Isixhosa Tonology: An Optimal Domains Theory Analysis Farida Cassimjee
1998 (1st printing) 2004 (2nd printing) LINCOM EUROPA
Published by LINCOM GmbH.
1998 (1st printing). 2004 (2nd printing).
LINCOM GmbH GmunderStr.35 D-81379Muenchen
[email protected] http://home.t-online.de/home/LINCOM.EUROPA www.lincom-europa.com webshop: lincom.at All rights reserved, including the rights of translation into any foreign language. No part of this book may be reproduced in any way without the permission of the publisher. Printed in E.C. Printed on chlorine-free paper Die Deutsche Bibliothek - CIP Cataloguing-in-Publication-Data A catalogue record for this publication is available from Die Deutsche Bibliothek (http://www.ddb.de)
Contents PREFACE
Chapter 1
i
OPTIMAL DOMAINS THEORY
1
1.0. Introduction 1.1. Optimally Theory 1.2. Optimality Domains Theory 1.3. OTD and the analysis of Bantu tone
1 2 7 12
Chapter 2
24
THE LONG FORM OF THE PRESENT TENSE
2.0. Introduction 2.1. Toneless verb stems 2.2. High-tones verb stems 2 3 On the tonology of depressor consonants 2.4. The inclusion of an object prefix into the long present verb
24 24 36 S1 69
Chapter 3
88
3.0. 3.1. 3.2. 3.3. 3.4. 3.5.
MORE VERBAL FORMS WITH LEXICAL TONE
Introduction Short form of the present tense Interaction of the short present form with a following word (nominal) Interaction of the short present verb with a preceding word The perfective form of the verb Two High-toned verb prefixes
Chapter 4
GRAMMATICAL TONE
88 88 104 114 118 134 151
4.0. Introduction 4.1. Negative present 4.2. The negative of the /sa/ tense 4.3. The negative potential 4.4. Negative past 4.5. Negative infinitive 4.6. ka-negative 4.7. The short form of the perfect 4 8. Participial 4.9. Participial of the /sa/ tense 4.10. Participial perfect 4.11. Imperative 4.12. Plural imperative 4.13. Remote past tense 4.14. Conclusion
151 151 172 179 185 187 197 201 207 216 219 223 228 229 233
REFERENCES
235
Preface
The writing of this book was stimulated by the radical reorientation of phonology that was initiated by the distribution of two manuscripts, Optimality Theory (Prince and Smolensky, 1993) and Prosodic Morphology (McCarthy and Prince, 1993a) in the spring of 1993, as well as by the notion of "domains" of High tone-spreading developed in Kisseberth (1994). The first draft was completed in the Fall of 1994. Various factors prevented me from immediately undertaking a final revision of that manuscript, and the emergence of "Correspondence Theory" in Optimality Theory in 1995 prompted a decision to totally rewrite the manuscript. This revision was delayed while, in collaboration with C. Kisseberth, I worked on a very long paper developing a general account of Bantu tone based on what we refer to as Optimal Domains Theory. This paper, "Optimal Domains Theory and Bantu tonology: a case study from Isixhosa and Shingazidja" is included in Hyman and Kisseberth (1998). It was not until 1997 that I was finally able to return to work on this book. While I recognize that too much time has elapsed (when I completed the original version, I thought it would be the first extended analysis of a language in Optimality Theoretic terms!), I have no doubt that the final product is better than if I had rushed into print. This book would not exist were it not for three people: Mbulelo Jokweni and his wife Koleka, and my husband. Chuck Kisseberth. Mbulelo and Koleka provided an immense amount of data on their respective dialects of Isixhosa. Furthermore, Mbulelo provided an abundance of insights deriving from his own research into Isixhosa phonology (see Jokweni 1995). I thank them both for their efforts in behalf of this project. Chuck has guided my research on Bantu phonology (as he has guided so many other African students) from the beginning, and his work on domains of High-tone spreading was the starting point for this study of Isixhosa tonology. Most of what is good about this work derives directly from these three people; the failings are entirely my own.
CHAPTER ONE OPTIMAL DOMAINS THEORY 1.0. Introduction. In this book we shall explore in some detail the tonology of the verb in Isixhosa1, one of the major languages of the Republic of South Africa and a member of the Nguni subgroup of the Bantu language family. Other Nguni languages include Isizulu, Siswati, Zimbabwean Ndebele, Southern (Transvaal) Ndebele, Northern (Transvaal) Ndebele, Phuthi, Nhlangwini, etc. Isixhosa, like most languages spoken by a large number of people dispersed over an extensive area with a complex social and political history, has considerable variation (though this variation has received scant attention in the research literature). Tone is not something that is represented in the written form of the language, and thus it is not an aspect of the language that has been at all "standardized". This study is based on the speech of Mbulelelo Jokweni [=M], who comes from Mzimkhulu, and his wife Koleka [=K], who comes from Matatiele - both representing Transkei varieties of Isixhosa. Mbulelo's speech may reflect Bhaca roots, while Koleka's speech (in her assessment) reflects a Hlubi background. Bhaca and Hlubi are two of the Nguni subgroups dispersed as part of the Mfecane (i.e. the upheaval of Nguni peoples occurring during the period of Zulu expansion in the nineteenth century); the status of Bhaca and Hlubi as independent and identifiable languages today is a matter requiring research. Both Mbulelo and Koleka regard their own speech patterns as part of the Isixhosa speech continuum. Other speakers of Isixhosa also consider M and K to be speaking Isixhosa. The speech of both M and K exhibits the phenomenon we refer to as Depressor-Induced High Tone Shift [=DIHTS] - a phenomenon that has been (in the published literature) more associated with Isizulu than Isixhosa. While many forms of Isixhosa do not exhibit this phenomenon (and thus a not inconsiderable portion of this book on Isixhosa is irrelevant to such dialects!), the exact distribution of this feature in the Isixhosa continuum is unknown. We have, for example, noted its occurrence in the speech of a young man from the East London area, far removed from Mzimkhulu and Matatiele geographically. We hope that the present book will provide the foundation for a systematic exploration of Isixhosa tonal dialectology that might clarify not only the distribution of DIHTS but also the distribution of a number of other tonal patterns. We do not review here the history of research on Isixhosa tone. This book is based entirely on data collected from Mbulelo and Koleka (data which forms part of a much larger body of data collected from some ten speakers of Isixhosa, representing a variety of Isixhosa speech forms)2. We employ a theoretical framework radically different from any employed in the previous literature (which, in any case, is not extensive). Since we are not concerned here with providing a critique of structuralist linguistics, or rule-based autosegmental phonology, and since we are not here concerned with dialect comparison, we refrain from any discussion of previous research, theoretical or otherwise. The bibliography contains references to the most important (published) research on Nguni and Isixhosa tonology. Isixhosa nominal tonology does not differ in a substantive way from the verbal tonology. Space limitations preclude the possibility of providing a complete exposition of both verbal and nominal tonology; since the verbal tonology presents the most complexity, we have opted to focus on it. 3 These data were collected in conjunction with Mbulelo Jokweni and Chuck Kisseberlh and will eventually form part of a monograph on Isixhosa tonal dialectology that the three of us intend to write. 1
It is fair, we think, to say that the following monograph represents the most comprehensive study of Isixhosa verbal tonology available. Furthermore, it is cast within an explicit theory of phonology that attempts to characterize the phonological principles that permit a native speaker to select the appropriate pronunciation for any given utterance in her/his language. Nevertheless, space limitations have prevented us from including a discussion of every verbal construction in Isixhosa. We believe that the constructions that we do examine are sufficient to establish the major elements of the Isixhosa tonal system This monograph also represents the first attempt to provide a comprehensive account of the tonal structure of a language within the framework of Optimal Domains Theory [=ODT]. Optimal Domains Theory is a particular implementation of Optimality Theory [=OT] (cf. Prince and Smolensky 1993 and much subsequent literature). In this chapter we provide an overview of OT in general and ODT in particular. We hope that this overview will be sufficient for the reader to follow the analysis of Isixhosa developed in later chapters. A somewhat more developed introduction to ODT and Bantu tonology is found in Cassimjee and Kisseberth (1998). 1.1. OptimaHty Theory. Optimality Theory represents a radical departure from the Generative Phonology [=GP] paradigm that has guided phonological theorizing over the past three decades. In Generative Phonology, the phonological component of a speaker's grammar consists of a set of principles [=rufes] that map an underlying representation onto a phonetic representation. Rules are expressed as procedures (e.g. deletions, insertions, and feature changes) which take an input representation and yield an output representation. Rules are applied sequentially, thus there are potentially many intermediate representations between the underlying representation and the final phonetic representation. GP is thus a procedural, derivational modeL In GP, the speaker must acquire both the underlying representations of morphemes and the rules that specify the surface forms of these morphemes. A given alternation pattern in the shapes of morphemes, and a given generalization concerning the distribution of features, is generally the consequence of rules. Rules represent phonological truths that are valid for all (unexceptional, regular) structures at the point in the derivation where the rule applies. Rules are learned just in case the data require them. Any rule can be learned provided that it is expressible in terms of the algebra governing rules and representations. Rules do not exist in the grammar unless there is positive evidence that requires them. The view that Optimality Theory takes is quite different (the reader is of course referred to Prince and Smolensky (1993) as well as McCarthy and Prince (1993a,b; 1995) for a full account of the OT approach to phonology, our summary here merely attempts to sketch the leading ideas of the theory). In OT, phonological principles are expressed as constraints, not rules. A given pattern of alternation or a given distribution of a feature is not generally the direct consequence of a single constraint but rather the result of the interaction of a number of general constraints. For example, what in a rule-based model might be expressed as a rule inserting a vowel in a particular position will be seen in OT as the interaction of constraints on syllable shape — e.g. No Complex Codas, No Complex Onsets — with constraints barring epenthetic material, constraints requiring input consonants to be present in the output, etc. In other words, there is no direct relation between a particular "phenomenon" and a particular constraint in the way that the rule-based models suggest. Constraints are not procedures for converting one phonological representation into another, but rather are the means for selecting the best (=optimal) output candidate for any given underlying representation (input structure). A given output candidate either satisfies or fails to satisfy a given constraint If the constraint is satisfied, there is nothing further to be sakl. But if an output candidate fails to satisfy a constraint, the situation is a bit more complex. Violations 2
may either be non-gradient or gradient (in the sense that there may be different degrees of violation). For example, a constraint such as Onset says that a syllable must have an onset A syllable that has an onset satisfies the constraint. A syllable with two consonants in onset position does not better satisfy the constraint than a syllable with a single consonant in onset position. In both cases the constraint is simply satisfied. A syllable that lacks an onset violates the constraint. There is no relevant sense in which one candidate that lacks an onset can violate the constraint to a greater or lesser degree than some other candidate that lacks an onset. Violation of Onset is non-gradient. Now consider the constraint No Coda, which says that a syllable may not have, a coda consonant. A syllable that ends in a vowel satisfies the constraint A syllable that ends in a consonant violates the constraint. A syllable that ends in two consonants violates the constraint two times. We thus have a situation where there can be greater or lesser violation of a constraint. Violation of No Coda is gradient. Constraints do not represent absolute truths about optimal output candidates. Constraints are vlolable. A constraint exists in the grammar of a language even though it is sometimes or even often violated by optimal ouput forms. Constraints are violated in OT due to the existence of other, more highly ranked constraints. A less highly ranked constraint will be violated in order to better satisfy a more highly ranked constraint. For example, in a given language L it may be true that syllables generally have an onset. However, syllables with h as onset "elide" this h. Without going into any detail regarding the proper formulation of the constraint that bars h as an onset, we shall simply refer to it as *h-Onset. *h-Onset is more highly ranked than Onset in L. As a consequence, Onset will be violated in order to satisfy *h-Onset. There is thus a deep connection between constraint violation and constraint ranking. Let us spell out this approach in more detail. The grammar of a language consists of a set of constraints. The constraints are ranked. For any given underlying representation there is a set of possible output candidates (see below for discussion of how this set is defined). These output candidates are evaluated as to which one best satisfies the constraint system Evaluation works as follows. All potential outputs are evaluated in terms of each of the constraints. We can represent this evaluation in the form of a table. The candidate set will be given on the vertical axis, the constraints on the horizontal axis. The constraints will be given sequentially: more highly ranked constraints before less highly ranked constraints. However, since not all constraints are critically ranked, it will be useful to distinguish between constraints that are critically ranked and ones which are not. It is not possible to fully indicate all critical rankings in the table-format What we will do is separate the columns of constraints that are not ranked relative to one another with dotted lines; a constraint that is critically ranked with respect to at least one of the preceding (group of unranked) constraints will be separated from those constraints by a solid line. In the text we will make clear which constraints are critically ranked. If a candidate satisfies a constraint, the cell in the table for that constraint will be left empty. If a candidate violates the constraint, this violation will be indicated by an asterisk. If the constraint is multiply violated, each violation will be indicated by an asterisk. (Sometimes, in the interest of clarity, the phonological element that is in violation will be listed instead of the asterisk.) An exclamation mark will be placed after an asterisk to indicate that this violation is fatal with respect to the possibility that the candidate in question can be the optimal candidate. What makes a violation fatal? In Optimally Theory, a candidate x is less optimal than a candidate y if, for all the constraints (if any) more highly ranked than some constraint C, they are equal in their satisfaction of the constraints, but x fares less well with respect to C than y. The failure of x to match its competitor with respect to C means that x's violation (or greater violation) of C is fatal for its chance to be the optimal candidate. Saying that two or more candidates are "equal" in their satisfaction of a set of constraints means that they have exactly the 3
same evaluation (Le. they satisfy the same constraints and violate the same constraints to the same extent). A violation that serves to render a candidate nonoptimal (due to the existence of at least one alternative that is more optimal) is a fatal violation. In the interest of visual clarity, we shall shade all of the cells in a row following the occurrence of a fatal violation. This shading is meant to highlight the fact that once a fatal violation occurs, the extent to which that candidate violates the remaining constraints is entirely irrelevant. Shading is thus a visual mark of irrelevance. Some simple examples will help to make evaluation of output candidates clearer. Consider first an input /ha/ and the following two output candidates: a and ha. Given the two constraints *h-Onset and Onset (where *h-Onset dominates Onset), the following table will illustrate the evaluation of the two candidates by these two constraints.
Hi Candidates -»
*h-Onset
Onset
*
a
ha
•I
The arrow indicates what in fact is the optimal candidate in the language in question. We see that (of the two candidates being considered here), the candidate ha violates the most highly ranked constraint (*h-Ohset) while the other candidate, a, does not. The asterisk indicates that ha violates *h-Onset. The exclamation sign is meant to indicate that this violation is fatal with respect to the opportunity of ha to be the optimal candidate. Notice that both candidates violate a single constraint. But since ha violates the more highly ranked constraint, it is non-optimal. The notion "optimal output" has nothing to do with how many constraints are violated in total, nor does it have to do with the total number of constraint violations. The optimal output is the one that better satisfies the more highly ranked constraints than does any of its competitors. Now consider an input Aahi/. An unviolated *h-Onset predicts that in the optimal output the input h will not be an onset One way in which to insure this result is, of course, to delete the h. Another way to achieve this result is to delete the last vowel of Aahi/, thereby making it possible for h to surface as a coda: tab. Let us assume, for convenience, that in the hypothetical language under discussion, h is indeed a possible coda but that tah is not the optimal output rather U.I is optimal, where the dot indicates a syllable boundary. Thus there must be a highly ranked constraint that makes deletion of the final vowel in Aahi/ undesirable. In Optimalhy Theory, failure to pronounce an underlying element is treated as a violation of a constraint type referred to as Max (segment). Max (segment) is a member of what in OT is referred to as the Faithfulness family of constraints. (We should note that we follow here the Correspondence Theory approach to Faithfulness found in McCarthy and Prince (1995) rather than the Containment Theory approach used in Prince and Smolensky (1993). We will have more to say on this matter in the discussion of ODT.) Faithfulness constraints are, in effect, constraints that prefer output candidates to be Identical to the input. Max (segment) says that segments that are in the underlying representation ("sponsored'' by the underlying representation, to use some terminology suggested by McCarthy) should have a corresponding segment in the output. Max (segment) is, of course, a family of constraints - Max (vowel) and Max (consonant), where these are themselves families of constraints (e.g. Max (low vowel), Max (mid vowel). Max (high vowel), etc.). The different members of this family of constraints are independently rankable. Thus if Max (vowel) is ranked higher than Max (consonant), then the non-optimal candidate tah will violate the more highly
4
ranked constraint Max (vowel) while the optimal candidate ta.1 violates the more lowly ranked constraint Max (consonant). By having *h-Onset and Max (vowel) undominated and having *h-Onset dominate both Onset and Max (consonant), we guarantee that ta.i will fare better than ta.hi and also tah. The following tableau illustrates. (2) Candidates ta.hi tah -»ta.i
Max (vowel) *h-Onset
Max (cons)
•1
,,'i~tF:K
Onset
^.fc" * -.
•! ,*
"•
*
fcVp****1,
;-»'. i ^HHH *
This tableau illustrates one very important point. The total number of violations of constraints is irrelevant to the evaluation of the optimal output candidate. For example, the output candidate ta.nl has a single violation, whereas the optimal output ta.i has two violations. But the total number of violations does not matter. What is critical is that ta.hf violates the most highly ranked constraint, *h-Onset, while ta.i does not. Similarly, the candidate tah has a single constraint violation, but it violates the more highly ranked constraint Max (vowel) which ta.i satisfies perfectly. It does not matter that ta.i violates the more lowly ranked constraints Max (consonant) and Onset. Max (segment) is just one type of Faithfulness constraint. A second type of Faithfulness constraint is Dep (segment). Dep (segment) requires that every segment in the output correspond to a segment in the input. In effect, Dep (segment) bans epenthesis. Constraints like Onset, if ranked above Dep (segment), will be best satisfied by outputs containing epenthetic material that violates Dep (segment). This ban on epenthetic segments is just one aspect of a general ban against all phonological structure (the *Struc family of constraints). ''Struc simply says that it is better for there to be no "structure" (no features, no segments, no syllables, no feet, no prosodic words, no prosodic phrases) than for there to be structure. In other words, structure should not be included in an output unless required. There are various types of constraints which, if ranked above *Stmc, will force violations. Max (segment) of course requires that elements of underlying representation be pronounced in the output candidate. Thus Max (segment) if ranked above *Struc requires the presence of a segment, in violation of "Struc. And if segments are to be pronounced (as required by Max (segment)), they must be part of syllables (on the assumption that unsyllabificd material cannot be pronounced); thus syllables must be present in the output candidate, again in violation of "Struc. For the time being, we will eschew discussion of the OT treatment of featural Faithfulness (see McCarthy and Prince 1995). This is a central concern of ODT and we will examine the matter closely in the next section. We have mentioned two classes of constraints that are especially important in OT: the Faithfulness family of constraints. Max (segment) and Dep (segment), and the phonological constraints on syllable structure (e.g. Onset, No Coda, No Complex Onsets, No Complex Codas). There is a third type of constraint in OT that will be of particular importance to the analysis of tone. This is the Alignment family of constraints (cf. McCarthy and Prince 1993). The Alignment family of constraints requires that edges of various constituents (e.g. morphological categories, prosodic categories) be aligned with edges of other constituents. For example. (3)
Align the Right edge of a Foot with the Right edge of a Prosodic Word. 5
There is one particular point about the formulation of Alignment constraints that must be emphasized. The argument "the Right edge of a Foot" in (3) is to be understood as universally quantified, while the argument "the Right edge of a Prosodic Word" is to be understood as existentiaDy quantified. Thus (3) is to be understood as requiring that the Right edge of every foot be aligned with the Right edge of an Instance of a Prosodic Word. The following constraint, (4)
Align the Right edge of a Prosodic Word with the Right edge of a Foot.
on the other hand, is to be understood as requiring the Right edge of every Prosodic Word to be aligned with the Right edge of an instance of the category Foot. Alignment constraints are (usually considered to be) gradient constraints. They are satisfied when the two edges coincide. They are violated when there is misalignment between the edges. But misalignment is a matter of degrees. For example, in the case of (4) above, a Prosodic Word of the structure a(oo)a] is misaligned with the Right edge of a Foot by one syllable, whereas (oo)oo~] is misaligned by two syllables. The latter represents a greater violation of Alignment than the former. Let us now turn our attention to how we know what constitutes the set of output candidates for a given underlying representation. In OT, the output candidates are assumed to be defined by a function Gen(erator) which takes an underlying representation and (a) assigns every possible prosodic structure (e.g. syllabification, foot structure) to the output; (b) inserts morae, segments, and features freely, and (c) deletes morae, segments, and features freely - all of this subject only to the limitations imposed by the existence of inviolable well-formedness principles. Gen defines an indefinitely large set of possible candidate outputs for any given input form. It is of course not practical to try to list of all these candidates when discussing particular examples. The vast majority of the members of this set will obviously be nonoptimal. We will restrict our attention to what are the most likely candidates for any given input. We have not yet addressed one of the central aspects of OT. The constraint system is universal. Constraints are not like the rules of GP, which the speaker acquires on the basis of the data to which she is exposed while learning the language. The constraints are the expression of the Interaction of the communicative function of language with the articulatory and acoustic principles that underlie the human language faculty. The constraints are not learned. They are simply the array of principles that determine the sound pattern of human language. Grammars differ in terms of the set of initial representations. Grammars differ in terms of the ranking of the constraints. (Grammars also differ in terms of the precise interaction of the constraints with the lexicon and the morphological structure of the language, though we do not here pursue this matter in any detail.) Grammars do not differ in terms of the set of constraints themselves. Grammars do not lack a given constraint; they simply rank the constraint so lowly that it has little if any opportunity to participate in selecting an optimal candidate. It is an empirical question as to what constitutes the set of universal constraints. In many cases we do have some general idea concerning the basic nature of a given constraint, but the precise formulation of these constraints remains to be determined. And deepening knowledge of the phonological structure of the world's languages may lead to the identification of hitherto undetected (or very imperfectly understood) constraints. In a certain sense it is impossible to give an analysis of any language's grammar without knowing what the universal constraint set is. On the other hand, determination of the constraint set can only proceed by the intensive study of the phonological systems of the world's languages, phonological change, and the acquisition of phonology, as well as exploration into the articulatory and acoustic bases of human speech. Thus
6
there is no option but to attempt to explore the constraint systems of particular languages in the absence of a full understanding of the universal constraint set 1.2. Optimal Domains Theory Up until this point we have simply sketched the principal features of Optimality Theory. Let us now turn to the implementation of OT known as Optimal Domains Theory. The origins of ODT can be found in Kisseberth's (1994) analysis of Xitsonga, a Bantu language spoken in South Africa, where he attempts to develop an account of the extent of High tone spreading in Xitsonga that is independent of autosegmental representations and that allows a collection of independent principles to jointly define the precise extent of spreading. The rise of Optimality Theory permitted a further development of these ideas, resulting in a series of papers on harmony by Cole and Kissebenh (1995a,b,c, 1997) and in Cassimjee and Kisseberth's (1998) examination of Bantu tonology, with particular reference to Isixhosa and Shingazidja. ODT is in essence a proposal about featural phonology in OT (the seminal works in OT grappled primarily with "suprafeatural" problems — syllabification, stress, reduplication, epenthesis, etc.). The basic idea is simply this: a phonological feature F can be realized phonetically only by means of the organization of one or more segments/moras (whatever the appropriate feature-organizing unit is taken to be) into a "domain of segments" (henceforth, Fdomain) over which it (potentially) may find expression. A feature cannot be realized on a segment/mora unless that segment/mora is inside an F-domain. In ODT, whether a segment/mora inside an F-domain actually bears the property F is itself resolved by virtue of constraint ranking. Specifically, ODT proposes that there is a constraint family Express. Express (F) requires that all segments in an F-domain bear the feature F. A segment inside the F-domain satisfies Express provided it has an F-specification, regardless of whether this specification is an underlying one or one provided by Oen. But an output candidate may have segments inside the F-domain that do not bear a specification for F (Le. such representations are not prohibited by inviolable principles of phonology). Such segments do violate Express, but violation of Express will be optimal if there is some anti-expression constraint (Le. a constraint which bans particular features in particular contexts) ranked above Express. In other words, Express will be violated in order to better satisfy a constraint against a particular feature in a particular context. Given this conception of features and their realization, Faithfulness turns out to be a notion that requires deconstruction. Let us begin by briefly reviewing the treatment of featural faithfulness in McCarthy and Prince (1995). They propose that there is a family of constraints Idem (F) which require that a segment in the output be featurally identical to its correspondent in the input. If a segment in the input bears a feature (value) F that its correspondent in the output does not bear, this is a violation of Idem (F). If a segment in the output bears a feature (value) F that its correspondent in the input does not bear, this is also a violation of Idem (F). It seems then that just as Max (segment) and Dep (segment) are distinguished, it is also the case that there are two aspects to Idem (F), though McCarthy and Prince (1995) to do not provide separate labels. We might refer to these two aspects as Max (feature) and Dep (feature). This account of featural Faithfulness misses the fact that in the arena of features there are degrees of faithfulness. For example, suppose that there is an input sequence jk and j bears F but k does not; suppose further that while j's correspondent in the output does not bear F, k's correspondent does. We have here a violation of Max (F) and also of Dep (F). The correspondent ofy is in violation of Max (F); the correspondent of * is in violation of Dep (F). But surely this is a case of "displaced" contrast and represents a form of faithfulness that is not recognized as such by just counting up violations of Max (F) and Dep (F).
