=FUT= 'go' T w i l l go' 2SG 6 e d e w a a de i 2SG= =FUT='arrive' 'You will arrive'4 4 6 e lo w a a lo I I i i 2SG= =FUT='go' 'You will go' 3SG e e d e o m a a de 3SG= =FUT='arrive' 'S/he will arrive' e e lo 6 m a a lo • i i i 3SG= =FUT='go' 'S/he will go' It is apparent that the FUT marker harmonizes completely with its verbal host. Additionally, the 1SG and 3SG forms harmonize as well. On the other hand, the 2SG form does not harmonize. As expected, it is invariably produced as RTR. Note, however, that the FUT marker could conceivably have agreed with the RTR value of the 4 3 Future marking in Moba and Standard Yoruba bear little resemblance to each other both morphologically and phonologically. Therefore, these Standard Yoruba forms should not be viewed as direct cognates. However, note that there is no harmony in these Standard Yoruba future forms, as expected. 4 4 The 2SG future form is included to show that the future clitic is not simply a copy of the preceding vowel in Moba. 60 preceding 2SG clitic, but instead chooses to agree with the verbal root's tongue-root value, whatever it is. Finally, (71) illustrates that the combination of three clitics (a subject-marking proclitic, a NEG proclitic and a FUT proclitic) in Moba do in fact result in a harmonic surface form (I omit Standard Yoruba forms here).45 (71) Harmony With Three Clitics: SUBJECT + NEG + FUT M B Gloss Meaning a ke e de lPL=NEG=FUT='arrive' 'We will not arrive' a ke e je lPL=NEG=FUT='eat' 'We will not eat' This is illustrated using the 1PL proclitic. Note that for the same reason as in (69) above, there will not be any harmonic interaction between the subject proclitic and the negative proclitic. The subject forms of the 1SG and 3SG clitics occur with a high vowel and with no vowel respectively since the negative clitic is present. Therefore, there is no way to test whether three consecutive clitics would harmonize across the board. However, there is no reason to suspect that they should not if such data were possible to elicit. 3.5 R T R Harmony in Encl i t ics 3.5.7 RTR Harmony in Enclitics Another area of consideration is the enclitic domain. Both Moba and Standard Yoruba i have a small set of enclitics. Of these, we find object-marking enclitics that attach following transitive verbs. Their phonological form is very similar to their corresponding subject markers in (67) above. What we find is that the same forms used with the negative markers occur as enclitics in Moba, meaning we have a null 3SG object marker and a 1SG object marker with a high vowel. The 2SG marker is invariably RTR as it was in all of the preceding cases as well. This is interesting in itself, and it will figure prominently in the analysis in chapters four and five. The plural markers all contain low or high vowels and are thus unable to participate in RTR harmony. See appendix B for sequences of four clitics where the progressive marker is inserted as the fourth clitic. 61 (72) Enclitics: Object Markers with High-Tone Verbs' Obj. Clitic M B SY Gloss Meaning 1SG ade le mi ade le mi 'Ade' 'pursue'=lSG ' A . pursue(s/d) me' ade ko mi ade ko mi i 'Ade' 'teach'=lSG ' A . teaches/taught me' 1PL ade le a ade le w a 'Ade' 'pursue'=1 PL ' A . pursue(s/d) us' ade ko a i ade ko w a 'Ade' 'teach'=lPL ' A . teaches/taught us' 2SG ade le o ade le e 'Ade' 'pursue'=2SG ' A . pursue(s/d) you' ade ko o i i ade ko e i i 'Ade' 'teach'=2SG ' A . teaches/taught you' 2PL ade le in ade lee yin 'Ade' 'pursue'=2PL ' A . pursue(s/d) you all ' ade ko in i ade koo yin 'Ade' 'teach'=2PL ' A . teaches/taught you all ' 3SG ade le ade 1(e) e 'Ade' 'pursue'=3SG ' A . pursue(s/d) him/her' ade ko i ade k(q) o 'Ade' 'teach'=3SG ' A . teaches/taught him/her' 3 PL ade le an ade le w o n 'Ade' 'pursue'=3PL ' A . pursue(s/d) them' ade ko an ade ko w o n 'Ade' 'teach'=3PL ' A . teaches/taught them' As was the case in (69) above, there is no direct evidence available to test whether underlyingly A T R (or unspecified) enclitics are included in the harmonic domain for RTR harmony or not. This indeterminacy is becoming very familiar; Yoruba is a language with a relatively small morphological arsenal. Therefore, it is unsurprising that it might be difficult to find evidence for the domain-size of phonological processes (that are already phonologically limited) across a limited class of morphemes. Because of these morphological limitations, indeterminacy results. 3.5.2 Tonal OCP in Enclitics There is, however, one interesting observation concerning the tone of the enclitics. The data in (72) contain high-tone verbal hosts (le and kq). When considering the pattern with low-tone or mid-tone verbs, we find that these enclitics surface with high tone (with 4 6 See appendix B for a complete paradigm of these object enclitics combined with three auxiliaries and verbal hosts that vary for tone and tongue-root value. 62 the exception of the 3SG form - this discrepancy will be addressed shortly). This pattern is exemplified using two low-tone verbs (pe and kq) in (73) below. (73) Enclitics: Object Markers with Low-Tone Verbs Subj. Clitic M B SY Gloss Meaning 1SG ade pe mi ade pe mi 'Ade' 'call'=lSG ' A . call(s/ed) me' ade ko mi i ade ko mi i 'Ade' 'reject'=lSG ' A . reject(s/ed) me' 2SG ade pe 6 ade pe e 'Ade' 'call'=2SG ' A . call(s/ed) you' ade ko 6 ade ko e i i 'Ade' 'reject'=2SG ' A . reject(s/ed) you' 3SG ade pe ade pe e 'Ade' 'call'=3SG ' A . call(s/ed) him/her' ade ko • ade ko 6 i i 'Ade' 'teach'=3SG ' A . teaches/taught him/her' As can be seen in (73) above, the enclitics are underlyingly specified as high-tone. This high tone surfaces in tact when the enclitics follow a low- or mid-tone vowel in the verbal base. However, when they follow a high-tone vowel as they do in (72), all the enclitics except the 2PL marker surface with mid-tone instead. This is presumably due to an OCP constraint that prohibits adjacent high tones between the vowel in the verbal base and the enclitic. This is resolved normally via high-tone drop, such that the enclitic surfaces with mid tone (recall that mid tone is analyzed as the lack of tone in Yoruba). In the case of the 2PL marker, the high tone is preserved on the enclitic and a mid-tone copy of the vowel in the verbal base is inserted to prevent the OCP violation. Akinlabi and Liberman (2000) noted this OCP effect seen in (72) where adjacent high tones across a stem-enclitic boundary are prohibited. One proposal they have to account for the presence of adjacent high-tones within roots is that tonal spreading applies only within a word (tonal spreading cannot apply across the verb-enclitic boundary then). However, the OCP holds over a domain that spans the verb-enclitic boundary, and thus in order to prevent a violation of the tonal OCP, the clitic high tone deletes. Under this analysis, a low-tone or mid-tone verb would allow an underlying high-tone to surface on the enclitic, since the OCP is only violated by adjacent high tones. However, when the same enclitic is attached to a high-tone verb, the high-tone on the enclitic is deleted. The lone exception of course is the 3SG form, which is aberrant with respect to its vowel quality in both dialects. In Standard Yoruba, it appears to be a copy of the vowel in the verbal host, with high tone specified. This high tone drops when it attaches to a high-tone verb, however, in compliance with the general solution to overcome the OCP 63 violation that would be incurred by adjacent high-tones. In Moba, the entire enclitic is apparently deleted.47 On its own this would be fine, since this is simply another solution to resolve the OCP, albeit non-minimal. What is interesting about the 3SG object form in Moba is that the tone on the verb is altered, suggesting a phonetic conflation of two tones, such that a single 'derived' mid-tone is produced. Although, there is no instrumental analysis available of either the length or tone of the vowel in the verb in the 3SG form in Moba, it appears that this derived 'mid' tone (marked with a level bar over the vowel in (73): ade pe) is actually mid-way between a low and mid tone. This can be contrasted with the vowel in the 3SG form in the high-tone verb paradigm where a true mid tone is found (i.e. ade le in (72)). 3.5.3 Implications for Domains Object enclitics, then, provide no direct insight into their status in the RTR harmonic domain. While it is clear that they do not trigger leftward RTR harmony, it is not clear if they would participate in rightward RTR harmony since there are no underlyingly ATR enclitics. Direct evidence for the harmonic status of the enclitics is lacking, however it is still possible to make certain claims regarding the RTR harmonic domain by considering the domain for the tonal OCP. This section begins by illustrating the right-branching structure that must be present based on the tonal OCP facts. There are two possibilities regarding the type of constituency referred to in the RTR harmonic domain. Either both RTR harmony and the tonal OCP apply over domains in the same type of constituency or they do not. The first possibility (that the domains for the tonal OCP and RTR harmony are of the same type of constituency) would necessarily imply that the enclitics are included in the harmonic domain. The second possibility (that the domains for the tonal OCP and RTR harmony are of different constituency types) would be consistent with enclitics that are or are not included in the domain of RTR harmony. This discussion will be expanded upon in section 4.4, where a stronger case is made that the enclitics are in fact included in the harmonic domain. First, whatever the relationship is between the enclitic and verb, this morpheme boundary is clearly contained within a domain that enforces the tonal OCP. However, this same tonal OCP is not enforced across the proclitic-verb boundary. Examples abound of high-tone clitics that surface with their high-tone intact preceding high tone verbs (i.e. Moba: e de ; Standard Yoruba: 6 de 'S/he arrived'). This OCP violation is tolerated across the proclitic-verb boundary. This deletion is optional in Standard Yoruba. 64 (74) Domain for Tonal OCP (D-TN OCP) - Both Dialects a. Proclitics excluded b. Enclitics included ® [ ^ ® ] D - T N O C P 3SG='arrive' 'S/he arrive(s/d)' e [le m i ] D _ T N 0 C P 3SG='pursue'=lSG 'S/he pursue(s/d) me' c. Structure Implied based on Tonal OCP X Y = D-TN OCP [le mi] D . T N O C P Let us assume that this difference can be attributed to a tonal OCP constraint that applies over some domain that contains the verb-enclitic pair but not the proclitic-verb pair as shown in (74c) above. Then, assuming this domain also refers to the same type of constituent as the RTR-harmonic domain, whether it is syntactic or prosodic, the implication is that the enclitic must be part of the RTR harmonic domain in Moba. The fact that RTR harmony does apply across the proclitic-verb boundary and the tonal OCP does not, implies that the RTR domain is a superset of the tonal OCP domain in Moba. The only constituent in (74c) above that contains both the proclitic and the verbal base is X , which also necessarily contains the enclitic domain. This is summarized in (75) below. (75) Interaction of RTR-Harmonic Domain (D-RTR HRM) with Tonal- OCP Domain Assuming Domains Refer to Same Type of Constituency a. Only Possibility: Enclitics Included in D-RTR HRM X = D-RTR H R M Y = D-TN OCP [e [le m i ] D . T N 0 C P ] D -TN OCP- lD-RTR H R M 65 The structure in (75) above would imply that the enclitics are included in the RTR harmonic domain. Evidence that the enclitics do not harmonize would not be consistent with the structure in (75), necessarily implying that the RTR domain is not defined via the same constituency-type. Instead, a different type of constituent must be referred to by the RTR-harmonic domain. For example, given evidence against harmony in enclitics, and also, given evidence that the RTR domain refers to some prosodic constituent, together this would amount to evidence that the tonal OCP does not refer to any prosodic constituent (possibly implying a direct reference to syntax). This is summarized in (76) below where it is assumed that the tonal-OCP domain refers to a different type of constituency (constituents, X and Y on the left) than the RTR-harmonic domain (constituents X ' and Y ' on the right). (76) Interaction of RTR-Harmonic Domain with Tonal-OCP Domain Assuming Domains Refer to Different Type of Constituency a. Possibility 1: Enclitics Included in RTR harmonic domain (D RTR HRM) X X ' D-RTR HRM D-TN OCP le mi] D-RTR HRM or X ' D-RTR HRM Y [e le mi] D-RTR HRM 66 b. Possibility 2: Enclitics Not Included in D-RTR HRM X X ' e Y = D-TN OCP [le mi] D - T N O C P Le le] D - R T R H R M mi In (76a) above, we see that if it could be shown that enclitics were included in the RTR-harmonic domain, there are two possibilities: There is either a reference to two distinct domains in the same constituency type (as in (75)) or a reference to two domains in different constituency types (as in (76a)). However, if it could be shown that enclitics were excluded from the RTR-harmonic domain, then this is only consistent with possibility two (76b), implying that different constituency types are referred to by the RTR-harmonic domain and the tonal-OCP. A second point concerns the syntactic asymmetry between the enclitic position and the proclitic position. Syntactic structure is right branching in Yoruba, meaning that the enclitic is a sister to the verbal head. The proclitic will then necessarily occupy a syntactic position that is less closely related to the verbal head. Regardless of where we analyze the proclitic in syntax, it cannot possibly be in a closer relationship with the verb than the enclitic. (77) Syntactic Constituency is Right-Branching Therefore, any reference by the RTR domain to syntactic structure (direct or indirect) will necessarily include the enclitic position as well. On the other hand, if the RTR domain does not reference syntax in any way, it is fathomable that in Moba, the enclitic might not be included in the RTR domain, while the proclitic might be.48 This is possible but unlikely. It would involve separate domains for proclitics and enclitics. At this point in the analysis, it is possible that the proclitics could be parsed in the PrWd and enclitics in some prosodic category dominating the PrWd. This would X e le mi 67 Therefore, an underlyingly ATR enclitic that definitively does not harmonize with the verbal host is consistent only with an RTR-harmonic domain that crucially does not refer to syntactic structure. An underlying ATR enclitic that does harmonize with the verbal host is consistent with either a reference to syntactic structure or a lack of such a reference. The most interesting outcome would be one where Standard Yoruba and Moba differ in that Standard Yoruba enclitics harmonize with the verb and Moba enclitics do not (recall that the opposite is true for proclitics). This outcome would strongly suggest that the dialects differ with respect to a reference to syntax: Standard Yoruba has a harmonic domain that refers to syntax somehow (right-branching structure), while Moba has one that does not (left-branching structure). The conclusion that can be drawn from the above discussion is that if it can be shown that enclitics are not included in the harmonic domain in Moba, an argument can be made for a reference to a prosodic domain that does not refer to syntactic structure in any way. Likewise, we would then argue against that same type of structure as a possible domain for the tonal OCP by invoking the reasoning in the first argument above - implying a syntactically defined domain for the tonal-OCP. The other possible scenario, where evidence was found implying that the enclitic was part of the harmonic domain in Moba, would not say much at all about the nature of the type of domain that is being referred to in either dialect based on this alone. 3.5.4 Implications for OT Accounts Alignment-based accounts are compatible with the Moba pattern. A l l that is needed, in fact, is to redefine ALIGN(RTR, L, PrWd, L) in Pulleyblank's (1996) account to align with the left edge of the CIGp in Moba rather than the word. The difference between Moba and Standard Yoruba is exactly that: ALIGN(RTR, L, CIGp, L) is active in Moba, while ALIGN(RTR, L , PrWd, L) is active in Standard Yoruba. Additionally, the ALIGN(RTR, L, CIGp, L) must be further subdivided into constraints governing left- alignment of ATR and left-alignment of RTR. Specifically, ALIGN(RTR, L , X , L) must left-align with the CIGp, while ALIGN(ATR, L, X , L) must left-align with the PrWd. This is true because while RTR is found to spread leftward from roots to proclitics, thus overwriting underlying ATR values on proclitics, ATR does not extend into the clitic domain, thus allowing underlying RTR values on proclitics. The difference then would be encoded directly in the left-alignment constraint and all other facts of the analysis would remain the same. However, alignment as a harmony-driver in Yoruba has other problems, which have already been raised already. imply two levels of prosodic structure above the level of the PrWd, one for proclitics, one for enclitics. I will argue against this in section 4.4 however. 