7
ODT proposes a rather more complex view of feature! faithfulness, one that recognizes that there is imperfect faithfulness that is not complete infidelity! The following Faithfulness constraints are proposed: (5)
DomCor (F-sponsor) For every F-sponsor (Le. segment/more specified as F) in the input, there is a 'corresponding' F-domain in the output
(6)
Incorporate (F-sponsor) Every F-sponsor is in an F-domain.
(7)
Express (F) Every segment/mora in an F-domain should bear the feature F.
(8)
Basic Alignment BAL: Align the left edge of an F-domain with the left edge of the F-sponsor to which it 'corresponds'. BAR: Align the right edge of an F-domain with the right edge of the F-sponsor to which it 'corresponds'.
(We should note that yet another constraint can be motivated - Uniqueness (F-sponsor) - which says that there is at most one F-sponsor in an F-domain. See 1.3 below for some discussion, as well as Cassimjee and Kisseberth (1998), where the role of this constraint in the analysis of Shingazidja is shown to be critical.) Faithfulness to an underlying F-specifkation occurs when F is phonetically realized on the sponsor and no nonsponsor bears F. Let us review how, in ODT, each of the above constraints participates in achieving faithfulness. DomCor requires that there be a domain for each input F-specifkation. The presence of a domain is a necessary ingredient to faithfulness since a feature F cannot be realized phonetically on an element unless that element is inside an Fdomain. Incorporate says that an F-sponsor is required to be in a domain. The inclusion of the sponsor inside the F-domain is, in ODT, the only means for that sponsor to actually be realized with F. Since being in an F-domain is not sufficient for a segment/mora to realize F, there must also be a constraint that actually requires the elements in the domain to realize F. This is the role of the Express constraint BAL and BAR require that no element other than the sponsor be in the domain (BAL requires that the sponsor be at the left edge of the domain and BAR requires that the sponsor be at the right edge of the domain). BAL and BAR thos do the same work as McCarthy and Prince's Dep constraint - they prevent a feature from appearing other than on the sponsor. If (5)-(8) are all undominated constraints, then each underlying F-specifkation will result in a surface form where F is realized on each sponsor and only on the sponsor. We may refer to this as perfect faithfulness to the input. But very often one or more of these Faithfulness constraints is violated (due to being dominated by a Phonoconstraint) and only an imperfect 3
There are, of course, other possible ways of getting at the idea that only sponsors should be inside domains. The Basic Alignment version is motivated by the fact that there appear to be cases where a sponsor is prevented from being inside the domain, but the domain is formed as near to the sponsor as possible. Bask Alignment, being a constraint that can be gradlently violated, allows one to capture this fact The issue does not arise in the context of Isixhosa, thus a constraint that directly banned nonsponsors from being inside domains would work equally as well as Basic Alignment. 8
faithfulness is achieved. Various examples of imperfect faithfulness will be exemplified in this and subsequent chapters with respect to tonal phenomena. Let us now begin to look at how these Faithfulness constraints work to select optimal outputs. F-domains are-aspects of phonological structure, not different in principle from other sorts of domains that are formally part of phonological representations (e.g. syllables, prosodic feet). As such. Gen freely generates output candidates with all possible F-domain structures (including the absence of F-domains). In the optimal output, a domain will exist just in order to satisfy either (i) a Faithfulness constraint (DomCor or Incorporate are the two Faithfulness constraints that need a domain in order to be satisfied) or (ii) some Phonoconstraint that demands the presence of F (thereby requiring a featural domain, since a feature cannot exist in the output without a domain). Faithfulness or Phonoconstraints can have the effect of demanding a domain only if they are ranked above the member of the *Struc family of constraints that bans feature domains. Any domain not motivated by Faithfulness or Phonoconstraints will necessarily be nonoptimal by virtue of *Struc. The Faithfulness constraints Express and Basic Alignment are vacuously satisfied if there is no F-domain at alL If they are undominated, then an underlying F-specification will be realized on the sponsor. Violations of Express occur when a Phonoconstraint prohibiting the feature F in some context domainates Express. Violations of Basic Alignment arise when domains are either wider than the sponsor of F or do not include the sponsor. There are two functional motivations for wide domains. One functional motivation can be referred to as Articulator Stability (or Ease of Articulation). From the point of view of articulation, it is easier to maintain a a gesture rather than to shift back and forth between gestures. The second motivation can be referred to as Perceptual Enhancement. From a perception point of view, a feature's perceptibility is enhanced if the feature can be extended over more segments. Notice that Articulator Stability and Perceptual Enhancement both prefer outputs where a domain is wider rather than smaller. (See Cole and Kisseberth 1995a and Kaun 199S.) We take the view that Articulator Stability and Perceptual Enhancement are not constraints themselves, but rather the motivation for constraints that align F-domains with the edges of constituents such as the word or phrase. Thus we propose a category of Widescope Alignment constraints which seek to align the edges of every F-domain with the edges of prosodic constituents. If Widescope Alignment constraints are dominated by Basic Alignment, then the optimal candidate will have a narrow domain consisting of just the sponsor. If, however, a Widescope Alignment constraint dominates Basic Alignment, the optimal output will have wider domains. For example, we know that in certain languages a feature F, sponsored by a given segment, may extend its influence from that segment to the end of the word, or to the beginning of the word, or bidirectionally throughout the word. We list these three cases as (9a,b,c). (9)
a. Feature F propagates from sponsor to the end of the word. b. Feature F propagates from sponsor to the beginning of the word. c. Feature F propagates from sponsor to both the beginning and the end of the word
The subtypes in (9) can be characterized by appealing to the following two Widescope Alignment constraints: (10)
Align Word R: Align the Right edge of an F-domain with the Right edge of a Word. Align Word L: Align the Left edge of an F-domain with the Left edge of a Word. 9
Case (9a) is predicted by having Basic Alignment Left and Align Word R undominated, and having Align Word R dominate Basic Alignment R and having Basic Alignment Left dominate Align Word L. The tableau in (11) illustrates. (When either of two mutually unranked constraints would evaluate a candidate as nonoptimal, we place parenthesized exclamation marks in both columns.)
The undominated BAL constraint eliminates any candidate where the initial element a is inside the domain. The undominated Align Word R constraint eliminates any candidate where the domain is not aligned with the right edge of the word. The only candidate that survives these two constraints is the optimal one — a (b'c) - which violates the two lowly ranked constraints, BAR and Align Word L. The (9b) case will follow from having BAR and Align Word L undominated, and having Align Word L dominate BAL and BAR dominate Align Word R. The tableau in (12) illustrates.
02) input: a b' c (a'b'c") aMc
-WbOc1
a (b'c )
BAR *!
Align Word L
BAL *
*!
Align Word R ;.*:-
*(«>
%£•'••••:
4
* *(!)
Given this constraint ranking, BAR eliminates all candidates where the domain is aligned with the right edge of the word and Align Word L eliminates all candidates that fail to align the domain with the left edge of the word. The optimal output - (a ( br) c - violates the two most lowly ranked constraints, BAL and Align Word R. Case (9c) derives from a constraint ranking where Align Word R and Align Word L are undominated, with Align Word R dominating BAR and Align Word L dominating BAL. Given this ranking, nonsponsors that help extend the domain either to the Right or the Left edge of the word will be incorporated into the domain. (13) input: a b1 c -Xa'b'c 1 ) a(b*)c (a'bnc a (b'c1)
Align Word R
Align Word L
*(!)
*(!)
-
*!
BAL *
BAR *
* .
*i
*
.
'
•
•
.
•
•
-
We have seen that by ranking Widescope Alignment constraints and Basic Alignment, we can obtain differing patterns of domain structure. By ranking Widescope Alignment 10
constraints with Express we can also obtain the phenomena known in the harmony literature as opacity and transparency (cf. Cole and Kisseberth 1995a). We review this point below. Given that Express is a constraint, it is violable. Express may be violated as a consequence of a constraint that bans the combining of a feature F with a feature G (call this a combinatorial or clash constraint). For example, assume an ATR-domain. Express would demand that the feature ATR be linked to every ATR-bearing element in the domain (for our purposes, let us assume that only vowels bear ATR). But suppose that there is a constraint that bars the combination of ATR and Low. This constraint favors output candidates where low vowels are not ATR. If Widescope Alignment and Express are ranked above the clash constraint *[ATR,Low], then a domain will extend past both Low and High vowels to the edge designated. Harmony results, with neither opacity nor transparency. However, if Express and *[ATR,Low] outrank Widescope Alignment, we have a situation where the extent of the domain is restricted so that the Low vowel will not be included in the domain. This is opacity. If Widescope Alignment and *[ATR,Low] dominate Express, it will be better to incorporate the Low vowel in the domain but not express ATR on that voweL This is transparency. (See Cole and Kisseberth (1995a) for discussion and exemplification.) The ODT approach to harmony thus allows both transparency and opacity to be characterized in terms of the interaction of Express with Widescope Alignment. There is a problem with this analysis. A single language may (conceivably) have both transparent and opaque targets in connection with the same F-domain. This leads to an apparent paradox: Express must dominate the widescope-constraint to get opacity, but the reverse ranking is required to get transparency. Consider a hypothetical example. Suppose we have a language with ATR harmony which extends bidirectionally throughout the Word. Thus the ATR-domain will be affected by the two widescope-driving constraints Align Word R and Align Word L. For convenience, we will simply refer to these two constraints as Widescope Alignment. Now, suppose that Low vowels are opaque (Le. an ATR-domain triggered by an ATR-sponsor cannot be extended onto or across a Low vowel). Furthermore, suppose that schwas (=Placeless) vowels are transparent (they do not bear the ATR feature and the ATR-domain extends across them). *[ATR,Low] predicts that ATR will not be realized on Low vowels inside an ATRdomain. "'[ATR, Placeless vowel] predicts that ATR will not be realized on a Placeless vowel in a ATR-domain. But now we are in a bind. If Express is ranked above Widescope Alignment, then we predict that both Low vowels and Placeless vowels will be opaque. If Widescope Alignment is ranked above Express, then we predict that both Low vowels and Placeless vowels will be transparent. We thus have an apparent paradox. One possible way out of the paradox would involve what in the OT literature is known as the "explosion" of a constraint into a family of constraints. Suppose that Express [ATR] is actually a family of constraints, including the following two members: (14)
Express [ATR on Low vowels]
(15)
Express [ATR on Placeless vowels]
The various members of this constraint family need not be ranked in an identical fashion. Thus, if the clash constraints are undominated, and if Express [ATR, Low vowel] is ranked above Widescope Alignment, but Express [ATR, Placeless vowel] is ranked below Widescope Alignment, then we have a situation where Low vowels are opaque and Placeless vowels are transparent.The tableau in (16) illustrates. Underlining indicates [+ATR]. Consonants are ignored. 11
(16) input: i o 11 i a i *[ATR, Low] (JailiaD
*(!)
*[ATR, Express Placeless V] IATR,Low]
Express [ATR. Place less]
ai
*
M) •1
GsuDai ->G»Ui)ai
(isaiai) (iouiaD
Wide scope Alignment
*!
SiiippiH
•1
i s a!i
iadiDai
Furtherresearchisrequiredto determine whether this is a linguistically appropriate solution to the problem. The issue is, however, rather tangential to the tonal concerns of the present work. It will be useful to discuss one additional general matter before turning to the ODT treatment of tone. F-domains, we have suggested, have the same status as other types of domains (syllables, feet, prosodic words, prosodic phrases, etc.). A very common assumption about these other domains is that they are non-overlapping (a possible, but not universally agreed upon, exception to this being "ambisyllabicity", where a consonant is considered to be simultaneously part of two different syllables). The ban on overlapping structure has generally been viewed as an inviolable aspect of phonological structure. Weregardit, however, as a constraint: (17)
"Overlapping Prosodic Structure Prosodic structure (Syllables, Feet, F-domains, Prosodic-domains) do not overlap.
As a constraint, it is violable. Thus we consider it possible that there might be other constraints that would be best satisfied by an output candidate that violates "Overlapping Prosodic Structure. A possible example arises in our analysis of Isixhosa. 1.3. OTD and the analysis of Bantu tone. At this point, let us consider the application of ODT to tonal systems and more specifically Bantu tonal systems. We follow the general consensus that in Bantu languages, the lexical inputs to the phonology have specifications for High tone but no other tonal specifications (specifically, there are no Low tones in lexical inputs). In autosegmental phonology, it has been common to view tonal features as anchored to either the mora or the syllable. We believe that tone is specified on moras, and that the role of the syllable in tonal structure can be accounted for without claiming that tone is specified directly on the syllable. This will not however be a theme that we dwell on here; we simply assume that tone in Bantu languages is specified on moras. [Onset consonants, since they are not moraic, will play no role in our discussion immediately below. We will assume that High [Tone] Domains prefer to align with syllables, and thus when the vocalic nucleus of a syllable is inside such a domain, we will assume that the onset to that syllable is also in the domain. It should be noted that onset consonants in Isixhosa, as in other Nguni languages, can play a critical role in the tonal system. We deal with this at great length in Chapters Two through Four. To varying degrees, the location of High tone specifications in Bantu languages may be a function of the morphological structure. We will not be concerned here with the predictability of the location of some High tone specifications in Bantu. We rather are concerned with the phonological patterning of tones given that they are specified on particular moras. The reader 12
should not take these comments to mean that there is nothing theoretically interesting about how (predictable) tones come to be located on particular morae. The issue is of great import; it is just beyond the scope of our concerns here. Suppose that we have a mora that sponsors a High-tone specification in the underlying representation. In ODT, DomCor requires that this High tone be parsed into a High Domain so that it can be realized phonetically. Incorporate requires that the sponsor of the feature H be Internal to the HD. Basic Alignment requires that the HD's edges be aligned with the edges of the H-sponsor. The constraint Express requires the moras inside the High Domain to bear the feature High tone, in violation of *Stmc (which bars all features as well as all domain structure, etc.). As mentioned earlier, if all of these Faithfulness constraints are satisfied, we will have "perfect faithfulness" - a High tone will appear on each sponsor and there will be no other High-toned moras. The vast majority of Bantu languages seem to be only imperfectly faithful, sometimes to quite astonishing degrees. In the remainder of this chapter, we will sketch the most pervasive tonological phenomena in Bantu and show how these phenomena can be understood to be the consequences of the interaction of a relatively small number of universal phonological constraints interacting with the various (featural) Faithfulness constraints that ODT postulates. The most fundamental phenomenon in Bantu tonology is the extension of a High tone from the sponsor of High to one or more neighboring moraic elements. When the extension is just onto a neighboring moraic element, the phenomenon is often referred to as High Tone Doubling or (Local) High Tone Spreading. When the extension is not so restricted, it is usually referred to as (unbounded) High Tone Spreading. In ODT, as noted above, the extension of a feature is viewed as being functionally motivated. In particular, a feature is perceptually enhanced if it is extended over more than one bearer of the feature. Ease (or Stability) of articulation also enforces maintenance of an articulation over multiple elements. Let us consider first those Bantu languages where a High spreads one mora to the right (or left). We propose that this type of spreading represents a minimal violation of Basic Alignment, this minimal violation being necessitated by the existence of a constraint that requires a degree of unfaithfulness. The constraint that we propose is the following4: (18)
*Mono(moraic) HD H-Domains do not consist of a single mora.
*MonoHD demands in effect that the High tone achieve some increased perceptibility by virtue of expanding beyond the sponsor to include at least an additional mora. Satisfaction can, of course, be achieved by expanding either to the Left or the Right of the sponsor. Satisfaction of *MonoHD often can be achieved only by including a nonsponsor in the domain, thus violating Basic Alignment (since the HD will be misaligned with the sponsor). When Bask Alignment is an undominated constraint, there can be no spreading of a High tone. If *MonoHD is ranked above Basic Alignment Right, then it will be optimal to extend a HD from the sponsor to the following mora in order to satisfy *MonoHD. But Basic Alignment will be violated minimally - i.e. the HD will extend only one mora since extending the domain over additional moras will incur a greater misalignment of the right edge of the H-domain with the Hsponsor without yielding a better satisfaction of *MonoHD. If, on the other hand, *MonoHD 4
One might reasonably propose an alternative (positive) constraint requiring binarity in the High Domain. We suspect that the explanation for binarity in stress domains does not carry over to pitch domains and that the negative constraint given here is more likely to be correct. However, nothing in the present work rests on the assumption that *Mono(moraic) HD and not Binarity is the appropriate constraint.
13
outranks Basic Alignment Left, the High Domain will extend to the left (since vioalting BAL is countenanced if the result is a domain that is not monomoraic). Let us illustrate this analysis with a concrete example. Emakhuwa (a Bantu language spoken in Mozambique and Tanzania) has "doubling" to the right The Emakhuwa tonal pattern has been analyzed in detail in Cheng and Kisseberth (1979,1980,1981). We have underlined the sponsors of a H tone in our examples. The placement of three dots after the item is meant to indicate that the item is in medial position in the phrase. (The significance of this will be dealt with later.) In (19) we show that the mora after a H-sponsor is regularly High in Emakhuwa. (19)
imipl.. 'house* nttjindtf... lull' nr£re'rn£l&... 'eel' nthgnd6ro... 'a long rope' umfri.. 'to fall' uhjpdttha... 'to hunt' urjikun]ist..'to turn over' Emakhuwa, as illustrated above, requires the following constraint system:
(20)
DomCor, Incorporate, *MonoHD, Express, BAL BAR - dominated by *MonoHD
undominated
The following tableau illustrates. To simplify, we have used the label "Max" to stand for DomCor, Incorporate, and Express. (21) Candidates ulupattha ufllppanha uhj(pfl)ttha
•MonoHD
as^smmimmmasmmmmm§mm Mdncor)
u(lflpfoM) (uljDpattha u(lupa~)ttha • u(Ujpa)ttha
oo\plllllBHIIIII^I
-
TOl
"[(Express)
Notice that the candidate (ulupattha -with "doubling" to the left -- would be optimal were BAR undominated and *MonoHD ranked above BAL. Even if *MonoHD is ranked higher than BAR, there may be other constraints - more highly ranked than *MonoHD - which will in effect prevent the doubling. We present two cases here. The first case is exemplified by Emakhuwa. We have seen that there is evidence that a H tone "doubles" to the right in Emakhuwa.. But this does not occur when the mora to the right is final in the (intonational) phrase (~IP)5. 5 We should note that while the failure to double onto an IP-final mora is a quite general phenomenon in Emakhuwa, there is at least one dialect - Enlai, spoken in the Angoche district of Nampula province in Mozambique - where a High tone does double onto an IP-final mora (unpublished research of C. Kisseberth and Francisco Ussene Mucanheia).
14
(22)
injpa 'house' nttunda 'hiU' nrgre'mgla 'eel' uirgra 'to fall' urjkungsa 'to turn over'
We propose to account for this failure to "double" in terms of a constraint belonging to the family of constraints known as Nonfinality. (23)
Nonfinality (HD) The R edge of a HD may not be aligned with the R edge of an IP.
If Nonfinality (HD) dominates *MonoHD, then we correctly predict the Emakhuwa data, as shown by the following tableau.
gfiCandidates umora ume(rd) (um£)ra
Max
Non-finality (HD)
•l(DomCor) *!(Incorp)
*dfevta #^§* !l: a!
BAR *' a
•'
' t \ • •
-Mi(m<S)ra
u(mgrif)
BAL •MonoHD 1 '."/Ifr.M&i IW*?-. -- '
.
*!
a
Setswana (a Bantu language spoken in South Africa and Botswana) provides a second example where a constraint dominates *MonoHD and consequently makes "doubling" nonoptimal The data and analysis of Setswana are from Mmusi (1992). The following data set illustrates that Setswana "doubles" to the right. The mora sponsoring a High tone is again underlined. (25) gofd 'to die' gorfka' 'to buy' gob£l£ya 'to kill' goigisanya 'to live in harmony' gokhunimeletsa 'to cover for' cf. toneless verbs: gowa 'to fall' golema 'to plow' gotsamaya 'to walk' In (25), we see infinitive verb forms where, if the stem has a High tone on the first mora, then that High doubles onto the second mora. Doubling may also go from a prefix to another prefix. For example, the High of the subject prefix in jdlema '(s)he is plowing' where a High-toned subject prefix doubles onto the following /a/ prefix. The stem /lema/ in dilema is toneless. (26) below shows a High verb stem. Notice that there is no doubling from the subject prefix to the /a/.
13
(26)
£ a Tiki '(s)he is plowing' d. a bghlya '(s)he is killing'
What is going on here is a well-known phenomenon from Bantu: a High tone does not double onto a vowel that is followed by a High-toned vowel (an effect of the so-called "Obligatory Contour Principle" proposed in autosegmental phonology). We propose the following constraint: (27)
No Adjacent Edges Domain edges may not abut
If No Adjacent Edges is ranked above *MonoHD, then we will account for the Setswana data. We do not illustrate here the role of the Faithfulness constraints other than Basic Alignment Thus we only include candidates that do not violate these other Faithfulness constraints.
(2§1 Candidates (eaXboJa)ya (fi)a(ba)laya -> fo) a (bg.la)ya
BAL
No Adj Edges
•MonoHD
BAR
»*! *
lilllflllilil
*l a
So far we have illustrated the phenomenon of "doubling" of High tones in Bantu and we have also illustrated how certain constraints, Nonfinality (HD) and No Adjacent Edges, may lead to violations of *MonoHD, the constraint that forces doubling. We have also seen how the domination of BAR by *MonoHD accounts for the rightwards doubling in Emakhuwa and Setswana. Let us now turn to see how ODT accounts for the unbounded extension of High tone. In Xitsonga (a Bantu language spoken in Mozambique and South Africa) a H tone will "spread" (using autosegmental terminology) in an unbounded fashion (see Kissebcrth (1994) for discussion). Consider the following examples. The left-hand column shows toneless verb stems in the first person singular present tense, while the right-hand column shows the corresponding third person plural forms. As in many other Bantu languages, third person subject prefixes sponsor a High tone. (In the interest of accuracy, we should note that there is an /a/ prefix located between the subject markers, /ndzi/ and /vg/, and the verb stem in (29). The subject prefix vowel elides. We nevertheless, in the third person case, underline the surface /a/.) (29) ndzatirrha vftfcrha (work) ndzatsutsu:ma v(Jtsutsu:ma (run) ndzatk>mute:la vjtl6mtft£:la (fish) ndzixava nya:ma vjxiva' ny£:ma (buy meat) From these data, we see that the High tone of the third person subject prefix spreads through the penultimate syllable of the phrase (the last example cited shows that the spreading crosses from the verbal word to a following complement and therefore is not a word-level phenomenon). When a domain extends in an unbounded fashion towards the end of the word, ODT appeals to the constraint aligning the R edge of a domain with the R edge of a word. But 'word* is just one of the possible categories that a domain may align with. The "prosodic phrase" (=PP) is another possibility. We propose that in Xitsonga it is Align PP R that is active (i.e. highly enough ranked to have an effect on the selection of optimal outputs).
16
(30)
Align PP R Align the R edge of a HD with the R edge of a Prosodic Phrase.
(See Kissebeith (1994) for a pre-OT discussion of the notion "Prosodic Phrase" in Xitsonga.) If Align PP R is ranked above BAR, we will characterize the situation where a High tone spreads as far as possible towards the Right edge of a PP. In Xitsonga, Nonfuiality (HD) dominates Align PP R and thus prevents an IP-final mora from being included in the HD. This explains why a H spreads only to the penult and not to the final syllable in the examples in (29). The following tableau illustrates the analysis of Xitsonga. We do not include the Faithfulness family of constraints other than Basic Alignment Right and therefore do not consider candidates where such constraints are violated.
(3D Candidates (v6tkSmrite\l£) (vDtk>mute:la (v|tl6)mute:la (vftldmu) te:la -> (v|fi6mut&)la
Nonfinality(HD)
Align PP R
BAR tk>muie:lo
*i
tlomulte; Id mute:! la te:la!
la
•
-
tlo tlomu tlo mute:
This tableau illustrates clearly the role that gradient violation of Align PP R plays. The first candidate, which is the only one that satisfies Align PP R perfectly, is rejected on the basis of the more highly ranked constraint Nonfuiality (HD). It is simply impossible to satisfy both Nonfuiality (HD) and Align PP R perfectly. It is worse to violate the more highly ranked constraint (Nonfuiality (HD)) than the less highly ranked constraint (Align PP R). Of the remaining candidates, the most optimal is the one that best satisfies Align PP R - i.e. the last candidate, which is misaligned from the R edge of the PP by a single syllable. We have seen that Nonfuiality (HD) constrains the alignment of a HD with the R edge of the PP. The constraint No Adjacent Edge has a similar role. We can see this by considering examples where a verb is followed by a nominal complement with a High tone. (32)
ndzixava ta:ndzjj 'I am buying an egg' Vjixdva' ta:ndz4 'they are buying an egg' ndzixava mata:ndzi 'I am buying eggs' vfixdva" m£ta:ndzg 'they are buying eggs'
The HD initiated by the subject prefix of the verb clearly extends rightward. It cannot, of course, extend all the way to the end of the phrase without overlapping the HD containing the final vowel of the nominal. (In the course of our analysis of Isixhosa, we will address the issue of whether the ban on overlapping domains is an inviolable or a violable one. Whatever the answer, Xitsonga does not permit overlapping HD's.) Furthermore, (32) shows that the subject prefix's domain also does not extend onto the syllable in front of the HD in the nominal. Clearly, No Adjacent Edges constrains the alignment to the Right In this language, just as in Setswana. The following tableau illustrates the analysis of Xitsonga. As usual, we ignore those constraints that are not directly pertinent to the present discussion, and therefore we also ignore candidates that violate those constraints.
17
(33) NoAdjacent Edges Align PP R BAR Candidates (yixAvi mit£0(ndzal 1* ^R\p§lH§SH^p (vi)xava mata: (ndzd) xavamalta: ndza/ 0 vamata-.'.ndza/ 0 (v4xa)va mata: (ndzd) (vtfxivd) mata: (ndzi) mata: ndza!/ 0 xavama ta:ndza/ 0 -HvaxavintiUta: (ndzd) (In the column dealing with Align PP R, to the left of the slash we indicate the extent to which the first HD fails to be aligned with the R edge of the prosodic phrase; to the right of the slash, we indicate the fact that the second HD is perfectly aligned.) In the examples so far discussed, Express has been an unviolated constraint when it comes to m o n k elements in the domain. But there is another widespread phenomenon in Bantu tone systems that ODT treats as representing the situation where total satisfaction of Express is not achieved. Bantu languages are, in the autosegmental literature, regarded as either being spreading languages (a High tone on the sponsor is heard not only on the sponsor but one or more moras to the right or left) or shifting languages (the High tone is not heard on the sponsor but rather on some other mora to the right or left of the sponsor). Of course, shifting languages in some cases were analyzed as involving the same spreading as observed in the "spreading" languages, but with an additional phenomenon of the delinking of the High tone from all of the moras up to the surface site of the High tone. Let us turn our attention now to some examples of "shifting" languages. Let us begin with a shifting language where the High tone extends just one mora to the right. Kijita (see Downing 1990) represents a typical example. In Kijita, a High tone spreads (in autosegmental parlance) one mora to the right, but delinks from its original mora. Furthermore, the High does not spread onto an IP-final mora. (34)
okufwi'todie' kumugera T>y the river* okufwa. kdmugera 'to die by the river* okuf&na 'to see' okuPsnana 'to see one another1 okuPflni i:nyonyi 'to see a bird' cf. oku()uma 'to hit' okujJumana 'to hit one another'
A word with a final High tone, like okufwg (we ignore a phonetic rule that make a final H falling in Kijita), will retain the H on this syllable in IP-final position, but the H will "shift" onto the following syllable in IP-medial position (cf. okufwa kumugera). Verbs in Kijita are either Hightoned or toneless (cf. okuflgna vs. okufluma); the High toned verbs will have the High tone dock to the first stem syllable (cf. okup^na). When the first stem syllable is penult in the IP, then the H will remain on that syllable, since the H will not shift onto the final syllable of the IP. However, once the word is IP-medial, then shift (Lc. spreading/delinking) can occur. This 18
explains the contrast between okupgna and okupjmd i:nyonyi as well as between okuJJgna and okuflgnana. Part of the ODT analysis of Kijita is immediately evident. Clearly, Nonfinality (HD) dominates *MonoHD, and *MonoHD dominates the BAR part of Basic Alignment. In other words, it has the same system as Emakhuwa. In Kijita what we must explain is why a High tone Is expressed on the second mora in the domain and not the first. We propose the following analysis. Domains are headed phonological structures - either Right- or Left-headed. We suggest that HD's in Kijita are Right-headed. We then propose that there is a constraint: (33)
*(High tone, nonhead)
which bans realizing High tone on nonheads. If (35) dominates Express (H) in Kijita, it will prevent a High from being realized on the first (nonhead) mora in a bimoraic domain. We will achieve the effect of "spread" and "delink". Notice that in order for this analysis to succeed, *MonoHD must dominate Express (H). If the ranking were the reverse, then it would be better to have perfect satisfaction of Express even if that could only be achieved by having a HD that contains only the head (thus violating •MonoHD). By ranking *MonoHD over Express, we require the presence of a mora in the HD that does not realize High tone. This represents the same kind of interaction that constitutes "transparency" in harmony systems (see the preceding section). Kijita illustrates the role of (35) in a system where *MonoHD is the only widescopedriving constraint ranked above BAR. Mijikenda (the data cited here is from work in progress by C. Kisseberth and myself) represents a language family where (35) is operative in conjunction with much wider domains than the bimoraic domains of Kijita. Mijikenda is like Xitsonga in that a H tone that originates to the left in the word will surface on the penultimate mora (of the Prosodic Phrase). However, whereas in Xitsonga all the moras from the point of origin to the penult vowel are High, in Mijikenda only the penult vowel is High. (36)
ninarima ninagula ninalamusa ninavumikiza ninagula nyama ninalamusa muganga
yunarfma yjHiagtila yunalamusa yu.navumikfza yjmagula nydma yjjnalamusa mugringa
(cultivate) (buy) (greet) (agree) (buy meat) (greet doctor)
On the basis of these data, we assume that Align PP R outranks BAR in Mijikenda, and that Nonfinality (HD) outranks Align PP R. Up to this point, there is no difference between Mijikenda and Xitsonga. The difference between the two systems will derive from the fact that in Xitsonga, Express dominates *(High, nonhead), while in Mijikenda the reverse ranking obtains. Furthermore, Align PP R must dominate Express in Mijikenda (whereas in Xitsonga these constraints are not necessarily ranked, due to the fact that Express is an undominated constraint). As a consequence, the domain will be as wide as possible (even though Express (H) is grossly violated and only the head of the domain is realized on a High tone). The tableau in (37) illustrates. (Note that sometimes a candidate violates two different constraints that are unranked with respect to one another, and either violation could be regarded as fatal to that candidate's chances to be optimal. In such cases we enter "I" after both violations but place the symbol in parentheses.)
19
(37) Candidates (yunalamusa) (yAnihfmusf) (y]j)nauunusa (y£ndl4mtf)ga -Hyunalamu)sa
•(H.nonhead)
Nonfinality
•1
*(!)*** *!**
Align PP R
BAR
Express
SgSSlSSti l*i«^-*SSMBSll:ipS^
lt| : itlSlMl^nM3^KiPlli : l*ip mnnniwssKiisasssss? ftilSliliSl£fiM»iiSllM;Sij®M^^^8^i *(l)
| sa
| nalamu
| ***
]
So far we have examined, from the ODT perspective, the pervasive Bantu phenomena of "spreading" and "delinking" of High tones. In so doing, we have had occasion to appeal to a Nonfinality constraint and a No Adjacent Edges constraint. These two constraints have to do with two other very pervasive aspects of Bantu tonology: phenomena having to do with the Right edges of words/phrases and phenomena having to do with the avoidance of adjacent High tones the so-called Obligatory Contour Principle (=OCP). For our present purposes, it is not necessary to explore in any more detail the role of right edge effects in Bantu tonology. However, it will be useful to look in a bit more detail at the ODT treatment of the Obligatory Contour Principle. We shall argue that the OCP is a family of constraints. Bantu provides examples of the various members of this family. We have already seen one expression of the OCP in ODT: namely, the constraint No Adjacent Edges. In the examples discussed earlier, we saw how No Adjacent Edges, if ranked above widescope-driving *MonoHD and Align Word/PP R constraints, expresses the case where the OCP prevents the spreading of High tones when spreading would produce a violation of the OCP. Another manifestation of the OCP in Bantu tone is the phenomenon that is referred to as fusion. In autosegmcntal parlance, "fusion" is the case where two or more High tones are fused into a single, multiply-linked High tone. In ODT terms, "fusion" is a case where two Highsponsors are placed into a single HD (violating Domain Correspondence, which requires a domain corresponding to each sponsor, but satisfying Incorporate). The evidence for fusion involves showing that two separate High-sponsors behave as though they are included in a single domain. Isixhosa represents a language where there is unambiguous evidence for fusion, so we will not here attempt to motivate fusion - rather, we will show how an appropriate ranking of the constraints will make fusion an optimal output. Assume a hypothetical language that lacks any spreading (Le. the constraints driving wide domains are ranked below Basic Alignment). Now, given an input nak|bona in this language, the perfectly faithful output would be na(k{)(bg)na. If No Adjacent Edges is ranked above DomCor, it will be better to minimally violate DomCor if by so doing we can avoid adjacent domain edges. Thus the ranking of No Adjacent Edges over DomCor will prefer a fused output such as na(kfb£)na over the perfectly faithful one. An undominated Incorporate constraint will prefer this fused output na(kfbg)na over na(kfjbgna. The latter has the same DomCor violation as the former, but in addition violates Incorporate and thus will be nonoptimaL The following tableau illustrates: (38) Candidates nakjbana naflcfjbpna ->na(kM)na na(kf)(b6)na
Incorporate •I* •1
No Adjacent Edges I DomCor
l^PSIWIiPffll T?y20
Fusion Is not the only way in which an OCP violation can be avoided. One of the best known phenomenon in Bantu tonoktgy is "Meeussen's Rule", which (in a procedural model) has the effect of deleting a H tone adjacent to a High tone. In ODT, Meeussen's rule is interpreted as a case, like fusion, where an OCP constraint (e.g. No Adjacent Edges) dominates DomCor. However, they differ in that the Meuussen's Rule phenomenon leaves one of the two H-sponsors outside a HD (though, of course, an independent principle may lead to its being incorporated into a HD - see the discussion of Emakhuwa below). How do we select as optimal an output that reflects Meeussen's Rule rather than fusion? We propose that the burden falls upon another Faithfulness constraint. Uniqueness, given below in (39). (39)
Uniqueness (F-sponsor) Each F-domain includes a single F-sponsor.
If Incorporate (H-sponsor) outranks Uniqueness, it will be better to incorporate multiple Hsponsors into some HD than to fail to incorporate one or more H-sponsors. This ranking thus leads to "fusion". To avoid "fusion", Uniqueness must dominate Incorporate (H-sponsor). Thus it will be better to fail to incorporate a H-sponsor if by so doing Uniqueness is better satisfied. To get fusion we must have the constraint rankings: (40)
Incorporate > Uniqueness OCP (e.g. No Adjacent Edges) > DomCor
to get Meeussen's Rule rather than "fusion", we must have the rankings: (41)
Uniqueness > Incorporate OCP (e.g. No Adjacent Edges) > DomCor Let us illustrate. Recall our hypothetical input nakibona in a language where High tones do not spread. We have seen how the ranking in (40) will yield na(kfbg)na as optimal In (42), we see how the rankings in (41) will yield as optimal either na(kf)bgna or nak[(b^)na. (42) Candidates nakjbena na(kfjbg.na na(kjb£)na naki(b£)na na(k£)(b6)na
No Adj Edges
Uniqueness
Incorporate
**! * •I * *!
DomCor
*
^3S38KiiS3S *
mmmmim?8$-ll§lill8f8
From this tableau, it is apparent that in the case of Meeussen's Rule, we have a directionality problem - which of the High tones is not parsed into a domain? Since Isixhosa does not make use of the Meeussen's Rule strategy, we do not undertake here a discussion of the problem of getting at directionality in Optimality Theory. One possible analysis is that, given the two alternatives na(k[)bo.na and naki(bg)na, the choice between these alternatives falls upon whether the language prefers domains to be aligned with the left or the right edge of the word We leave this matter for further research.
21
We indicated above that the OCP wilt turn out to be a family of constraints. So far we have mentioned only No Adjacent Edges. There is evidence that No Adjacent Edges is insufficient and that there must a second OCP constraint, one that we refer to as No Adjacent Sponsors. This constraint is given in (43). (43)
No Adjacent Sponsors *F-sponsor)(F-sponsor
This constraint bans adjacent High Domains, but just in case the sponsors are also adjacent. Emakhuwa provides evidence that No Adjacent Sponsors is needed in addition to No Adjacent Edges. We have already seen that in Emakhuwa, *MonoHD dominates BAR, resulting in the doubling of High tones. Nonfinality (with reference to the IP) dominates *MonoHD. However, there is clear evidence that in Emakhuwa No Adjacent Edges does not dominate *MonoHD. This is shown by the fact that a High tone "doubles" onto a mora that is in turn followed by another High-toned mora. In the following examples we, as usual, underline the H-sponsors. We do not here attempt to justify the claim that the underlined morae are sponsors. The reader is referred to the papers by Cheng and Kisseberth (1979,1980,1981) for argumentation. (44)
kingtfllma'... kingoijpl.. kinASrgkdngsl-
1 will cultivate...' 1 will bathe...' 'I will turn over...'
But just because No Adjacent Edges is violated does not mean that there are no OCP effects in Emakhuwa. The examples in (45) represent one situation where two H-sponsors are adjacent in the input. (45)
kahglfma... 1 cultivated...' kahfrgkuniisl.. 'I turned over...'
While the second H-sponsor is realized with a High tone, this High tone is simply the consequence of "doubling" from the first sponsor. If both H-sponsors triggered the formation of a HD, we would expect ka(hgj(lima)... and the pronunciation *kahgl{mi... In order to achieve the correct pronunciation, DomCor must be violated; there must be just one domain, not two: ka(hojj)ma... From these data it is clear that there must be some constraint that is ranked above DomCor (H) that makes it better to fail to parse a High into a HD than to violate that constraint. The issue is simply: what is the constraint in question? It cannot be No Adjacent Edges, as the following tableau illustrates.
/urujcunusa.../ u(ry)ku(nusa)... u(rujcu)(n]isa)... -»*u(ryku)nu.sa
•MonoHD
No Adj Edges
DomCor
*! *l *
Ranking No Adjacent Edges above DomCor incorrectly predicts the pronunciation ur&ktinusa... To obtain the correct pronunciation, urgktingssi..., No Adjacent Edges must be ranked below DomCor (H). Thus we need another OCP constraint, No Adjacent Sponsors, to be ranked above 22
DomCor (H). No Adjacent Sponsors has no bearing on an example like u(rgk(i)(ngxi)... and thus permits both H's in this example to be parsed; it does have a bearing on an example like ka(h£r£lku(n£x£)...and will prevent the second of the successive sponsors from being parsed into its own unique domain. We will not undertake here an elaboration of the details of the constraint system in Emakhuwa that will yield examples like ka(hgr£}ku(n£xa)... as optimal since our primary concern is to show that the OCP must be a constraint family, and that No Adjacent Edges and No Adjacent Sponsors are likely two members of this family. There are many additional aspects of Bantu tonal structure that merit discussion; but our purpose in this chapter has been to provide a sufficiently full introduction to Optimal Domains Theory to lay the groundwork for our discussion of the tonal pattern of Isixhosa. We now turn to Isixhosa.