68 One final issue with an alignment-based account is that it needs to be adjusted to be extendable to other dialects of Yoruba. For example, Ife allows harmony across high vowels and Ekiti allows high vowels to undergo active retraction. Recall that these effects were captured in an alignment-based approach (Pulleyblank 1996) by defining local alignment, which is satisfied by inserting features. Thus, in a form like odide, the final RTR vowel only incurs one violation of left-alignment by virtue of the inserted feature, while in odide, left-alignment incurs the usual two violations. Of these issues raised above, one involves a well-motivated interpretation of the basic alignment family. The need to adopt local alignment is motivated in order to capture the dialectal variation in Yoruba. It is motivated by any harmony system where transparency is attested. This would appear to be the escape hatch for ALIGN in these cases. If, however we do not want to use gradient alignment constraints, as McCarthy (2003) suggests, then we are at a crossroads. This is true because we cannot adopt a categorical alignment constraint that captures the facts of Yoruba RTR harmony. Instead, Chapter 5 presents an account that utilizes alignment not as a harmony-driving constraint, but as a method to map prosodic structure onto morphological structure. The constraints ALIGN(PrHd, R, PrWd, R) and ALIGN(PrWd, R, ROOT, R) will enforce that the prosodic head is right-aligned with the morphological root, so that this root-final vowel can be referred to formally via OT constraints. As in the alignment-based account, a stem-control-based account would handle the difference between the domain sizes for RTR harmony in Moba and Standard Yoruba by setting the domain of application for the active constraints. For example, the AGREE(ATR) constraint would apply over the PrWd in Standard Yoruba. This would exclude the clitics from the harmonic domain and they would thus be allowed to contrast freely for ATR/RTR. However, the facts of Moba Yoruba are less clearly accounted for in the stem-control account. The constraint set used by Bakovic (2000) is unable to account for the pattern seen in the clitics. AGREE(ATR) does not distinguish between ATR and RTR, and therefore there is no way to simultaneously enforce harmony targeting the A T R proclitics and to prevent harmony targeting the RTR proclitics. If we set the domain for AGREE(ATR) to a domain, X , that includes proclitics, this would correctly predict that the A T R proclitics harmonize with their verbal hosts, but it would incorrectly predict that the RTR proclitics should do the same. This is illustrated below in (78) and (79).49 It is assumed here that an output-output correspondence is established between a 'stem' consisting of the verbal host of cliticization and an 'affixed form' that is essentially a cliticized form. SA-IDENT(ATR) can then refer to this correspondence in the same way it did with true affixed forms. If this assumption is not made, the facts of the clitics must be explained via some independent method in the stem-control account. 69 (78) A T R Proclitics Harmonize in Moba Yoruba Stem: [je] le j e / SA-IDENT(ATR) AGREE(ATR) IO-IDENT(ATR) a. e je *! b. e je * (79) RTR Proclitics Incorrectly Predicted to Harmonize in Moba Yoruba Stem: [se] /6 s e / SA-IDENT(ATR) AGREE(ATR) IO-IDENT(ATR) a. 6 se *! ®° b. 6 se c. 6 se *! i • ' . : : j . ' - ^ ,~ , . The enclitic environment provides us with the only definitive test where we might actually expect rightward harmony in a stem-control account. Note that the enclitic facts might be consistent with stem-control theory, in that at least there is no leftward harmony triggered by the 2SG enclitic when it is attached to a verb with a mid A T R vowel. However, stem-control theory makes the incorrect prediction that an RTR enclitic should not be able to surface faithfully following an A T R verbal base. This is illustrated in (80) below. (80) RTR Enclitics Incorrectly Predicted to Harmonize Stem: [le] / a d e le o/ SA-IDENT(ATR) AGREE(ATR) IO-IDENT(ATR) ^ a. ade le o *! ^ b. ade le o c. ade le o i i *! It is impossible within the stem-control account to both allow A T R proclitics to harmonize and to disallow RTR proclitics and enclitics to harmonize. If we restrict the domain of AGREE(ATR) to exclude the clitic domain, we need some other mechanism to enforce harmony in A T R clitics, as can be seen in (78) above. If we allow the domain to include the clitics, then we need an additional mechanism to disallow harmony in the RTR clitics, as can be seen above in (79) and (80). 70 On the other hand, the alignment-based account taken strictly does not necessarily allow for spreading onto an enclitic. If right-alignment to the root means perfect alignment such that spreading one vowel beyond the edge of the root incurs a single violation of right-alignment (if over-alignment is worse than perfect alignment, in other words), then this account predicts that mid ATR enclitics should not harmonically alter to agree with their verbal hosts. Unfortunately, the lack of A T R mid vowel enclitics in Moba does not allow us to test these predictions. One final point concerning the enclitics is in order. If the 3SG enclitic in Standard Yoruba is assumed to exhibit RTR harmony (which it does only by virtue of the fact that it is apparently a total copy of the vowel quality of the verbal root) then we have apparent evidence against the right-alignment-with-the-root account. However, even if no extra violation of alignment is incurred by an over-aligned domain, those RTR and ATR feature values that fall outside the usual domain for harmony are protected via faithfulness constraints in any case. Therefore, an over-aligned candidate could never win because it would incur an extra faithfulness violation (assuming faithfulness is ranked high enough). The only way the 3SG enclitic (or any enclitic or proclitic in Standard Yoruba) could be targeted in this position at the exclusion of all other clitics, ATR and RTR alike, is if it were underspecified for featural content. Faithfulness would still be relevant in preventing insertion of featural material and in preserving the high-tone mora. However, by ranking DEPLINK-ATR and DEPLINK-RTR sufficiently low, there is nothing to stop spreading of the root node of the preceding root vowel. In fact, progressive assimilation is a well-attested phenomenon in Yoruba when two vowels come into hiatus (as was seen in the pattern of Cl-deletion in section 3.2.4). Therefore, it is not surprising that progressive assimilation might be called on to fill in the values of this featurally unspecified morpheme. An account utilizing prohibition-type constraints is more successful than a stem- control account in its ability to capture cross-dialectal variation typologically and it does not come with the problems that an alignment-based account does concerning gradient evaluation. This account is presented in Chapter 5. It posits MAXLlNK-type faithfulness constraints that apply over different prosodic domains ranked with *FG-type sequence prohibition constraints that also apply over these same prosodic domains. By ranking constraints that apply over a tighter prosodic domain above those that apply over a wider prosodic domain, the pattern seen in Moba Yoruba can be accounted for. 71 3.6 R T R Harmony Outside the Verba l Domain 3.6.1 RTR Harmony and Adverbials While the above data show that RTR harmony is triggered by verbal hosts on proclitics in Moba, it turns out that this is not strictly a property of verbs. Adverbials that occur preceding verbs but following the proclitics are examined in this section. We fully expect that any class of lexical items that constitutes a morphological root should act as a potential trigger of RTR harmony under any of the OT accounts we have seen. This prediction is borne out as can be seen by attaching the 1SG, 3SG, NEG, and FUT proclitics to these adverbials. I do not offer Standard Yoruba versions for 1SG FUT forms since they do not parallel the Moba future structure in question. (81) Adverbials: RTR Harmonic Triggers M B (SY) Gloss e tete de e papa de e tete lo i e papa lo 6 tete de 3SG= 'early' 'arrive' 6 p a p a de 3SG= 'in the end' 'arrive' 6 tete lo 3SG= 'early' 'go' 6 p a p a lo 3SG= 'in the end' 'go' ke tete de ko tete de NEG= 'early' 'arrive' ke papa de ko p a p a de NEG= 'in the end' 'arrive' ke tete lo ko tete lo NEG= 'early' ' go' ke papa lo ko p a p a lo NEG= 'in the end' 'go' me e tete de me e papa de me e tete lo me e papa lo 1SG=FUT= 'early' 'arrive' 1SG=FUT= 'in the end' 'arrive' 1SG=FUT= 'early' 'go' 1SG=FUT= 'in the end' 'go' Meaning 'S/he arrived early' 'S/he arrived in the end' 'S/he went early' 'S/he went in the end' 'S/he didn't arrive early' 'S/he didn't arrive in the end' 'S/he didn't go early' 'S/he didn't go in the end' T will arrive early' T will arrive in the end' T will go early' T will go in the end' In the above Moba forms, the clitics harmonize with the tongue root values of the adverbials, tete and papa. As expected, no such harmony is seen in Standard Yoruba. Again, this harmony extends across sequences of two clitics in Moba. The RTR-value of the verb is inconsequential. This result follows if we assume that the adverbials are morphological roots. A stem-control account would posit that the 'prefixal' material must agree with the root (the final vowel in the adverbial). In an alignment-based 72 account, the RTR value of the adverbial root must align with the right edge of the adverbial, since it is a root. This right-alignment overrules any left-alignment in Pulleyblank's (1996) account since ALIGN(RTR, R, ROOT, R) » ALIGN(RTR, L, PrWd, L). Therefore, it is more important to align a root value of RTR with the right edge of a root than it is to left-align to whatever domain left-alignment refers to. Each occurrence of a root signifies a 'reset' for RTR harmony then. 3.6.2 Root versus Non-Root Status The question that arises next, is what exactly gives a lexical item status as a morphological root. In Yoruba, verbs, nouns and adverbials act as roots, while affixes do not. Clitics are interesting in that when they contain ATR mid vowels, they act like affixes in Moba since they undergo harmony. However, the 2SG form, which contains an RTR mid vowel, superficially acts like a root in that it 'resets' the harmonic domain. The fact that there is a split where ATR clitics are targeted in harmony but not RTR clitics, suggests that in fact there is another explanation. By splitting the harmonic driver into separate constraints, one referring to ATR, the other referring to RTR (by splitting AGREE into two separate constraints, *ATR-RTR and *RTR-ATRfor example or alternatively by splitting ALIGN into ALIGN(ATR) and ALIGN(RTR)) we can specify different domains for two separate harmony-driving constraints. This split is necessary since it is clear that RTR harmony can apply in the clitic domain (see tableau (78) above) but that ATR harmony cannot (see tableaux (79) and (80) above). Under this view, there is nothing particularly special about the 2SG marker other than it is specified underlyingly RTR and the harmony driver over the clitic group only enforces RTR harmony and not ATR harmony. By splitting the harmony-driving constraint in this way, it is possible to capture this pattern seen in the clitic domain in Moba (this will be demonstrated in section 5.1). Assuming clitics and affixes are not roots, we still can ask what it is about them that prevents them from being analyzed as roots. Prosodically speaking, Yoruba has a strict templatic requirement for verbs to be minimally (and maximally in most cases) CV. Many of the subject proclitics are C V , however and this has not elevated them to root- like status. However, subject proclitics are not generated in the same syntactic positions as verbs. Therefore, they should not be subject to the same minimal requirements as verbs are. Instead, they occupy nominal positions, and therefore, they are subject to the prosodic requirements that the Yoruba noun must meet: They must be minimally V C V . Since the clitics do not meet this V C V requirement, they are not roots. Additionally, by considering the strong subject pronoun forms in Yoruba, which do meet the minimal V C V requirement, we can see a clear example of a contrast between roots and non-roots. The following data is from Pulleyblank (1986:46). 73 (82) NP Coordination in Standard Yoruba [taiwo ati kehinde] lo ki i Taiwo and Kehinde go greet her 'Taiwo and Kehinde went to greet her' [emi ati kehinde] lo ki i I and Kehinde go greet her 'Kehinde and I went to greet her.' * [ m o ati kehinde]. . . I and Kehinde... * [keh inde ati mo]. . . Kehinde and I... *[mo ati 6]... I and he... *[6 ati mo]... he and I... It is possible to co-ordinate the full V C V form of the pronoun, but it is not possible to co-ordinate the weak form of the pronoun (the clitic). This is the basis for an analysis of these weak pronouns as clitics rather than as full NP's (Bamgbose 1966, 1967; Pulleyblank 1986). This provides another aspect where root and non-root morphemes differ.5 0 We have seen evidence, both within phonology and in syntax, that non-root morphemes like clitics and affixes are treated differently than root morphemes are. This motivates a split between roots and non-roots that is crucial in terms of defining the harmonic domain of RTR. While non-roots seem to act as targets in leftward RTR harmony, the right edges of roots seem to align with the right-edge of an RTR harmonic domain. In cases where a non-root also resists being targeted in RTR harmony, such as the 2SG form noted above in Moba, this will be analyzed as following from interactions of violable constraints in an OT account. In the analysis presented in chapters 4 and 5, it is assumed that roots project well-formed prosodic words, while non-roots do not. This will be instantiated via an ALIGN constraint that aligns the right-edge of a root with the right edge of a PrWd. On the other hand, non-roots will not be subject to alignment 5 0 These strong forms could be contrasted in terms of their harmonic status ideally. In this situation, we would expect that the strong pronouns with a final A T R mid vowel would not harmonize with a following verb, by virtue of their root status. However, there are no such ATR-mid-vowel final forms. The 2SG form contains a final mid RTR vowel (just as the corresponding clitic form does), however we would not expect to find harmony here for the same reasons we do not find it with the clitic form - only ATR mid vowels are targeted by tongue-root harmony in this position. 74 constraints such as these. Independently motivated prosodic constraints will then handle other aspects of the prosodic structure, such as where to parse non-roots and what constraints are placed on the minimal and maximal PrWd, foot or syllable. These topics are discussed in detail in the next chapter. The discussion above is compatible with a system where prosody rather than morphology is crucial in determining the RTR-domain. We have seen that prosodic categories and morphological categories have tended to coincide in Yoruba. To the extent that evidence for morphological or syntactic constituency is lacking in some cases in Yoruba, a speaker might be forced into referring directly to emergent prosodic domains in order to build the kinds of generalizations that result in systematic phonological patterns like RTR harmony. We have seen many examples where the harmonic domain cannot be determined. With a lack of morphological cues, a speaker is forced to refer to an emergent prosodic constituent that is consistent with the language data they are hearing. Such constituents are shown to exist in the following chapter. Ola (1995) argues for prosodic structure in Standard Yoruba based on a number of phenomena. These arguments will motivate an account that uses prosodic structure rather than morphological structure in accounting for the differences between the RTR harmonic domain in Moba and Standard Yoruba. 75 Chapter 4 - Analysis of Domain Size 4.1 Introduction The basic pattern of RTR harmony exhibited above in Moba differs from Standard Yoruba in that its domain includes the class of proclitics, while in Standard Yoruba these proclitics fall outside the harmonic domain. This difference regarding the clitics is a result of the presence of a prosodic category, 'clitic-group' (CIGp), that dominates the PrWd. This can be accounted for in two conceivable ways. Hypothesis one states that the PrWd includes clitics in Moba but not in Standard Yoruba. RTR harmony refers to the PrWd in both Moba and Standard Yoruba in this account (the CIGp is redundant and possibly absent in this hypothesis). This would involve a difference in what syntactic category is mapped onto the PrWd in each dialect. Hypothesis two states that the PrWd does not include clitics in Moba or Standard Yoruba. The RTR-harmonic domain would then refer to the CIGp in Moba, and the PrWd in Standard Yoruba. Under this hypothesis, the domain difference is not a result of a difference between the syntax-prosody mapping. Instead, different prosodic categories are referred to in each dialect and this underlies the difference in domain-size. In order to evaluate which of these two hypotheses is correct, independent processes showing similar domain effects (at the word-level) need to be evaluated in Standard Yoruba and Moba. One possible candidate is nasal harmony. The prediction under hypothesis one is that if nasal harmony refers to the PrWd in each dialect (as it does for RTR harmony), a similar split should be seen where proclitics are included in the nasal domain in Moba, but not in Standard Yoruba. Under hypothesis two, however, there is no reason to expect that the two dialects are referring to the same prosodic category to begin with. The nasal harmonic domain might refer to any prosodic constituent in this account. Although, nasal harmony is used as an example here, the general expectation under hypothesis one is that any process that also refers to the PrWd in both Moba and Standard Yoruba should show the same split where proclitics are included in the domain for that process in Moba but not in Standard Yoruba. The goal is then to find such processes that fit this description. 