23
CHAPTER TWO THE LONG FORM OF THE PRESENT TENSE 2.0. Introduction. We will initiate our study of the Isixhosa tonal system by examining the tonal pattern of the (long form of the) present tense. The morphological structure of this tense is fairly straightforward. Then is an obligatory subject marker at the beginning of the form, followed by a prefix that has the underlying phonological shape ya (this prefix cannot readily be identified as the "present tense marker" since it does not occur in the so-called "short form'' of the present tense dealt with in Chapter Three), followed by the verb stem, which always ends in the "final vowel" a in this tense. An object prefix may be intercakated between ya and the verb stem. The verb stem consists of a verb root plus one or more verbal "extensions"; these extensions, while extending the length of the verb stem, do not have any direct effect on the tonal pattern of the verb (Le. they fall into the category of "toneless" elements in the system; this notion is developed immediately below). 2.1. Toneless verb stems. It is by now fairly standard practice in Bantu studies to view Bantu languages as revolving around a fundamental contrast between High tone and the absence of High tone (tonelessness), rather than a contrast between High and Low tone. In the present study, we follow this standard practice without any attempt to argue the point. The Isixhosa data at every point support the standard view. There are two principal types of verb stems in Isixhosa: (a) toneless verb stems and (b) High verb stems. The characteristic feature of the toneless type is that they do not contribute a High tone to the input representation. The High verbs do. Since the toneless type is the simplest type to describe, we will begin our study with them. We begin by examining cases involving toneless verb stems in the present tense with first and second person subject prefixes (rjdi 'I'. U 'you (sg.)\ sj 'we', nj 'you (pL)'). These subject prefixes are the only ones in Isixhosa which do not contribute a H tone to the underlying representation. (The significance of the underscoring of certain consonants and vowels in our examples will be explicated later.) Examples in this study (if not followed by ...) are transcribed as though they constitute a complete utterance. In particular, our transcriptions reflect the automatic lengthening that affects the penultimate syllable of a word that stands in utterance-final position. We indicate a lengthened syllable by doubling the (vocalic) nucleus of that syllable. Forms which are specific to the speech of Mbulelo Jokweni will be labeled [M], while those specific to his wife Koleka will be marked [K], Forms left unmarked can be assumed to be equally valid for both varieties. Examine the examples in (1): 0)
ndiyacaciisa 'I am explaining' rjdjyaxoleela 'I forgive' rjdiyachukuirdisa 'I am provoking' rtdiyahlambeela 'I am washing for' 24
ndiyaffoniisa 1 am dancing' ndiyajjkeela 'I am turning aside' odiyalondglooga 'I am keeping safely' ndiyerxJcc'a (from: /ndi-ya-endela/) 'I am marrying at' ndiyandlflflla (from: /ndi-ya-andlala/) 'I am making the bed' s.iyalindeela 'we are waiting for' s.iyayumeela 'we are permitting' sjyamoneela 'we are jealous' s.iyalunfliisa 'we are putting right' s.iyagu£U\j1a 'we are scraping off s.iyaflibiseela 'we are throwing' s.iyaflp.et»b80qeet»h.3 'we are hitting repeatedly' sjyabaliseela 'we are narrating to' njyabaliisa 'you pi. are narrating' njyashukumiisa 'you pL are shaking' niyanamatheliisa 'you pL are cementing' niyakhohlakaleela 'you pi. are being cruel for' niyaqononoodiisa 'you pi. are making clear, emphasizing' niyabanekiseela 'you pi. are illuminating for1 There are no surface High tones in these forms, which suggests that there are no underlying High tones. If there are no underlying High tones, then it is reasonable to conclude that not only are the verb stems and subject prefixes toneless, but the prefix ya is also toneless. There are two different positions available as to what tonal information is present in the phonetic representation of the examples in (1). One position is that only High tones are represented in phonetic representation (just as only High tones are represented in the underlying representation). Syllables that are not High simply have no tonal specification. Languagespecific principles of phonetic implementation then determine exactly what the pitch shape of a sentences is, and in so doing they specify what the realization of any given toneless syllables is. The second position is that syllables that are not specified with a High tone in the output are specified as Low. The principles of phonetic implementation then determine the actual phonetic realization of both High and Low syllables. We shall assume the former approach, though admittedly largely ignoring any analysis of the phonetic implementation. Since the examples in (1) have no High tone in underlying representation, DomCor (H) does not necessitate the appearance of a High Domain in the optimal output candidate. Consequently, the constraint *Struc will reject all candidate outputs where Gen has provided a HD. In order for structure provided by Gen to be optimal, it must be needed in order to satisfy some constraint that is more highly ranked than *Struc. There is no constraint in Isixhosa that is ranked high enough to require, in the case of (1), a HD to be present in the output. So far we have cited examples of trisyllabic and longer verb stems of the toneless class in the first and second person forms of the long present tense. Substituting a third person subject prefix (e.g. bfi [cL 2], ^J [cl. 10]) provides some complications ("[cl. 2]" and "[cl. 10]" refer to
25
the traditional numbering of Bantu noun classes and the morphemes that "agree" with these noun classes). bayac&ciisa bayaltirjsijsa bayamoneela bayaxoieela baye^deeja bayindjaala zjyahtombeela zjyabiliisa bayagjbfseela bayachuktimiisa zjyabaliseela juyashukrimiisa zjyanamatheliisa bayaqonontfrjdjjsa bayakhohlaktfleela bayabanekiseela
(explain) (repair) (be jealous) (forgive) (marry at) (make the bed) (wash for) (narrate) (throw) (provoke) (narrate to) (shake) (cement) (emphasize) (be cruel for) (illuminate for)
Notice that these forms do have a High tone. We saw above that the ya prefix is toneless and that these particular verb stems are toneless. It follows that the source of the High tone in the words in (2) must be the third person prefix (ba, zj). However, even though the third person subject prefix is the underlying source of the High tone in these words, it does not bear the surface High tone - rather the antepenult syllable bears the High tone. Before entering into the development of an analysis of the data in (2), we must elaborate a bit on the phonetic pattern of these examples. We have transcribed the antepenult syllable in these words with a High tone. This transcription reflects what seems by far the clearest aspect of the interpretation of these data: from a phonetic point of view, the pitch peaks on the antepenult syllable. This syllable has, in the literature, always been regarded as High-toned. We believe this to be correct. The phonetics of the penult syllable requires comment This syllable is lengthened when the verb stands in the appropriate position in the sentence. In many of the above examples, this lengthened penult syllable will display a very clear falling pitch. One might very well have transcribed such words as follows: (3)
bayacAcfisa bayax6le*ela etc.
We have chosen not to do so. This railing pitch is entirely predictable from the High tone on the antepenult syllable: Le. a lengthened penult is predictably falling in pitch after a preceding High tone. Furthermore, there are true falling tones on penultimate syllables in the language which may in turn be preceded by a High-toned syllable (cf. Chapter Four, for example), and these falling-toned syllables contrast phonetically with the type of descent in pitch found in the present
26
examples. We therefore relegate the sort of descent in pitch observed in the examples in (2) to the phonetic implementation component of the grammar. Actually, there is a restriction on this phonetic falling pitch that merits comment. No fall appears when the penult has an onset consonant that belongs to the set of consonants commonly known in the Bantu literature as "depressor" consonants. In our transcription, we underscore these consonants throughout this study; since the nucleus of syllables with depressor onsets is depressed, we have also underscored the nucleus. Some examples: (4)
bayend.e.fila [*bayendfola] bayandJaala [*bayandjaala] bayahWirjljeela [*bayahla'mt}|$e>] bayaqonon6ndi.isa[*bayaqonon
*!
1 1
M)
*(!) *
ba(»; '
ie~
' %*
Nonfinal
.
Align Word BAR R yacuciis a" vacacii sa yaca cu iJ ca en sal ¥S vacK.il 1 sa Cliia ywa 4
••
•
•
At this juncture in our development of the constraint system, DomCor (H), Incorporate (H-sponsor), Avoid Prominence and BAL can all be considered undominated. Candidate (g) in (13), bajacadisa (which violates DomCor (H) and Incorporate (H-sponsor), and candidate (f), bg(yaca)ciisa (which violates Incorporate (H-sponsor) as well as BAL), are immediately rejected as nonoptimal due to their multiple violations of undominated constraints.. Candidates (a)-(e) are equal with respect to three of the undominated constraints - namely, DomCor (H), Incorporate (H-sponsor), and BAL. However, candidate (a), where all the syllables in the word following ba. are gathered together into the HD, possibly violates Avoid Prominence and certainly violates Nonfinality (HD) and is thus nonoptimal. Candidate (b), where all but the last syllable is included in the HD, violates the undominated constraint Avoid Prominence and is thus rejected as non-optimal. Candidates (c)-(e) all satisfy the undominated constraints as well as Nonfinality (HD). All three of these candidates do violate Align Word R. Nevertheless, they differ in the extent to which they violate Align Word R. The Right edge of the Word in the optimal representation (c) is misaligned with the HD by virtue of the two syllables cil sa, whereas candidates (d) and (e) are misaligned by three (ca cil sa) or four (ya ca cil sa) syllables. Consequently, candidates (d) and (e) fail to be optimal It is, of course, totally irrelevant that the last candidate is the only one to satisfy the lowest-ranked constraint, BAR. Since Align Word R prefers (bgyaca)dlsa to *(bg)yacacilsa, it is immaterial whether a lower ranked constraint is violated by the former and satisfied by die latter. We have now postulated a set of constraints that will evaluate as optimal an output candidate where a HD extends from the sponsor through the antepenult. Recall that in ODT, a constraint Express (H) requires the feature H to be realized on each element in the HD. If Express (H) is undominated, then it will be the case that each element in the domain is specified as High-toned. There is, however, in Isixhosa a constraint that dominates Express (H) - namely, the same constraint that we saw operating in Kijita and Chidigo in Chapter One: (14)
*(H,nonhead)
(We propose that HD's in Isixhosa, as in Kijita and Chidigo, must be Right-headed.)
31
Given that *(H,nonhead) dominates Express (H), it will be better if nonheads are not specified High than for them to be specified High in accordance with Express (H). If Express (H) were ranked above Align Word R, then it would be better to have domains where all the anchors are specified with a H tone than to have widescope domains. In other words. Express (H) would be better satisfied by *(bA)yacadlsa than by the correct (bgyac6)disa. Thus it is necessary that Align Word R be ranked higher than Express (H). We review again the constraint system we have motivated. (15)
DomCor (H), Incorporate (H-sponsor), BAL, Syllable Alignment,Avoid Prominence, *(H, nonhead) - undonrinated *Struc - dominated by DomCor (H) and/or Incorporate (H-sponsor) Nonfinality (HD) - dominated by Incorporate (H-sponsor) Align Word R - dominated by Nonfinality (HD), Avoid Prominence BAR - dominated by Align Word R Express (H) - dominated by Align Word R and *(H, non-head)
The following tableau illustrates how, given the domain structure selected as optimal (see (13)), the correct surface tone shape is predicted by the two constraints involving expression. (16)
Candidates -» (bayaca*)ciisa (bayaca)ciisa (bayicOciisa (biyac£)ciisa (b£yaca)ciisa
*(H, nonhead)
Express (H) *•
*!
v . 1*1 1
* • •
"
*! •
••i
i
* -
nu beenZH
•
*
n&i
becnza
No Adjacent Edges dominates *MonoHD, as shown above. Transitivity of constraint ranking makes the following prediction: given that NAE dominates •MonoHD, and given that •MonoHD dominates Align Word R, then NAE also dominates Align Word R. This ranking is critical, of course, only if Align Word R is formulated so as to demand that all HD's try to align with the R edge of a word. While the tableau in (42) omits any reference to •Overlapping, we should note that this constraint must dominate NAE - otherwise it would be optimal to avoid adjacent domains by creating an overlapping domain structure. We have not yet encountered a situation where there are adjacent High-sponsors and thus we have no basis for determining the interaction of No Adjacent Edges with DomCor (H) and 41
Incorporate (H-sponsor). Adjacent sponsors will arise later, so the matter of the interaction of No Adjacent Edges (and the other OCP-constraint cited in Chapter One, No Adjacent Sponsors) with DomCor (H) and Incorporate (H-sponsor) will eventually have to be determined. The constraint system as it now stands: (43)
DomCor (H), Incorporate (H-sponsor), BAL, Syllable Alignment, *(H, nonhead), •Overlapping - undominated •Struc - dominated by DomCor (H) and/or Incorporate (H-sponsor) Nonfinality (HD) - dominated by Incorporate (H-sponsor) No Adjacent Edges -dominated by "Overlapping *MonoHD - dominated by Nonfinality (HD), No Adjacent Edges Avoid Prominence - dominated by Syllable Alignment, *MonoHD Align Word R - dominated by No Adjacent Edges, Nonfinality (HD), Avoid Prominence BAR - dominated by Align Word R Express (H) ~ dominated by Align Word R, *MonoHD and *(H, non-head)
At this juncture let us go back to an issue raised at the end of our discussion of toneless verb stems in the long present. We discussed there the possibility of claiming that the HD consists just of the syllable that expresses a High tone. This sort of analysis would necessitate Express (H) being ranked above Incorporate (H-sponsor), and we argued that this ranking cannot be maintained (given transitivity of ranking). But suppose that there were some analysis whereby one could claim that surface structure domains were in fact limited to the syllables that manifest the High tone (call this the WYSIWYG analysis - i.e. the "what you see is what you get" analysis). Let us consider what such an analysis would have to say about the data from High verb stems -- specifically, the data involving a High-toned subject prefix. WYSIWYG would have to explain why the subject prefix High does not surface on the syllable /ya/ in a form like bdyanyinyith-'lkiisa- Our analysis invokes No Adjacent Edges to account for this phenomenon. In WYSIWYG, the domain in the verb would be around the syllable /the/ Consequently, the syllable /ya/ would not be next to another domain. No Adjacent Edges would not be relevant. So what constraint must the WYSIWYG analysis invoke? The constraint would have to appeal to the fact that the stem-initial syllable /nyi/ sponsors a High tone in the input. Thus the following constraint would have to be involved: (44)
*[H-mora in output] [mora in output whose correspondent in the input is High-toned]
This constraint would replace the No Adjacent Edges constraint that we invoked. Notice that this constraint introduces reference to the input structure into this member of the OCP constraint family. Our analysis, through domain structure, does not necessitate reference to the input. Whether this result of our analysis is decisive in its support depends of course on the ultimate determination of the answer to the question: which constraints have access to the input and how? But WYSIWYG is not yet home free even assuming (44). It must also explain why the High tone does not surface on say the second nyi syllable in WyanyinyitbJelciisa. Our analysis explained this in terms of "Overlapping. But that analysis is not available to WYSIWYG. To effect the right results, WYSIWYG would have to propose something like the following: (45)
given an input a (which is a H-sponsor) located to the Left of b (which is a H-sponsor), an output mora bearing a High tone corresponding to the High tone on a may not have b 42
as its input correspondent nor may it have any mora located to the Right of b as its input correspondent (45) would prevent the High tone of the subject prefix from appearing on the first vowel of the verb stem (since this vowel's underlying correspondent is H-toned) or any vowel to the Right of the initial vowel of the verb stem (since these vowels have input correspondents to the Right of the input correspondent of the initial vowel of the verb stem). (45) depends on a theory where correspondence of features can be invoked independently of segments (contra the McCarthy and Prince (1995) view of Correspondence Theory). But regardless of whether this is theoretically desirable, (45) seems to be making a strikingly wrong prediction. For example, (45) predicts that if an epenthetic vowel (remember, such vowels have no input correspondent) appeared in a verb stem before the antepenult High tone, then the subject prefix H could indeed appear on it since there would be no violation of (45). Why would there be no violation of (45)? Because (45) bars a H only on vowels which have a correspondent in the input! We believe that this is a ludicrous claim and thus (45) cannot be correct. WYSIWYG appears to us to end in a quagmire while offering no advantages to the analysis that we have developed (other than a more transparent relationship between "phonetic representation" and overt sound than our analysis offers). We will thus leave WYSIWYG to others who might find it attractive to see whether it can leap out of the quagmire and offer real insights. The ODT account of Isixhosa sketched above is, in our opinion, a relatively elegant analysis where there are various faithfulness constraints that refer to the input structure and other phonological constraints (such as No Adjacent Edges, *MonoHD, Nonfinality (HD), Avoid Prominence, Align Word R, *(H, non-heads)) which do not. We shall assume this analysis throughout the remainder of this study. We shall also provide considerable positive evidence in support of domains where only the final mora bears High tone. Let us now return to the Isixhosa data. All our examples of High verb stems have been consonant-initiaL Vowel-initial verbs are somewhat intractable in that what would appear to be the most elegant solution to a portion of the data turns out not to extend neatly to the full range of data. The full range of data seems to necessitate a certain degree of stipulation. In the hope that future research may clarify the issues, we will discuss the full range of data from the dialects we have studied, and discuss possible interpretations of these data, while admitting that in the end there is much uncertainty about the proper analysis. Recall that the vowel of the syllable /ya/ is not realized when it stands in front of a vowel-initial verb stem, /ya/ and the initial vowel of the verb stem merge into a single syllable. We begin our discussion with forms where the subject prefix is toneless and the verb stem has four vowels. (46)
njyonw4biisa ndjyome1ee?a ajyakhe'laana rjdjyalilphaala (idjyomlleela (|djyal4thiisa sjyohlukaana njyahlukiniisa
(make happy) (strengthen) (build for one another) (break down) (be strong) (cause to see where pointing) (part, differ) (disunite)
43
If these data were the only data, there would be nothing to be explained: assuming that the HD starts with the High-sponsor, we would expect the domain to extend to the antepenult syllable, resulting in the observed pronunciations: e.g. ni(yonw£)biisa. There is however a second pronunciation available for the above items, one in which there is a High tone on the penultimate syllable: (47)
rjdjyomelWga., ndjyalathtfsa, siyohluk&na, etc.
These variant forms require explanation. This variation seems to be connected to a phenomenon attested in other Nguni languages (see Downing (1995) for discussion of Siswati data) whereby stem-initial vowels are disregarded (not "counted") by the phonology. The data in (47) seem to reflect a HD from which the initial vowel is excluded, resulting in structures like niyoXnwabfQsa (similar to consonant-initial cases like nfyafbonfQsa). If it is correct to interpret the data in (47) as excluding the initial vowel of the verb stem from the HD, then a constraint is required that outranks BAL and Incorporate (H-sponsor) because these two constraints prefer HD's whose left edge is aligned with the sponsor, Le. the initial vowel of the verb stem. What could that constraint be? One possibility is a constraint which says that the L edge of a HD may not be aligned with the L edge of a vowel-initial verb stem. (48)
Noninitiality (HD) The Leftmost mora in a HD may not be initial in a verb stem.
This constraint is very stipulative in nature, since it bans alignment of the HD with an entity (=*vowel-initial verb stem) that does not have any particular status in prosodic or morphological structure. In particular, (48) does not take advantage of the major distinguishing factor between consonant-initial and vowel-initial verb stems: when the stem starts with a consonant, that consonant is the Left edge of a syllable; when the stem starts with a vowel, that vowel must join with a preceding prefixal consonant to form a syllable. An alternative analysis would invoke the notion of a prosodic stem, distinct from the morphological notion stem (parallel to the distinction between prosodic word and word). Given the notion prosodic stem, we could propose a constraint requiring the prosodic stem to be aligned with the Left edge of a syllable. (49)
Align the L edge of a Prosodic Stem with the L edge of a Syllable.
Assuming that (49) dominates a general constraint (SO) aligning the edges of a Prosodic Stem with the edges of a Morphological Stem, we could get the Prosodic Stem to omit the initial vowel of all vowel-initial stems? (50)
Align the edges (L,R) of a Prosodic Stem with the edges (L,R) of a Morphological Stem.
5
(49) potentially might be satisfied by having the vowel-initial verb stem "annex" the preceding prefix consonant to fonn a Prosodic Stem rather than by omitting the Initial vowel. The reason that there is "contraction" rather than "expansion" of the Morphological Stem conceivably could be due to a requirement that all elements in a Prosodic Stem must also be in a Morphological Stem. We will, however, leave this matter for future research. 44
To complete this particular story, we would require a means of getting the prosodic stem to affect the construction of HD's. The constraint may be as simple as (51): (51)
Align the Left edge of a HD with the Left edge of a (Verbal) Prosodic Stem.
(HD's in nouns may be located internally to prosodic stems, whereas this does not happen in the case of verbs in the long form of the present tense. Verb tenses dealt with later introduce additional complexity.) Another possibility has to do with the predictable location of the sponsor in High verb stems: the sponsor is always the first mora. But perhaps it is not the Morphological Stem that is relevant but rather the Prosodic Stem. In other words, perhaps the Prosodic Stem can play a role in how the High tone gets located in the input to the phonology. If so, then there is no need to stipulate that the HD aligns with the Prosodic Stem as in (48). Basic Alignment forces a HD to align with the sponsor, and it the role of the Prosodic Stem is to determine what mora sponsors High tone. In the following discussion, we will assume an analysis where the Prosodic Stem functions to locate the sponsor. However, nothing about the subsequent discussion ultimately rests on the correctness of this assumption. Given this assumption, the data in (47) may be analyzed as instances where (49) yields a Prosodic Stem which excludes the initial vowel, with the consequence that the High-sponsor is the second vowel of the Morphological Stem and not the first. Recall, however, that (47) represents just one variant; the pronunciations in (46) are also possible. How are we to understand this variation? The answer that immediately suggests itself is that while (49) may dominate (50), in which case the initial vowel of the Morphological Stem is outside the Prosodic Stem, it is (optionally) possible for (50) to dominate (49), in which case the initial vowel of the Morphological Stem is included in the Prosodk Stem. The idea that variability in the optimality of output candidates is nothing more than variability in constraint ranking is an attractive one. However, evidence from forms with High subject prefixes will cast serious doubt on the viability of taking the data in (46) as representing a case where the first vowel of the morphological stem is the sponsor. The data will suggest that in both (46) and (47), the constraint (49) is obeyed: the Prosodic Stem always is aligned with a syllable and the initial vowel is always excluded from the Prosodic Stem If this is the case, then some other account of the variation between (46) and (47) is required. From a purely descriptive point of view, the variation between (46) and (47) could be viewed as a variation between whether *MonoHD is respected (as in (47)) or not (as in (46)). This description of the data is available only on the assumption that in both cases the initial vowel of the verb stem is not the High-sponsor (a point of view that we indicated above will be supported by later data). But why should there be variation between whether *MonoHD is respected or not? There is, of course, another case where *MonoHD is not respected. Recall the variation in K's dialect between trisyllabic stems like /bfintlsa/ and stems like /billiisa/. This variation has its origins in a vowel-length contrast that originally existed in Nguni (the stem in /bo.niisa/ had a short first vowel, while the stem in /billiisa/ had a long first vowel). It is not clear that there is any connection between these examples from K and the variation in (46) and (47). Nevertheless, the only descriptive account of (46) and (47) that we can suggest is that •MonoHD is optionally upheld in the case of vowel-initial High verb stems. (We have seen that vowel-initial toneless verb stems behave normally with respect to *MonoHD.) When a stem is longer than in (46) — e.g. n.jyonwablsaana the Prosodic Stem (which excludes the initial vowel of the Morphological Stem) will be at least four syllables and will
45
behave Just like four syllable or longer consonant-initial High verb stems and have a HD that extends to the antepenult syllable. They pose no problem. Let us look now at trisyllabic vowel-initial High verb stems with a toneless subject prefix: (52)
njyoyifka ndjyahltfula ndjyalfia'tha ndjyoyltsa rjdjyaphtMla ndjyonwda'ba
(fear) (separate) (point at) (defeat) (break) (be happy)
If we adopt an analysis where (49) is always obeyed (i.e. the Prosodic Stem is always aligned with a syllable and does not include the initial vowel), then the examples in (52) have a bisyllabk Prosodic Stem. The HD will necessarily consist just of the sponsor (the second vowel of the Morphological Stem) due to the high ranking of Nonfinality (HD): Qdjya(hl]}tf)la. If we adopt an analysis where (49) is variably ordered with (SO), either ranking of those constraints will lead to the same pronunciation: n,djya(hlj}il)la if (49) dominates (50), n,dj(ya.hl\lii)la if (50) dominates (49). In the former case, as noted above, Nonfinality (HD) prevents the domain from extending to the final vowel; in the latter case, *MonoHD demands that the domain extend to the penult, but Nonfinality (HD) prevents it from extending further. The data in (52) thus do not help us in deciding between a variable ranking of (49) and (50) versus an absolute ranking of (49) over (50). Bisytlabic vowel-initial stems are particularly interesting. Once again we find that there are two pronunciations possible, but these two pronunciations do not seem to be in free variation: our consultants prefer one pronunciation for certain lexical items, the other pronunciation for other lexical items, and either pronunciation for yet other lexical items. Examination of the published Isixhosa literature reveals similar variation in other dialects. One pronunciation type is shown in (53). (S3)
njyaakhfi ndjyeemfi
(build) (stand) (fear) (spoil)
ndiyeeyfl
ndiyooni
The second pronunciation type is sho (54)
rjdjyodna sjydVSma ndjy&ba rjdjy«si ndjyfidkha
(spoil) (dry) (steal) (know) (build)
In an analysis where (49) is variably ranked with respect to (50), the above data are very straightforwardly accounted for. The lexical items in (53) would be specified as having (49) dominate (50). As a consequence, the Prosodic Stem would consist just of the last syllable of the 46
Morphological Stem. The final vowel would therefore be the sponsor of a High tone. The result would be a HD that includes just the last vowel of the word (violating Nonfinality (HD) in order to be faithful to the underlying High). The lexical items in (54), on the other hand, would be specified as requiring the ranking of (SO) above (49). Consequently, the initial vowel of the verb stem would be Initial in the Prosodic Stem and would be the High-sponsor. Nonfinality (HD) prevents the domain from extending through the final vowel. If we assume an invariable ranking of (49) above (50), then the data in (S3) are entirely expected. However, to account for the data in (54), we would need to invoke some constraint C that (for the relevant lexical items) outranks (49), a constraint that would have the consequence that (49) would be violated in order to better serve it. Recall that (49) and (50) are constraints on the formation of a Prosodic Stem, and that the Prosodic Stem plays a role in determining the location of the High-sponsor. It would appear that C could be a member of the Nonfinality family of constraints, one that prefers a sponsor not to be word-final. If Nonfinality (Sponsor) is dominated by (49) for the items in (53), then the Prosodic Stem will align with a syllable and the sponsor will be word-final. If, however, Nonfinality (Sponsor) dominates (49) in the case of the lexical items in (54), then it will be better to not have a word-final sponsor even if this means that the Prosodic Stem will not be aligned with a syllable. Given this explanation for the data in (53) and (54), these data would not help to determine whether (49) and (50) are variably ranked or whether (49) is invariably ranked above (50): the correct results can be derived under either analysis. As indicated earlier, forms where a High-toned subject prefix is used in conjunction with a vowel-initial High verb stem provide significant evidence on the ultimate analysis of these stems. Let us turn our attention now to these forms. We begin with cases where the Morphological Stem has four vowels. There are two pronunciations available. (55)
Ifyonwl&biisa b£yom!61eeza. biyakh!6Iaana brfyalluphaala b4yoml6leela bdyalla'thiisa bdyohllif kaana lfyahlukldniisa
(make happy) (strengthen) (build for one another) (breakdown) (be strong) (cause to see where pointing) (part, differ) (disunite)
as well as the alternative forms: ba'yomel!6e'za, tayalathliisa, bdyohluklia'na, etc If we accept the analysis that (49) is invariably ranked above (50), then the surface tone pattern simply shows the effects of No Adjacent Edges (which will keep the HD corresponding to the subject prefix High from extending onto the initial vowel of the Morphological Stem) combined with variation as to whether the stem HD is required to obey *MonoHD or not. If, on the other hand, we appeal to variable ranking of (49) and (50), we encounter a problem — though the problem only emerges clearly when additional principles of Isixhosa tonology are understood. We shall see later that when two adjacent HD's might be expected, one of two things may occur: (a) the adjacent domains may both appear in the output, but all of the syllables in the second domain will be pronounced High on the surface (and not just the last syllable of the domain), setting aside disrupting effects caused by depressor consonants; or (b) all 47
of the syllables that would be in the two domains will appear inside a single domain (in which case only the final syllable will realize a High tone on the surface). Now, notice that if (SO) dominates (49), then we would at first glance expect a domain structure where there are adjacent domains: (bg)(ysnwa)bilsa. This structure is, however, incompatible with the observed pronunciation — namely, a High on the heads of the two HD's. As stated in the above paragraph, such structures in the language actually produce *bdyt)nw
(g)a# - optimal (aa)# a(ft)#
we need Nonfinality (HD) to be more highly ranked than DomCor (H) and Incorporate (Hsponsor) so that (133a), which docs not violate Nonfinality (HD) but does violate both DomCor (H) and Incorporate (H-sponsor), can be more highly valued than (133b), which violates DomCor (H) once, and (133c), which violates both DomCor (H) and Incorporate (H-sponsor) once. So we have a major dilemma. Nonfinality (HD) must be ranked above Incorporate (Hsponsor) in order to explain the selection of (&)a# as optimal, but it must be ranked below Incorporate (H-sponsor) to explain the optimality of a(g)#. We see no way out of this dilemma without exploding Nonfinality (HD) into two separate constraints: (134) (a)
(b)
Nonfinality (polysyllabic HD) *(...oa)# Nonfinality (HD) *)#
81
The flrat member of the Nonfinality family, (134a), says that is nonoptimal for a polysyllabic domain to be aligned with the Right edge of the word. The second member, (134b), is more general and disfavors any domain at the Right edge of the word. Given this recognition of two constraints, we can locate (134a) above Incorporate (Hsponsor) and (134b) below Incorporate (H-sponsor). This ranking will allow us to select all the appropriate outputs. We demonstrate this below with sample tableaux for several critical cases. First consider the case of njyawdtftya.
-> niya(waa)tya |
[
|^
|
[^
In this tableau we see that both members of the Nonfinality constraint family play a role in choosing njyawMtya as optimal The constraint on polysyllabic domains rules out the fused domain in nlya(waatyg). The more general constraint favors niya(waa)tya over niyawaa(tya). Notice, incidentally, that there is no evidence yet of the relative ranking of (134b) and DomCor. In Chapter Three we will find evidence that (134b) must dominate DomCor (H). Consider next the derivation of njyaaty 1 (136) Candidates niyaatya ni(yaa)tya
No Edges
Adj (134a)
Incorporate (134b)
•I
§^Pl£iij£iiSlij&&
•1
*
-»niyaa(tya)
DomCor
BR
The highly ranked constraint on polysyllabic word-final domains has no relevance to this case, where the sponsor is the last mora of the word. The relatively lowly ranked general Nonfinality constraint has no effect since all the alternative candidates are rejected by the higher ranked Incorporate (H-sponsor). Ranking either (134a) or (134b) above *MonoHD and Align Word R will account for the fact that widescope domains cannot be extended to include word-final syllables. We will assume that it is (134a) that constrains the extension of a domain onto the final syllable. Is it motivated, on cross-linguistic grounds, to see Nonfinality as a constraint family that includes both (134a) and (134b)? We are not aware of any clear supporting evidence, so consider it worthwhile to consider alternatives. There is one alternative to the explosion of Nonfinality that is certainly worth considering. Borrowing from Odden's (1982) analysis of similar data from Shambaa, we might propose the following constraint: (137) ^Stressed Non-heads Non-head syllables in a HO may not be stressed. (137) would reject niya(waatyg) not because it has a word-final polysyllabic HD but rather because it has a nonhead syllable that is stressed (prominent). We leave open the issue of 82
whether 'Stressed Non-heads or an exploded Nonfinality constraint is responsible for the rejection of niya(woatyg), but come back to the issue later. Below we give the set of constraints governing HD structure and the realization of High tone that we have so far developed. (We do not repeat here the RO-structure constraints discussed above.) (138) Constraint System A: Syllable Alignment, Express (H, head of HD), 'Overlapping, Nonfinality (polysyllabic domain) or 'Stressed Non-heads, BAL, Plateau—undominated *(H, non-head) - dominated by Express (H, head of HD), Plateau Incorporate (H-sponsor) — dominated by Nonfinality (polysyllabic domain) or 'Stressed Non-heads Uniqueness — dominated by Incorporate (H-sponsor) No Adjacent Edges -dominated by 'Overlapping No Adjacent Sponsors — (it is not yet possible to determine whether No Adjacent Edges or No Adjacent Sponsors dominates DomCor (H) and thus results in "fusion"; indeed, as yet there is no evidence that No Adjacent Sponsors plays a critical role in the system, while there is evidence that No Adjacent Edges must dominate 'MonoHD and Align Word R) - dominated by 'Overlapping DomCor (H) - dominated by No Adjacent Edges (alternatively No Adjacent Sponsor); later we will see that Incorporate (H-sponsor) also dominates DomCor (H) 'Struc - dominated by DomCor (H) and/or Incorporate (H-sponsor) Nonfinality (HD) - dominated by Incorporate (H-sponsor) 'MonoHD - dominated by Nonfinality (polysyllabic HD), No Adjacent Edges Avoid Prominence - dominated by Syllable Alignment, 'MonoHD Align Word R ~ dominated by No Adjacent Edges, Nonfinality (polysyllabic HD), Avoid Prominence BAR - dominated by Align Word R Express (H) -- dominated by 'MonoHD, Align Word R, *(H, non-head) Constraint System B: (71a) - undominated Nonfinality (HD) - dominated by (71a) (73) - dominated by (71a) *(H, Depressed Syllable) (=(70)) - dominated by Nonfinality (HD) No Adjacent Edges — dominated by (70) Syllable Alignment - dominated by (70) *(71b) - dominated by (71a), (70) (Recall that (71a-b) refers to the two aspects of Domain-Faithfulness: (71a) requires that elements incorporated into a HD in the output of A should be incorporated into the corresponding HD in the output of B; (71b) requires that elements that are not in a HD in the output of A should not be in a HD in the output of B. (73) bars depressed syllables from being inside a HD.) Let us return briefly now to the examples in (131). The pronunciations cited are certainly valid pronunciations in Isixhosa. But we should note that there is another pronunciation available: sjyaylitya. sjyaydakha etc. In the context of the present chapter, it is not really possible to understand this variation. We will simply note that in Chapter Four, we present
83
evidence that penultimate Falling tones on a lengthened vowel (which are not the result of da DIHTS phenomenon) are to be understood as an IP-final tonal pattern that correlates with I word-final High tone in medial position. Examples like sjyaylitya have an altemath* pronunciation sjyayiityd under certain circumstances (see Jokweni 1995), suggesting that thek analysis is entirely parallel to the analysis of the similar data discussed in Chapter Four. We have not yet illustrated the case where a depressed object prefix is placed in front of i High verb stem A depressed object prefix does not produce any difference from the case when the object prefix is non-depressed. The reason for this is simple: depression does not interfere in any way with "fusion" (Le. the placement of two or more H-sponsors into a single HD). 4 A few examples illustrate: (139) njyaglftya (on niyaaiitya) niyazibddna s.iyaabonffsa sjyazjbonfsiisa
(eat) (see) (show) (see clearly)
We repeat examples from (131) which show that fusion also occurs regardless of the presence of a depressed syllable in the stem: (140) sjyayigawUiila sjyawajjm^ela
(chop) (hide)
Depressed syllables both in the object prefix and the stem likewise do not interfere with fusion: (141) sjyagigflwiitfla sjyaaisjme'61a
(chop) (hide)
In order for depressed syllables to block fusion, there would have to be some high-ranked constraint (connected to depressed syllables) that would be better served by avoiding fusion. We have postulated a constraint that bars a depressed syllable inside a HD. But there is no way in which adjacent HD's rather than fused HD's will lead to a better satisfaction of this constraint. In any case, the ban on depressed syllables inside a HD is so lowly ranked in the A system that it has no effect on the formation of domains. We have now explained the fusion of object prefixes with High-toned verb stems. But we have only looked at such configurations when the subject prefix is toneless. In (142) below we show the forms that occur when there is a High-toned subject prefix: 4
It is perhaps worth noting that the need to expand a HD so as to avoid a depressed High tone does not trigger fusion. For example, in K we haveftiya'fumtta'naand In M we have Kiytffif mMiM but in neither dialect do we have *&|yar\imia*na, which would be the fused form. So, how do we explain the fact that depressed syllables do not trigger fusion In the B system? In principle, the DIHTS could lead to fusion. But it does not In IsUhosa. We can explain this by the high ranking of Domain-Faithfulness In the B system: we assume that Domain-Failhnilness (specifically, the aspect of Domain-Faithfulness that requires any mora In an input HD to be In the corresponding HD In the output) outranks No Adjacent Edges. In other words, It is more Important for the output of B to be faithful to the HD structure of the input to B than It is to avoid adjacent High Domain Edges in the output of B.
84
(142) Ilyawlditya (or: Hyaawta'atya) HyawabtWka llyabamteima bdyabablddna biyabayltftfsha llyawabluliisa [K] / liyawabullfisa [M] biyababon'iisa biyawabonltsiisa Hyawl&ba Ifyawt&Ekha Iiyawoylifka IlyonwWbiisa Hyahlukla'niisa
(eat) (place) (invite) (see) (wake up) (greet) (show) (see clearly) (steal) (build) (fear) (make happy) (differ from)
At expected, we see that No Adjacent Edges requires that the domain initiated by the subject prefix not extend onto the syllable in front of the object prefix. Turning to the case where the High subject prefix is depressed (and the object prefix not depressed): (143) ny&v/Aitya [M], giydwtfdtya [K] Xjyaw4b«ka [M], zjya>ab!6eka [K] %jy6wa,me'£ma [M], gjydwamle'e'ma [K] XJya>4biSMisa [M], jjyaVablilliisa [K] giydwibdnifsa [M], giyAwabonlffsa ZJydwdbdnlsiisa [M], zJyAwabonlfsiisa SJyaw«ba [M], zjyiwle^ba [K] Ziyiwdyiika [MJ, zjyiwoylilka [K] SJy4w6nwdbiisa [M], ziyiwonwldbiisa [K\
(eat) (place) (invite) (greet) (show) (see clearly) (steal) (fear) (make happy)
(143) shows, when there is a depressed subject prefix, its H will shift to the prefix /ya/. In the case of K, there will be downstep between the H tone on /ya/ and the head of the following HD (and there will be no protection of non-heads following /ya/). This all is due to the fact that in K there is a register domain separation between /ya/ and the object prefix. In M, on the other hand, the shifted H on /ya/ will not be separated from a following HD-head by downstep and there will be protection of following non-head syllables. This is due to the fact that in M the /ya/ and the following object prefix are in the same RD. However, when the stem has a depressed syllable, as in (144), that depressed syllable will initiate a RD due to Depressed Syllable Alignment. Cohesion prevents stem syllables in front of the depressed syllable from being incorporated into a preceding RD, while in M, the HD-mora Incorporation (Prefix) constraint will force the object prefix to be included in the preceding RD. Plateau will force the object prefix to appear with a High tone. In K, the object prefix is toneless. The difference is attributable to the fact that in K prefixes are not incorporated into a preceding RD. (But see Chapter Three for more discussion.)
85
(144) ay«ytflall<ela[M],iriy4yigal!e-elB[Kl zjyiwflgflwlilula [Ml, zjyAwagawlritfla [K] zjyawfthengUlsa [Ml, zjyiwithenglilsa [K] zjyAbdsebenzjsIlela [M], zjytfbaseberjzjsllela [K]
(pour) (chop) (sell) (cause to work for)
These examples in no way differ from the cases discussed above where a High-toned subject cooccurs with a High-toned object prefix and a toneless verb stem. The object prefixes in (143) and (144) were not depressed. If the object prefix is depressed, as in (145), then in both K and M the object prefix will not be in a RD with /ya/ due to the high ranking of Depressed Syllable Alignment, which (as reformulated) bars a depressed syllable from being included in the same RD as a preceding syllable. Since neither the object prefix nor any of the material following the object prefix can be in the same RD as the /ya/, then there must be a downstep between /ya/ and the head of the next HD and there can be no protection of following non-heads.
(145) wy^sUitya
ziytaib.'ddna zjydzjbonlifsa
jaytoioawiiNiia siydsithengl.Usa zjya'zjbonlisiisa
(eat) (see) (show) (chop) (sell) (see clearly)
zjydzjsebenzjsle'ela (make work for) We have now reviewed the tonal behavior of (High-toned) object prefixes in front of both High and toneless verb stems, with both High and toneless subject prefixes. This essentially completes our study of the long form of the present tense. The object-prefix data in every way substantiate the analysis developed earlier; their sole peculiar property is that in M they may be organized into a preceding RD, in violation of Cohesion, and thus surface as High due to Plateau. In Chapter Three we will see that it is not just object prefixes that have this character, but subject prefixes and noun class prefixes as welL In the next and concluding section of this chapter, we discuss a specific point about phonetic implementation. IS. A matter of phonetic interpretation. We have ignored a certain point up until now that needs to be made explicit. When non-heads are not realized as H due to the constraint (146) *(H. non-head) the fact of the matter is that they are realized at a relatively raised pitch level if they are not preceded by a depressed syllable. This is particularly noticeable in phrase-initial position when the non-head is immediately followed by the head. Thus in an example such as baytfliima from /bsyaliima/, the initial syllable may be quite raised in pilch. As the number of non-heads increase, it is perhaps more obvious that these syllables are lower than the head in pitch.
86
Given this phonetic raising of the non-heads, there is a phonetic contrast between bayacdciisa (where the subject prefix sponsors a H tone) and sjyabjileela [K] (where the •object prefix is toneless). The initial syllables in the former are raised in comparison with the latter. The underscoring of the vowel of the subject prefix in the latter case is meant to indicate the relative fewness of these syllables. AH that is meant by this underscoring is that the subject prefix in these cases has a lowering effect analogous to that triggered by depressed consonants. This marking does not represent a claim that these syllables are necessarily "depressed" in any other sense. (The "depressing" nature of the underscored subject prefixes has been recognized by various researchers dealing with Nguni languages; phonetic research is badly needed to determine whether there is anything other than pitch that would lead one to conclude that these syllables are in fact depressed.) The initial syllables in sjyabilleela are not part of a HD in contrast to the initial syllables in bayaclciisa. The former has the domain structure siya(bji)leela whereas the later has the domain structure (bgyaca)disa. We suggest that phonetic interpretation is sensitive to domain structure. Specifically, we suggest that toneless non-heads are relatively higher in pitch than toneless mora that arc not parsed into domains. Given this analysis, there is no need to consider the subject prefix in sjyabilleela to be in fact depressed. The details of the phonetic implementation require careful investigation. No raising of non-heads in a phrase-initial HD seems to occur when there is a depressed syllable in that HD. Furthermore, it is not clear that there is raising in HD's that are not phrase-initial.
87
CHAPTER THREE MORE VERBAL FORMS WITH LEXICAL TONE 3.0. Introduction. In Chapter Two, we examined the tonology of the long form of the present tense. This tense provided us whh an overview of many of the complexities of the Isixhosa tonal system. In this chapter we continue our exploration of the tonal pattern of tenses where the verb stem continues to display the lexical tone patterning noted in Chapter Two (Le. the obvious differentiation into toneless and High verb stems). 3.1. Short form of the present tense. In this section we will examine the tonology of the short form of the present tense. The short form of the present tense differs from the long form in lacking the prefix /ya/. As a consequence, in the short form, the subject prefix abuts directly the verb stem. Roughly speaking, the long form of the present tense is used when the verb is final in its clause; the short form is used when the verb is non-finaL. One point needs attention immediately. In our analysis up to this point, we have formulated Nonfinalhy to refer to the word. We have also formulated our widescope-driving alignment principle (e.g. Align Word R) to refer to the word. There are Bantu languages where HD's extend beyond the limits of the word, and where Nonfinalhy refers to the phrase rather than the word (more precisely, where Nonfinalhy (Phrase) is ranked highly and helps choose optimal outputs, while Nonfinalhy (Word) is lowly ranked and had no significant effect). In Isixhosa, however, we do not generally find domain structure crossing word boundaries. The word edge continues to be significant even when looking at phrases. This observation will be important since the short form of the present tense is never used in phrase-final position but only phrase-medially. We will first examine the shape of the verb as it would appear in phrase-initial position (and ignoring its interaction whh a following word). Later we will look at the interactions of the verb with a preceding word as well as with a following word. In the non-third person short present tense form of toneless verbs, there is of course no overt High tone present in the verb. (1)
n.djlwa.../5jlwa... nd(5nzn./sGnza.... ndjwa... /sjwa... ndipheka..7s.jphekB... ndjlima.../s;jlima... ndjhleka.../s,jhleka... ndjwisa..y§jwisa... ndjlwisa.../$jlwisa... ndjbalisa.../sjbalisa...
(fight) (do) (fall) (cook) (cultivate) (laugh) (drop) (cause to fight) (narrate) 88
fldjcacisa.../ sjcacisa... (explain) nd.jlunflisa.../«luriaisa» (repair) tldjkhohlakala.../ sjyakhohlakala...(be cruel) ndiflibisela.../ sjajbisela... (throw) ndjmangalela.../ sjmangalela... (accuse) Nothing needs to be said concerning the above forms. There is no mora that sponsors a H in the underlying representation, consequently these verbal forms lack any HD-structure. When there is a High-toned subject prefix, and the verb stem is monosyllabic, we find data as in (2): (2)
Mlwa... b4wa... Wrn.ti»... WnjCB...
(fight) (fall) (dig) (do)
These data show, as mentioned above, that Nonfinality (polysyllabic HD) in Isixhosa refers to the word rather than, say, the Intonational Phrase, since tbe HD initiated by the subject prefix cannot extend to the word-final syllable even though that syllable is phrase-medial. BisyUabic toneless stems used in conjunction with a High subject prefix appear in (3): baphgka... babala... baytfla... bagtfla... bahieka... bacula... bag&a... ballma...
(cook) (count) (close) (be sick) (laugh) (sing) (bemad) (cultivate)
In these data we see that the HD initiated by the subject prefix extends through the penult syllable. This could reflect the fact that *MonoHD dominates Avoid Prominence. However, since the penultimate syllable is not lengthened in the medial position, it is not apparent whether this extension of the HD through the penult is actually a violation of Avoid Prominence. If there is no violation of Avoid Prominence involved, then the extension of the domain is the optimal satisfaction of *MonoHD, Nonfinality, and Align Word R. The data involving bisyllabic verb stems based on monosyllabic roots is of some interest in connection with the issue of whether Avoid Prominence is at work in the short form of the present tense. (4) biwisa... or bawisa... (drop) lilwela... or ulw61a... (fight for) bdhlisa... or bahlisa... (lower) Recall that we have suggested that these verb stems are exceptions to *MonoHD. This exceptionality will explain the pronunciations like bdwisa... as long as Avoid Prominence is in 89
fact (at least optionally) in effect in these short forms of the present tense. If Avoid Prominence were not in effect, then it would not matter whether these stems were exceptions to *MonoHD since the constraint Align Word R would (in the absence of Avoid Prominence) require the HD to extend as far right as possible. The following tableau illustrates that if Avoid Prominence were inapplicable, due to the absence of prominence, we would generate bawflsa as the (only) optimal output.