4.2 Prosodic Structure in Standard Yoruba Before testing the above hypotheses, an outline of the arguments for prosodic constituency in Yoruba is presented. The prosodic status of clitics can only be analyzed 76 once this basic picture is presented. Ola (1995) has highlighted evidence for syllable- structure, foot-structure, and word-structure in Yoruba. These arguments are outlined below. Constraints in OT are posited to uphold this prosodic structure and to map it onto morphological structure so that prosodic domains can be defined. These prosodic domains must be defined so that they can be properly referenced by the constraints that will be used in an OT account of RTR harmony in Moba Yoruba. 4.2.1 Syllable Structure in Standard Yoruba Consider the process of Cl-deletion in Standard Yoruba that was described in section 3.2.4 above. In V C V C V nouns, C l can optionally (and sometimes obligatorily) be deleted to render V V C V nouns. This is shown again in (83) below, for two of the examples that were presented in (63) above. (83) Cl-Deletion in V C V C V nouns in Standard Yoruba Full Form After Cl-deletion Gloss erupe eepe 'earth' o w u r o oo rq 'morning' However, when we consider V C V nouns with a medial 'r,' we find that in this context, Cl-deletion is not an option. (84) Cl-Deletion Blocked in V C V nouns in Standard Yoruba (from Ola 1995) Full form After Cl-deletion Gloss on *o i 'head' ara * a a 'thunder' oro * o o 'pain' oro i i * o o • i 'wealth' This ban on Cl-deletion in V C V forms (but not V C V C V forms) seems to rest on the fact that the surface form must retain at least one consonant. Notice that both the full form and the unattested form with Cl-deletion both have two vowels (or two moras). Ola (1995) sees this ban as evidence that all syllables must contain onsets in Yoruba and that all words must minimally contain one syllable. The constraint, ONSET is undominated in Standard Yoruba under this view, and is ranked above PARSER, a), which forces moraic elements to be parsed under a syllable. Onset-less syllables are analyzed as being nuclear 77 moras licensed not by the syllable but directly by the word. Therefore, the distinction between onsetless vowels (which are moraic but not syllabic) and onset-full vowels (which are syllabic) is captured.51 4.2.2 Foot and PrWd Structure in Standard Yoruba Concerning foot structure in Standard Yoruba, Ola (1995) demonstrates the relevance of binary feet in reduplicative templates. One such reduplicative template is the ideophone reduplication signifying 'disorderliness'. A tonal melody of H M L M is mapped onto a reduplicated C V C V base. This reduplication is illustrated below. (85) Standard Yoruba Ideophone Reduplication signifying "disorderliness" (from Ola 1995) Base Reduplicated Form Gloss ja la ja lajala 'moving shabbily' balu balubalu 'unsteady movement' yele ye leye le 'carelessly' wuru wu ruwuru 'disorderly' rada radarada 'sluggish' boro bo roboro 'open and drippy' As can be seen above, a C V C V base undergoes total reduplication and the tonal melody of the reduplicative process is mapped onto the resulting form. This reduplicative process does not extend, however, to bases with more than two syllables. 5 1 Ola (1995) presents evidence in addition to that shown here that are consistent with obligatory CV structure in Standard Yoruba syllables. Among this evidence, she cites loan verb truncation, word-initial morpheme structure conditions on high tone, nasality and high back vowels, distributive reduplication and vowel hiatus resolution. 78 (86) Standard Yoruba Ideophone Reduplication - Unattested Forms (from Ola 1995) Base Reduplicated Form Gloss repe te * repe te repe te 'bulky (soft)' g b a r a g a d a * g b a r a g a d a g b a r a g a d a 'falling' We find that only disyllabic bases are possible candidates for this type of reduplication. Ola argues that this process is governed by a requirement on the reduplicant to be a well- formed foot. Since feet can be at most two syllables, this explains the data in (86). The constraint, BlNARY(Ft, a) must dominate PARSE(a, Ft) in order to militate against feet with more than two syllables. 5 2 Ola further argues for right-headed feet (as opposed to non-headed morphological feet). The pattern of Cl-deletion is again central to this hypothesis. In V C V C V nouns containing two 'r's, it is the first V that deletes and not the second V . (87) Cl-Deletion in Standard Yoruba: V C V C V Nouns with two r's Full Form After Cl-deletion Gloss oror i oori * oo i *oroi 'mausoleum' From this pattern, Ola deduces that Cl-deletion is allowed only in the case that it can still preserve a right-headed foot. Under this hypothesis, feet must obligatorily be iambic. Additionally, parse constraints enforce that syllables are preferentially parsed in feet and that feet are preferentially parsed under a PrWd. The constraints PARSE(Ft, PrWd) and PARSE(cr, Ft) enforce this. As for the size of the PrWd in Standard Yoruba, Ola provides arguments (that I will not include here) that it is minimally a single foot and maximally a pair of binary feet. A CV verb then constitutes that minimal PrWd, since feet must minimally contain one head syllable and a PrWd must minimally contain one foot. The upper limit of a PrWd containing maximally two binary feet is supported by the fact that Standard Yoruba roots can be four syllables in length at most. Additional arguments in Ola (1995) for the presence of foot-structure in Standard Yoruba include agentive reduplication, numeral distributive reduplication and back harmony. 79 4.2.3 Prosodic Constituency in Moba Yoruba It remains to be shown where Moba fits into this prosodic picture. The data in (64) above demonstrate that Moba allows glide deletion in V C V nouns to yield a V V form. In Standard Yoruba, this would violate minimality requirements since there is no way to syllabify an onsetless vowel (Ola 1995). However, if PARSE(p., a) were ranked above ONSET in Moba, so that onsetless vowels are possible syllable nuclei, we can account for this. In the account following, therefore, it is assumed that Moba and Standard Yoruba differ only in that Moba allows onsetless syllables. The relative ranking of ONSET and PARSE(p, a) would handle this straightforwardly as explained above: In Moba, the ranking PARSER, or) » ONSET is found, thus allowing onsetless syllables. In Standard Yoruba, the reverse ranking, ONSET » PARSER, a) holds, thus preventing onsetless moras from being syllabified. This would account for the pattern of w-deletion seen in Moba and not in Standard Yoruba straightforwardly. However the fact that Moba does not exhibit the pattern of Cl-deletion and assimilation that is seen in Standard Yoruba remains unexplained. In order to capture the domain effects seen in RTR harmony in Moba using prosodic constituency, it is necessary to formally define the prosodic categories. The prosodic head (PrHd) is central to the OT analysis offered in the next chapter. It is assumed that every PrWd contains a head foot and that every foot contains a head syllable. The constraints, HEAD(PrWd)=Ft and HEAD(Ft)=a force every PrWd to contain a head foot and every foot to contain a head syllable. This, in turn, forces every PrWd to contain a single prosodic head that is the head syllable in the head foot of that PrWd. In order to enforce right-alignment of the prosodic head with the PrWd, the constraints, RIGHTMOST and RHTYPE=I are posited. RIGHTMOST is essentially a special kind of alignment constraint that says that the head foot of a PrWd is right aligned with the right-edge of that PrWd (Prince and Smolensky 1993). RHTYPE=I forces feet to be iambic, or right-headed. Together these two constraints enforce right-alignment of the PrHd with the PrWd. This set of constraints then defines a position, the PrHd, as the rightmost syllable in the PrWd. Finally, this prosodic constituency just described is mapped onto morphological structure via the constraint ALIGN(PrWd, R, ROOT, R). This set of constraints allows an OT account to refer to the root-final vowel by referring to the PrHd. Constraints that refer to this position (such as those posited for Ife and Ekiti Yoruba in section 2.4) can now be formalized as constraints referring to the prosodic head position. Recall that the harmony-via-prohibition account for Standard Yoruba utilized [ M A X - R T R ] R O O T m order to preserve root values of RTR. An account is proposed in chapter 5 that utilizes a similar constraint that refers to the PrHd, rather than 80 the root. This positional faithfulness constraint, [MAX-RTR] P r H d does not require any version of an OCP constraint though in order to rule out multiple root values on the surface. In the prosodic account, there is exactly one PrHd per PrWd and so the job of the OCP in the root domain is accomplished via constraints on prosodic structure. Under this prosodic framework described above, a V C V noun in Yoruba (both Moba and Standard Yoruba presumably) would consist of a single binary foot. (88) Prosodic Constituency for V C V nouns in Yoruba (Ola 1995). This representation holds of lexical items that contain one single root and possibly also of forms with prefixes (prefixes might instead constitute instances of nuclear moras linking directly to the PrWd). However, in Standard Yoruba the domain for RTR harmony must be the PrWd. It could not be the foot, since there are trisyllabic roots with harmony across all three syllables in both dialects and since feet are binary. If the domain exceeded the PrWd in Standard Yoruba, then we would expect the proclitics to harmonize with their verbal hosts; but they do not. Unlike the other prosodic categories, the CIGp is not necessarily binary- branching. The fact that we can stack multiple clitics, one on top of the other, implies either an iterative domain or a non-binary domain. Iterativity is only useful in that it can uphold a binary branching structure. There is nothing forcing the CIGp to be a binary- branching category though. In other cases, the maximum size of a prosodic category is binary: PrWd can contain at most two feet and feet can contain at most two moras (or syllables).53 These properties are defined via violable constraints. I assume that the Feet are assumed to be binary either with respect to moras or syllables. I make no claims as to which of these categories are referred to, as this does not effect the analysis presented. 4.2.4 On the Prosodic Status of Clitics PrWd uc 81 constraints, BlNARY(PrWd, Ft) 5 4 and BINARY(Ft, \i(lo)) dominate PARSE(Ft, PrWd) and PARSE(a, Ft) respectively and that this delimits the maximum size of the PrWd and Foot respectively. However, in the case of the CIGp, there is no evidence for an upper limit since multiple clitics can stack one on top of the other. In this case, PARSE(PrWd, CIGp) dominates any constraint that delimits the maximum size of the CIGp (such as BINARY(ClGp, PrWd) for example), allowing the CIGp to branch more than twice. This is illustrated in (89) below. (89) CIGp is not Subject to Binary-Branching Requirements T rWd Nuc V e 1PL=NEG FUT 'We will not eat' Of note in the structure in (89) above is the fact that the negative proclitic constitutes a well-formed syllable, since it has an onset. Given the prosodic analysis above, the undominated constraints PARSE(cr, Ft) and PARSE(Ft, PrWd) would imply that this proclitic should project a PrWd category of its own. However, non-roots such as the negative proclitic do not constitute well-formed prosodic words in Yoruba as was outlined in section 3.6.2.55 In order to prevent non-roots from projecting a PrWd category, some constraint must dominate PARSE(Ft, PrWd) that militates against non- roots that project a PrWd category. We have seen one constraint that can do exactly this: ALIGN(PrWd, R, ROOT, R). By ranking ALIGN(PrWd, R, ROOT, R) above PARSE(Ft, PrWd), non-roots are prevented from projecting a PrWd category. Any PrWd that is not right aligned with the right-edge of a root would incur a violation of this constraint. Since the negative marker is not a root, the structure in (89) is selected optimally ahead of Note that this two-foot restriction applies only to roots. Affixed forms can exceed two feet. 5 5 Whether or not they constitute well-formed feet is also debatable. In (89) above, I assume they do, although there is nothing to say that a non-root should constitute a well- formed foot. The main point is that it cannot constitute a well-formed PrWd. 82 a structure where the negative marker projects a PrWd. A foot would only be required to project a PrWd if this PrWd can be right aligned to a root, as is the case with the verbal root in (89). Now turning to the RTR-harmonic domain, in Standard Yoruba, it maps onto prosodic structure as shown in (90) below. On the other hand, we have two possible mappings of the harmonic domain onto prosodic structure in Moba. These correspond to the two hypotheses set forward at the outset of this chapter. (90) Standard Yoruba: Prosodic Constituency of Clitics V [C proclitic verb 83 (91) Moba Yoruba: Prosodic Constituency of Clitics Hyp. 1: Domain=PrWd Same domain as SY; Parsing of Clitics different CIGp [p[wd]D (92) Moba Yoruba: Prosodic Constituency of Clitics Hyp. 2: Domain=ClGp Same parsing of clitics as SY; Domain different [ClGp]D PrWd [V C V] D proclitic verb As was previously noted above, these two hypotheses make different predictions concerning the difference in the size of domains of other processes that also refer to the PrWd. Namely, hypothesis one predicts that other processes that refer to the same 84 prosodic domain in Moba and Standard Yoruba, should exhibit exactly the same dialectal split that is seen in the RTR harmonic behaviour of the proclitics. Under hypothesis two, we can expect harmonic domains to coincide only by chance. With each process that is attested to exhibit the dialectal proclitic split, this can be taken as positive evidence for hypothesis one. This is first tested on the domain for nasal harmony in the next section. 4.3 Nasal Ha rmony Nasal harmony is another instance of harmony that occurs in both Moba and Standard Yoruba. Unfortunately, it is not ideal for our purposes however, because in Standard Yoruba nasal harmony is syllable-bound, but in Moba Yoruba, it extends beyond the syllable. Since the PrWd is not a possible domain for nasal harmony in Standard Yoruba, it does not meet the requirement that the domain be the PrWd in both dialects. This requirement is needed in order to test hypothesis one for nasal harmony. However, it is a separate question, interesting in itself, to ask if the harmonic domain of RTR harmony and nasal harmony in Moba coincide. If it can be shown that nasal harmony does not extend to the proclitic domain but that it nonetheless applies minimally over the PrWd, then this, in itself, would argue against hypothesis one. Under this hypothesis, the proclitics are included in the PrWd by virtue of their being included in the RTR harmonic domain. However, if nasal harmony can independently be shown to apply over the PrWd, a contradiction emerges. How could the proclitics simultaneously occur in the PrWd but not be included within a nasal harmonic domain that is minimally the PrWd? If, on the other hand, it is found that the proclitics are contained in the nasal harmonic domain, this is consistent with either hypothesis above. 4.3.1 Syllable-bound Nasal Harmony Roots in both Moba and Standard Yoruba exhibit nasal harmony that is triggered by nasal vowels. Recall that only high and low vowels have nasalized counterparts in Moba and that in Standard Yoruba the mid vowel, 6 can occur as an allophone of a. Nasal harmony is thus restricted phonologically much as RTR harmony is. Unlike RTR harmony, consonants play a more central role in this system. Liquids and glides are nasalized when preceding a nasal vowel in both dialects. This behaviour is shown below (phonetic transcriptions are given for nasal harmony). 85 (93) Syllable-bound Nasal Harmony in Moba and Standard Yoruba M B SY Gloss ogu ogu 'twenty' ami ami 'sign' i b a d a i b a d a the city - Ibadan d f a o f a 'trouble' e f i e f i 'elephant' wa / a wo / *5 'measure' d j i lgbo o p l g b o a place in Lagos5 6 As can be seen in the final three examples above, the glides /y / and /w/ and the liquid, Ix] are nasalized when they precede a nasal vowel. The liquid /I/ is in complementary distribution with the phoneme /n/. /I/ is never found to occur preceding a nasal vowel and likewise /n / is never found to occur preceding an oral vowel. 4.3.2 Nasal Harmony across Syllable Boundaries in Moba The data in (93) are consistent with a pattern of syllable-bound nasal harmony (or merely assimilation) since preceding vowels apparently are not targeted in nasal harmony. When considering high vowels, however, we can see the first signs of vowel harmony. (94) Nasal Harmony Targets High Vowels in Moba M B SY Gloss a g u t a a g u t a 'sheep' d l ^ p l t a o l ^ p i t a 'historian' i n u i n u 'stomach' f I f l r i f i 'grating' (gerundive reduplicant) u r i o i f i w o 'four hundred' The data in (94) show that nasal harmony is strictly syllable-bound (and therefore strictly local) in Standard Yoruba. While any nasal vowel is a potential trigger, only high vowels The phoneme /y / (IPA glide /j/) has a palatal nasal allophone, [n] when it precedes a nasal vowel in both Moba and Standard Yoruba. 