(5) Candidates (bawisa)... -» (bawfysa... (bsjwisa...
Nonfinality
•MonoHD
Avoid Prom
^A%nWordR
[irrelevant]
sa
I* wi!sa
[*]
In order to generate bdwisa..., we need Avoid Prominence to be (optionally) applicable and to dominate Align Word R:
(6) Candidates (bawisa)... (bgwQsa... -»(ba)wisa...
Nonfinality
*MonoHD
!*
Yrwj^^nn^K^m %%"^?k&3^M?$£$%3^ | wisa
[*]
(Recall that we enclose in brackets violation marks that are inoperative due to the item in question being an exception to the constraint.) To conclude: we need Avoid Prominence to be relevant in the short present form in order to have the evaluation in (6), but we also need it to be optionally inapplicable (presumably due to the fact the penult syllable is no longer actually prominent in medial position) in order to have the evaluation in (5). Given the assumption that bawisa... reflects the surface lack of prominence on the penult and that bdwisa--- reflects an abstract prominence on the penult, then our analysis of verbs derived from monosyllabic roots as exceptions to *MonoHD yields the correct results. Turning to trisyllabic or longer stems, we find the HD initiated by the subject prefix extends to the antepenult (in the case where the penult is taken to be prominent) or to the penult (in the case where the verb is treated as lacking prominence and thus Avoid Prominence is irrelevant). (7)
babaMisa... or babalisa... baedcisa... or bacaclsa... baluflflisa... or balungisa... baxolela... or baxolila... banabulula... or banabuldla... bajacdjtela... or bajacujtffla... bahlehtezeja... or bahlehlez#la... baqonondrjdjsa... or baqononondjsa... baqhathanisela... or baqhathanisgla...
90
(narrate) (explain) (repair) (forgive) (stretch the body) (go about in rags) (retreat hastily) (emphasize) (be dishonest)
If the antepenult syllable is depressed, then we find the DIHTS phenomenon. There will then be only one pronunciation available: (8)
bayymela... badantsa... bamangal6la... oahlajjulula... baphajjamtsa... baphengultfla...
(permit) (dance) (accuse) (spread) (disturb) (search out)
There is only one optimal output due to the fact that two different evaluations yield the same output. Let us see why this is so. Take the case where the penult syllable is considered as prominent Then the HD structure will be (ba.vu)mela... as a consequence of the A system, but (bgvume)la... as a consequence of the B system. In other words, the High will be shifted to the penult in the final output. Now take the case where the penult syllable is not considered to be prominent. The HD structure will be (ba.vume)la... as a consequence of the A system, and the B system will not alter this in any way. The result will be a penult High tone. Thus it does not matter whether Avoid Prominence is applicable or not; the same form will result in either case. If the antepenult syllable is depressed and the penult syllable is also depressed, then we would expect two different pronunciations to be available - and this is what we find. (9)
bayagdduka... or bayagodtfka... bahlunfllizela... or bayahlungujgla...
(go home) (shake the head)
The reason for the two pronunciations should be clear. If Avoid Prominence is operative, then the output of the A system will be (bahlungu)zela... The depressor in the penult will, however, mean that even in the output of the B system the HD-structure will be the same. Thus we predict the form bahlurjgtf2g]a.. But if Avoid Prominence is inapplicable because die penult is not prominent, then the A system will yiekl (bn,hlungU2e)la... and the B system will not alter this in any fashion. Thus we predict the pronunciation bahlurigu^gla-Up until this point we have set aside examples where the subject prefix is depressed. Such examples are for the most part indistinguishable from the case where the subject prefix is not depressed, since the head of the HD is not the subject prefix in most cases. (10)
fcflwa...
(fight) (close) (be sick) (laugh) (sing) (drop) (make fight) (forgive) (get used to) (explain) (dance)
siytfia... zigtfla... KJhlgka... iiciila... jjwfsa... Sllwlsa... Kix61ela... or aixotela... SJqhilisa. '• or juqhelisa... jjcacisa... or jjyacacisa... aid»n1sa... 91
SJyumeia... Cjnabuiula... or zjnabulula... Ojactizela... or gjjacuftila... Zjhlehl6;eja... or zihlehleg£la... z.jmangaje'la... zjphaza.misa... Siqonon6ndisa— or siqononoridjsa... Uqhathanlsela... or sjqhathanis61a...
(permit) (stretch the body) (go about in rags) (retreat hastily) (accuse) (disturb) (emphasize) (be dishonest)
We see that in the one case where the head may be the subject prefix — namely, in the case of verbs like /wisa/ (cf. bdwisa—) - the DIHTS phenomenon is observed. As a consequence, ZJwlsa... is the only available pronunciation, in contrast to bdwisa-Zbawlsa... The most important thing that we have learned from the toneless verb stem in the short form of the present is that there is optionally with respect to whether the penult vowel of the verb Is considered to be prominent or not. The consequence of this is that we have alternative forms available, since when Avoid Prominence is not operative, Align Word R may enforce a wider domain than otherwise. We will see this same phenomenon in the High verb stems. In the case of the short present tense form of High verbs (with a toneless subject prefix), some discussion is required. If the verb is a monomoraic H verb, in phrase-medial position the H tone is not retained.
(11)
n.djtya..Vsitya... ndjdWsidla. ndikha... Odipha...
(eat) (cat) (draw water) (give)
How are we to account for the absence of the word-final High tone in these examples? Recall that at the end of Chapter Two, we proposed that Nonfinality (polysyllabic HD) is ranked higher than the other, more general version of this constraint, Nonfinality (HD). This ranking explained why certain final High tones could not be incorporated into a HD, if to incorporate them would have required a polysyllabic HD aligned with the R edge of the word. At the same time, it allowed those final Highs that could be incorporated into a monosyllabic wordfinal HD to be optimal. Now we see that in phrase-medial position, the final H observed in the long form n.diy&atya' is missing in the short form rjdttya... There are two possible explanations for this: either monosyllabic word-final domains are not allowed in phrase-medial position or their High is not allowed to be expressed. We believe that there is evidence that the domain must exist but the H is not expressed The relevant data involves words of the structure CVCVCV. In phrase-final position, such words would have die HD structure (CV) C W (CV) due to the fact that No Adjacent Edges outranks *MonoHD . But now consider the phrase-medial position. Assume that no domain structure were built over the word-final H-sponsor. Then No Adjacent Edges could not prevent the HD initiated by the first H-sponsor from extending to the penult Thus it would be predicted that the optimal domain structure would be (CVCV)CV. This, however, is incorrect A noun such as fmiityid 'straps,' when it is phrase-medial, is pronounced Imitya... and not *imltya— This medial pronunciation suggests that there must be a word-final
92
HD to explain the role that No Adjacent Edges plays in preventing the extension of the initial HD in tmitya... We conclude then that the data in (11) involves not the failure to parse the H-sponsor into a domain, but rather the failure of Express (H) to be satisfied — Le. the existence of some constraint that outranks Express (H). The relevant constraint is a form of Nonfinality, but one which is specifically about realization of High tone and not about domain structure. (12)
Nonfinality (H) The final mora in a representation may not bear H.
(12) is of course a constraint family, referring to either "word"- or "phrase"-final elements. In Isixhosa the "wonT-final version of this constraint is highly enough ranked to have impact. However, in Isixhosa, this constraint impacts only on word-final mors in medial position in the phrase. Word-final moras are High-toned when they are also phrase-final. To get at the difference in behavior between phrase-medial and phrase-final High tones in Isixhosa, we propose a solution similar to the one that was adopted in our account of "protection" in Chapter Two. We assume that Express (H) is a family of constraints. Specifically, while in general it is best to express the feature throughout the domain, it may be even better to express the feature in particular positions. Thus we earlier proposed that expressing High on heads was a separate, independently rankable constraint from the general Express (H) constraint. Now we suggest another member of the Express (H) family. (13)
Express (H, phrase-final mora)
If (13) dominates (12), while (12) dominates Express (H, head), then we will express H on phrase-final heads but not on phrase-medial heads. This account is adequate only if (13) is plausible — Le. if there is real cross-linguistic evidence that phrase-final position is somehow a "stronger" (more favored) bearer of High tone that phrase-medial position. The immediate problem with this claim is that there seems to be cross-linguistic evidence for phrase-final position to be associated with the lowest pitch level in the phrase! Thus (13) seems to be contrary to a better-motivated universal principle. However, contradictory constraints are not necessarily inconsistent with the OT model; so it may be that (13) can in the end be sustained. We do not, however, have any specific evidence of a cross-linguistic nature supporting a preference for High tone in phrase-final position in contrast to phrase-medial position. In the absence of such evidence, (13) represents a dubious move. However, since we have no better explanation, we will assume an undominated (13). Let us summarize, then, our system of realization constraints: (14)
Express (H, phrase-final mora) ~ undominated Nonfinality (Word-final H) - dominated by Express (H, phrase-final mora) Express (H, head) - dominated by Nonfinality (Word-final H) Plateau -- dominated by Express (H, head) *(H,non-head) - dominated by Plateau, Express (H, head) Express (H) - dominated by Nonfinality (H), *(H, non-head)
The tableaux in (IS) illustrate how the two different pronunciations (phrase-medial and phrasefinal) will be selected under this analysis. These tableaux are confined to the constraints concerning expression and assume the optimal domain structure:
93
(15)
fldiyaatya" Domain structure: ndiyaa (tyg)
rjdjtya... Domain structure: ndi (tyg)™ Candidates
Express (H, phrase- Nonflnality final) (Word-final H)
ndi(tya)... -> ndi (tya)».
j*
Express (NHLheodJt^
P^^^^^^^^^^ *
As noted above, there is evidence for a preference for IP-final moras to be low in tone. Thus it is not difficult to find languages with data that are just the opposite of Isixhosa. For example, in Mijikenda there are no word-final High tones in phrase-final position, but word-final High tones can be realized in phrase-medial position. Mijikenda could be analyzed as follows: Nonfinality (Phrase-final H) is active in the language and dominates the entire family of Express (H) constraints. The Nonfinality (Word-final H) constraint is inactive, being lowly. While the line of analysis that we have developed does allow us to account for both Isixhosa and Mijikenda, we have not attempted to explore what the exact range of systems is that our analysis predicts as possible. Thus we do not want to claim too much for our analysis. We do believe that the idea of seeing Expression as a constraint family is plausible. It is also plausible to propose the existence of a hierarchy of positions that favor expression. Our analysis of the data in (11) claims that phrase-final position is one of the positions that may be promoted to a higher ranking than the general expression constraint. There may be evidence from the languages of the world to support this idea (e.g. cases where phonological contrasts are tolerated phrase-finally and not elsewhere), but we have not yet searched out an array of such evidence. Thus this treatment of the data in (11) must be considered speculative at best. Let us now return to the presentation of the surface forms of High verb stems in the short present (when there is a toneless subject prefix). Consider next Disyllabic High verb stems: (16)
ndjbdna.../sjb6na... HdttMngaVsjtMn.gft. ndistila.../s,jsrila... ndjb£ka.../s,jbe'ka...
(sec) (buy) (wipe) (place)
ndidlftia..vMW»-
(play)
If the verb stem is bimoraic and H, the H is linked to the first stem mora in the short form as in the long form. This is of course completely predictable, since Nonfinality (polysllabic HD) continues to delimit the scope of a HD regardless of the medial position. Next consider trisyllabic or longer H verb stems:
94
(17)
siseWrjja. ndjgaWlasibonisa... ndjfum&na... tldiphuliphula.../ ndiphulaphula... s.ingxab«la«iy5igx»bala«i. Odjqaqimb.jsa.../ndjqaqamb.{sa... Sljsebenseja... / sjsebenj^la... ndiphambanisa... sjbonakalisa...
(work) (pour) (show) (get) (listen) (straddle) (make ache) (work for) (turn upside down) (make visible)
Nothing much needs to be said about these data. The High tone of the verb stem behaves the same in the short form as in the long form, except for the possibility of Avoid Prominence being rendered inapplicable due to the verb (optionally) lacking prominence. In those speech varieties where /bujisa/ retains the H on the first stem mora in the long form, the short form may show this same property (as long as Avoid Prominence is inapplicable due to the verb lacking prominence): (18)
odibulisa.... or ndjbulisa... ndifufldisa... or ndifundjsa...
(greet) (teach)
Vowel-initial High verb stems, as usual, require some discussion. Recall our analysis of such stems from Chapter Two. We assume that the constraint that wants a Prosodic Stem to be aligned with a syllable (2:49) is more important than the constraint that wants a Prosodic Stem to be aligned with the Morphological Stem (2:50). This means that the H-sponsor is going to be the second stem mora rather the first. However, in the case of VCV stems, while some steins do have a final H-sponsor as expected, others have a penult H sponsor. Another complicating factor is that VCVCVCV show variation as to whether *MonoHD is enforced or not. In (19) we illustrate the pronunciation of various V-initial stems in the short form of the present tense: (19)
nakha... (build) ndoma... (dry up) ndgna... (spoil) (We have not observed pronunciations like *ntfkha..., *rjdd"na... even though a penult H can be observed in phrase-final position.) Hdaliisa... ndoylsa...
(herd) (defeat) rjdQnwaba... (be happy) tldala'tha... (point at) naliliikana..7 nahlukina... (differ from) ndalu'sela.../ ndahislla... (herd for) nonw4bisa.../nonwabisa... (make happy)
9J
The VCV roots are uniformly treated like CV roots (Le. the Prosodic Stem is always aligned with the syllable rather than the Morphological Stem). As a consequence, the final vowel is the Hsponsor and is organized into a HD; but expression of the H on the head of this domain is barred in phrase-medial position for the same reason that we have Qdjtya-.. and not *ndjty4.. We do not know why (in the speech of our consultants at least) a VCV stem never permits the first vowel to be the sponsor in short forms but does in long forms. The trisyllabic and longer stems in (19) are consistent with assuming that the second stem mora is the H-sponsor as a result of an undoirdnated (2:49). Recall that we suggested in Chapter Two that V-initial verbs are optionally exceptional to *MonoHD. This has no consequence for a VCVCV stem: the second stem vowel is the H-sponsor and Nonfinality (HD), which dominated •MonoHD in any case, will prevent the HD from extending onto the final syllable. Longer verb steins have two possible pronunciations. These two pronunciations follow from the fact that (a) in the short form of the present tense the penult is only optionally considered prominent (thus Avoid Prominence may or may not be relevant) and (b) in the case of V-initial stems, *MonoHD may be respected or not A pronunciation such as nflhldkana— is predicted if the penult is considered prominent and *MonoHD is inoperative, while ns)hlukdna>» is predicted either by virtue of *MonoHD being enforced or by virtue of the penult not being regarded as prominent Let us now consider High verbs with High subject prefixes in the short form of the present tense. We begin with bisyllabic stems. We include both depressed and non-depressed subject prefixes since no difference results from this contrast (20)
bab::*-:M
*i
ifetoiiiPi 96
Candidate (e) is rejected since it violates the OCP; (a), (c), and (d) are rejected since they violate Incorporate (H-sponsor). (b) is the only remaining candidate. The data cited do not in fact show that the HD is initiated by the subject prefix. In other words, (c) would explain the surface forms as well as (b). We shall see later in this chapter that a candidate such as (b) is required in order to characterize "protection" facts (see Chapter Two for a discussion of Plateau and the phenomenon of protection). There is one additional point that bears (repeated) emphasis. With respect to the fusion of domains, depressed syllables play no role whatsoever. In Chapter Two, we saw that the object prefix High and the verb stem H fuse regardless of whether either is depressed -• e.g. njyayiggwiliila and njyazjthengt]sa b o th s n o w fusion. Similarly, we find fusion in badjtfla... just like in basildla... Depressors play no role at all in determining the HD-structure of the output of the A system. Their role is restricted to the B system. Next consider trisyllabic and longer H stems with a H subject prefix. (22)
baseb£nza., zjseWnza. bagajela..., zjgalela... babonlsa..., zjbonisa... bafumdna..., zjfumAna... bafundjsa..., zjfundjsa... boyika..., zqyika... babonlsisa... or babonisisa... baphulAphula... or baphulaphtila... zjsebdnjela... or zjsebenz£1a... baphambanfsa..., zjphambanfsa... lonwibisa... or lonwabisa... lahlukdnisa... or lahlukanisa... babonak&lisa..., zjbonakalisa... bashumay6zana..., sjshumayezana... bashiyiselana..., bashiyiselina...
(work) (pour) (show) (get) (teach) (fear) (see clearly) (listen) (work for) (turn upside down) (make happy) (disunite) (make visible) (preach to each other) (compete in running) Again, we see that there is a single surface High tone which appears in the location where the verb stem High would be expected to appear. The subject prefix High does not surface. These data (like the data in (20)) are consistent with either an analysis where the two High-tone anchors are fused into a single domain or an analysis where the first High tone simply fails to trigger a HD. We analyze the data in terms of fusion. The tableau in (23) illustrates babonakaltsa--, but we omit from consideration all of those constraints that guarantee that the domain will extend as far Right as possible (e.g.. Nonfinality (polysyllabic HD), Avoid Prominence, Align Word R) since they are irrelevant to the main point. (23) babonakallsa.. Candidates (a) bflbanakalisa... •(b) (babgnakali)sa... (c) ba (bpnakali)sa (d) (bajbanakalisa (e) (baj(bonakali)sa
OCP
Incorporate
*!
DomCor
¥h®&$9•
•1
(U)(wa)tya ••-»(lDwa(tyft)
The correct output can be achieved if Nonfinality (HD) dominates DomCor (H)
(33JL Candidates
OCP
ljwfltya. (ijwatya) »(Uwa)tyi lj(wfl) tya. (Ipwatya
Nonfinality (polyHD
Incorporate Nonfinality DomCor (HD) (H)
*!*• •I
a B I O S rt^>
•i* *i*
** ?
&^#fo£-&
(a(wa)(tva) • I * (lD(wi)tya (Ij)wa(tya)
I.* ' f ''J. V -I
From this we can see that Incorporate (H-sponsor) must dominate DomCor (H) as a result of transitivity (since we know that Incorporate (HD) must dominate Nonfinality (HD) in order to guarantee that a word-final H-sponsor will trigger formation of a domain). The evaluation of Uwabeeka— demonstrates that independently of Nonfinality (HD), Incorporate (H-sponsor) must dominate DomCor (H).
Here we have conclusive evidence that Incorporate (H-sponsor) must dominate DomCor (H). The optimal candidate (l{wgbee)ka has two violations of DomCor (H) while the non-optimal candidate (li)wa.(bee)ka has only one — but the non-optimal candidate has a violation of
102
Incorporate (H-sponsor) that the optimal candidate does not have. Thus by ranking Incorporate (H-sponsor) over DomCor (H) we can generate the correct output Extension of the HD onto the penult in examples like liwabonltsa... cannot be explained through an appeal to *MonoHD; output-output faithfulness seems to be at work here. To emphasize this point, we show in (35) the evaluation ignoring *MonoHD and Avoid Prominence for the moment: (35) Candidates
OCP
Nonfinality (polyHD)
Incorporate Nonfinality (HD) (H-
DomCor
sponsor)
liwflbgniisa (ID(wj)(bonii)sa
•I*
(lp(wa)bflniisa
*l
(lflwflbflniisa (Ijwabojniisa
r
*Viig»,.
•p; {$ t
."M • 'li j ' n ' i ffi.i
"I*
.TVi*..:
(liwabgnii)sa
(li)wfl(bQnii)sa
•I
:
• >
•«'•
'•»"»&•.-»
We want Oiwabflnil)sa to be optimal, but in the above evaluation it has the same evaluation as (Hwajbojniisa. *MonoHD is equally satisfied by both of these candidates. Consequently, it would seem that Avoid Prominence should prefer (Hwabpjnlsa... over (liwa.bgni)sa... (although the latter would also be a possible pronunciation due to the fact that the penult vowel in short forms may be treated as lacking prominence, thereby permitting the HD to extend through it). But we do not want fllwflbojnllsa... to be judged optima] under any circumstances. The pronunciation *liwab6nisa... is not possible. The idea behind an output-output constraint is that the optimal form (!lwa.botti)sa... wants its stem to have the same domain membership as in a surface form such as ni(bo.ni)sa (where there is no High-toned subject prefix and no object prefix and where •MonoHD drives the High Domain onto the penult syllable). In other words, (liw§bflnl)sa... mimics the domain structure of output forms where the stem's tone is unobscured by other factors (a preceding High tone). While we believe this approach is on the right track, there are many questions that would need to be addressed in order to develop it in detail We do not deal with this matter further. At this juncture, let us repeat the set of constraints and their dominance relationships that we have motivated for HD-structure and expression (restricting ourselves to the A system). (36)
BAL, Syllable Alignment, "Overlapping, Nonfinality (polysyllabic HD), Express (H, phrase-final mora) - undominated Nonfinality (Word-final H) - dominated by Express (H, phrase-final mora) Express (H, head) - dominated by Nonfinality (Word-final H) Plateau - dominated by Express (H.head) *(H,nonhead) - dominated by Plateau, Express (H, head) No Adjacent Edges —dominated by "'Overlapping No Adjacent Sponsors (possibly an alternative to No Adjacent Edges as the OCP constraint that drives fusion in Isixhosa), but not yet independently motivated) - dominated by "Overlapping Incorporate (H-sponsor) - dominated by Nonfinality (polysyllabic HD) Nonfinality (HD)— dominated by Incorporate (H-sponsor) 103
DomCor (H) - dominated by No Adjacent Edges and/or No Adjacent Sponsors, Incorporate (H-sponsor), Nonfinality (HD) Uniqueness -- dominated by Incorporate (H-sponsor) •Structure - dominated by DomCor (H) and/or Incorporate (H-sponsor) •MonoHD - dominated by No Adjacent Edges, Nonfinality (polysyllabic HD) Avoid Prominence - dominated by Syllable Alignment, *MonoHD Align Word R - dominated by *Structure, Nonfinality (polysyllabic HD), Avoid Prominence BAR - dominated by Align Word R Express (H) - dominated by *MonoHD, Align Word R. *(H, non-head) 3.2. Interaction of the short present form with a following word (nominal). In this section we consider die interaction of a short present form of the verb and a following nominal (including the infinitive form of the verb, which is structurally a nominal) in order to (a) expand briefly on the discussion of the DIHTS phenomenon, (b) demonstrate the generality of the phenomenon of "protection" discussed earlier, and (c) provide evidence in favor of our claim in the preceding section that the short present tense form involves fusion of multiple H-sponsors into a single HD where only the head of the HDrealizesthe High tone. We do not undertake here any detailed discussion of sentence phonology in Isixhosa (see Jokweni 1995 for an analysis) and limit ourselves to matters that are directly pertinent to our analysis of the word-level tonology. The aspect of Isixhosa sentence phonology that is of some relevance here is the fact that the final vowel of the verb will elide in front of a vowel-initial complement that is in the same "phrase" as the verb (we are not here concerned with defining the phrase domain). Nominals in Isixhosa, as in all Bantu languages, are sorted into noun classes. Noun class membership is Indicated by prefixes. The prefixal structure of Isixhosa nominals involves two independent elements*, a preprefix and a prefix. In some cases these two elements may be fused phonologically. In certain contexts, the preprefix may be omitted. But all the examples we discuss here contain the preprefix. This preprefix is important for our concerns because it is always specified with a High tone. This High tone behaves much like the High tones we have been discussing - it initiates a HD that will extend to the Right under appropriate circumstances. The first class of data that we consider involves a verb that has a depressed final syllable and a following nominal that would, in its isolation form, realize a High tone on the first syllable: e.g. Icuuba 'tobacco', ithaagga 'pumpkin', illuuth.'f 'stick,' etc. The contraction of the depressed syllable of the verb with the initial High-toned syllable of the noun produces the results shown in (37), where we place the verbal form n.djthe'nga 'I am buying' in front of nouns with an initial High tone. (37)
ndjth6ng jciiiuba 'I am buying tobacco' (cf. icuuba 'tobacco') Qdjthgng jthld&rjga 'I am buying a pumpkin' (cf. tthaanga 'pumpkin') ndjthe'ng ul!tiuth!( 'I am buying a stick' (cf. dluuthti 'stick') ndjthgng jmltipltl 'I am buying guns' (cf. fmiipld 'guns') ndjthlng utyw!a*al!ti 'I am buying beer' (cf. dtywaallS 'beer') rjdjthlng ubldoyla' 'I am buying wool' (cf.tibooy.'a*'wool') 0dith«00 jq!aand!'lisa and n.dithdrjd u.k?ddjb!oo)(si) (if Syllable Alignment dominates No Adjacent Edges) or a(no)o(si) (if No Adjacent Edges dominates Syllable Alignment). If the former HD-structure is optimal, both HD's will be gathered into a single RD due to NARDE and there will be no downstep between the High tones on the heads of the two HD's. If the latter HDstructure is optimal, NARDE will be irrelevant and therefore there will be two separate RD's and downstep between the heads of the two HD's. Let us now turn our attention to High verb stems in the negative present. The following data show that when the verb stem is lexically High (and trisyllabic or longer), just a single High tone surfaces, and that High tone is always on the penult syllable (the monosyllabic stem case is different and will be dealt with later). This pattern holds whether the subject prefix is depressed or not and whedier the stem contains a depressed syllable or not. (31)
arjdifum4£ni andjsebee'iKi asithenfltfsi aniboniisi anigawtitili akakhathajtfjgki akaphakameeli akangsabaUa'iu akapliamtjanfisi akasebengggli akabonakalifsi abasebenjjsela^ni andjphulaphuuli abafunflgke'eli
(get) (work) (sell) (show) (chop down) (be distressed) (stand above) (straddle, take long strides) (turn upside down) (work for) (make visible) (work together for) (listen) (make noise)
161
These forms are analogous to toneless verb stems in that they have a penult High tone. They differ from toneless verb stems in lacking any trace of the High from the subject prefix. Why is there this difference? We propose first that, in underlying representation, these verbs have their lexical H associated with the first stem mora. The toneless verb stems of course lack such a H tone. Second, we propose that these verb stems also have a grammatical High tone on the second mora. The input for anifuma'a'ni then, is /a-nj-fumani/. This input has three adjacent Highsponsors. We would expect adjacent H-sponsors inside a verb stem to be fused into a single HD due to No Adjacent Sponsors (recall that in the perfect, an object prefix H-sponsor and the verb stem H-sponsor are fused into a single domain even though in general the perfect exceptionally fails to obey No Adjacent Sponsors). Thus we expect /fuZ and /maa/ to be part of the same HD. However, to obtain the correct surface forms, it is necessary that the subject prefix H-sponsor InU also be organized into that same HD. We must have the HD-structure a(nifu.maa)ni if we are to have a surface form where the only High tone is on th penult syllable. In order to obtain this result. No Adjacent Sponsors must be able to force a violation of DomCor (H) and thereby drive th subject prefix H-sponsor into the same domain as the verb stem's H-sponsors. The behavior of High verbs in the negative present is thus just like the behavior of High verbs in the short present: No Adjacent Edges drives fusion. The only problematic aspect of this is that we have seen that in the case of toneless verb stems in the negative present. No Adjacent Sponsors does not drive fusion. We suggested that these forms had to be exceptional to No Adjacent Sponsors so as to prevent fusion from occurring. We are thus faced with a complicated pattern of exceptionality: the grammatical High tone found in toneless verb stems in the negative present does not fuse with the subject prefix H-sponsor, but the lexical High tone on the first mora of the High verb stems does fuse. Since it is not clear whether there is an explanation for this pattern, we will not pursue the matter further here. We will simply assume that it is indeed No Adjacent Sponsors that drives the fusion of the three H-sponsors in a(nlfuma,a)nl into a single domain. Given that these sponsors are organized into a single domain, the only surface High tone will be located on the head of this domain - i.e. the penult syllable. Bisyllabic verb stems are entirely analogous. (32)
anib6dni, andjb
(eat) (see) (show) (listen)
The subject prefix H and /ntf£/ fuse into a single HD, while the object prefix H, the lexical H of the verb stem, and the grammatical H tone (if there is one) all fuse into a single HD, except in the monosyllabic stem case (which behaves parallel to the monosyllabic lexically toneless verb stem). The data in (88) show that ^ g £ / protects all of the non-heads in the following fused domain. Protection occurs here because there is a single RD initiated by/ngif. In (89), (89)
abarig a depressed syllable will prevent the formation of a single RD, as in andjm4ngalel!4a'riga; Look at High verb stems next. We find that there is "fusion" between the subject prefix and the verb stem, as well as between the lexical High of the verb stem and the grammatical High tone. As a consequence, we hear only one High tone, which is located on the penult. (93)
anityaanga anibdfinga anibek£dn.ga anibonis&irjga anigawulSinga anibonisis£6n,ga.
(eat) (steal) (put) (show) (chop) (see clearly)
1
In this section, we will not exemplify dialectal differences that have already been explored fully. Rather, we will cite just examples from M's dialect.
I8S
The inclusion of an object prefix yields familiar patterns. Consider first the toneless verb stems. Both the subject prefix and the object prefix surface High-toned (recall, from the negative present, that this is true just in M's dialect - K exhibits fusion in this case).3 We regard this as a case where "fusion" of the subject prefix and the object prefix into a single HD is barred. The object prefix H-sponsor Is generally not adjacent to the verb stem's grammatical High and thus no fusion is expected. (94)
anlwdwisltfaaga anlwdballdinga aniwdbalisliinga. anVwashukumisIfia'n.ga, am'wa'namathelisltfa'n.ga,
(drop) (count) (narrate) (shake) (cement)
Of course, the grammatical High tone is located on the first stem syllable when the suffixation of /angfl/ yields a bisyllabic stem, as in: antwalwda'n.ga (fight) and anlwdsa'aqga (roast). Here the object prefix H-sponsor is adjacent to the stem H-sponsor. The observed pronunciations are fully consistent with the claim that there is no fusion in this context (though if one wished to claim that fusion did occur, one could still explain the appearance of a High tone on the object prefix even though It would be a non-head, given fusion - by appeal to Plateau). When the object prefix is depressed, its High will "shift" onto the stem initial syllable if possible: (95) akdzjbla'lfsfia'nga (narrate) abAzjchl<Sktimts£6n,ga. (touch) abaaph.'a'za.misia'a'nga (disturb) akajsjqidnonondjsia'a'nga, (emphasize) but: akaj'.lyumelldSnga (permit) The data above reflect the consequences of this tonal shift in M (Le. the effects of Plateau except where a depressed syllable yields an RD-structure that prevents Plateau from affecting non-heads in the verb stem's HD). When the subject prefix is depressed, its HD will - as a result of the B system - overlap the HD of the object prefix (provided the object prefix is not depressed), resulting in the object prefix and not the subject prefix being a head. The result will be a toneless subject prefix. (96)
andjba"xoIel!ainga and.iylcacis!44fl0a andjzixid'le'lia'n.ga arjdjjiichltikiiinlsia'n.ga,
(forgive) (explain) (forgive) (touch)
Next let us consider object prefixes in conjunction with High verb stems in the negative past. We see here that the subject prefix and the object prefix and the stem High tones all "fuse": 9
Juat aa in the negative present, the subject prefix is phonetically higher than the preceding negative marker, but not as high as the object prefix. We have not attempted a phonological account of why the subject prefix is not fully high. 186
(97)
aniwatyafirjga. aniwamem£an.ga aniwabulis&drjga, aniwabonisisianga
(eat) (invite) (greet) (see clearly)
Since "fusion" affects these structures, the presence of a depressed subject prefix or a depressed object prefix has no affect on the tone pattern: an.djwatyaa'riga abazjtya'a'rig&; andi?lityaa'n,ga etc. Similarly, the presence of a depressed syllable in the verb stem makes no difference: ababasebenz,ola4n,ga BQdJwafheagjsa'a'rjga. and abasutiQ'n.fU show that fusion occurs regardless of depression. In conclusion, we can see that the negative past simply follows from the analysis already developed on the basis of the negative present. The negative infinitive, discussed in the next section, adds a new element to the analysis. 4.5. Negative infinitive. The negative infinitive consists of the preprefix /u/, which sponsors a High tone underlyingly, followed by the prefix /ku/, which in turn is followed immediately by the negative prefix /riga/ The verb stem, with the final vowel /i/, is located immediately after /nga/. The negative infinitive form of the verb exhibits a tone pattern that is quite different from the negative tone pattern discussed above. We cannot actually make sense out of the negative infinitive form unless we consider both the Intonational Phrase-final form and the Intonational Phrase-medial forms in parallel. We begin as usual with toneless verb stems. (98)
ukiingaxolleeli ukiingahlom!6eli ukiingalungl.ljsi ukdngayumteeli ukdngadanllisi ukungagi«!\juji ukiirjgachukumliisi ukiingaqononond.l.Usi
(forgive) (add) (repair) (permit) (dance) (scrape) (touch) (emphasize)
ukiingahlomelM kaJcuuhllg ukUngagiK«l!i kakuuhlle ukdngachukumisH kakuuhlle' ukiingaqononondjslf kakuuhlle ukiingaxolelli mmntliu ukdngahlomellti nt!d
(add well) (scrape well) (touch well) (emphasize well) (forgive anyone) (add anything)
Before we examine the stem tone, notice that there is a surface High tone on the prefix /ku/. (Recall from our discussion of the positive form of the infinitive in Chapter Three that the preprefix IvJ is the underlying anchor of the High tone, not the prefix /ku/). *MonoHD will 187
account for why the HD initiated by the preprefix must extend onto the prefix /ku/. The only factor that could block *MonoHD would be if No Adjacent Edges were violated. However, /noa/ is toneless; thus it will not prevent the HD from extending onto /ku/. However, there is one question that needs to be addressed: do we expect the High tone of the preprefix to extend even further to the Right! For example, assuming that the grammatical High tone in the negative stem is not located on the first stem syllable, should the HD initiated by the preprefix extend onto /008/7 Recall our discussion in Chapter Two of the constraint Align Word R, the constraint that is responsible for HD's trying to align all the way to the Right of the word. We noted that there are two different versions of this constraint that could be responsible for the extension of HD's to the right. We repeat these two versions here: (99) AUON (Word. R: HD, R) (100) AUON (HD, R; Word, R) The first of these constraints requires only that the Right edge of a Word be aligned with the R edge of a HD (a formulation such as this also requires 'Structure to be ranked above it, so that HD's are not introduced just in order to satisfy this constraint). The second requires that every HD aligns with the R edge of the Word - in other words, every HD goes as far to the Right as possible (e.g. consistent with 'Overlapping and No Adjacent Edges, etc.). If (99) is correct, then we don't expect the HD initiated by the /u/ to extend beyond two syllables, since the stem HD will serve to satisfy (99). On the other hand, if (100) is correct, then we would expect the HD initiated by /u/ to extend as far to the Right as possible. The data in (98), where the HD does not extend onto /nga/, suggest that (99) must be the correct form of Align Word R. We should point out, however, that the issue is not entirely straightforward; for example, the subjunctive form of the verb - omitted from discussion here due to space limitations - might be taken as providing evidence in favor of (100). On the other hand, nominal tonology- also not dealt with here ~ might be taken as supporting (99). We leave this matter for further research. Now let us turn to the matter of the grammatical High tone. The verb stems in (98) show a Falling tone on the penultimate syllable in IP-final position, but High tone on the final vowel in IP-medial position. Up until this point, we have only seen Falling tones in cases where the DIHTS phenomenon placed a High tone on the penult syllable. However, the Falling tone in (98) is entirely independent of DIHTS. We shall postpone deriving the IP-final form with the Falling tone until later, and concentrate instead on what the IP-medial form tells us. We will assume that the grammatical High in the negative infinitive is located on the second stem mora, just as it is in all the other negative tenses. The surface word-final High tone suggests that the HD initiated by the grammatical H must extend to the very end of the word. In other words, we must have a domain structure like (uku)ngachu(kjjniisi)... in the IP-medial case. This domain structure requires, however, that Nonflnality (polysyllabic HD) not be enforced in the negative infinitive. (Recall that Avoid Prominence is not enforced in any of the tenses with a stem grammatical H tone.) Thus Align Word R will demand the extension of the HD all the way to the end of the word since there is nothing to prevent this wide domain. The correct surface form results from (uku)ngachu(ku.mlsi)... since only the heads of the two HD's would be expected to surface with a High tone. Let us consider next bisyllabic (consonant-initial) toneless verb stems.
188
(101) uktingaliimlf ukiingahleekM
(cultivate) (laugh)
ukdng»gc88dU
(«>ast)
ukiingayaalll ukiingalimlf k£kuuhll6
(close) (cultivate well)
Notice that here a final High tone appears in both IP-final and IP-medial position. These data would suggest that the grammatical High tone links to the final syllable when the verb is bisyllabic (unlike the negative present, where the grammatical High links to the first stem syllable). If we adopt this analysis, then the motivation for why the grammatical H is able to be located on the final vowel could reside in the fact that all nonfinality effects are absent in the negative infinitive (including the one that is at work in determining the location of the grammatical High tone).. However, there is a complication. Those verb stems that are exceptions to *MonoHD show a Falling tone on the penult syllable in IP-final position and a final High tone in IP-medial position: (102) uktirtgftwtftsi ukiingalwliisi ukiirigahllfisi ukiSngawisli k&kuuhlt6
(drop) (make fight) (lower) (to drop well)
The fact that verb stems exceptional to *MonoHD show this behavior suggests that somehow •MonoHD is involved in the contrasting behavior of /limi/ and /wisi/ in the negative infinitive. But it is not all clear how *MonoHD could play a role since the IP-medial form demonstrates the HD does extend to the final syllable (and it is Align Word R in any case that drives a wide domain here regardless of whether ""MonoHD is enforced). To account for the Falling tones in (102) in the same way that we account for the Falling tones in (98) above, it will be necessary to assume that the grammatical High tone in uktlQgaViisi is located on the first stem syllable. We will develop the details of the analysis below. There is no synchronically-motivated way to explain why /limi/ has the grammatical High on the last syllable while /wisi/ has it on the first syllable. We suspect that the historical origin of the contrast may reside in verb stems like /wisi/ originally having had a bimoraic initial syllable (compare Chimwiini, which retains prato-Bantu vowel length, where bisyllabic verbs derived from "monosyllabic" roots regularly have a long initial syllable (Kisseberth, p.c.)). If so, then the observed pattern may have been due to the parallel trimoraic structure of/cacisi/, /xobli/, /wiisi/ as compared to the bimoraic structure of /limi/. We will simply assume, then, that while ordinary bisyllabic stems have the grammatical High tone located on the final syllable, the bisyllabic stems derived from "monosyllabic" stems have the grammatical H located on the first stem vowel (for synchronically inexplicable reasons). The situation is a bit more complex. Vowel-initial verb stems that are trisyllabic might be expected to have bisyllabic Prosodic Stems (due to the effects of alignment of the Prosodic Stem with the syllable). If so, they would be expected to parallel bisyllabic consonant-initial verb stems, but they do not regularly do so:
189
(103) uku^gojiduli (possible, but fess preferred: ukriaggluulli) ukiirjgadJta'Bli (possible, but less preferred: uktingandJftalH)
(stretch) (make bad)
It seems then that in the negative infinitive, the Prosodic Stem prefers to align with the Morphological Stem (meaning that a VCVCV stem will be treated as trisyllabic) rather than with the syllable (in which case a VCVCV stem will be treated as bisyllabic). We do not have an explanation for this fact Monosyllabic stems would be expected to have the grammatical H located on their only syllable, and this is what we find (both in IP-final and IP-medial position). (104) ukilngaalwtf ukiingalwli k£kuuhl!e' ukiingechl! ukiingefinslt uktingaangU uktfngQQsli
(fight) (fight well) (go down) (do) (kiss) (roast)
Notice in (104) that vowel-initial bisyllabic verb stems (whose Prosodic Stems are actually monosyllabic if alignment with the syllable dominates alignment with the Morphological Stem) behave in analogous fashion to CV stems like /lwi/ and never like CVCV stems. In our analysis, they uniformly prefer the Prosodic Stem to be aligned with a syllable rather than the Morphological Stem. There Is one point about the IP-medial data cited at various points above that requires mention. Notice that we find word-final High tones in IP-medial position. Recall from Chapter Three, however, that word-final High tones delete in medial position in the short form of the present tense. The problem here has to do with the construction of phonological phrases in Isixhosa (see Jokweni 1995 for extensive discussion). Word-final High tones delete only when they are medial in the phonological phrase. Negative verbs are always at the end of a phonological phrase (just as in Chimwiini (Kisseberth, p.c.)), but are of course not always IPfinaL While the placement of the grammatical High in the negative infinitive has the complexities noted above, given that placement we can predict the surface forms in IP-medial position without difficulty. The remaining challenge is to account for the Falling tone in the IPfinal form of trisyllabic and longer verbs, as well as in bisyllabic verbs derived from "monosyllabic" roots. At first glance, one might think that the Falling tone is directly connected to the fact that it appears on a lengthened voweL This is the basis, for example, of Khumalo's (1988) account of similar data in Isizulu. We can, however, prove that the Falling tone is not connected to the length but rather to the IP-final position. The evidence for this claim derives from Jokweni's (199S) study of phrasal phonology in Isixhosa. He shows that lengthening is not restricted to IP-final position. When a word that has the alternation ...FH/...0H (as the negative infinitive has) is in a lengthening position that is not IP-final, then the ...OH pattern occurs. One such environment is illustrated in (105): (105) amantombajtfana dyabdlmfca 'the girls are leaving' andjbonfinga mantornbflza.ana' ecdweeni 'I did not see any girls'
190
On the basis of Jokweni's evidence, we conclude that the Fall is directly connected to IPfinal position and not to length. It is not, however, clear exactly how we should use IP-finality to derive the correct results. We can sketch two lines of attack. One line of attack is to use IPfinality to constrict the HD so as to extend only onto the first mora of the penult syllable. The second line of attack is to use IP-finality to somehow influence how High tone surfaces in an IPfinal polysyllabic stem. We begin with the first line of attack. Let us assume that besides a Nonfinality (polysyllabic HD) which refers to the word, there is also another version that refers to the Intonational Phrase. We shall identify these two constraints as W-Nonfinality (polysyllabic HD) and IP-Nonfinality (polysyllabic HD). In general, IP-Nonfinality (polysyllabic HD) has little work to do. The word-level constraint bars the creation of polysyllabic word-final HD's. Thus it is only in a tense like the negative infinitive, where W-Nonfinality (polysyllabic HD) is inoperative that we can actually see an effect from IP-Nonfinality (polysyllabic HD). But still this does not give us an explanation for why (y.ku)ngaxo(l£)ell should be optimal rather than (u.ku)ngaxo(lee)U since both of these avoid an IP-final polysyllabic HD1 In other words, there has to be more to the story than just an IP-Nonfinality effect. A constraint that would make the correct choice between (uku)ngaxo(le)eli and (jiku)ngaxo(l£e)li is: *O")HDO]D> - essentially our Avoid Prominence constraint (as originally formulated), only restricted to the Intonational Phrase. The problem, of course, is that we have seen that Syllable Alignment must dominate such a constraint - e.g. fldjyabdd'na not *tldiyab. If we invoke such a constraint, and make no other assumptions, then we would in fact derive the incorrect surface form 'ukilrjgftxoleeli ~ Le. we would fail to realize the High tone on the HD-head and all the non-heads would be prevented from being High due to the constraint *(H, non-head). What we need to derive is an optimal output where a H cannot be realized on the final mora, but *(H, non-head) is minimally violated. The only cases so far where *(H, nonhead) has been violated is as a consequence of Plateau. We need to find some other way to force a violation of *(H, non-head). We propose (very tentatively) that Express (H, head) should (a) be ranked above *(H, non-head) and (b) be regarded as a gradient constraint. In other words, if *(...H)]n> is undominated and prevents full satisfaction of Express (H, head), then the minimal violation of Express (H, head) will be optimal even if it means that a violation of *(H, non-head) occurs. We assume that the minimal violation would be the closest mora in the HD. At first glance, this doesn't look like it would help since 'ukdn.gaxolee'li is also bad. But *ukilrjgaxolee1i c o n t a u l s a Rising tone on a bimoraic syllable, not a tonal shape allowed in the language.9 Thus we can simply postulate that there is a constraint against OH over two moras and rank it above Express (H, head) as well. This means that the optimal output (minimally violating Express (H, head)) will be the desired ukUngsxoKeli The constraint *(.. ,H)]n> is simply a member of the Nonfinality family of constraints and thus seems to us entirely reasonable, as in the constraint against Rising tones. The only somewhat unexpected claim in this analysis is that Express (H, head) is a gradient constraint. However, there is a critical problem with the idea that Express (H, head) is a gradient constraint that dominates *(H, non-head). While negative infinitive forms alternate between (...OH) in IPmedial position and (...FO) in IP-final position, there are other forms (nominate, affirmative verb tenses- see below) which appear as (...00) in IP-medial position and (...FO) in IP-final position. These forms have (..00) due to the principle that disallows word-final H tone to be expressed in medial positions in the phonological phrase. Recall that we observed earlier that negative forms are treated as being at the end of a phonological phrase and thus escape the loss of a final High tone. The problem, then, is this: if Express (H, head) is gradient and outranks *(H, non-head), why doesn't a High tone surface on the penult syllable in phrase-medial position rather than (.. .00)7 The appearance of a High tone on a mora other than the HD-head seems to be limited to IP-final position. The idea of "exploding" constraints will, of course, permit us to wriggle out of the problem. We say that it is only Express (H, IP-final head) that outranks *(H, non-head) and
3 It Is die case thai Rising pitch occurs on a short mora (hot is (a) High-toned and (b) has a depressed onset. Bui this does not detractfromdie ill-farmedness of die structure under discussion.