86 are potential targets. Low and mid vowels are not targeted in nasal harmony, on the other hand. Another feature of Moba nasal harmony is that obstruents are transparent. This is seen in the first two examples of (94) where nasal harmony occurs across a voiceless obstruent. 4.3.3 Nasal Harmony Beyond the Root Concerning proclitics, we find again that Yoruba does not provide us with the relevant context for checking the harmonic status of the proclitic position in nasal harmony. This problem is not surprising, however; we are limited to underlying high oral vowels preceding verbs with nasal vowels. This is a highly restricted environment and unfortunately, there are no examples of proclitics in Yoruba that contain high oral vowels (see (67) and (68) above - the 2PL proclitic is high but underlyingly nasal: in). Since nasal harmony is strictly leftward (as can be seen in Moba: egusi, *egusi 'melon / a food made from seeds of melon'), we would not expect nasal clitics to trigger rightward nasal harmony onto their verbal hosts anyway. However, one potential test concerning clitics can be conducted. Given leftward nasal harmony onto high oral vowels, can high nasal vowels in an enclitic spread leftward onto a verbal host? If enclitics were to trigger harmony onto their verbal hosts, then this would be consistent with a common reference for nasal and RTR domains (whether it be the CIGp or the PrWd). This, of course, presupposes that the enclitics and proclitics are parsed in the same prosodic domain in Moba. (95) Enclitics and Nasal Harmony M B SY Gloss Meaning e s i mi 6 s i mi 3SG='bury'=lSG 's/he buried me' e s i 0 6 s i s 3SG='bury'=2SG 's/he buried you' e s i i 5 s i 3SG='bury'=2PL 's/he buried you all' e s i a 6 s i wo 3SG='bury'=3PL 's/he buried them' The data in (95) above illustrate that the nasal/oral quality of an enclitic does not spread leftward onto the host verb. The vowel in the verb, si, is invariably oral in Moba and is invariably nasal in Standard Yoruba (the 2SG oral enclitic was included to show that it is not the case that nasal harmony is applying in Standard Yoruba - the verb is invariably nasal in this dialect). If root-faithfulness (or something else that protects the nasal/oral quality of the root-final vowel) protects the nasal/oral quality of the verbal base, then given the leftward direction of harmony, this is unsurprising. The nasal harmonic domain might still be any possible constituent, CIGp or PrWd included. 87 There are other morpheme boundaries where an oral high vowel-nasal vowel sequence might arise. This situation is found in WH-constructions. (96) WH-words: Evidence of Nasal Harmony across Word Boundaries M B : 5 r i k i n i SY: o r i k i n i Gloss: 2SG 'see' W H FOC Meaning: 'You saw what?' This piece of evidence implies that in Moba, there is nasal harmony across a clitic- boundary. Dechaine (2002) has argued that the FOC particle is a phrasal enclitic. This then amounts to an enclitic triggering harmony onto a morpheme. Regardless of the status of the morpheme targeted, the very fact that a clitic is triggering nasal harmony implies that nasal harmony is attested in the clitic domain, whether this is the PrWd, the CIGp, or even some higher P-Cat. A second context where nasal harmony is triggered by a clitic is found in the proclitic domain. (97) 2PL Subject Proclitic Triggers Nasal Harmony a. M B : b a i n i k i i se e SY: b a j i n i k i s Je e Gloss: 'like this' FOC COMP 2PL='do'=3SG Meaning: 'Do it like this!' b. M B : e bewe w i k i a na a se na Gloss: lSG='askfor' say COMP lPL='beat X'=1PL 'then"beatX' Meaning: 'He asked his comrades to flog him and he was flogged' Example (97a) provides an example where the aforementioned 2PL enclitic actually triggers nasal harmony on a preceding complementizer. The complementizer appears with an oral vowel normally as is seen in (97b). Again, regardless of the prosodic status of the complementizer, this is an example of nasal harmony triggered in the clitic domain, whether this is the PrWd or the CIGp. One final example illustrates that nasal harmony can actually occur in a rightward direction in the clitic domain in Moba Yoruba. This is seen in the progressive marker, I, which is normally analyzed as an auxiliary. When this progressive marker occurs following a nasal vowel in a subject proclitic, it is nasalized. This is in stark contrast to the pattern seen in the clitic domain in RTR harmony, where proclitics always agree with the tongue-root value of a following root vowel. The evidence for rightward nasal 88 harmony is shown below in (98). The progressive marker surfaces as oral normally, as can be seen when in the 3SG form below in (98d). However, when it is preceded by a nasal vowel in a proclitic, as it is in (98a, b, and c) it surfaces with a nasal vowel. (98) Rightward Nasal Harmony in the Clitic Domain a. M B : mi i de SY: mo n de Gloss: lSG=PROG 'arrive' Meaning: T am arriving' b. M B : I I de SY: e n de Gloss: 2PL=PROG 'arrive' Meaning: 'You all are arriving' c. M B : a i de SY: wo n de Gloss: 3PL=PROG 'arrive' Meaning: 'They are arriving' d. M B : i i de SY: 6 n de Gloss: 3SG=PROG 'arrive' Meaning: 'S/he is arriving' It appears, then, that nasal harmony does apply in the clitic domain. Clitic-to-root nasal spreading is not allowed due to high-ranking constraints enforcing root-identity (or possibly PrHd identity) with respect to nasal/oral quality. This would imply that the difference between the WH-marker, ki in (96) above (and also the complementizer in (97) above) and a C V verb is that the former is not a root, while the latter is. Root faithfulness prevents the verbal bases in (95) above from harmonizing with a nasal enclitic. However, if ki is not analyzed as a root, then as long as it is within the nasal harmonic domain (which it apparently is), it will participate in harmony. These same arguments hold for the progressive marker, i , which is not also not a root. 4.3.4 Summary of Nasal Harmony - Implications for Domain-Size Nasal harmony in Moba differs from its syllable-bound Standard Yoruba cousin in that it extends beyond morpheme-boundaries and it can be triggered in the clitic domain. This 89 amounts to positive evidence that at least in Moba, the nasal domain and the RTR domain both include clitics, while in Standard Yoruba they do not. Unfortunately, Standard Yoruba does not provide a test for the main hypotheses that were posited at the beginning of this chapter, since nasal harmony does not apply in the PrWd. 4.4 Clitics and Prefixes in Moba: Implications of a Domain Mismatch The 2SG proclitic form in Moba is exceptional in that it provides a single exception to the rule that proclitics harmonize in this dialect. This is repeated below in (99). (99) RTR Proclitics Surface Faithfully in Moba Yoruba 6 se 2SG='do' 'you(sg.) do/did' 6 je 2SG='eat' 'you(sg.) eat/ate' A T R Proclitics Harmonize in Moba Yoruba e s e 3SG='do' 's/he does/did' e je 3SG='eat' 's/he eat/ate' As was discussed previously, this can be analyzed by viewing the 2SG form as just another clitic that happens to be specified as RTR. However, when considering the class of prefixes in Yoruba, even though there is a relatively large variety available, not one of these demonstrates non-harmonic behaviour. (100) RTR and A T R Prefixes Harmonize in Moba Yoruba a. de de 'to hunt' o d e ode 'hunter' I I I I * o d e * o d e i • b. jou j o w u 'to be jealous' d jou o jowu 'a jealous person' *6j6u *6j6wu Assuming the Richness of the Base hypothesis, this implies that prefixes, no matter what underlying tongue-root value they may have, are always targeted in RTR harmony. This same statement cannot be applied to the proclitics however. The fact that 90 the 2SG form surfaces with RTR, while the other mid-vowel clitics behave harmonically (presumably because they are underlyingly ATR) implies that the clitics cannot be in the same domain as the prefixes. These prefixes must be parsed into the PrWd rather than the foot since there are numerous cases where they are attached to disyllabic roots. Feet are maximally binary in Yoruba and therefore, this prefix must be parsed directly to the PrWd. This implies that the proclitics are not included in the PrWd in Moba. Hypothesis 1 is thus disproved in favour of hypothesis two by this argument and the CIGp is therefore the domain for RTR harmony. This is consistent with what was found in the case of nasal harmony in Moba, coincidentally, that the CIGp is the domain for nasal harmony as well. (101) RTR Harmonic Domain Mapped onto Prosodic Structure in Moba and Standard Yoruba Moba Yoruba Standard Yoruba i [ClGp]D CIGp proclitics prefixes proclitics prefixes Regarding the prosodic status of enclitics, it is not entirely clear where they are parsed in the structure in (101) above. Since harmony is strictly leftward and root-final values of RTR and ATR are protected, there is no evidence of their RTR harmonic status. However, recall from section 3.5.2 that the tonal OCP effect was observed to apply in some domain that includes verbs and enclitics, to the exclusion of proclitics. This amounts to evidence that whatever type of constituency is referred to by the domain for the tonal OCP, it is one where the verb-enclitic pair forms a constituent to the exclusion of the proclitic-verb pair. The only way that the tonal OCP could refer to a prosodic domain would be if the enclitics were included in the PrWd. This corresponds to the structure in (75) above, repeated in (102) below with prosodic categories filled in this time (X in (75) is the CIGp, Y is the PrWd). 91 (102) Interaction of RTR-Harmonic Domain with Tonal-OCP Domain Assuming Domains Refer to Same Type of Constituency a. Enclitics Included in D-RTR HRM (Unlikely) CIGp = D-RTR HRM PrWd = D-TN OCP K [(3 [16 ITI j ] D _ T N O C p ] D - R T R H R M On the other hand, if the enclitic was parsed in the CIGp and not the PrWd, no prosodic category could refer solely to the verb-enclitic pair, so that the proclitic is excluded from that constituent. The tonal OCP must in this case refer to a non-prosodic domain. However, this directly contradicts the prosodic analysis in section 4.2.2, where it was stated that the prosodic head is right-aligned with the edge of the PrWd. In this case, the enclitic would constitute the prosodic head, and therefore, we would expect it to trigger harmony leftward onto the verbal base. If we accept the arguments in section 4.2.2 pertaining to right-alignment of the prosodic head with the PrWd, and we accept that the PrWd is right-aligned with the morphological root, then we must exclude the enclitics from the PrWd domain. Instead, they should be parsed in the CIGp, implying a distinct type of domain for the tonal OCP. Given that the tonal OCP must refer to a non-prosodic domain, then, of the two possibilities afforded it in (76) (from section 3.5.3 above), possibility two (76b) is very unlikely because it states that the enclitics are not included in the RTR-harmonic domain. This would imply an extra level of prosodic structure above the CIGp with no added gain - this is not a likely scenario. There is no reason to add a level of prosodic structure without any evidence of empirical gains. This is illustrated in (103a) below. Possibility one ((76a) from section 3.5.3 above repeated as (103b) below) is a more likely scenario - one where it is assumed that the enclitics are contained in the RTR harmonic domain.57 Without any evidence to the contrary, we must assume that (103b) is the correct representation for these domains in Moba. 5 7 Note that the alternative structure for the RTR Harmonic Domain in (76a) (where the PrWd contains the verb and enclitic and not the proclitic) is not appropriate. This is because we have already established that the prosodic structure of Yoruba designates the vowel in the verb as the prosodic head of a right-headed PrWd by right-aligning the PrWd with the root and then by right-aligning the PrHd with the PrWd. 92 (103) Interaction of RTR-Harmonic Domain with Tonal-OCP Domain Assuming Domains Refer to Different Type of Constituency a. Assuming Enclitics Not Included in D-RTR H R M (Unlikely) X X ' [le m i ] d . T N O C P [e ' e ] D . R T R H R M mi b. Assuming Enclitics Included in D-RTR H R M (Likely) X CIGp = D-RTR HRM One possibility afforded the tonal OCP domain that is consistent with the structure illustrated in (103b) above, is a direct reference to syntax. This would allow the enclitic to be parsed in any given prosodic domain. Tonal processes are often grammatically conditioned in any case. Therefore, a direct reference to a syntactic domain might be appropriate here anyway. 4.5 Summary In order to account for the fact that proclitics were included in the RTR harmonic domain in Moba, but not in Standard Yoruba, there were two hypotheses put forward. In the first, a common prosodic domain was referred to, but this domain in turn referred to different syntactic constituents. In the second hypothesis, this difference was attributed to the fact that the dialects refer to different prosodic constituents that were mapped to a common syntactic constituent. An outline of the prosodic structure of Standard Yoruba was presented based on Ola (1995). The CIGp was posited as the prosodic category dominating the PrWd in Standard Yoruba, and therefore the clitics were necessarily parsed in the CIGp. Assuming that the harmonic domain is the PrWd in Standard Yoruba, this explained why prefixes were included in the domain and clitics were not. 93 In order to test the hypotheses concerning domain references, nasal harmony was considered. The pattern seen in Standard Yoruba is syllable-bound, however, and therefore it was not possible to discern between the two hypotheses. It was possible to demonstrate that the CIGp operates as the active harmonic domain in Moba, however. Finally, based on evidence internal to Moba, it was possible to rule out hypothesis one. The fact that clitics can occur invariably as RTR on the surface, but that prefixes cannot implied that only an analysis where proclitics in Moba are not in the same domain as the prefixes is possible. Since the first hypothesis posited the PrWd as the common domain in each dialect (by virtue of the clitics being parsed in the PrWd in Moba but not in Standard Yoruba), this could no longer be the case. Otherwise, we would have expected symmetric behaviour between clitics and prefixes in Moba. The harmonic domain is thus the CIGp in both RTR harmony and nasal harmony in Moba. This had implications for the tonal OCP domain, since it interacts with the domain for RTR harmony. It was not possible for the tonal OCP domain to be prosodically defined, since this would have implied that the enclitics should be included in the PrWd domain. This was not compatible with the view that the PrWd is right aligned with the right-edge of the root. Further, it was more likely that the enclitic is parsed in the CIGp, and is thus contained in the RTR-harmonic domain. This was true since there is no evidence that the enclitics are excluded from the harmonic domain; and this does not constitute any motivation for a prosodic category dominating the CIGp that is otherwise unmotivated. 9 4 Chapter 5 - An OT Account for RTR Harmony in Moba Yoruba This chapter posits an OT analysis that accounts for the pattern seen in RTR harmony in Moba Yoruba. It utilizes the basic idea of the harmony-via-prohibition account in that constraints militating against certain feature sequences are utilized to drive harmony. It capitalizes on prosodic constituency as defined in chapter four in order to enforce constraints more rigorously in certain prosodic domains. This account is extended to account for Ife, Ekiti, and Standard Yoruba in section 5.2. These accounts given for Ife and Ekiti were actually proposed in section 2.4. The only missing piece to the puzzle was the formalization of the root-final position as the prosodic head. 5.1 An OT Account for RTR Harmony in Moba Yoruba In order to capture the pattern of Moba RTR harmony, faithfulness and harmony-driving constraints are split into constraints holding in separate domains. Among these are the PrWd domain, the CIGp domain and the prosodic head (denoted the PrHd from this point on). This account adapts the constraint set used in the harmony via prohibition account in Pulleyblank (2002). This is done in the same way as the analyses of Ekiti and Ife Yoruba in section 2.4. These accounts utilized constraints that referred to the root-final vowel, [MAX-RTR] R t F i n a l and [Hl /ATR] R t F i n a l . These constraints can now be formally defined since the PrHd has been defined as the right-aligned head syllable of the PrWd, which is in turn right aligned with the morphological root. The domains of the constraints, M A X - RTR and HI/ATR can be set to apply only in the PrHd position. [MAX-RTR] P r H d and [Hl/ATRJprHd are restricted to the PrHd domain. By ranking these domain-restricted constraints above the general constraints, HI/ATR and MAX-RTR, we can capture positional effects and domain-specific effects that were seen in Ife and Ekiti Yoruba. In the present account for Moba Yoruba, the faithfulness constraints, MAXLINK- ATR and MAXLINK-RTR are used instead of M A X - F (see section 2.4.2 for an explanation). While M A X - F does not incur any violation for the re-association of an underlying feature, F, MAXLINK-F does. There will be one crucial case where it is necessary to allow re-association between underlying RTR values that occur in the PrHd. In this case, [MAX-RTR] P r H d is used in addition to the MAXLINK constraints. The basic faithfulness constraints that will be used in this account are [ M A X L l N K - A T R ] a G p and [MAXLINK-RTR] c l G p . Since they apply over the largest prosodic domain, the CIGp, these constraints necessarily also enforce faithfulness over tighter prosodic domains such as the PrWd and the PrHd since these are both contained in the CIGp. Therefore, constraints will only be split into domain-specific pairs when it is necessary to enforce a constraint in a tighter domain. Otherwise, the constraint will simply apply over the CIGp. 95 Low vowels are obligatorily RTR. This is captured by ranking LO/RTR and MAX-LO above [ M A X L l N K - A T R ] c l G p . This is illustrated below in tableau (104). A form with an underlying A T R low vowel (a) is given to show that such a form could never surface. Candidate (104a) is ruled out since the A T R low vowel surfaces faithfully. This incurs a fatal violation of LO/RTR. Candidate (104c) is ruled out since the underlying +low value of the initial vowel is deleted. This incurs a fatal violation of MAX-LO. (104) Low Vowels are RTR /ade/ LO/RTR MAX-LO [ M A X L I N K - A T R ] C ] G p a. ade *! ••4-.-.J:*- , , . •> . . , .•»!'• , ; 4 K - ^ b. ade c. ede *! Similarly, high vowels are obligatorily ATR. This is captured by ranking HI/ATR and MAX-HI above [MAXLlNK-RTR] C I G p . This is illustrated in tableau (105) below. A form with an underlying RTR high vowel (i) is given to show that such a form could never surface. Candidate (105a) is ruled out since the RTR high vowel surfaces faithfully. This incurs a fatal violation of HI/ATR. Candidate (105c) is ruled out since the underlying +high feature of the initial vowel is deleted. This incurs a fatal violation of MAX-HI. (105) High Vowels are ATR /a t i / HI/ATR MAX-HI [ M A X L l N K - R T R ] C I G p a. ati *! «* b. ati c. ate *! The pattern of leftward RTR harmony can be accounted for by ranking the sequence prohibition constraint, *ATR-C 0 -RTR above the faithfulness constraint, MAXLINK-ATR but below MAXLINK-RTR. Since leftward RTR harmony applies across proclitics as well as within the root, the domain for these constraints is the CIGp. Therefore, the ranking in (106) below will enforce leftward RTR harmony within the entire CIGp domain. This is demonstrated below in tableau (106) with a hypothetical underlying A T R proclitic (Proclitics are marked as such in tableaux as PCI). The faithful candidate (106a) is ruled out because it incurs a fatal violation of [*ATR-C 0 -RTR] c l G p . Since [MAXLINK-RTR] C I G p outranks [ M A X L I N K - A T R ] a G p , it is optimal to delete the ATR feature on the clitic rather than the RTR feature on the root-vowel. Candidate (106c) exhibits rightward ATR harmony and is therefore ruled out because of a fatal 96 violation of [MAXLINK-RTR] C I G p . This allows the RTR harmonic candidate (106b) to surface optimally. (106) Leftward RTR Harmony in the CIGp Domain: [ M A X L l N K - R T R ] c l G p , [*ATR-C 0 -RTRl C I G p >> [MAXLINK- A T R ] C I G p le (PCI) j e / [ M A X L l N K - R T R ] C I G p [*ATR-C 0 -RTR] C I G p [ M A X L l N K - A T R ] c l G p a. e je *! & ^ b. e je c. e je *! There is a potential problem with the above ranking concerning the pattern seen with RTR enclitics. Recall that RTR enclitics surface faithfully and that they do not trigger harmony onto root A T R vowels (i.e. ade le q 'Ade pursued you (sg.)'). However, given the ranking in (106) above, we might expect leftward A T R harmony triggered by the enclitic onto the ATR root vowel. This can be ruled out by ranking the domain-specific faithfulness constraint, [ M A X L l N K - A T R j P r W d above the harmony driving constraint, [*ATR-C 0 -RTR] c , G p . This would allow leftward RTR harmony onto ATR proclitics as in (106), but it would not allow leftward RTR harmony onto root ATR vowels. This is illustrated in (107) below. Candidate (107b) fatally violates [MAXLINK- A T R ] P r W d . Candidate (107c) fatally violates [ M A X L l N K - R T R ] c l G p . The optimal candidate is the faithful candidate (107a). The violation of [*ATR-C 0 -RTR] a G p is tolerated since faithfulness constraints outrank the harmony-driver. (107) Disharmony Tolerated with RTR Enclitics Following A T R Root Vowels: ( [MAXLINK-ATR] P r W d Needed) [ M A X L l N K - A T R ] P r W d , [ M A X L l N K - R T R ] c l G p » [*ATR- C 0 - R T R ] a G / l e o i (EnCl)/ [ M A X L I N K - A T R ] ^ [ M A X L l N K - R T R ] c l G p [*ATR-C 0 -RTR] C I G p a. le o i b. le o • i *! c. le o *! However, this is not the complete story. By protecting A T R vowels in the PrWd to such an extent, leftward RTR harmony within the PrWd domain is blocked. Tableau (108) below illustrates this. The harmonic candidates (108b) and (108c) fatally violate 97 the faithfulness constraints, [ M A X L l N K - A T R ] P r W d and [ M A X L l N K - R T R ] c l G p respectively. This predicts that sequences of ATR mid vowels followed by RTR mid vowels should be attested since candidate (108a) is optimally selected under the ranking offered thus far. (108) Leftward RTR Harmony Blocked in PrWd /ebo/ [ M A X L l N K - A T R ] P r W d [ M A X L l N K - R T R ] c l G p [*ATR-C 0 -RTR] c l G p a. ebo i b. ebo i i *! r c. ebo *! This effect can be undone though by ranking the domain-specific constraint, [*ATR-C 0 -RTR] P r W d above [MAXLINK-ATR | P r W d . This would essentially enforce rightward A T R harmony in the PrWd. Again, this is potentially problematic since high vowels do not trigger rightward harmony. This problem is repaired by introducing a non- high condition on the constraint, P A T R . NONHI-Q-RTR. NONHI], so that sequences of high vowels followed by RTR vowels do not incur violations of this constraint. This is exactly analogous to the proposition in Pulleyblank (2002) where the non-low condition was introduced into the *RTR-C 0 -ATR constraint in order to allow sequences of low vowels followed by mid ATR vowels. By introducing the non-high condition, rightward A T R harmony is only exhibited in sequences of mid vowels and not high-mid vowel sequences. The non-high condition on the harmony driving constraint, r*ATR. NONHI-C 0-RTR. N O N H l ] P r W d allows sequences of high vowels followed by RTR vowels to surface faithfully. This is illustrated in (109) below. Candidate (109a) is disharmonic and would fatally violate the r*ATR. NONHI-C 0-RTR. NONHl] P r W d constraint if it weren't for the non-high condition. Instead, this faithful candidate surfaces as is and the harmonic candidates (109b) and (109c) incur fatal violations of [MAXLINK-ATR] P r W d and [MAXLINK-RTR| C I G p . (109) High Vowels do not Trigger Rightward ATR Harmony: (Non- high condition Needed in *ATR-C 0 -RTR) /ile/ t P A T R . NONHI-C 0-RTR. NONHT] r T J [MAXLINK- A T R ] P r W d [MAXLINK- R T R ] n G n ®* a. ile b. ile *! c. ile i *! The ranking established thus far actually enforces leftward RTR harmony triggered by low vowels. An underlying form with a sequence of a high vowel, mid ATR vowel and low vowel (i.e. i-e-a) would be problematic without the non-high condition. 98 This is illustrated in (110) below. Recall that in (104) above, it was demonstrated that LO/RTR and M A X - L O dominated [MAXLINK - A T R ] c l G p . Since the CIGp is a domain that spans the entire PrWd, this implies that LO/RTR and M A X - L O also dominate the constraint, [MAXLINK - A T R ] P r W d . This rules out the harmonic candidates (110c) and (HOd) below, which satisfy l*ATR. NONHI-C 0-RTR. NONHl] P r W d , but fatally violate LO/RTR and M A X - L O respectively. Candidate (110a) is ruled out due to a fatal violation of r*ATR. NONHI-Q-RTR. NONHf | P r W d due to the sequence of the mid ATR vowel and low vowel. Note that if the non-high condition were not introduced, the optimal candidate (110b) would also have violated this sequence prohibition constraint. The non- high condition allows leftward harmony triggered by low vowels then since candidate (110a) would have been selected optimally since it incurs one fewer violation of [ M A X L l N K - A T R ] P r W d . (110) Leftward RTR Harmony Triggered by Low Vowels / i - e - a / r*ATR. NONHI-C 0-RTR. N O N H l k W H LO/RTR i M A X - L O [MAXLINK - A T R ] P r W d a. i-e-a *! ^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂^̂ f b. i-e-a [ c. i-e-e * ! : * ' " d. i-e-e Returning to the clitics now, recall that RTR mid vowel proclitics can surface preceding an ATR mid vowel in the root. This disharmony is tolerated presumably because leftward A T R harmony only applies in the PrWd and not in the CIGp. Within the PrWd (in prefixes and root-internally), however, this disharmony is not tolerated. The constraint, *RTR-C 0 -ATR militates against such mid vowel sequences.58 Leftward ATR harmony must be blocked in the CIGp but must be allowed in the PrWd. This is enforced by splitting *RTR-C 0 -ATR into two constraints, one applying over the PrWd and the other applying over the CIGp. By ranking [*RTR-C 0 -ATR] P r W d above [MAXLINK-RTR] c l G p , we can allow leftward A T R harmony in the PrWd domain. By ranking [*RTR-C 0 -ATR] c l G p b e l o w [ M A X L l N K - R T R ] c l G p , leftward A T R harmony is blocked in the proclitic domain, allowing RTR proclitics to surface faithfully. Additionally, note that [ M A X L l N K - A T R ] P r W d must also dominate [*RTR-C 0 -ATR] c l G p in order to rule out rightward RTR harmony from the proclitic onto the root vowel. In fact, 5 8 Recall from section 2.4 that this constraint was formulated with a non-low condition in Pulleyblank (2002) so that sequences of low vowels followed by mid A T R vowels were allowed, but not sequences of mid RTR vowels followed by mid ATR vowels (*[RTR, NONLQl-C r rATR. NONLOIV This condition is not needed in this account. In fact, this non-low condition would actually make incorrect predictions concerning certain trisyllabic sequences. Therefore, while a non-high condition is necessary in this account, a non-low condition is not used. 99 by ranking [ M A X L I N K - A T R | P r W d above [*RTR-C 0 -ATR] P r W d , this ranking will prevent low vowels from triggering rightward RTR harmony generally. This ranking is illustrated in (111) below. Candidate (11 lb) incurs a fatal violation of [MAXLINK-ATR] P r W d . The candidate with leftward A T R harmony (11 lc) is ruled out because of a fatal violation of [MAXLINK-RTR] C I G p . This allows the faithful disharmonic candidate (11 la) to be optimally selected. Note that this optimal candidate escapes a potentially fatal violation of [*RTR-C 0 -ATR] P r W d since the RTR-ATR sequence is external to the PrWd domain.59 ( I l l ) Leftward A T R Harmony Blocked in the CIGp 16 (PCI) d e / [MAXLINK- A T R ] P r W d [*RTR-C 0- A T R ] P r W r i [MAXLINK- R T R ] n n n [*RTR-C 0- ATRl n n „ ^ a. 6 de . ... -Vwlhilj b. 6 de *! c. 6 de *! Within the PrWd, this ranking enforces leftward A T R harmony so that RTR mid vowel - ATR mid vowel sequences are disallowed. This is illustrated in (112) below. The harmonic candidate that exhibits rightward RTR harmony (112b) fatally violates [MAXLlNK -ATR] P r W d . The faithful candidate (112a) fatally violates [*RTR-C 0- ATRlprwd- The ranking of [*RTR-C 0 -ATR] P r W d above [MAXLINK -RTR] c l G p allows candidate (112c) to optimally surface. (112) Leftward A T R Harmony Enforced in the PrWd / e j e / [MAXLINK- A T R ] P r W d [*RTR-C 0- A T R | P r W d [MAXLINK- R T R ] c l G n [*RTR-C 0- A T R l n f i n a. eje *! -———- yy v.. *. b. e je *! iK - ••7 - ^ c. e je Low vowels are prevented from triggering rightward RTR harmony via the ranking already established. This is illustrated in (113) below. Candidates (113b) and (113c) are ruled out due to fatal violations of LO/RTR and M A X - L O . The RTR- However, note that this crucially relies on the root vowel being specified as ATR. If the root were unspecified, it would actually satisfy [ M A X L l N K - A T R ] P r W d and would be targeted in rightward RTR harmony that is triggered by the proclitic. Therefore, this account relies on the assumption that all root vowels are underlyingly specified either ATR or RTR. 100 harmonic candidate (113d) is ruled out due to a fatal violation of [MAXLINK-ATR] F This allows candidate (113a) to be selected optimally, despite the disharmony. This illustrates that there is no need for a non-low condition on the constraint, [*RTR-C 0- A T R ] P r W d . (113) Low Vowel-Mid ATR Sequences are Allowed / a t e / LO/RTR MAX-LO [ M A X L l N K - A T R ] P r W d [*RTR-C 0- A T R ] P r W d ^ a. ate * b. a te *! c. ete i d. ate i *! One major problem concerns the pattern of relative alignment seen in both Moba and Standard Yoruba. Recall from section 3.2.1 that an ATR/RTR contrast can exist either following or preceding a high vowel. While there is a pressure for a 'root RTR feature' to be 'right-aligned', this requirement is not absolute. When a rightmost high vowel prevents an underlying RTR feature from being perfectly right aligned with the root, it can still surface, as long as it is as right aligned as possible. This pattern of relative alignment is in jeopardy given the ranking in (112) above, where any A T R vowel will trigger leftward A T R spreading in the PrWd. How could an ATR/RTR contrast ever exist preceding a high vowel (which is necessarily ATR) within the PrWd? A solution is proposed that utilizes the constraint [MAX-RTR] P r H d . By protecting an RTR feature that occurs underlyingly on the rightmost vowel (the PrHd vowel), and allowing this RTR feature to re-associate freely on the next available anchor, relative alignment, and not absolute alignment, is enforced. [ M A X - R T R ] P r H d must be ranked above [*RTR-C 0 -ATR] P r W d in order to enforce re-association in cases where a final high vowel is underlyingly RTR. This would still enforce neutralization of an ATR/RTR contrast preceding a high vowel - the contrast will only exist in the PrHd. Disyllabic sequences where an RTR mid vowel precedes a high vowel then, are essentially cases where the high vowel, and not the mid vowel, was actually underlyingly RTR 6 0 and this RTR feature is preserved to avoid a fatal violation of | M A X - R T R ] P r H d . Since high RTR 6 0 Note that the stem-control account also assumes that the high vowel is underlyingly RTR in these cases. While it relies on positing an opaque level of representation to enforce this, the present account simply utilizes the flexibility of the M A X - F type constraint in allowing re-association of an underlying RTR feature. No opaque level of representation is needed in the present account. This comes at the cost of relying on autosegmental representations though, something the stem-control account did not need to do. 101 vowels are not allowed in any position, including the PrHd, HI/ATR, and MAX-HI must dominate [MAX-RTR] P r H d . Given this, the only way to avoid such a violation is to re- associate the RTR feature onto the preceding vowel. By ranking [MAX-RTR] P r H d above [*RTR-C 0 -ATR] P r W d , the disharmonic sequence of a mid-RTR vowel followed by a high vowel must be tolerated in order to avoid a fatal violation of [MAX-RTR] P r H d . This is illustrated below in (114) (subscripts are used to show correspondences between linking sites for underlying RTR segments and their surface correspondents). The faithful candidate (114a) incurs a fatal violation of HI/ATR. Candidate (114d) likewise incurs a fatal violation of MAX-HI. The ATR-harmonic candidate (114c) satisfies these constraints but deletes the underlying RTR value in the prosodic head. This incurs a fatal violation of [MAX-RTR] P r H d . Candidate (114b) re-associates the underlying RTR feature of the PrHd so that it avoids a violation of [MAX-RTR] P r H d and is thus selected optimally. (114) RTR Mid Vowels Preceding High Vowels /e,bi,/ HI/ATR MAX-HI [MAX-RTR] P r H d r*RTR-C n -ATRl P r W d a. eibij *! f b. e=bi c. ebi • ; d. ebe *! It should be noted that this account relies on an underlying form where both vowels are RTR. An underlying form with a single RTR vowel followed by a high vowel would actually surface with two ATR vowels in violation of [MAX-RTR] P r H d . Tableau (115) below demonstrates this. (115) A Single Underlying RTR Feature is Deleted in a Mid-High Sequence /6 ,bi/ HI/ATR MAX-HI [MAX-RTR] P r H d [MAXLINK- A T R ] P r W d [*RTR-C 0 - A T R ] P r W d a. ejbi *! b. djbi *! **• c. ebi d. ebe 11 *! - ; , ; ' * v 4* " v . -'--f An underlying form with a sequence of a mid ATR (or unspecified) prefix followed by a high RTR vowel provides crucial evidence in the relative ranking of [MAX-RTR] P r H d and [ M A X U N K - A T R ] P r W d . In this case, an RTR feature in the prosodic 102 head is either re-associated or deleted, depending on the mutual ranking of [MAX- RTR] P r H d and [ M A X L l N K - A T R ] P r W d . However, recall in section 2.3.3 that it was shown that prefixal vowels preceding high root vowels can contrast for ATR/RTR. This was argued to be a result of a re-association of an underlying RTR value that was present in the high root vowel. This situation can only exist if [MAX-RTR] P r H d dominates [MAXLINK-ATR] P r W d . Tableau (116) below illustrates this. Candidates (116a) and (116d) are ruled out due to fatal violations of HI/ATR and MAX-HI respectively. Candidate (116c) is ruled out due to a fatal violation of [MAX-RTR] P r H d . Therefore, the disharmonic candidate, (116b) is selected optimally even though it violates [MAXLINK- A T R ] P r W d . (116) Re-Association of RTR onto a Mid-Vowel Prefix: [MAX-RTR] P r H d » [MAXLINK-ATR] /6 (Pfx) muj/ HI/ATR MAX-HI [MAX-RTR] P r H d [MAXLINK- A T R ] P r W d [*RTR-C 0- A T R ] P r W d a. obUj *! l̂lliiliiilliiillilllllillfP^ ^ b. Oiimu c. o m u *! d. 6:mo 11 *! Recall that the constraint, [*RTR-C 0 -ATR] P r W d was originally posited with a non- low condition. The present account does not need to posit this non-low condition, however. In fact, this condition must be left off this constraint. If it were included, it would not allow an ATR/RTR contrast to exist on a medial mid vowel that is flanked by a low vowel on the left and a high vowel on the right. While [MAX-RTR] P r H d allows re- association of an underlying RTR feature, it does not specify which vowel it should dock onto. When an underlying RTR feature from a final high vowel is given a choice between docking onto an initial low vowel or a medial high vowel, other constraints will decide which of these vowels is a better potential linking site. In the constraint ranking that has been built so far in this analysis, a trisyllabic form with a low-mid-high sequence of three underlying RTR vowels would surface with an RTR mid vowel. This is the only case where an actual underlying RTR feature is preserved non-finally. This is illustrated in (117) below. Both candidates in (117) preserve the underlying RTR from the PrHd by re-associating it onto the initial low vowel rather than the medial mid vowel. However, since both candidates necessarily incur a single violation of [*RTR-C 0 -ATR] P r W d , the lower ranked faithfulness constraint, [MAXLINK- R T R ] c l G p plays an active role. The optimal candidate (117a) incurs one less violation of [ M A X L I N K - R T R ] c l G p and is thus selected optimally. 