192
not the general constraint Express (H.head). But this added complication makes one wonder whether this approach is indeed on target We have developed two quite different analyses of the Falling tone in the negative infinitive. We do not find the evidence for either compelling and thus do not consider the matter in any way closed. For convenience, we will assume the second analysis, where there is a HD aligned with the Right edge of an IP, and the expression system is responsible for the appearance of a penult Falling tone. Up until this point, we have only examined toneless verb stems in the negative infinitive. Let us look now at High verb stems. Consider trisyllabic or longer stems first. (107) uktingafiny!e>sj uktingafum!fiani uktingasenglfoli ukiingagawltiuli ukdngagawulii ka'kuuhlle'
(shorten) (get) (milk for) (chop) (chop well)
These data are indistinguishable from toneless verb stems of the same length. We propose that the lexical H is on the first stem mora and the grammatical H is located on the second stem mora (just as in the case of the other negative tenses). The OCP constraints lead to "fusion". The inapplicability of Nonfinality (polysyllabic HD) means that the HD will extend through the final syllable. The Falling tone will emerge in IP-final position in the same way as it does in the toneless verb case. Bisyllabic stems are shown in (108): (108) ukiingathteenfli ukiingaslduli ukiingadJUisi ukringgy.'iisi uloSrjgasulIf kSkuuhlle'
(buy) (wipe) (feed) (defeat) (wipe well)
These data are of course quite distinct from the toneless bisyllabic verbs. The reason is clear. There is a lexical High tone on the first mora of the stem. If there is also a grammatical High tone on the final mora (as we have assumed is true for toneless verb stems), we would expect that the two H-sponsors would fuse into a single HD. The result will be the domain structure (uJ(ii)nga(bopni). But even if we assumed that the grammatical High tone fails to locate to the final vowel, we would still expect the HD initiated by the lexical High tone on the first mora to extend through the second syllable (since Nonflnality (polysyllabic HD) is inoperative in this tense). Given the domain structure (uku)nga(bopni), w e expect a penult Falling tone in IP-final position and that is what we find. Monosyllabic stems surface with a High tone (there is only one place for the grammatical H tone to be located, and that is the same mora that bears the lexical H tone): (109) uktlngaatyli ukiingaabM ukungaakhli
(eat) (steal) (build)
uktingaty!! kalcuuhlle'
193
(eat well)
Notice VCV stems are treated identical to CV roots. There is no variant available where the Prosodic Stem is aligned with the Morphological Stem rather than the syllable. The inclusion of an object prefix provides nothing new in the case of toneless verb stems. The object prefix is of course High-toned, and the HD that it initiates does not extend onto the following syllable (due to No Adjacent Edges): (110) uktingaytlliim.'f ukiingay!fy.a.al!l ukilngablitxolle'eli ukdngaylihlomteeli ukdngayHgusliJuli ukiingay!lchukum!1isi ukifngay!fqononondj.tjsi
(cultivate) (close) (forgive) (add) (scrape) (touch) (emphasize)
When the verb stem is monosyllabic, we have evidence that the object prefix and the verb stem do "fuse" into a single domain that includes the final syllable: (111) ukiSngaylilwi ukdngayilwf kakuuhl!4
(fight) (fight well)
Thus while the OCP generally does not control the interaction of the object prefix H-sponsor and the verb stem H-sponsor in negative forms of toneless verb stems, here it does. Given that there is fusion, and given that Nonfinality (polysyllabic) is inoperative in the negative infinitive, we end up with the domain structure (jjku)nga(yi(Iwi). The result will be a Falling tone when the word is IP-final and a final H tone in IP-medial positioa When the verb stem is /VCV/, we find a similar pattern to (111) due to the fact that alignment of the Prosodic Stem with the syllable gives a monosyllabic Prosodic Stem in the case of a VCV Morphological Stem: ukdrjga.y'o'osi (roast), No discussion is required. When the object prefix is depressed and the stem does not have a depressed initial syllable, we find the H "shifted" away from the object prefix onto the first stem syllable. (112) uktingazjlliimlf ukringaftjxiolltfeli [K], ukringajixldMeli [M] ukdngazjhltdmie'eli [K] ukdngasibMHisi [M] ukiingflSQnwIa'blisi [M] ukdngatich!ukum!iisi [K], ukringajuchiVikumfisi [M] ukiingasiq!(5nonoi)(Jl.tisi [K] ukiSnga&bla'lfse'eli [M]
(cultivate) (forgive) (add) (narrate) (make happy) (touch) (emphasize) (narrate to)
In the case of a bisyllabic stem like /liimj/, the shift is onto the lengthened, toneless penult syllable, which produces a Falling tone. When the stem is trisyllabic, we find that there is downstep in K between the shifted High and the Falling tone on the penult while in M there is no downstep. When the stem is quadrisyllabic (or longer), the shift of the High onto the first stem syllable does not result in "protection" in K but does in M. This difference between K and M has 194
been analyzed extensively in terms of whether NARDE dominates RD-Faithfulness (as in M), or whether RD-Faithfulness dominates NARDE (as in K). The only point that needs to be made is that the data in (112) above provides clear cut evidence that we were correct in assuming that the grammatical H tone is on the second stem mora in the negative infinitive, just as in other negative forms. If the verb stem begins with a depressed syllable, the H of the object prefix cannot shift: (113) ukiingaz!jg«?!tfwli (scrape) ukringazjtvumle'eli (permit) If the verb stem has an internal depressed syllable, the object prefix H can shift onto the initial vowel of the verb stem, but in both K and M there will there be a RD break between that High tone and the head of the stem HD - thus both dialects will show downstep between the two heads and no protection of nonheads in the second HD: (114) uktfngaz.iphla'zamliisi ukiingaz.iq!6nonond!.lisi
(disturb) (emphasize)
We have up to this point largely ignored the toneless verb stems with an object prefix included. It preferred to use the form with an object prefix only in these verbal forms with a complement following, our following:
issue of the IP-medial pronunciation of is important to note that our consultants IP-final position. When requested to use consultants gave pronunciations like the
(115) ukdogsyilwll kfikuuhHg (cf. ukun,gay!iilwi) ukiingazilllmlt kSkuuhlle [K], ukiingaziMml kdkuuhllfi [M] (cf. ukurjgflzilllimi) ukdngaylfguzul'l kfikuuhlle" (cf. uktirjgay!fgu.z!tiuli) ukilngay!iqononond.is!i k6kuuhl!e' (ukringayllqononondj.ijsi) ukUngazjqdnonondjsit kdkuuhlM (cf. ukunga.zjqldnononfm.jsi
(fight well) (cultivate well) (scrape well) (emphasize well) (emphasize well)
While these data are a bit artificial, they do reflect the fact that if the forms with an object arc used in IP-medial position, then a final H tone appears and not a penult Fall. The example ukilggazjltimM kdkuuhlle [K] but ukdngazjlMmi kdkuuhlM [M] is particularly interesting. In the IP-final form, ukdQga^jflimll, the shift of the H from the object prefix did not create adjacent HD's in either K or M. The domain structure for this item is (uku)nga(zili)i(rn1). In neither K nor M will there be any violation of NARDE when these HD's are located each in its own Register Domain. But in medial position, there is no length on the penultimate syllable. The HD structure will thus be (uku)nga(zili)(mi). In M, this structure will trigger a single RD including both of the last two HD's (since NARDE dominates RD-Faithfulness). As a consequence, there is no downstepping in uktingazjlifmf kikuuhl!e\ In K, however, where RDFaithfulness is more important than NARDE, faithfulness to the RD-structure of the output of the A system will require two adjacent RD's in the output of the B system Thus there will be downstep: ukrin,ga.zj{im!l ka"kuuhl!e\ High verb stems with an object prefix are our next topic. Here we find a rather startling new situation. We have not yet encountered a case where the H of the object prefix and the
195
lexical H of the verb stem fail to "fuse". But this is just what will explain the negative infinitive forms below: (116) ukilngaylfbdoni ukiingayllbe'eki uktingabidyiisi
(see) (place) (defeat)
Were fusion in effect we would expect *uktiqgQyitfdoni. To obtain the correct form, we must say that in this morphological structure the sequence of an object prefix H and a verb stem High is an exception to the OCP constraint No Adjacent Sponsors. However, fusion does affect the lexical High-sponsor and the grammatical High-sponsor in the verb stem. We shall sec later that this failure to get fusion of the object prefix H-sponsor and the verb stem H-sponsor is idiosyncratic to the negative infinitive; specifically, it is not a general aspect of all tenses where HD's extend through the last syllable of the word. If "fusion'' does not occur, then we should see the effects of Plateau when the verb stem is longer. This is exactly the case: (117) ukringaytfbdnlisi ukiingayltbdnislisi ukiiogayllfiimfiani ukiingab!£16m4n1isi
(show) (see clearly) (get) (connect with)
The HD structure will be (jiku) nga (yj) (boniisi). Since we have two adjacent HD's (and no disrupting depressed syllables), in both K and M there will be a single RD. Plateau will affect the nonheads in the verb stem. If the verb stem has an initial depressor or an internal depressor, then we will have two RD's in both K and M and there will be no Plateau effect. (118) uk\Sngay!igalteeli ukdngayllthengUjsi ukdngayltsebenglXisi
(pour) (sell) (use)
The object prefix in the negative infinitive constitutes a HD in the output of the A system (due to the failure of fusion to group the object prefix H with the verb stem H). When the object prefix is depressed, the B system will demand that this HD extend onto the initial syllable of the stem (creating an overlapping domain structure) if that stem does not have an initial depressor. This means that the first stem syllable will be the head of a HD: (jiku)nga(zj (bg)nakalii)si. The H tone on the verb stem's initial syllable will serve to protect non-heads in the verb stem's HD, unless a subsequent depressed syllable invokes a new RD. (119) shift:
shift and protection: shift and no protection:
ukungasithleengi ukiinga%js!tfuli ukdngszifuma'ani ukdnga^jbdndkilf isi uktinga%jth!6ng!.tisi ukdngsxisle'benzl.usi 196
(buy) (wipe) (get) (make visible) (sell) (make work)
no shift:
ukiingajfoldcindi uktfrjgazltgalie'eli
(hate) (pour)
The negative infinitive of a H monosyllabic stem, with an object prefix attached, surfaces with a penult Fall: ukdrjgayHityi (a depressed object prefix makes no difference: ukdggg2?,ljty0- In other words, its tonal shape is not different from the toneless monosyllabic verbs. We noted that in the case of bisyllabic and longer H verb stems, it is necessary to block fusion of the object prefix in the negative infinitive. But in the monosyllabic case we need to allow fusion (the normal case) to occur. This, combined with the fact that Nonfinality (polysyllabic HD) is inoperative in the negative present, will produce a word-final bisyllabic HD that will yield a Falling tone on the penult of the verb when it is IP-final (and a OH partem when the verb is medial: ukilQggyityH— )• As matters stand at present, the exceptional behavior of the object prefix in the negative infinitive based on bisyllabic or longer H verb stems is mysterious. 4.6. ka-negative. In this section we complete our survey of negative tenses. The construction we examine here has the prefix /ka/ located after the subject prefix; otherwise it is similar to the negative present in terms of morphological structure. It can be glossed as "hasn't yet". In this tense we find a tone pattern in the stem that is the same as in the negative infinitive. However, as we will see below, the tone pattern of the object prefix in conjunction with High verb stems is not the same as in the negative infinitive (at least in the speech of our consultants). Toneless verb stems, as shown in (120), are identical to their negative infinitive shape: (120) ab£kaxol!6eli abdkafikelle'eli anikacornbulitfuli anikadanliisi anikago.dlViu.ki akdkaqononondj.lisi anikayokojjej liisi akikapheeklf ak£kaya.aj!f abikawiiisi anikaalwli
(forgive) (reach) (become untied) (dance) (go home) (emphasize) (decorate with many colors) (cook) (close) (drop) (fight)
From an analytical point of view, then, the major new aspect of these data is the fact that H tone (which we have argued is contributed by the subject prefix in the negative tenses) is unable to extend onto the following /ka/ (which itself does not appear to contribute any H tone to the representation). One can imagine various approaches to this problem. Perhaps the simplest analysis is to propose that the subject prefix in the negative is an exception to *MonoHD (Lc. does not trigger a mark if it is in violation of *MonoHD). This is possible because in no construction have we found it necessary for "MonoHD to induce the HD initiated by the subject prefix to expand on the basis of needing to satisfy *MonoHD. The subject prefix H has fused with a following H-sponsor, but this fusion is of course not triggered by *MonoHD. Of course, if we adopt this line of analysis, then we do not need to appeal to the grammatical High on the
197
second stem mora to explain the failure of the subject HD in aba*xoltf£li to expand rightwards. We have other evidence (the protection facts) to argue for the grammatical High being on the second stem more; thus our analysis of the location of the grammatical High tone is not dependent on using this tone to explicate the apparent violation of *MonoHD in aba'xolle'liWhen the subject prefix is depressed, the H will shift onto the /ka/ syllable. Let us examine the consequences of this shift. When the stem is monosyllabic, the /ka/ is lengthened and is realized with a Falling tone. (121) andjkaalwll, azjkaalwlt
(fight)
In medial position, this Falling tone will be replaced by a level High. In M, there will not be downstep between the penult and the final High tones, while in K there will be: (122) andjktfwi...[M].ar,djklilwli...[K]
(fight)
When the depressed subject prefix precedes a (regular) trisyllabic toneless verb stem, then the shift onto /ka/ does not produce any abutting domains. (123) andjk*baal!t andjkdpheekll andjkayaalli
(count) (cook) (close)
However, in the /wisaZ-type stem, there will be abutting HD's in the output of the B system, but with the usual difference in RD structure between M and K: (124) azjkiwfisi [M], a%ik£w!fisi [K] andjkdhlfisi [M], andjkdhl!fisi [K]
(drop) (lower)
In the case of trisyllabic or longer stems, the shift away from the depressed subject prefix will not yield adjacent domains and thus there will be two distinct RD's in both K and M. (125) andjk4xol!e'eli, azjk4xol!e*eli andjkaTikelleeli, azjkafikelteeli andjkfcornb.ultu'uli, az.jkdsom|)u1!iiuli andjkdqononondj.ljsi andjkddanllisi, azjkd
(drop) (count) (cultivate) (sing)
The affixation of /e.e/ yields a trisyllabic stem The initial mora of the verb stem is clearly not the sponsor of a High tone - witness an example like ba'limie'e... The second mora of the stem (the initial element in /e.e/) not only sponsors the grammatical High tone, but it is also the surface bearer of the High tone (since the HD cannot extend onto the final syllable). We see the expected difference between K and M with respect to depressed High subject prefixes. The shifted High in M leads to a single RD and thus there is no downstep: jybale'e..-, while in K there are two RD's and thus downstep: giba'lie'e-.. When the stem is "monosyllabic", whether CV or VCV, we find that the grammatical High tone is docked to the first mora in the stem This, of course, is the same pattern as observed in the negative present. (139) njlwe'e..., Iilw£e..., ailwe'e... ndgs6e..., b6s£e..., jose'e... fldirazfo-, begz!..., liwabonisisle... njwanyinyithekisee..., liwanyinyithekisee... njwakhee..., liwakhec... njwoyikee..., liwoyikee... njwonwabisee..., liwonwabisee...
(eat) (wipe) (cultivate) (greet) (see clearly) (make slippery) (build) (fear) (make happy)
The fact that there is a single High tone on the surface shows unequivocally that a High subject prefix fuses with the fused domain of the object prefix and the verb stem. We have a single HD and thus a surface H on the head of that domain. It does not matter if the verb stem has a depressed syllable, or if the object prefix is depressed, or if the subject prefix is depressed. Fusion is invariable. A few examples: (147) njyiJMndge..., bayizondjie..., zjyizondge... njzJZQndfc..., baz.jzp.ndge..., sizjzgndge... njyiga.wulee..., bayiggwulee..., zjyigawulee... njzjgawutee..., bazjgawulee..., zjzjgawul6e... 206
(hate)
(chop)
HJyitherjflisfie..., bayitherjg.js6e..., ziyitherjgise'e... nisitherjaistfe..., baz.itheno.is6e..., z.izjther)flis«e...
(sell)
In this section, we have argued that the short perfect has the same stem grammatical High tone (on the second mora) as the negative tenses. In the next section we encounter a case where there is a grammatical High tone located on the subject prefix and not on the verb stem. 4.8. Participial. In this section we examine the tonal pattern of the participial form of the verb. The structure of the participial can be summarized as follows: SPp+Stem, where SPP is segmentally identical to the general subject prefixes except in the case of CL 1 (where in the participial we have /el rather than /W). Tonally, SPP is always High-toned in input forms. When the verb stem is toneless, the High of the subject prefix surfaces according to the same pattern as the subject prefix H in the present tense. We demonstrate this point by quickly surveying verb stems of different syllabic lengths. If the verb stem is monosyllabic, it is augmented by the syllable /si/ in the participial. The H of the subject prefix ends up associated with this syllable - the effect of *MonoHD. Nonflnality (polysyllabic HD) will bar the domain from extending onto the final syllable. Since the domain is always larger than the subject prefix, whether the subject prefix is depressed or not is immaterial. (148) nisiilwa, esitlwa esilwa, rjdislfwa sis&nza, ndjsWnza. us&nga. rjdjsWnga
(fight) (fall) (do) (kiss)
These data demonstrate that the subject prefixes like /u/ 2 sg., /ni/ 2. pi. and /si/ 1 pi., which in other tenses are toneless, are High-toned in the participial. It is not uncommon in Bantu languages to find tenses where prefixes are uniformly High or uniformly toneless. The small group of bisyllabic verb stems which are exceptions to *MonoHD continue to be exceptions in the participial: (149) Iwiisa, ndiwlisa e"hlaala, rjdjhldala eliliisa, ndinltisa
(drop) (sit) (let go down)
Since the monosyllabic HD in an example like (ejwiisa does not violate *MonoHD (due to the exceptionality of the verb stem), Avoid Prominence will prefer (£)wiisa to *(ewii)sa. The fact that the subject prefix constitutes a HD in the output of the A system means that, in the B system, a depressed subject prefix will trigger the extension of the HD onto the next mora: (ndihli)isa. We thus have a contrast between ghllisa and rj yJYaoge kfa'kuuhlle' yjxdbe kldkuuhlte' bacmlkumise k!4kuuhl!6 bggubungeje kitfkuuhlle' ba.q6nonondjse k!dkuuhl!6
'sharpen it welll' 'narrate it welll' 'scrape it welll' 'mix it welll' 'mix it well!' 'touch it well!' 'cover them well!' 'emphasize them welll'
4.12. Plural imperative There is a plural form of the imperative which involves adding an enclitic element /ni/ at the end of the verbal form. (208) shows toneless verb stems in the plural imperative. (208) yiw&ni wisddni balMni yaWfai balisdtfni df|nis£dni gwnjIMni shukumisAdni
'(pi) fall!' '(pL) drop!' '(pL) count!' '(pL) closer '(pi.) narrate!' '(pi) dance!' '(pL) scrape!' '(pL) shake!'
Clearly, the same grammatical H tone is present and Nonfinality (polysyllabic HD) is enforced. There are no complexities that require discussion. High verb stems are also unproblematic - highs fuse into a single domain that does not include the final syllable. A High tone is heard only on the head of the fused domain. (209) yityMni bomtitni fumdini seben^Sni phakamis&ni bonisisiini shumayeja'flni camagush&ni (ja.ndaluka'iini fundekelfia'ni
'(pL) eat!' '(pi.) seel' '(pL)getl' '(pi.) work!' '(pL)liftup!' '(pL) see clearly!' '(pi) preach!' '(pL) propitiate the elders!' '(pL) call aloud!' '(pi) make noise!'
The object prefix in the plural imperative has imposed depressed, just like in the singular imperative. As a consequence, the "shifted" High on the first stem syllable will (in M but not K) protect the non-heads in the following domain and there will be no downstep between the shifted H and the verb stem H. A depressor consonant in the stem will, of course, force two RD's and thus prevent protection and result in downstep. Some examples with toneless verb stems. 228
(210) wabAKe'ni [M] wawls«ni [M] y.jba*ifs&ni [M] washtiktim1s66ni [M] wanAmithglisleni [M] bffyu.mel!e'6ni bath6nfljs!dfini bfiqdnonondjsitfni
(count) (drop) (narrate) (shake) (cement) (permit) (sell) (emphasize)
High verb stems with an object prefix in the plural (211) wa,tyWni wgbon66ni wa.fuman£gni yjbonis&ni
(eat) (see) (get) (show) (chop) (work for) (lift up)
y.igawuletoi wasebenjele'e'ni wa.phakamisge'ni
There is simply fusion between the object prefix H-sponsor and the lexical H-sponsor in the verb stem and the grammatical H-sponsor in the verb stem No discussion is needed. 4.13. Remote past tense. In this section we shall examine another affirmative verb tense which involves a grammatical High tone. We refer to this tense as the "remote past". The remote past involves the placement of the element /aV between the subject prefix and the verb stem. (This element is fully long when in the penult syllable. In pre-penult positions, this element has, ideally speaking, some length. In actual speech the length may be difficult to perceive. This element also has a Falling tone. As we will see, there is an environment where this Falling tone may disappear.) As usual, toneless verbs serve as our initial object of investigation. Monosyllabic verb stems have a H on their only syllable. The tense marker /£/ is fully long in this case due to the usual penultimate lengthening. (212) n4alw!i y6an!a* biatshld
ndiembifi rjdflangla: n1fll«!