103 (117) RTR Contrast Preserved Medially in Low-Mid-High Sequences /a,-ej- i k / [MAX-RTR] P r H d [MAXLINK- A T R ] P r W d [*RTR-C 0- A T R J P r W d [ M A X L l N K - R T R ] c l G p ^ a. a k - e r i * * b. a k -e- i * However, if the constraint, [*RTR-C 0 -ATR] P r W d were to include the non-low condition, it would actually select candidate (117b) instead. This is illustrated in (118) below. While candidate (118a) still incurs a violation of the sequence prohibition constraint due to an e- i sequence (both non-low), candidate (118b) does not since an a-e sequence is allowed. The initial vowel is low thus allowing candidate (118b) to surface. Since all three vowels are underlyingly RTR, there is no way to force the medial vowel to ever surface with an RTR feature (without raising [ M A X L l N K - R T R ] a G p , that is). This situation is essentially neutralization of an ATR/RTR contrast in a position where in fact, one should exist. Therefore, the non-low condition cannot be included on the constraint, [*RTR-C 0 -ATR] P r W d if we are enforcing relative alignment. However, note that in languages where absolute alignment is enforced, this non-low condition is essential in enforcing non-final neutralization of an ATR/RTR contrast. (118) Non-Low Condition Predicts Neutralization of ATR/RTR Contrast Medially in Low-Mid-High Sequences / a r e r i k / [MAX-RTR] P r H d [MAXLINK- A T R ] P r W d [*RTR, NONLO-C 0 - ATR. NONLOL,,,,, [ M A X L l N K - R T R ] c l G p a. a k - e r i *! ^ b. a k -e- i Finally, the account must be able to account for opacity of high vowels that are flanked by two mid vowels. This effect falls out of the ranking offered thus far, as is shown in (119) below. The RTR-harmonic candidate (119a) fatally violates HI/ATR. Candidate (119d) fatally violates [MAX-RTR] P r H d . The other candidates all retain at least one RTR feature, and it is assumed that the PrHd value is one of these (it doesn't matter which). Candidate (119b) attempts to preserve the RTR feature of the initial vowel. This candidate is ruled out though by \ * R T R - Q - A T R l P r W d . The opaque candidate (119c) is correctly selected. 104 (119) Opacity of High Vowels / e w u r e / i i HI/ATR [MAX-RTR] P r H d | * R T R - C 0 - A T R l P r W d [MAXLINK- RTR | n r a. ewu re *! ... ..i:..'.,;v-.. b. ewure i i î ^BliSiiiiiiM &° c. ewure d. ewure *! An interesting question concerns the implications of the above analysis regarding a hypothetical enclitic with a mid ATR vowel. Recall from section 3.5 that there are no examples of ATR enclitics in either dialect, and so it was not possible to test whether or not rightward RTR harmony would spread onto ATR enclitics. In the account above, we find a case of indeterminacy. The constraints, [ *RTR-C 0 -ATR] a G p and [MAXLINK- A T R ] C I G p could not be ranked based on the language data evidence. However, their mutual ranking would decide whether rightward RTR harmony should proceed in the clitic domain. The constraint, [*RTR-C 0 -ATR] c , G p , could actually be satisfied via leftward A T R harmony, but recall that since RTR proclitics resist A T R harmony, it was necessary to rank [ M A X L l N K - R T R ] C I G p above [*RTR-C 0 -ATR| c l G p and leftward ATR harmony is thus blocked in the clitic domain. Rightward harmony could only occur from root to enclitic then if [*RTR- C 0 - A T R ] c l G p were to dominate [ M A X L l N K - A T R ] a G p . This is illustrated in (120) below. (120) Indeterminacy of Participation of ATR Enclitics / je e (EnCl)/ [ M A X L l N K - R T R ] c l G p [*RTR- Q r A T R ] C I G p [ M A X L l N K - A T R ] C I G p ? a r a. je e * b. je e *! ? , 3 P c. je e * The final ranking for Moba Yoruba is illustrated schematically below in (121). 105 (121) Final Constraint Ranking for Moba Yoruba RTR Harmony HI/ATR MAX-HI f*ATR. NONHI-Q-RTR. NONHT] r , V J LO/RTR M A X - L O [MAXLlNKj-ATR] JPrWd [*RTR- C - A T R I PrWd [ M A X L l N K - R T R ] C I G p [ * R T R < - A T R ] C I G p f*ATR. NONHI-C 0-RTR. N O N H l ] c l G p [ M A X L l N K - A T R ] c l G p 5.2 Dialectal Variation: Ife, Ekiti, and Standard Yoruba In the previous section, an OT analysis was presented for Moba Yoruba that accounted for the pattern of harmony seen in that dialect. This analysis is extendable to three other dialects. Recall that Standard Yoruba differs from Moba in that it does not allow leftward RTR harmony in the CIGp domain. This is easily accounted for by reversing the ranking of [ M A X L l N K - A T R ] c l G p and f*ATR. NONHI-C 0-RTR. NONHl] c l G p . Tableau (122) below illustrates this. Rightward ATR-harmony is ruled out since candidate (122c) violates [MAXLlNK-RTR] C I G p fatally. The RTR-harmonic candidate (122b) is ruled out because it fatally violates [ M A X L I N K - A T R ] c l G p . The faithful candidate, (122a) is selected optimally even though it is disharmonic. 106 (122) No Harmony in Proclitics in Standard Yoruba 16 (PCI) j e / [ M A X U N K - R T R ] C I G p [MAXLINK-ATR] C I G p r*ATR. NONHI- C 0 -RTR, N O N H l ] n f i n ®° a. 6 je 1 ~ ••'•LIE b. 6 je *! ^^^^^^^^^^^^^^^^^^ c. 6 je *! Since all other facts of Standard Yoruba are identical to the facts of Moba, the final ranking for Standard Yoruba is identical to Moba with the exception of the ranking reversal illustrated in (122) above. The final ranking for Standard Yoruba is given in (123) below. (123) Final Constraint Ranking for Standard Yoruba RTR Harmony HI/ATR MAX-HI I*ATR. NONHI-Q,-RTR. NONHM P r V . d LO/RJR M A X - L O [ M A £ R T R ] P r H d [MAXLlNKfA"TR] P r W d [*RTR- Q - A T R 1 PrWd [ M A X L l N K - R T R ] c l G p [ * R T R - r j ^ A T R ] C I G p [ M A X L l N K - A T R ] C I G p f*ATR. NONHI-C 0-RTR. NONHl] C I G p The accounts offered in section 2.4 for Ife and Ekiti Yoruba are summarized below. Since there is no data to provide us with an insight into the clitic-behaviour in these dialects, the prohibition and faithfulness constraints apply generally in order to derive the patterns seen within the word. Additionally, I assume identical prosodic status with respect to the PrHd, so that the PrHd is aligned with the right edge of the root in all dialects. First, the final ranking is given for Ife Yoruba, which exhibits transparency of 107 high vowels and absolute right-alignment of the RTR feature with the right edge of the root. (124) Final Constraint Ranking for Ife Yoruba RTR Harmony HI/ATR MAX-HI LO/RTR M A X - L O [MAX^RTR ]p r H d *ATR. NONHl-'oo-RTR. NONHI *RTR. N O N L O ^ A T R . NONLO MAXLINK-RTR MAXLINK-ATR Ekiti Yoruba, meanwhile, had high vowels that actively participated in RTR harmony, but didn't trigger it. Additionally, absolute right-alignment of the RTR feature with the right edge of the root was seen. (125) Final Constraint Ranking for Ekiti Yoruba RTR Harmony The accounts offered above capture the effects seen in four dialects of Yoruba without encountering many of the problems that other existing accounts have. 108 Chapter 6 - Conclusion Moba Yoruba differs from Standard Yoruba with respect to RTR harmony in that proclitics are included in the harmonic domain in the former dialect but not in the latter one. I have argued that this is due to a reference to a different harmonic domain that includes clitics (the CIGp) in Moba. In Standard Yoruba, the harmonic domain is the PrWd instead. Two hypotheses were offered to explain this pattern. The first posited that the harmonic domain referenced the same prosodic constituent, the PrWd, but allowed clitics to be parsed in the PrWd in Moba, and in the CIGp in Standard Yoruba. This would essentially require an indirect reference to syntax, mapping the PrWd onto different syntactic constituents in the two dialects. The alternative hypothesis was to directly refer to the CIGp in Moba and to the PrWd in Standard Yoruba as the domains of reference for a harmony-driving constraint. In order to test these hypotheses, nasal harmony was examined. While there is evidence that in Moba, the domain of nasal harmony also includes clitics, the facts of Standard Yoruba didn't enable a proper test of the hypotheses: nasal harmony is syllable- bound in Standard Yoruba. Fortunately, an argument could be made based on evidence internal to Moba RTR harmony. There is a single exception in the 2SG clitic. This clitic does not harmonize in Moba. Instead, it surfaces invariably as RTR. However, in the case of prefixes, these harmonize invariably in both dialects. For this reason they are undoubtedly parsed in the PrWd in both dialects. However, since there was a single exception in the class of Moba clitics, the clitics could not also be parsed in the PrWd if this single exception is to be accounted for in the phonology. This amounted to evidence that the harmonic domain refers to the CIGp in Moba and the PrWd in Standard Yoruba. This account of RTR harmony in the clitic domain had implications for three existing accounts of RTR harmony, an alignment-based account (Pulleyblank 1996), a stem-control account (Bakovic 2000) and an account utilizing prohibition constraints on features (Pulleyblank 2002). The alignment-based account required only a few minor provisos, but otherwise could extend to include the facts of Moba. However, since alignment relied on positing gradiently evaluated constraints to enforce harmony, a situation that is theoretically undesirable, the alignment-based account was abandoned in favour of an analysis that uses prohibition type constraints. The stem-control account failed to handle the facts of Moba RTR harmony in the clitic domain. Arguably, prosodic structure, and not morphological structure is responsible for the apparent dominance of the right-edge in Yoruba roots. Ola (1995) proposed a theory of prosodic constituency based on independent observations for 109 Yoruba that is virtually identical to the inside-out morphological constituency that Bakovic assumes. However, the lack of evidence for morphological constituency in at least some V C V nouns and the wealth of evidence for prosodic structure that holds in all V C V nouns (for example) argued for an inside-out reference to prosodic constituency instead. A unique account was then proposed that capitalized on the status of a head syllable that is right aligned via independent prosodic constraints with the morphological root. Positional constraints were posited that refer to this position, rather than ones referring to morphological structure. It succeeded not only in capturing the Moba and Standard Yoruba patterns of RTR harmony, but was also able to account for the patterns of Ife and Ekiti Yoruba. 110 References Adetugbo, Abiodun. 1967. The Yoruba language in western Nigeria: its major dialect areas. Doctoral dissertation, Columbia University. Akinkugbe, O. O. 1978. The comparative phonology of Yoruba dialects: Isekiri and Igala. Doctoral dissertation, University of Ibadan. Akinlabi, Akinbiyi, and Mark Liberman. 2000. The tonal phonology of Yoruba clitics. In Clitics in phonology, morphology and syntax, ed. Birgit Gerlach and Janet Grijzenhout, 31-62. Amsterdam: John Benjamins. Archangeli, Diana, and Douglas Pulleyblank. 1989. Yoruba vowel harmony. Linguistic Inquiry 20:173-217. Archangeli, Diana, and Douglas Pulleyblank. 1994. Grounded phonology. Cambridge, M A : MIT Press. Awoyale, J. O. Y . 1974. Studies in syntax and semantics of Yoruba nominalizations. Doctoral dissertation, University of Illinois at Urbana-Champaign. Bakovic, Eric. 2000. Harmony, dominance and control. Doctoral dissertation, Rutgers University. Bakovic, Eric and Colin Wilson. 2000. Transparency, strict locality, and targeted constraints. In Proceedings ofWCCFL 19, ed. Roger Billerey and Brook Lillehaugen, 430-456. Cascadilla Press, Somerville, M A . Bamgbose, Ayo. 1966. A grammar of Yoruba. Cambridge, M A : Cambridge University Press. Bamgbose, Ayo. 1967. Vowel Harmony in Yoruba. Journal of African Languages 6:268-273. Benua, Laura. 1995. Identity effects in morphological truncation. In University of Massachusetts occasional papers in Linguistics 18: Papers in Optimality Theory. ed. J. Beckman, S. Urbanczyk, and L. Walsh, 77-136. Burzio, Luigi. 1996. Surface constraints versus underlying representation. In Current trends in phonology: models and methods, ed. J. Durand and B. Laks, 97-122. Paris: University of Salford Publications.. Dechaine, Rose-Marie. 2002. Decomposing focus: evidence from Yoruba. Paper read at "Triggers for Movement" Workshop, Tilburg. Delano, I.O. 1969. A dictionary of Yoriibd monosyllabic verbs. Institute of African Studies, University of Ife. Fresco, Edward Max. 1970. Topics in Yoruba Dialect Phonology. Studies in African Linguistics, Supplement to Volume 1. Gafos, Adamantios. 1999. The articulatory basis of locality in phonology. New York: Garland. Ito, Junko and Armin Mester. 1992. Weak Layering and Word Binarity. Ms., University of California, Santa Cruz. Ito, Junko, Armin Mester, and Jaye Padgett. 1995. Licensing and underspecification in Optimality Theory. Linguistic Inquiry 26: 571-613. I l l McCarthy, John. 1995. Extensions of faithfulness: Rotuman revisited. Ms., University of Massachusetts, Amherst. [Available on Rutgers Optimality Archive, ROA- 110.] McCarthy, John. 1998. Sympathy and phonological opacity. Phonology 16: 331-99. McCarthy, John. 2003. OT constraints are categorical. Phonology 20: 75-138. McCarthy, John and Alan Prince. 1993. Generalized alignment. In Yearbook of Morphology, ed. Geert Booij and Jaap van Marie, 79-153. Dordrecht: Kluwer. [Available on Rutgers Optimality Archive, ROA-7.] Myers, Scott. 1995. OCP effects in Optimality Theory. Ms., University of Texas, Austin. Ni Chiosain, Maire and Jaye Padgett. 2001. Markedness, segment realization, and locality in spreading. In Segmental Phonology in Optimality Theory: Constraints and Representations, ed. Linda Lombardi. New York: Cambridge University Press. [Available on Rutgers Optimality Archive, ROA-503.] Ola, Olanike. 1995. Optimality in Benue-Congo prosodic phonology and morphology. Doctoral dissertation, University of British Columbia. Orie, Olanike Ola. 2001. An alignment-based account of vowel harmony in Ife Yoruba. Journal of African Languages and Linguistics 22:117-143. Orie, Olanike Ola. 2003. Two harmony theories and high vowels patterns in Ebira and Yoruba. The Linguistic Review 20:1-35. Prince, Alan and Paul Smolensky. 1993. Optimality Theory: constraint interaction in generative grammar. Report. New Brunswick, NJ: Rutgers University Center for Cognitive Science. Pulleyblank, Douglas. 1986. Clitics in Yoruba. Syntax and Semantics 19:43-64. Pulleyblank, Douglas. 1996. Neutral vowels in Optimality Theory: a comparison of Yoruba and Wolof. Canadian Journal of Linguistics 41:295-347. Pulleyblank, Douglas. 2002. Harmony drivers: no disagreement allowed. In Proceedings of the Twenty-eighth Annual Meeting of the Berkeley Linguistics Society, 249-267. Berkeley Linguistics Society, Berkeley, California. 112 Appendix A - Constraint Definitions A formal definition of each constraint-type used in this thesis is given generally with one example for clarification. Below the definitions, specific instantiations of each constraint that are used in the analysis above are listed. When shortened or alternate names are used for constraints these are included in brackets following the constraint. When constraint types are restricted to domains, this is denoted by a subscript following square brackets around the constraint. When a constraint is not restricted to a domain, it is assumed to apply generally across-the-board. • *[aF, p*G...] x: No root node in domain X can dominate both [ctF] and [f3G] specifications on the surface. The constraint, *[+Hl, RTR] would incur a single violation for every root node (on the surface) that simultaneously dominates both [+HI] and [RTR] specifications. Examples: *[+Hl, RTR] (HI/ATR), *[+LO, ATR] (LO/RTR), [Hl /ATR] P r H d ([Hl/ATR] R l F i n a l ) • r*faF, pG. . . l -Y -ryH. 6I...]]X: A violation is incurred for every segment in domain X that is specified as [aF, |3G...] that is in the relation, Y with a segment that is also in domain X that is specified as [yH, 51...]. The relation Y is one of proximity: it can require adjacency, (Y=0), require adjacency between vowels so that intervening consonants are allowed (Y=C 0 ) or it can allow any number of intervening segments (Y=oo) . The constraint *RTR-C„-ATR incurs one violation for every A T R segment that is preceded by an RTR segment (allowing only consonants to intervene). Examples: *RTR-C 0 -ATR; *ATR-C 0 -RTR; *RTR-c*-ATR; *RTR -oo -ATR; *RTR; NONLO-Q-ATR. NONLO: *RTR-oo-RTR; *ATR. NONHl-oo-RTR. NONHI; *RTR. NONLO-oo-ATR. NONLO; *ATR. NONHl-oo-RTR. NONHI; *1ATR. N Q N H I - C Q - R T R . NONHl] c l G p ; r*ATR. NONHl-C 0 -RTR. NONHl] P r W d ; [*RTR- C 0 - A T R ] P r W d ; [*RTR-C 0 -ATR] C I G p • AGREE(F): Adjacent segments must have the same value of the feature, F. Violations of AGREE(ATR) are incurred per distinct pair of adjacent segments that do not have the same value for ATR (i.e. either both must be A T R or both must be RTR). Examples: AGREE(ATR) • ALIGN(Cat l , Edgel , Cat2, Edge2): Edgel of Catl and Edge2 of Cat2 are required to coincide. Catl and Cat2 can be any morphologically or prosodically defined 113 category. Pulleyblank (1996) extends this definition to include autosegmental features as potential categories following Myers (1995). Violations are incurred gradiently in Yoruba, one for every root node that is contained in Cat2 that intervenes between Edgel of Catl and Edge2 of Cat2.6 1 ALIGN(RTR, R, Root, R) incurs violations for every root node that is not RTR within the Root, that follows an RTR span. This definition is based on that given by McCarthy and Prince (1993). Examples: ALIGN(RTR, L, PrWd, L), ALIGN(RTR, R, Root, R), ALIGN(RTR, L, CIGp, L), ALIGN(ATR, L, PrWd, L), ALIGN(PrWd, R, ROOT, R) • L O C A L I G N ( C a t l , E d g e l , Cat2 , Edge2): Edge 1 of Catl and Edge2 of Cat2 are required to align. Catl and Cat2 can be any morphologically or prosodically defined category Pulleyblank (1996) extends this definition to include autosegmental features as potential categories following Myers (1995). Violations are incurred gradiently in Yoruba, one for every root node that is not linked to any Catl that intervenes between Edgel and Edge2 of the two categories in question. LOCALIGN(RTR, L , PrWd, L) would incur one violation per every root node that is both not linked to an RTR feature and that intervenes between the left edge of some RTR span and the left edge of the PrWd. Note that this constraint is defined such that over-alignment of a Catl feature with respect to a Cat2 feature could never incur a violation; only underalignment could incur violations of this constraint. Examples: LOCALIGN(RTR, L, PrWd, L) • E D G E M O S T : This is essentially a special alignment constraint that aligns the head foot either with the right edge of the prosodic word (RIGHTMOST) or with the left edge of the prosodic word (LEFTMOST). One violation of RIGHTMOST is incurred for every prosodic word that does have its head foot right-aligned with its right edge. Example: RIGHTMOST • R H - T Y P E = X : Every foot must be X-headed. If X=I, then feet are iambic or right- headed. If X=T, then feet are trochaic or left-headed. One violation of RH-TYPE=I is incurred per left-headed foot that occurs. Example: RH-TYPE=I 6 1 This definition tolerates over-alignment of Catl with respect to Cat2 - it is not symmetric then. If we want a symmetric ALIGN constraint that militates against both over- and under- alignment, we would remove the 'contained in Cat2' condition. This results in the following definition: 'Violations are incurred gradiently, one for each root node that intervenes between Edgel of Catl and Edge2 of Cat2. This incurs violations for any misalignment, without regard to the difference between over- and under- alignment. 114 [DEP-aF]x: For all featural occurrences [aF] that are linked to segments contained in domain X on the surface, these featural occurrences must be present underlyingly. DEP-RTR incurs one violation for an RTR feature that appears on the surface but that was present underlyingly. Examples: DEP-RTR [DEPLINK-aF]x: Every root node in domain X that is linked (directly or indirectly) to a feature value [aF] on the surface, must also be linked to a feature value [aF] underlyingly. DEPLINK-RTR incurs one violation for every root node that is linked to an RTR feature on the surface that was not also linked to an RTR feature underlyingly. Examples: DEPLINK-RTR [MAX-aF]x: For all featural occurrences, [aF] which are underlyingly linked to segments contained in domain X , these featural occurrences must be present on the surface. [MAX-RTR] P r H d incurs one violation for an RTR feature that was linked underlyingly to a segment in the PrHd domain, but that is not linked to any segment in the output (that is deleted). Examples: MAX-RTR, [MAX-RTR] P r H d ([MAX-RTR] R l F i n a , ) , [ M A X - R T R ] R 0 0 T , [ M A X - A T R ] R 0 0 T , M A X - L O , MAX-HI [MAXLlNK-ctF]x: Every root node in domain X that is linked (directly or indirectly) to a feature value [aF] in the input must have an output correspondent root node that is also linked to a feature value [aF]. [ M A X L l N K - R T R ] a G p incurs one violation for every root node in the CIGp domain that is linked to RTR underlyingly, but that does not have an output correspondent that is linked to RTR. Examples: MAXLINK-RTR, MAXLINK-ATR, [MAXLINK-RTR] C I G p , [MAXLINK- A T R ] ^ , [ M A X L l N K - A T R ] P r W d IO-IDENT(F): Given correspondence between segments in an underlying form and the surface form, all corresponding segments must have the same value for [aF] in the underlying form and in the surface form. Violations are incurred, one for each segment that has a correspondent in both the underlying form and the surface form, where the [aF] value is not identical. For example, IO-IDENT(ATR) incurs one violation for every segment that has a correspondent in the underlying form and the surface form that does not have identical A T R values. Examples: IO-IDENT(ATR) (IO)-ID(ATR)), [IO-IDENT(ATR)] R O O T (ROOT- IDENT(ATR), RT-ID(ATR)), IO-IDENT(HI) (IO-ID(HI) 115 SA-IDENT(F): Given correspondence between segments in an affixed form and its corresponding stem, all corresponding segments must have the same value for [ccF] in the stem and in the affixed form. Violations are incurred, one for each segment that has a correspondent in both the stem and the affixed form, where the [aF] value is not identical. For example, SA-IDENT(ATR) incurs one violation for every segment that has a correspondent in the stem and affixed form that does not have identical ATR values. Examples: SA-IDENT(ATR) (SA-ID(ATR)) •-IDENT(F): Given a correspondence between a sympathetic form and a surface form, all corresponding segments must have the same value for [aF] in the sympathetic form and in the surface form. Violations are incurred, one for each segment that has a correspondent in both the sympathetic form and the surface form, where the [aF] value is not identical. For example, • - IDENT(ATR) incurs one violation for every segment that has a correspondent in the sympathetic form and the surface form that does not have identical A T R values. Examples: • - IDENT(ATR) (• - ID (ATR)) PARSEfX, Y): The constituent X must be linked to a constituent, Y that occupies a tier dominating it. PARSEfp,, o) incurs one violation for every mora that is not linked to a syllable. Examples: PARSE(u., a), PARSE(Ft, PrWd), PARSE(a, Ft), PARSE(PrWd, CIGp) BlNARYfX, Y): The constituent X must be linked to two constituents Y that occupy a lower tier. BlNARY(Ft, a) incurs one violation for every foot that is not linked to exactly two syllables (this is a categorical definition). Examples: BlNARY(Ft, a), BlNARY(PrWd, Ft), BlNARY(Ft, \i) [OCP-RTR]R O O X: One violation is incurred per root-RTR feature that is preceded by another root-RTR feature. Since this constraint applies only in the root domain, it does not incur violations for segment-level occurrences of RTR (i.e. for those attached to low vowels in order to satisfy segmental markedness constraints). ONSET: One violation is incurred for every syllable without an onset 116 Appendix B - Clitic-Aux-Verb Paradigm *A11 data in this appendix is transcribed phonetically using IPA conventions. High T o n e / A T R de (arr ive) Subject Proclitic + auxes MB SY l . S G me de / mi de mo de 2.SG 5 de o de 3.SG e de 6 de l . P L a de a de 2.PL I de e de 3.PL a de wo de l . S G , NEG *mi ke de fj ko de 2 . S G , NEG o ke de o ko de 3 . S G , NEG ke de ko de l . P L , NEG a ke de a ko de 2.PL, NEG 1 ke de e ko de 3.PL, NEG a ke de wo ko de •Throughout this appendix, sequences of adjacent low vowels fol l lowed by high vowels result in the high tone of the second vowel becoming a rising tone; this is a phonetic process. l . S G , FUT me e de n 6 de 2 . S G , FUT o e de o 6 de 3 . S G , FUT e e de 6 maa de l . P L , FUT a e de a 6 de 2.PL, FUT I e de e 6 de 3.PL, FUT a e de wo 6 de l . S G , N E G , FUT mi ke e de *§ ko n i l de 2 . S G , N E G , FUT o ke e de o ko n i l de 3 . S G , N E G , FUT ke e de ko n i l de l . P L , N E G , FUT a ke e de a ko n i l de 2.PL, N E G , FUT I ke e de e ko n i l de 3.PL, N E G , FUT a ke e de wo ko n i l de *nll is the orthographic convent ion for this morpheme; however, sounds like only one vowel (unconf i rmed phonetically) 117 High Tone/ATR de (arrive) Subject Proclitic + auxes MB SY l.SG, PROG mi i d e mo n d e 2.SG, PROG o i d e o n d e 3.SG, PROG i i de 6 n d e l .PL, PROG a i de a n d e 2.PL, PROG i i de e n d e 3.PL, PROG a i d e wo n de * The above paradigm can have a habitual or progressive reading l . S G , PROG, NEG mi k e i d e 3 k i i d e 2.SG, PROG, NEG o k e i de o k i i de 3.SG, PROG, NEG k e i d e k i i d e l .PL, PROG, NEG a k e i de a k i i d e 2.PL, PROG, NEG i k e i de e k i i d e 3.PL, PROG, NEG a k e i d e wo k i i d e * The above paradigm can have only a habitual reading (no progressive reading available) l . S G , PROG, FUT me e d e n 6 maa d e 2.SG, PROG, FUT o e de o 6 maa de 3.SG, PROG, FUT e e de j 66 maa d e l .PL, PROG, FUT a e d e a 6 maa d e 2.PL, PROG, FUT i e de s 6 maa d e 3.PL, PROG, FUT a e de wo 6 maa d e l . S G , PROG, FUT, NEG mi k e n i e d e f) k o n i i maa d e 2.SG, PROG, FUT, NEG o k e n i e d e o k o n i i maa d e 3.SG, PROG, FUT, NEG k e n i e de k o n i i maa d e l .PL, PROG, FUT, NEG a k e n i e d e a k o n i i maa d e 2.PL, PROG, FUT, NEG i k e n i e d e . e k o n i i maa d e 3.PL, PROG, FUT, NEG a k e n i e de wo k o n i i maa de 118 High T o n e / R T R se / J E (change (money)) Subject Procl i t ic + auxes MB S Y l .SG me se / mi se mo Je 2.SG o se o Je 3.SG e se o Je l.PL a se a Je 2.PL 1 se e Se 3. PL a se wo li l .SG, NEG mi ke se rj ko Se 2.SG, NEG o ke se o ko S E 3.SG, NEG ke se ko Sc l.PL, NEG a ke se a ko J E 2.PL, NEG 1 ke se E ko Je 3.PL, NEG a ke se wo ko li • l .SG, FUT me e se n 6 Se 2.SG, FUT d e se o 6 Se 3.SG, FUT e e se 6 maa Se l.PL, FUT a e se a o Se 2.PL, FUT I e se e 6 Se 3.PL, FUT a e se wo 6 Se l .SG, NEG, FUT mi ke e se Q ko n i i Se 2.SG, NEG, FUT o ke e se o ko n i i Se 3.SG, NEG, FUT ke e se ko n i i Se l.PL, NEG, FUT a ke e se a ko n i i Se 2.PL, NEG, FUT i ke e se e ko n i i Se 3.PL, NEG, FUT a ke e se wo ko n i i Se *Note that in MB, n i l can also be used as the FUT marker as is seen below: l .SG, NEG, FUT mi ke n i i se 2.SG, NEG, FUT o ke n i i se 3.SG, NEG, FUT ke n i i se l.PL, NEG, FUT a ke n i i se 2.PL, NEG, FUT i ke n i i se 3.PL, NEG, FUT a ke n i i se 119 High Tone/RTR se / Je (change (money)) Subject Proclitic + auxes MB SY l . S G , PROG mi 1 se mo ii Je 2.SG, PROG o i se o n Je 3.SG, PROG • • ' • I l se o n Je l . P L , PROG a I se ' P ' a n Je 2.PL, PROG 1 I se ' P ' e n Je 3. PL, PROG a 1 se wo i i Je l . S G , PROG, NEG mi ke i se rj k i i Je 2 .SG, PROG, NEG o ke i se o k i i Ss 3.SG, PROG, NEG ke i se k i i Ss l . P L , PROG, NEG a ke i se a k i i Ss 2.PL, PROG, NEG 1 ke i se e k i i Ss 3.PL, PROG, NEG a ke i se wo k i i Ss l . S G , PROG, FUT me e se n 6 maa Ss 2.SG, PROG, FUT 6 e se o 6 maa Ss 3.SG, PROG, FUT e e se j6 6 maa Ss l . P L , PROG, FUT a e se a 6 maa Ss 2.PL, PROG, FUT 1 e se s 6 maa Ss 3.PL, PROG, FUT a e se wo 6 maa Ss l . S G , PROG, FUT, NEG mi ke n i e se fj ko n i i maa Ss 2 .SG, PROG, FUT, NEG o ke n i e se o ko n i i maa Ss 3.SG, PROG, FUT, NEG ke n i e se ko n i i maa Ss l . P L , PROG, FUT, NEG a ke n i e se a ko n i i maa Ss 2.PL, PROG, FUT, NEG i ke n i e se e ko n i i maa Ss 3.PL, PROG, FUT, NEG a ke n i e se w5 ko n i i maa Ss 120 High Tone/ATR le (pursue) Object Enclitic + auxes MB SY l.SG ade l e mi ade l e mi 2.SG ade l e o ade l e e 3.SG ade l e ade 1(e) e l.PL ade l e a ade l e wa 2.PL ade l e i ade l e e j i i 3.PL ade l e a ade l e wo l.SG, NEG ade ke l e mi ade ko l e mi 2.SG, NEG ade ke l e o ade ko l e e 3.SG, NEG ade ke l e ade ko 1(e) e l.PL, NEG ade ke l e a ade ko l e wa 2.PL, NEG ade ke l e i ade ko l e e j i i 3.PL, NEG ade ke l e a ade ko l e wo l.SG, FUT ade ee l e mi ade joo l e mi 2.SG, FUT ade ee l e o ade j 66 l e e 3.SG, FUT ade ee l e ade j 66 1(e) e l.PL, FUT ade ee l e a ade j 66 l e wa 2.PL, FUT ade ee l e i ade j 66 l e e j i i 3.PL, FUT ade ee l e a ade j 66 l e w5 l.SG, NEG, FUT ade ke n i e l e mi ade ko n i i maa l e mi 2.SG, NEG, FUT ade ke n i e l e o ade ko n i i maa l e e 3.SG, NEG, FUT ade ke n i e l e ade ko n i i maa 1(e) e l.PL, NEG, FUT ade ke n i e l e a ade ko n i i maa l e wa 2.PL, NEG, FUT ade ke n i e l e i ade ko n i i maa l e e j i i 3.PL, NEG, FUT ade ke n i e l e a ade ko n i i maa l e w5 Note: ade is a first name in Yoruba. 121 High Tone/ATR le (pursue) Object Enclitic + auxes MB SY l.SG, PROG ade i l e mi ade n l e mi 2.SG, PROG ade i l e o ade n l e e 3.SG, PROG ade i l e ade n 1(e) e l.PL, PROG ade i l e a ade fi l e wa 2.PL, PROG ade i l e I ade n l e e n i 3.PL, PROG ade i l e a ade n l e w5 l.SG, PROG, NEG ade ke i l e mi ade k i i l e mi 2.SG, PROG, NEG ade ke i l e o ade k i i l e e 3.SG, PROG, NEG ade ke i l e ade k i i 1(e) e l.PL, PROG, NEG ade ke i l e a ade k i i l e wa 2.PL, PROG, NEG ade ke i l e I ade k i i l e e n i 3.PL, PROG, NEG ade ke i l e a ade k i i l e wo l.SG, PROG, FUT ade ee e l e mi ade j 66 maa l e mi 2.SG, PROG, FUT ade ee e l e o ade j 66 maa l e s 3.SG, PROG, FUT ade ee e l e ade j 66 maa 1(e) e l.PL, PROG, FUT ade ee e l e a ade j 66 maa l e wa 2. PL, PROG, FUT ade ee e l e 1 ade j 66 maa l e e j i i 3.PL, PROG, FUT ade ee e l e a ade j 66 maa l e wo l.SG, PROG, FUT, NEG ade ke n i e l e mi ade ko n i i maa l e mi 2.SG, PROG, FUT, NEG ade ke n i e l e o ade ko n i i maa l e s 3.SG, PROG, FUT, NEG ade ke n i e l e ade ko n i i maa 1(e) e l.PL, PROG, FUT, NEG ade ke n i e l e a ade ko n i i maa l e wa 2.PL, PROG, FUT, NEG ade ke n i e l e i ade ko n i i maa lee j i i 3.PL, PROG, FUT, NEG ade ke n i e l e a ade ko n i i maa l e wo 122 High Tone/RTR ko (teach) Object Enclitic + auxes MB SY l . S G ade ko mi ade ko mi 2.SG ade ko o ade ko e 3.SG ade ko ade k(6) o l .PL ade ko a ade ko wa 2.PL ade ko 1 ade koo j i i 3.PL ade ko a ade ko wo l . S G , NEG ade ke ko mi ade ko ko mi 2.SG, NEG ade ke ko o ade ko ko e 3.SG, NEG ade ke ko ade ko k(6) o l .PL, NEG ade ke ko a ade ko ko wa 2. PL, NEG ade ke ko i ade ko k5o j i i 3.PL, NEG ade ke ko a ade ko ko wo l . S G , FUT ade ee k5 mi ade j 66 ko mi 2.SG, FUT ade ee ko o ade j 66 ko e 3.SG, FUT ade ee ko ade j 66 k(6) o l.PL, FUT ade ee ko a ade j 66 ko wa 2.PL, FUT ade ee ko i ade j 66 koo j i i 3.PL, FUT ade ee ko a ade j 66 ko w5 l . S G , NEG, FUT ade ke n i ko mi ade ko n i i ko mi 2.SG, NEG, FUT ade ke n i ko o ade ko n i i ko e 3.SG, NEG, FUT ade ke n i ko ade ko n i i k(6) o l.PL, NEG, FUT ade ke n i ko a ade ko n i i ko wa 2.PL, NEG, FUT ade ke n i ko i ade ko n i i koo j i i 3.PL, NEG, FUT ade ke n i ko a ade ko n i i ko wo 123 High Tone/RTR k5 (teach) Object Enclitic + auxes MB SY l . S G , PROG ade i ko mi ade fj k5 mi 2.SG, PROG ade i ko o ade fj k5 e 3.SG, PROG ade i ko ade fj k(6) o l .PL, PROG ade i ko a ade fj ko wa 2.PL, PROG ade i ko i ade fj koo j i i 3.PL, PROG ade i ko a ade fj ko wo l . S G , PROG, NEG ade ke i ko mi ade k i i ko mi 2.SG, PROG, NEG ade ke i ko o ade k i i ko s 3.SG, PROG, NEG ade ke i ko ade k i i k(6) o l .PL, PROG, NEG ade ke i ko a ade k i i ko wa 2.PL, PROG, NEG ade ke i ko i ade k i i koo n i 3.PL, PROG, NEG ade ke i ko a ade k i i ko wo l . S G , PROG, FUT ade ee e ko mi ade j 66 maa ko mi 2.SG, PROG, FUT ade ee e ko o ade j 66 maa ko e 3.SG, PROG, FUT ade ee e ko ade j 66 maa k(6) o l .PL, PROG, FUT ade ee e ko a ade j 66 maa ko wa 2.PL, PROG, FUT ade ee e ko i ade j 66 maa koo j i i 3. PL, PROG, FUT ade ee e ko a ade j 66 maa ko wo l . S G , PROG, FUT, NEG ade ke n i e ko mi ade ko n i i maa ko mi 2.SG, PROG, FUT, NEG • ade ke n i e ko o ade ko n i i maa ko e 3.SG, PROG, FUT, NEG ade ke n i e ko ade ko n i i maa k(6) o l .PL, PROG, FUT, NEG ade ke n i e ko a ade ko n i i maa ko wa 2.PL, PROG, FUT, NEG ade ke n i e ko i ade ko n i i maa koo n i 3.PL, PROG, FUT, NEG ade ke n i e ko a ade ko n i i maa ko wo 124 Mid Tone/ATR se / Je (do) Subject Proclitic + auxes MB SY l . S G me se / ml se mo S© 2.SG o se o Se 3.SG e se 6 Se l .PL a se a Se 2.PL 1 se e Se 3.PL a se wo Se l . S G , NEG ml ke se rj ko Se 2 .SG, NEG o ke se o ko Je 3.SG, NEG ke se ko Se l .PL, NEG a ke se a ko Se 2.PL, NEG 1 ke se s ko Je 3.PL, NEG a ke se w5 ko Se l . S G , FUT me e se n 6 Se 2 .SG, FUT o e se o 6 Se 3.SG, FUT e e se 6 maa Se l .PL, FUT a e se a 6 Se 2.PL, FUT 1 e se e 6 Je 3.PL, FUT a e se wo 6 Je l . S G , NEG, FUT mi ke e se fj ko n i i Se 2 .SG, NEG, FUT o ke e se o ko n i i Se 3.SG, NEG, FUT ke e se ko n i i Se l .PL, NEG, FUT a ke e se a ko n i i Se 2.PL, NEG, FUT i ke e se 8 ko n i i Se 3.PL, NEG, FUT a ke e se wo ko n i i Se 125 Mid Tone/ATR se / Je (do) Subject Proclitic + auxes MB SY l . S G , PROG ml 1 se mo ii Je 2.SG, PROG o i s e o n Se 3.SG, PROG i i s e o n Se l .PL, PROG a i s e a n Se 2. PL, PROG 1 1 s e s n Se 3.PL, PROG a 1 s e wo n Se * The above paradigm can have a habitual or progressive reading l . S G , PROG, NEG mi k e i s e rj k i i Se 2.SG, PROG, NEG o k e i s e o k i i S e 3.SG, PROG, NEG k e i s e k i i Se l .PL, PROG, NEG a k e i s e a k i i Se 2.PL, PROG, NEG I k e i s e e k i i Se 3.PL, PROG, NEG a k e i s e wo k i i S e * The above paradigm can have only a habitual reading (no progressive reading available) l . S G , PROG, FUT me e s e n 6 maa Se 2.SG, PROG, FUT o e s e o 6 maa Se 3.SG, PROG, FUT e e s e j 6 6 maa Se l .PL, PROG, FUT a e s e a 6 maa Se 2.PL, PROG, FUT 1 e s e e 6 maa Se 3.PL, PROG, FUT a e s e wo 6 maa Se l . S G , PROG, FUT, NEG mi k e n i e s e f) k o n i maa Se 2.SG, PROG, FUT, NEG o k e n i e s e o k o n i maa Se 3.SG, PROG, FUT, NEG k e n i e s e k o n i maa Se l .PL, PROG, FUT, NEG a k e n i e s e a k o n i maa Se 2.PL, PROG, FUT, NEG i k e n i e s e s k o n i maa Se 3.PL, PROG, FUT, NEG a k e n i e s e wo k o n i maa Se 126 Mid T o n e / R T R l o (go) Subject Proclitic + auxes MB SY l .SG m l l o / m e l o mo l o 2 .SG 5 l o o l o 3 .SG e l o 6 l o l.PL a l o a l o 2.PL 1 l o e l o 3.PL a l o wo l o l .SG, NEG m l k e l o fj k o l o 2 .SG, NEG o k e l o o ko l o 3 .SG, NEG k e l o k o l o l . P L , NEG a k e l o a ko l o 2.PL, NEG 1 k e l o e ko l o 3.PL, NEG a k e l o wo k o l o l .SG, FUT m e e l o n 6 l o 2 .SG, FUT o e l o o 6 l o 3 .SG, FUT e e l o 6 m a a l o l.PL, FUT v v T a e l o a 6 l o 2.PL, FUT 1 e l o e 6 l o 3.PL, FUT a e l o wo 6 l o l .SG, NEG, FUT m i k e e l o fj k o n i i l o 2 .SG, NEG, FUT o k e e l o o k o n i i l o 3 .SG, NEG, FUT k e e l o k o n i i l o l.PL, NEG, FUT a k e e l o a ko n i i l o 2.PL, NEG, FUT 1 k e e l o e k o n i i l o 3.PL, NEG, FUT a k e e l o wo ko n i i l o *Note that in MB, n i l can also be used as the FUT marker as is seen below: l .SG, NEG, FUT m i ke n i i l o 2 .SG, NEG, FUT o ke n i i l o 3 .SG, NEG, FUT ke n i i l o l.PL, NEG, FUT a ke n i i l o 2.PL, NEG, FUT i ke n i i l o 3.PL, NEG, FUT a ke n i i l o 127 Mid Tone/RTR lo (go) Subject Proclitic + auxes MB SY l . S G , PROG mi i l o mo n l o 2.SG, PROG v ' T 0 1 l o o n l o 3 .SG, PROG 1 1 l o 6 n l o l .PL, PROG N ' 1 a i l o a n l o 2.PL, PROG i i l o e n l o 3.PL, PROG a 1 l o wo n l o l . S G , PROG, NEG mi ke i l o fj k i i l o 2.SG, PROG, NEG o ke i l o o k i i l o 3 .SG, PROG, NEG ke i l o k i i l o l .PL, PROG, NEG a ke i l o a k i i l o 2.PL, PROG, NEG i ke i l o e k i i l o 3.PL, PROG, NEG a ke i l o wo k i i l o l . S G , PROG, FUT me e l o n 6 maa l o 2.SG, PROG, FUT 6 e l o o 6 maa l o 3 .SG, PROG, FUT e e l o j 66 maa l o l .PL, PROG, FUT a e l o a 6 maa l o 2.PL, PROG, FUT i e l o e 6 maa l o 3.PL, PROG, FUT a e l o wo 6 maa l o l . S G , PROG, FUT, NEG mi ke n i e l o fj ko n i i maa l o 2.SG, PROG, FUT, NEG o ke n i e l o o ko n i i maa l o 3 .SG, PROG, FUT, NEG ke n i e l o ko n i i maa l o l .PL, PROG, FUT, NEG a ke n i e l o a ko n i i maa l o 2.PL, PROG, FUT, NEG i ke n i e l o e ko n i i maa l o 3.PL, PROG, FUT, NEG a ke n i e l o w5 ko n i i maa l o 128 Mid Tone/ATR se / Je (hurt / implicate) Object Enclitic + auxes MB SY l . S G ade se mi ade Je mi 2.SG ade se 6 ade Je e 3.SG ade see ade J(e) e l .PL ade se a ade Je wa 2.PL ade se i ade Je j i i 3.PL ade se a ade Je wo T h i s mid-high sequence on the vowel in the verb is a mid-high contour on a single vowel througout this appendix l . S G , NEG ade ke se mi ade ko Je mi 2.SG, NEG ade ke se 6 ade ko Je s 3.SG, NEG ade ke se ade ko J(e) e l .PL, NEG ade ke se a ade ko Je wa 2.PL, NEG ade ke se i ade ko Je j i i 3.PL, NEG ade ke se a ade ko Je wo l . S G , FUT ade ee se mi ade j o o Je mi 2.SG, FUT ade ee se 6 ade j66 Je z 3.SG, FUT ade ee se ade j 66 J(e) e l .PL, FUT ade ee se a ade j66 Je wa 2. PL, FUT ade ee se i ade j66 Je j i i 3. PL, FUT ade ee se a ade j66 Je wo l . S G , NEG, FUT ade ke n i e se mi ade ko n i i maa Je mi 2.SG, NEG, FUT ade ke n i e se 5 ade ko n i i maa Je e 3.SG, NEG, FUT ade ke n i e see ade ko n i i maa J(e) e l .PL, NEG, FUT ade ke n i e se a ade ko n i i maa Je wa 2.PL, NEG, FUT ade ke n i e se i ade ko n i i maa Je j i i 3.PL, NEG, FUT ade ke n i e se a ade ko n i i maa Je wo 129 Mid Tone/ATR se / Je (hurt / implicate) Object Enclitic + auxes MB SY l . S G , PROG ade i se mi ade ii Je mi 2.SG, PROG ade i se 6 ade n Je s 3.SG, PROG ade i see ade n J(e) e l .PL, PROG ade i se a ade n Je wa 2.PL, PROG ade i se i ade n Je n i 3.PL, PROG ade i se a ade n Je wo l . S G , PROG, NEG ade ke i se mi ade k i i Je mi 2.SG, PROG, NEG ade ke i se 6 ade k i i Je s 3.SG, PROG, NEG ade ke i see ade k i i J(e) e l .PL, PROG, NEG ade ke i se a ade k i i Je wa 2.PL, PROG, NEG ade ke i se i ade k i i Je j i i 3.PL, PROG, NEG ade ke i se a ade k i i Je wo l . S G , PROG, FUT ade ee e se mi ade j66 maa Je mi 2.SG, PROG, FUT ade ee e se 6 ade j66 maa Je e 3.SG, PROG, FUT ade ee e see ade j 66 maa J(e) e l .PL, PROG, FUT ade ee e se a ade j66 maa Je wa 2.PL, PROG, FUT ade ee e se i ade j66 maa Je j i i 3.PL, PROG, FUT ade ee e se a ade j66 maa Je wo l . S G , PROG, FUT, NEG ade ke n i e se mi ade ko n i i maa Je mi 2.SG, PROG, FUT, NEG ade ke n i e se 6 ade ko n i i maa Je e 3.SG, PROG, FUT, NEG ade ke n i e see ade ko n i i maa J(e) e l .PL, PROG, FUT, NEG ade ke n i e se a ade ko n i i maa Je wa 2.PL, PROG, FUT, NEG ade ke n i e se i ade ko n i i maa Je n i 3.PL, PROG, FUT, NEG ade ke n i e se a ade ko n i i maa Je wo 130 M i d ro (to feed (( T o n e / R T R reedily like a baby)) O b j e c t E n c l i t i c + a u x e s M B S Y l . S G a d e r o mi ade r o mi 2.SG a d e r o 6 ade r o e 3 .SG a d e r o o ade r o 6 l .PL a d e r o a ade r o wa 2.PL a d e r o 1 ade r o j i i 3. PL a d e r o a ade r o wo l . S G , NEG a d e k e r o ml a d e k o r o mi 2 .SG, NEG a d e k e r o 5 a d e k o r o e 3 .SG, NEG a d e k e r o * a d e k o r o 6 l .PL, NEG a d e k e r o a a d e k o r o wa 2.PL, NEG a d e k e r o 1 a d e k o r o j i i 3.PL, NEG a d e k e r o a a d e k o r o wo *The vowel in the verb ro seems like it might be optional in SY (consultant unsure) l . S G , FUT a d e ee r o mi a d e j 66 r o mi 2 .SG, FUT a d e ee r o 6 a d e j 66 r o e 3 .SG, FUT a d e ee r o * a d e j 66 r o 6 l .PL, FUT a d e ee r o a a d e j 66 r o wa 2. PL, FUT a d e ee r o I a d e j 66 r o j i i 3.PL, FUT a d e ee r o a a d e j 66 r o wo *The vowel in the verb ro seems like it might be optional in SY (consultant unsure) l . S G , NEG, FUT a d e k e n i r o mi a d e k o n i i r o mi 2 .SG, NEG, FUT a d e k e n i r o 6 a d e k o n i i r o e 3 .SG, NEG, FUT a d e k e n i r o o * a d e k o n i i r o 6 l .PL, NEG, FUT a d e k e n i r o a ade k o n i i r o wa 2. PL, NEG, FUT a d e k e n i r o i ade k o n i i r o j i i 3.PL, NEG, FUT ade k e n i r o a ade k o n i i r o wo *The vowel in the verb ro seems like it is obligatory in SY (but consultant unsure) 131 M i d T o n e / R T R to feed (greedily like a baby)) O b j e c t E n c l i t i c + a u x e s M B S Y l.SG, PROG ade i ro ml ade n ro mi 2.SG, PROG ade i ro 6 ade n ro e 3.SG, PROG ade i roo ade n ro 6 l.PL, PROG ade i ro a ade n ro wa 2.PL, PROG ade i ro 1 ade n ro j i i 3.PL, PROG ade i ro a ade n ro wo l.SG, PROG, NEG ade ke i ro mi ade k i i ro mi 2.SG, PROG, NEG ade ke i ro 6 ade k i i ro e 3.SG, PROG, NEG ade ke i roo ade k i i r(o) 5 l.PL, PROG, NEG ade ke i ro a ade k i i ro wa 2.PL, PROG, NEG ade ke i ro i ade k i i ro n i 3.PL, PROG, NEG ade ke i ro a ade k i i ro wo l.SG, PROG, FUT ade ee e ro mi ade j 66 maa ro mi 2.SG, PROG, FUT ade ee e ro 6 ade j 66 maa ro e 3.SG, PROG, FUT ade ee e roo ade j 66 maa r(o) 6 l.PL, PROG, FUT ade ee e ro a ade joo maa ro wa 2.PL, PROG, FUT ade ee e ro i ade j 66 maa ro j i i 3.PL, PROG, FUT ade ee e ro a ade j 66 maa ro wo l.SG, PROG, FUT, NEG ade ke n i e ro mi ade ko n i i maa ro mi 2.SG, PROG, FUT, NEG ade ke n i e ro 6 ade ko n i i maa ro e 3.SG, PROG, FUT, NEG ade ke n i e roo *ade ko n i i maa r(o) 6 l.PL, PROG, FUT, NEG ade ke n i e ro a ade ko n i i maa ro wa 2.PL, PROG, FUT, NEG ade ke n i e ro i ade ko n i i maa ro j i i 3.PL, PROG, FUT, NEG ade ke n i e ro a < ade ko n i i maa ro # 5 *Mid tone on verb ro only present in careful speech in SY 132 L o w T o n e / A T R fo (jump) S u b j e c t P r o c l i t i c + a u x e s M B S Y l . S G me f o / m l f o mo f o 2.SG o f o o f o 3.SG e f o 6 f o l .PL a f o a f o 2.PL I f o s f o 3.PL a f o w5 f o l . S G , NEG m i k e f o fj k o f o 2.SG, NEG o k e f o o k o f o 3.SG, NEG k e f o k o f o l .PL, NEG a k e f o a k o f o 2.PL, NEG 1 k e f o e k o f o 3.PL, NEG a k e f o wo k o f o l . S G , FUT me e f o i i 6 f o 2.SG, FUT o e f o o 6 f o 3.SG, FUT e e f o 6 maa f o l .PL, FUT a e f o a 6 f o 2.PL, FUT I e f o £ 6 f o 3.PL, FUT a e f o wo 6 f o l . S G , NEG, FUT m i k e e f o fj k o n i i f o 2.SG, NEG, FUT o k e e f o o k o n i i f o 3.SG, NEG, FUT k e e f o k o n i i f o l .PL, NEG, FUT a k e e f o a k o n i i f o 2.PL, NEG, FUT i k e e f o c k o n i i f o 3.PL, NEG, FUT a k e e f o wo k o n i i f o 133 Low Tone/ATR ro (think) Subject Proclitic + auxes MB SY l.SG, PROG mi i fo mo n fo 2.SG, PROG o i fo o n fo 3.SG, PROG i i fo o n fo l.PL, PROG a i fo a n fo 2.PL, PROG i i fo 8 n fo 3.PL, PROG a i fo wo n fo * The above paradigm can have a habitual or progressive reading l.SG, PROG, NEG mi ke i fo fj k i i fo 2.SG, PROG, NEG o ke i fo o k i i fo 3.SG, PROG, NEG ke i fo k i i fo l.PL, PROG, NEG a ke i fo a k i i fo 2.PL, PROG, NEG i ke i fo e k i i fo 3.PL, PROG, NEG a ke i fo wo k i i fo * The above paradigm can have only a habitual reading (no progressive reading available) l.SG, PROG, FUT me e fo ii 6 maa fo 2.SG, PROG, FUT o e fo o 6 maa fo 3.SG, PROG, FUT e e fo j 66 maa fo l.PL, PROG, FUT a e fo a o maa fo 2.PL, PROG, FUT i e fo s 6 maa fo 3.PL, PROG, FUT a e fo wo 6 maa fo l.SG, PROG, FUT, NEG mi ke n i e fo fj ko n i maa fo 2.SG, PROG, FUT, NEG o ke n i e fo o ko n i maa fo 3.SG, PROG, FUT, NEG ke n i e fo ko n i maa fo l.PL, PROG, FUT, NEG a ke n i e fo a ko n i maa fo 2.PL, PROG, FUT, NEG i ke n i e fo s ko n i maa fo 3.PL, PROG, FUT, NEG a ke n i e fo wo ko n i maa fo 134 Low T o n e / R T R we (swim) Subject Proclitic + auxes MB SY l . S G me we / ml we mo we 2.SG 5 we o we 3.SG e we 6 we l . P L a we a we 2.PL 1 we e we 3.PL a we wo we l . S G , N E G ml ke we 9 ko we 2.SG, N E G o ke we o ko we 3.SG, NEG ke we ko we l . P L , N E G a ke we a ko we 2.PL, NEG I ke we e ko we 3.PL, NEG a ke we wo ko we l . S G , FUT me e we n 6 we 2.SG, FUT 6 e we o 6 we 3.SG, FUT e e we 6 maa we l . P L , FUT a e we a 6 we 2.PL, FUT 1 e we e 6 we 3.PL, FUT a e we w5 6 we l . S G , N E G , FUT ml ke e we 0 ko n i i we 2.SG, N E G , FUT o ke e we o ko n i i we 3.SG, N E G , FUT ke e we ko n i i we l . P L , N E G , FUT a ke e we a ko n i i we 2.PL, N E G , FUT 1 ke e we e ko n i i we 3.PL, N E G , FUT a ke e we wo ko n i i we *No te that in MB, m i can also be used as the FUT marke r as is seen be low: l . S G , N E G , FUT mi k e n i i we 2.SG, N E G , FUT o k e n i i we 3.SG, N E G , FUT k e n i i we l . P L , N E G , FUT a k e n i i we 2.PL, N E G , FUT i k e n i i we 3.PL, N E G , FUT a k e n i i we 135 L o w T o n e / R T R we (swim) S u b j e c t P r o c l i t i c + a u x e s M B S Y l . S G , PROG mi 1 ws mo n we 2.SG, PROG o i we o n we 3.SG, PROG i i we 6 n we l .PL, PROG a i we a n we 2.PL, PROG i i we e n we 3.PL, PROG a i we wo ri we l . S G , PROG, NEG mi ke i we fj k i i we 2.SG, PROG, NEG o ke i we o k i i we 3.SG, PROG, NEG ke i we k i i we l .PL, PROG, NEG a ke i we a k i i we 2.PL, PROG, NEG i ke i we e k i i we 3.PL, PROG, NEG a ke i we wo k i i we l . S G , PROG, FUT me e we n 6 maa we 2.SG, PROG, FUT o e we o 6 maa we 3.SG, PROG, FUT e e we j 66 maa we l .PL, PROG, FUT a e we a 6 maa we 2.PL, PROG, FUT i e we e 6 maa we 3.PL, PROG, FUT a e we wo 6 maa we l . S G , PROG, FUT, NEG mi ke n i e we fj ko n i i maa we 2.SG, PROG, FUT, NEG o ke n i e we o ko n i i maa we 3.SG, PROG, FUT, NEG ke n i e we ko n i i maa we l .PL, PROG, FUT, NEG a ke n i e we a ko n i i maa we 2.PL, PROG, FUT, NEG i ke n i e we e ko n i i maa we 3.PL, PROG, FUT, NEG a ke n i e we wo ko n i i maa we 136 L o w T o n e / A T R k'pe (call) O b j e c t E n c l i t i c + a u x e s M B S Y l.SG ade k'pe mi ade k'pe mi 2.SG ade kpe 5 ade k'pe e 3.SG *ade kpe ade k'pe e l.PL ade kpe a ade k'pe wa 2.PL ade k'pe i ade k'pe j i i 3.PL ade k'pe a ade k'pe w5 *Derived mid-tone on vowel in kpe in MB is lower than mid tone but higher than low tone l.SG, NEG ade ke kpe mi ade ko kpe mi 2.SG, NEG ade ke k'pe 6 ade ko k'pe e 3.SG, NEG ade ke k'pe ade ko kpe e l.PL, NEG ade ke k'pe a ade ko k'pe wa 2.PL, NEG ade ke k'pe i ade ko k'pe jii 3.PL, NEG ade ke k'pe a ade ko k'pe w5 l.SG, FUT ade ee k'pe mi ade j 66 k'pe mi 2.SG, FUT ade ee k'pe 6 ade j 66 k'pe e 3.SG, FUT ade ee k'pe ade joo k'pe e l.PL, FUT ade ee k'pe a ade j 66 k'pe wa 2.PL, FUT ade ee k'pe i ade j 66 k'pe j i i 3.PL, FUT ade ee k'pe a ade j 66 k'pe wo l.SG, NEG, FUT ade ke n i e k'pe mi ade ko n i i maa k'pe mi 2.SG, NEG, FUT ade ke n i e k'pe 6 ade ko n i i maa k'pe e 3.SG, NEG, FUT ade ke n i e k'pe ade ko n i i maa k'pe e l.PL, NEG, FUT ade ke n i e k'pe a ade ko n i i maa k'pe wa 2.PL, NEG, FUT ade ke n i e k'pe i ade ko n i i maa k'pe j i i 3.PL, NEG, FUT ade ke n i e k'pe a ade ko n i i maa k'pe wo 137 L o w T o n e / A T R kpe (call) O b j e c t E n c l i t i c + a u x e s M B S Y l . S G , PROG ade i kpe ml ade ri lepe mi 2.SG, PROG ade i kpe 6 ade ri k~pe z 3.SG, PROG ade i kpe ade ri kpe e l .PL, PROG ade i kpe a ade ri kpe wa 2.PL, PROG ade i kibe I ade ri k̂ pe n i 3.PL, PROG ade i kpe a ade ri kpe wo l . S G , PROG, NEG ade ke i k~pe mi ade k i i £pe mi 2.SG, PROG, NEG ade ke i £pe 6 ade k i i £pe z 3.SG, PROG, NEG ade ke i kpe ade k i i kpe e l .PL, PROG, NEG ade ke i £pe a ade k i i £pe wa 2.PL, PROG, NEG ade ke i k~pe i ade k i i k~pe j i i 3.PL, PROG, NEG ade ke i kpe a ade k i i l̂ pe wo l . S G , PROG, FUT ade ee e k~pe mi ade joo maa lepe mi 2.SG, PROG, FUT ade ee e kpe 6 ade j 66 maa l̂ pe z 3.SG, PROG, FUT ade ee e kpe ade j 66 maa kpe e l .PL, PROG, FUT ade ee e l̂ pe a ade j 66 maa kpe wa 2.PL, PROG, FUT ade ee e kpe i ade j 66 maa kpe j i i 3.PL, PROG, FUT ade ee e k̂ be a ade j 66 maa kpe wo l . S G , PROG, FUT, NEG ade ke n i e kpe mi ade ko n i i maa k~pe mi 2.SG, PROG, FUT, NEG ade ke n i e kpe 6 ade ko n i i maa kpe e 3.SG, PROG, FUT, NEG ade ke n i e k~pe ade ko n i i maa Itpe e l .PL, PROG, FUT, NEG ade ke n i e ikpe a < ade ko n i i maa k~pe via 2.PL, PROG, FUT, NEG ade ke n i e k^pe i ; ade ko n i i maa k~pe n i 3.PL, PROG, FUT, NEG ade ke n i e kpe a < ade ko n i i maa kpe wo 138 L o w T o n e / R T R ko (reject) MB lade ko mi ade ko mi 2.SG ade ko 6 ade ko e 3.SG ade ko ade ko 6 l .PL ade ko a ade ko wa 2.PL ade ko i ade ko j i i 3.PL ade ko a ade ko wo .SG, NEG ade ke ko mi ade ko ko mi 2.SG, NEG ade ke ko 6 ade ko ko e 3.SG, NEG ade ke ko ade ko ko 6 l .PL, NEG ade ke ko a ade ko ko wa 2.PL, NEG ade ke ko i ade ko ko j i i 3.PL, NEG ade ke ko a ade ko ko wo l . S G , FUT 2.SG, FUT ade ee ko mi ade ee ko 6 ade joo ko mi ade joo ko e 3.SG, FUT ade ee ko ade j 66 ko 6 .PL, FUT ade ee ko a ade j 66 ko wa 2.PL, FUT ade ee ko i ade j 66 ko j i i 3.PL, FUT ade ee ko a ade j 66 ko wo SG, NEG, FUT ade ke n i ko mi ade ko n i ko mi SG, NEG, FUT ade ke n i ko 6 ade ko n i ko e .SG, NEG, FUT ade ke n i ko ade ko n i ko 6 PL, NEG, FUT ade ke n i ko a ade ko n i ko wa PL, NEG, FUT ade ke n i ko i ade ko n i ko j i i PL, NEG, FUT ade ke n i ko a ade ko n i ko wo 139 L o w T o n e / R T R ko (reject) O b j e c t E n c l i t i c + a u x e s M B S Y l.SG, PROG ade i kd mi ade rj ko mi 2.SG, PROG ade i ko 6 ade o ko s 3.SG, PROG ade i ko ade fj ko 5 l.PL, PROG ade i ko a ade rj ko wa 2.PL, PROG ade i ko 1 ade rj ko j i i 3.PL, PROG ade i ko a ade r) ko w5 l.SG, PROG, NEG ade ke i ko mi ade k i i ko mi 2.SG, PROG, NEG ade ke i ko 5 ade k i i ko s 3.SG, PROG, NEG ade ke i ko ade k i i ko 5 l.PL, PROG, NEG ade ke i ko a ade k i i ko wa 2.PL, PROG, NEG ade ke i ko i ade k i i ko j i i 3.PL, PROG, NEG ade ke i ko wo ade k i i ko wo l.SG, PROG, FUT ade ss s ko mi ade j 66 maa ko mi 2.SG, PROG, FUT ade ss s ko 6 ade j 66 maa ko s 3.SG, PROG, FUT ade ss s ko ade j 66 maa ko 6 l.PL, PROG, FUT ade ss E ko a ade j 66 maa ko wa 2.PL, PROG, FUT ade E E E ko i ade j 66 maa ko j i i 3.PL, PROG, FUT ade ss E ko a ade j 66 maa ko wo l.SG, PROG, FUT, NEG ade ke n i s ko mi ade ko n i i maa ko mi 2.SG, PROG, FUT, NEG ade ke n i s ko 6 ade ko n i i maa ko s 3.SG, PROG, FUT, NEG ade ke n i s ko ade ko n i i maa ko 6 l.PL, PROG, FUT, NEG ade ke n i s ko a ade ko n i i maa ko wa 2.PL, PROG, FUT, NEG ade ke n i s ko i ade ko n i i maa ko j i i 3.PL, PROG, FUT, NEG ade ke n i s ko a ade ko n i i maa ko w5 140

Cite

Citation Scheme:

    

Usage Statistics

Country Views Downloads
United States 10 0
China 2 0
Canada 1 0
France 1 0
City Views Downloads
San Francisco 4 0
Sunnyvale 3 0
Beijing 2 0
Unknown 1 7
Mountain View 1 0
Redmond 1 0
Vancouver 1 0
Ashburn 1 0

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}

Share

Share to:

Comment

Related Items